Patent Application
20190296548
The present invention relates to a method of loadflow computation in power flow control and voltage control for an electrical power system.
The present invention relates to power-flow/voltage control in utility/industrial power networks of the types including many power plants/generators interconnected through transmission/distribution lines to other loads and motors. Each of these components of the power network is protected against unhealthy or alternatively faulty, over/under voltage, and/or over loaded damaging operating conditions. Such a protection is automatic and operates without the consent of power network operator, and takes an unhealthy component out of service by disconnecting it from the network. The time domain of operation of the protection is of the order of milliseconds.
The purpose of a utility/industrial power network is to meet the electricity demands of its various consumers 24-hours a day, 7-days a week while maintaining the quality of electricity supply. The quality of electricity supply means the consumer demands be met at specified voltage and frequency levels without over loaded, under/over voltage operation of any of the power network components. The operation of a power network is different at different times due to changing consumer demands and development of any faulty/contingency situation. In other words healthy operating power network is constantly subjected to small and large disturbances. These disturbances could be consumer/operator initiated, or initiated by overload and under/over voltage alleviating functions collectively referred to as security control functions and various optimization functions such as economic operation and minimization of losses, or caused by a fault/contingency incident.
For example, a power network is operating healthy and meeting quality electricity needs of its consumers. A fault occurs on a line or a transformer or a generator which faulty component gets isolated from the rest of the healthy network by virtue of the automatic operation of its protection. Such a disturbance would cause a change in the pattern of power flows in the network, which can cause over loading of one or more of the other components and/or over/under voltage at one or more nodes in the rest of the network. This in turn can isolate one or more other components out of service by virtue of the operation of associated protection, which disturbance can trigger chain reaction disintegrating the power network.
Therefore, the most basic and integral part of all other functions including optimizations in power network operation and control is security control. Security control means controlling power flows so that no component of the network is over loaded and controlling voltages such that there is no over voltage or under voltage at any of the nodes in the network following a disturbance small or large. As is well known, controlling electric power flows include both controlling real power flows which is given in MWs, and controlling reactive power flows which is given in MVARs. Security control functions or alternatively overloads alleviation and over/under voltage alleviation functions can be realized through one or combination of more controls in the network. These involve control of power flow over tie line connecting other utility network, turbine steam/water/gas input control to control real power generated by each generator, load shedding function curtails load demands of consumers, excitation controls reactive power generated by individual generator which essentially controls generator terminal voltage, transformer taps control connected node voltage, switching in/out in capacitor/reactor banks controls reactive power at the connected node.
Control of an electrical power system involving power-flow control and voltage control commonly is performed according to a process shown in
Overload and under/over voltage alleviation functions produce changes in controlled variables/parameters in step-60 of
It is obvious that loadflow computation consequently is performed many times in real-time operation and control environment and, therefore, efficient and high-speed loadflow computation is necessary to provide corrective control in the changing power system conditions including an outage or failure of any of the power network components. Moreover, the loadflow computation must be highly reliable to yield converged solution under a wide range of system operating conditions and network parameters. Failure to yield converged loadflow solution creates blind spot as to what exactly could be happening in the network leading to potentially damaging operational and control decisions/actions in capital-intensive power utilities.
The power system control process shown in
The prior art and present invention are described using the following symbols and terms:
Ypq=Gpq+jBpq: (p−q) th element of nodal admittance matrix without shunts
Ypp=Gpp+jBpp: p-th diagonal element of nodal admittance matrix without shunts
yp=gp+jbp: total shunt admittance at any node-p
Vp=ep+jfp=Vp∠θp: complex voltage of any node-p
Vs=es+jfs=Vs∠θs: complex slack-node voltage
Δθp,ΔVp: voltage angle, magnitude corrections
Δfp,Δep: imaginary, real part of complex voltage corrections
Sp=Pp+jQp: net nodal injected power, calculated
ΔPp+jΔQp: nodal power residue or mismatch
RPp+jRQp: modified nodal power residue or mismatch
RIp+jIIp: net nodal injected current, calculated
ΔRIp+jΔIIp: nodal injected current residue or mismatch
RRIp+jRIIp: modified nodal current residue or mismatch
SSHp=PSHp+jQSHp: net nodal injected power, scheduled/specified
Cp=1∠Φp=Cos Φp+jSin Φp: Unitary rotation/transformation
m: number of PQ-nodes
k: number of PV-nodes
n=m+k+1: total number of nodes
q>p: node-q is connected to node-p excluding the case of q=p
[ ]: indicates enclosed variable symbol to be a vector or matrix
LRA: Limiting Rotation Angle, −48° for invented models
PQ-node: load-node, where, Real-Power-P and Reactive-Power-Q are specified
PV-node: generator-node, where, Real-Power-P and Voltage-Magnitude-V are specified
Vs≈VB≈VN: Slack-node voltage magnitude, Base value, and Nominal value of voltage magnitude are very closely similar, and therefore, they can be used interchangeably. However, in the following development only Vs will be used. Particularly, in the treatment of loadflow problem with distributed slack-node, there is no specific slack-node and VB or VN can be used.
Prior art method of loadflow computation of the kind carried out as step-30 in
It is a primary object of the present invention to improve convergence and efficiency of the prior art Super Super Decoupled Loadflow computation method under wide range of system operating conditions and network parameters for use in power flow control and voltage control in the power system. A further object of the invention is to reduce computer storage/memory or calculating volume requirements.
The above and other objects are achieved, according to the present inventions, Patel Loadflow (PL-1 & PL-2), Patel Super Decoupled Loadflow (PSDL-YY1 & PSDL-YY2), Y matrix—Patel Loadflow (YPL-1 & YPL2), Sparse Z or C−1 matrix—Patel Loadflow (SZPL or SCIPL), Guass-Seidel-Patel Loadflow (GSPL) Methods and their many variants, for loadflow calculation for Electrical Power System. In context of voltage control, one of the inventive system of PSDL-YY2 and others listed in the above methods of loadflow computation is used for Electrical Power system consisting of plurality of electromechanical rotating machines, transformers and electrical loads connected in a network, each machine having a reactive power characteristic and an excitation element which is controllable for adjusting the reactive power generated or absorbed by the machine, and some of the transformers each having a tap changing element, which is controllable for adjusting turns ratio or alternatively terminal voltage of the transformer, said system comprising:
The method and system of voltage control according to the preferred embodiment of the present invention provide voltage control for the nodes connected to PV-node generators and tap changing transformers for a network in which real power assignments have already been fixed. The said voltage control is realized by controlling reactive power generation and transformer tap positions.
One of the inventive methods of defining and solving loadflow computation models PL-1, PL-2, PSDL-YY1, PSDL-YY2, YPL-1, YPL-2, SZPL or GSPL can be used for voltage control in Electrical power System. For this purpose real and reactive power assignments or settings at PQ-nodes, real power and voltage magnitude assignments or settings at PV-nodes and transformer turns ratios, open/close status of all circuit breaker, the reactive capability characteristic or curve for each machine, maximum and minimum tap positions limits of tap changing transformers, operating limits of all other network components, and the impedance or admittance of all lines are supplied. A decoupled loadflow system of equations {(28) and (29)} or {(30) and (31)} is solved by an iterative process until convergence. During this solution the quantities which can vary are the real and reactive power at the reference/slack node, the reactive power set points for each PV-node generator, the transformer transformation ratios, and voltages on all PQ-nodes nodes, all being held within the specified ranges. When the iterative process converges to a solution, indications of reactive power generation at PV-nodes and transformer turns-ratios or tap-settings are provided. Based on the known reactive power capability characteristics of each PV-node generator, the determined reactive power values are used to adjust the excitation current to each generator to establish the reactive power set points. The transformer taps are set in accordance with the turns ratio indication provided by the system of loadflow computation.
For voltage control, system of PSDL-YY2 or others and many variants listed in the above Methods of Loadflow computation can be employed either on-line or off-line. In off-line operation, the user can simulate and experiment with various sets of operating conditions and determine reactive power generation and transformer tap settings requirements. A general-purpose computer can implement the entire system. For on-line operation, the loadflow computation system is provided with data identifying the current real and reactive power assignments and transformer transformation ratios, the present status of all switches and circuit breakers in the network and machine characteristic curves in steps-10 and -20 in
The present inventive system of loadflow computation for Electrical Power System consists of, one of the Patel Super Decoupled Loadflow: YY2-version (PSDL-YY2) or PSDL-X′X′, or others listed in the above Methods characterized in that 1) single decoupled coefficient matrix solution requiring only 50% of memory used by prior art methods, 2) the presence of transformed values of known/given/specified/scheduled/set quantities in the diagonal elements of the gain matrices [Yf] and [Ye] of the decoupled loadflow sub-problems, and 3) transformation angles are restricted to maximum of −0° to −90° (say, −48°) to be determined experimentally, 4) PV-nodes being active in both RI-f and The sub-problems, PQ-node to PV-node and PV-node to PQ-node switching is simple to implement, and these inventive loadflow computation steps together yield some processing acceleration and consequent efficiency gains, and are each individually inventive, and 5) modified real and imaginary current mismatches at PV-nodes in case of PSDL-YY1, SSDL-YY, HSSDL-YY, ESSDL-YY or their generalized variations PSDL-B′B′, SSDL-B′B′, HSSDL-B′B′, ESSDL-B′B′ are determined as RRIp=(epΔPp)/[Kp(ep2+fp2)] and RIIp=(−fpΔPp)/[Kp(ep2+fp2)] in order to keep gain matrices [Yf] and [Ye] symmetrical. If the value of factor Kp=1, the gain matrices [Yf] and [Ye] becomes unsymmetrical in that elements in the rows corresponding to PV-nodes are defined without transformation or rotation applied, as Yfpq=Yepq=−Bpq. It is possible that Patel Super Decoupled methods can be formulated in polar coordinates by simply replacing correction vectors [Δf] and [Δe] in equations (28) and (29) and subsequently followed equations by correction vectors [Δθ] and [ΔV]. However, it will not be easy to have single gain matrix model, because [ΔV] for PV-nodes is zero and absent.
A loadflow computation is involved as a step in power flow control and/or voltage control in accordance with
Node-6 is a reference/slack-node alternatively referred to as the slack or swing-node, representing the biggest power plant in a power network. Nodes-4 and -5 are PV-nodes where Generators are connected, and nodes-1, -2, and -3 are PQ-nodes where loads are connected. It should be noted that the nodes-4, -5, and -6 each represents a power plant that contains many generators in parallel operation. The single generator symbol at each of the nodes-4, -5, and -6 is equivalent of all generators in each plant. The power network further includes controllable circuit breakers located at each end of the transmission lines and transformers, and depicted by cross markings in one-line diagram of
A goal of the present invention is to provide a reliable and computationally efficient loadflow computation that appears as a step in power flow control and/or voltage control systems of
Short description of other possible embodiment of the present invention is also provided herein. The present invention relates to control of utility/industrial power networks of the types including plurality of power plants/generators and one or more motors/loads, and connected to other external utility. In the utility/industrial systems of this type, it is the usual practice to adjust the real and reactive power produced by each generator and each of the other sources including synchronous condensers and capacitor/inductor banks, in order to optimize the real and reactive power generation assignments of the system. Healthy or secure operation of the network can be shifted to optimized operation through corrective control produced by optimization functions without violation of security constraints. This is referred to as security constrained optimization of operation. Such an optimization is described in the U.S. Pat. No. 5,081,591 dated Jan. 13, 1992: “Optimizing Reactive Power Distribution in an Industrial Power Network”, where the present invention can be embodied by replacing the step nos. 56 and 66 each by a step of constant gain matrices [Yf] and [Ye], and replacing steps of “Exercise Newton-Raphson Algorithm” by steps of “Exercise PSDL-YY1 or PSDL-YY2 or YPL-1 or YPL-2 or SZPL or GSPL Computation” in places of steps 58 and 68. This is just to indicate the possible embodiment of the present invention in optimization functions like in many others including state estimation function. However, invention is being claimed through a simplified embodiment without optimization function as in
In
Each of the transformers T1 and T2 in
The indications provided by function steps 14, and 22 are supplied to a function step 42 in which constant gain matrices [Yf] and [Ye], or [Y] or [Z] or [C]−1 of any of the invented PSDL-YY1 or PSDL-YY2 or YPL-1 or YPL-2 or SZPL or GSPL models are constructed, factorized or inverted and stored. The coefficient matrices [Yf] and [Ye], or [C] or [C]−1 or [Z] are conventional tools employed for solving PSDL-1 or PSDL-2 or CPL-1 or CPL-2 or SZPL models defined by equations {(28) and (29)} or {(30) and (31)} or (56) or (58) or (69) or (70) of a power system. [C] is the most general representation of all possible matrices involved in the solution of linear and non-linear simultaneous algebraic equations. [C] could be the Jacobian, approximated Jacobian, constant Jacobian, approximated constant decoupled Jacobian in case of Newton-Raphson based approaches. It could be the coefficient matrix, approximated coefficient matrix, constant coefficient matrix, approximated constant decoupled coefficient matrix in case of Patel Numerical Method (PNM) based approaches described as preferred embodiments in this application. [C]−1 when fully inverted is the full matrix. However, it can be made sparse by storing and processing only selected elements, and it becomes approximation of fully inverted [C]−1.
Indications of initial reactive power, or Q on each node, based on initial calculations or measurements, are provided by a function step 22 and these indications are used in function step 24, to assign a Q level to each generator and motor. Initially, the Q assigned to each machine can be the same as the indicated Q value for the node to which that machine is connected.
An indication of measured real power, P, on each node is supplied by function step 32. Indications of assigned/specified/scheduled/set generating plant loads that are constituted by known program are provided by function step 34, which assigns the real power, P, load for each generating plant on the basis of the total P, which must be generated within the power system. The value of P assigned to each power plant represents an economic optimum, and these values represent fixed constraints on the variations, which can be made by the system according to the present invention. The indications provided by function steps 32 and 34 are supplied to function step 36 which adjusts the P distribution on the various plant nodes accordingly. Function step 38 assigns initial approximate or guess solution to begin iterative method of loadflow computation, and reads data file of operating limits on power network components, such as maximum and minimum reactive power generation capability limits of PV-nodes generators.
The indications provided by function steps 24, 36, 38 and 42 are supplied to function step 44 where inventive PSDL-YY1 or PSDL-YY2 or YPL-1 or YPL-2 or SZPL or GSPL model solution is carried out, the results of which appear in function step 46. The loadflow computation yields voltage magnitudes and voltage angles at PQ-nodes, real and reactive power generation by the reference/slack/swing node generator, voltage angles and reactive power generation indications at PV-nodes, and transformer turns ratio or tap position indications for tap changing transformers. The system stores in step 44 a representation of the reactive capability characteristic of each PV-node generator and these characteristics act as constraints on the reactive power that can be calculated for each PV-node generator for indication in step 46. The indications provided in step 46 actuate machine excitation control and transformer tap position control. All the loadflow computation methods using inventive PSDL-YY1 or PSDL-YY2 or YPL-1 or YPL-2 or SZPL or GSPL computation models can be used to affect efficient and reliable voltage control in power systems as in the process flow diagram of
Particularly inventive PSDL-YY1 or PSDL-YY2 or CPL-1 or CPL-2 or SZPL or GSPL models in terms of equations for determining elements of vectors [RI′], [II′], [ΔRI′], [ΔII′], [I], [ΔI] and elements of coefficient matrices [Yf] and [Ye], or [C] or [Z] are described followed by computation steps of corresponding methods are described.
The presence of values of known/given/specified/scheduled/set quantities in the diagonal elements of the coefficient matrix [Yf] and [Ye], or [C] or [Z], which takes different form for different methods, is brought about by such formulation of loadflow equations. The said quantities in the diagonal elements in the coefficient matrices improved convergence and the reliability of obtaining converged loadflow computation.
The slack-start is to use the same voltage magnitude and angle as those of the reference/slack/swing node as the initial guess solution estimate for initiating the iterative loadflow computation. With the specified/scheduled/set voltage magnitudes, PV-node voltage magnitudes are adjusted to their known values after the first P-θ iteration. This slack-start saves almost all effort of mismatch calculation in the first P-f iteration. It requires only shunt flows from each node to ground to be calculated at each node, because no flows occurs from one node to another because they are at the same voltage magnitude and angle.
The following inventions are based on the Patel Numerical Method-1 (PNM-1) originally propounded by this inventor in 2007 in his international patent application no. PCT/CA2007/001537 and consequent granted patents CA 2661753 and U.S. Pat. No. 8,315,742. The invented class of methods of forming/defining and solving loadflow computation models of a power network are the methods that organize a set of nonlinear algebraic equations in linear form as a product of coefficient matrix and unknown vector on one side and all other terms on the other side or the corresponding mismatch vector on the other side, and then solving the linear matrix equation for unknown vector in an iterative fashion.
Propounding Statement of Patel Numerical Method-2 (PNM-2)
Preliminary investigations suggest that Patel Numerical Method may potentially produce monotonous convergence, and therefore may be amenable to acceleration factors unlike Newton-Raphson method.
Patel Loadflow-1 (PL-1)
The PL-1 Model comprises eqns. (1) to (9)
The equations (1) to (7) represents linearized global solution of the nonlinear loadflow equations. Local nonlinearity can be handled by introduction of self-iterations as per equations (8) to (9).
[fp(sr+1)](r+1)=[(RIp/Bfpp)(sr)](r) (8)
[ep(sr+1)](r+1)=[(IIp/Bepp)(sr)](r) (9)
Equations (8) to (9) are solved independently for each node, and can be performed simultaneously in parallel for all the nodes. Equations (2) and {(8) and (9)} are solved in sequence. In other words linear global solution followed by non-linear local (nodal) solution by self-iterations, or non-linear local (nodal) solution by self-iterations followed by linear global solution.
The PL-2 model comprises eqns. {(11) and (12)} or (14), (5), (15) to (20), and {(21) to (24)} or {(25) to (26)}.
Equations (15) and (16) provides alternative expressions of real and imaginary current mismatches where ΔQp=0.0 at PV-nodes. An alternative definition of PL-2 model can be provided by defining ΔRIp of (15) and ΔIIp of (16) as the subtraction of the terms containing specified values from the calculated values that would make ΔRID and ΔIID defined by eqns. (15) and (16) and elements of [C] defined by eqns. (17) to (20) negative.
It can be seen that diagonal elements of the coefficient matrix [C] are changing with changing values of (ep2+fp2), and therefore, requiring time consuming re-factorization of [C] in each iteration. To avoid re-factorization, it is proposed to make [C] constant by using (es2+fs2), the slack-node voltage values, instead of (ep2+fp2) in equations (19) and (20) requiring factorization of [C] only once in the beginning of the iteration process.
The equations (10) to (20) represents linearized global solution of the nonlinear loadflow equations. Local nonlinearity can be handled by introduction of self-iterations as per equations {(21) to (24)} or {(25) to (26)}. It is possible to expand in detail all equations involving self iterations as in equations (21), (23), (25), (26), (54), (55), (66), (67) etc. in the following.
[Δfp(sr+1)](r+1)=[(ΔRIp/Bfpp)(sr)](r) (21)
[fp(sr+1)](r+1)=[fp(sr)](r)+[Δfp(sr+1)](r+1) (22)
[Δep(sr+1)](r+1)=[(ΔIIp/Bepp)(sr)](r) (23)
[ep(sr+1)](r+1)=[ep(sr)](r)[Δep(sr+1)](r+1) (24)
Equations {(21) to (24)} or {(25) to (26)} are solved independently for each node, and can be performed simultaneously in parallel for all the nodes. Equations {(11) and (12)} or (14), and {(21) to (24)} or {(25) and (26)} are solved in sequence. In other words linear global solution followed by non-linear local (nodal) solution by self-iterations, or non-linear local (nodal) solution by self-iterations followed by linear global solution.
[fp(sr+1)](r+1)=[(ΔRIp/Bfpp)(sr)](r) (25)
[ep(sr+1)](r+1)=[(ΔIIp/Bepp)(sr)](r) (26)
In a class of super decoupled loadflow models, each super decoupled loadflow model comprises a system of equations {(28) and (29)} or {(30) and (31)} differing in the definition of elements of [ΔRI′], [ΔII′], [RI′], [II′], and [Yf] and [Ye]. It is a system of equations for the separate calculation of imaginary part of and real part of complex voltage or its corrections. [C′] is the transformed coefficient matrix.
In this scheme {(28) and (29)} or {(30) and (31)} are solved alternately with intermediate updating. Each iteration involves one calculation of {[ΔRI′] or [RI′]} and {[Δf] or [f]} to update [f] and then one calculation of {[ΔII′] or [II′]} and {[Δe] or [e]} to update [e]. The sequence of relations {(32) to (35)} or {(36) to (37)} depicts the scheme.
[Δf]=[Yf]−1[ΔRI′] (32)
[f]=[f]+[Δf] (33)
[Δe]=[Ye]−1[ΔII′] (34)
[e]=[e]+[Δe] (35)
[f]=[Yf]−1{[ΔRI′] or [RI′]} (36)
[e]=[Ye]−1{[ΔII′] or [II′]} (37)
The scheme involves solution of system of equations {(28) and (29)} or {(30) and (31)} in an iterative manner depicted in the sequence of relations {(32) to (35)} or {(36) to (37)}. This scheme requires calculation for each half iteration because {[ΔRI′] and [ΔII′]} or {[RI′] and [II′]} is calculated always using the most recent imaginary part of and real part of complex voltage values, and it is block Gauss-Seidel approach. The scheme is block successive, which imparts increased stability to the solution process, and it in turn improves convergence and increases the reliability of obtaining solution.
The PSDL-YY1 model comprises equations{(32) to (35)} or {(36) to (37)} & (38) to (50).
Two new versions of SSDL-YY are provided. One is Hybrid SSDL-YY (HSSDL-YY) and the other is Efficient SSDL-YY (ESSDL-YY). The HSSDL model comprises eqns. (32) to (35), (38a), (38b), (39a), (39b), (40a), (40b), (41a), (41b), (42) to (45), (46a), (47a), and (48) to (50). The ESSDL-YY model comprises eqns. (32) to (35), (38c), (39a), (39b), (40a), (40b), {(41c) and (41d)} where QSHp replaced by Qp0 calculated value at initial estimate at PV-nodes, {(42) and (43)} with approximate versions of {(15) and (16)} where QSHp replaced by Qp (calculated value) at PV-nodes, and {(48) and (49)}.
Branch admittance magnitude in (38), (38a), (38b), (38c) is of the same algebraic sign as its susceptance. Rotation angles are to be determined as per (48) and (49), and could be restricted to the maximum anywhere −0 to −90 degrees to be determined experimentally. There can be many possible variations of PSDL, HSSDL, and ESSDL models, and the one variation being their generalized versions PSDL-B′B′, HSSDL-B′B′, and ESSDL-B′B′ where B′ symbolizes suceptance matrix transformed, B′pq=Bpq+Gpq tan Φpq and tan Φpq=Gpq/Bpq. Also, the two versions PSDL-YY and PSDL-B′B′, HSSDL-YY and HSSDL-B′B′, and ESSDL-YY and ESSDL-B′B′ can be mixed in any possible combination. Corresponding transformed diagonal elements Bpp′ and transformed mismatches can easily be determined.
Slack-Start is use of the same voltage magnitude and angle as those of the slack-node for all nodes as an initial guess solution. With the specified magnitudes, PV-nodes voltage magnitudes are adjusted to their known values after the first half iteration. This start procedure referred to as the slack-start, saves almost all effort of mismatch calculation in the first P-f iteration as it requires only shunt flows to be calculated at each node.
where, Kp is defined in equation (50) which is initially restricted to the minimum value of 0.75 determined experimentally; however its restriction is lowered to the minimum value of 0.6 when its average over all less than 1.0 values at PV nodes is less than 0.6.
In super decoupled loadflow models [Yf] and [Ye] are real, sparse, symmetrical and built only from network elements. Since they are constant, they need to be factorized once only at the start of the solution. Equations {(28) and (29)} or {(30) and (31)} are to be solved repeatedly by forward and backward substitutions. [Yf] and [Ye] are of the same dimensions (m+k)×(m+k) when only a row/column of the reference/slack-node is excluded and both are triangularized using the same ordering regardless of the node-types.
Unlike the HSSDL and the prior art SSDL (Super Super Decoupled Loadflow, presented at Toronto International Conference—Science and Technology for Humanity—2009, pages: 652-659) methods, the PSDL methods are single matrix loadflow computations substantially reducing memory requirements, and since all nodes are active in the iterative process implementations of PQ-node to PV-node and PV-node to PQ-node switching is simple. The best possible convergence from non-linearity consideration could be achieved by restricting rotation angle to maximum of −0 to −90 degrees (say, −48 degrees) to be determined experimentally.
The steps of loadflow computation method, PSDL-YY1 method are shown in the flowchart of
The Patel Super Decoupled Loadflow-2 (PSDL-YY2) model comprises equations {(32) to (35)} or {(36) to (37)}, {(3), (4), (51), (52), (40c), and (53b)}, or {(42), (43) with approximate versions of (15) and (16), (40), and (53a)}, (39), and {(54) and (55)}. In (3), (4), (15), and (16): QSHp is replaced by Qp (calculated) for PV-nodes, and in (40) QSHp is replaced by Qp0 (calculated at initial estimate) for PV-nodes.
All equations other than (54) and (55) of the model PSDL-YY2 represents linearized global solution of the nonlinear loadflow equations. Local nonlinearity can be handled by introduction of self-iterations as per equations (54) to (55).
[fp(sr+1)](r+1)[{(RI′p or ΔRI′p)/Yfpp}(sr)](r) (54)
[ep(sr+1)](r+1)[{(II′p or ΔII′p)/Yepp}(sr)](r) (55)
Equations (54) to (55) are solved independently for each node, and can be performed simultaneously in parallel for all the nodes. Super Decoupled equations {(32) or (36), and (54)} and {(34) or (37), and (55)} are solved in sequence. In other words linear global solution followed by non-linear local (nodal) solution by self-iterations.
The steps of loadflow computation method, PSDL-YY2 method are shown in the flowchart of
Patel Loadflow model can be organized in coefficient matrix [C] based complex form, because it is not involved with any partial differentiation of original or mismatch functions. The model constitutes eqns. {(57) or (59)}, {(60) to (62)} or {(63) to (65)} or {(60a), (64), and (65a)}, {(66) or (67)}, and (68). It involves one solution of {(57) or (59)} followed by one solution of {(66) or (67)}, or one solution of {(66) or (67)} followed by one solution of {(57) or (59)}. However, {(66) or (67)} constitutes one equation for each node except the Slack-node, and equations for all the nodes can be solved in parallel, just like Gauss numerical method.
[ΔI]=[C][ΔV] (56)
[ΔV]=[C]−1[ΔI] (57)
OR
{[ΔI] or [I]}=[C][V] (58)
[V]=[C]−1{[ΔI] or [I]} (59)
Where, components of vectors [I] and [ΔI], and matrix [C] are defined in the following:
The equations (62), (65), and (68) provide elegant formulation for diagonal elements of the coefficient matrix [C] that suggest a mechanism for their numerical manipulations particularly useful when diagonal dominance issue arise in the presence of a capacitive series branch or an excessive capacitive compensation at a node. The factor Lp of different value can be applied separately to real and imaginary components of a diagonal element of [C]. Similar developments can be provided for Patel Super Decoupled Loadflow models and other loadflow models. Equations (66) and (67) and their expanded versions can also be written with factor L.
It can be seen that diagonal elements of the coefficient matrix [C] are changing with changing values of Vp, and therefore, values of (ep2+fp2) during iteration process requiring time consuming re-factorization of [C] in each iteration. To avoid re-factorization, it is proposed to make [C] constant by using (es2+fs2), the slack-node voltage values, instead of (ep2+fp2) in equations (62) and (65) requiring factorization of [C] only once in the beginning of the iteration process.
The steps of loadflow calculation by CPL-1 method are shown in the flowchart of
The steps of loadflow calculation by CPL-2 method are shown in the flowchart of
Matrix complex [C] or real [C] is the coefficient matrix based on original equations (functions) or organized as mismatch equations (functions) in the solution of linear or non-linear simultaneous algebraic equations. Inverses complex [C]−1 and real [C]−1 are correspondingly referred to as complex [Z]-matrix and real [Z]-matrix in this application. The complex [C]−1 and real [C]−1 can also represent admittance matrices respectively complex [Y]−1 and real [Y]−1. Complex [C]−1 and real [C]−1 are generalized representations, wherein real [C]−1 can be Newton-Raphson approach based Jacobian [J]−1 and its simplified approximations including inverses of decoupled or Super Decoupled matrices.
The complex [C] and real [C] are generally sparse matrices wherein many of its off diagonal elements are zeros. In order to save computation time and computer storage, processing of off-diagonal elements that are zeros can be avoided by sparsity preserving programming techniques. However, fully inverted complex [Z]-matrix and real [Z]-matrix are full matrices wherein no off-diagonal elements are zeros. Therefore, it is proposed to make complex [Z]-matrix and real [Z]-matrix sparse by selectively choosing off-diagonal elements that need to be stored and processed, and thereby introducing approximations. There are two extremes, one is to store and process only one element in each raw corresponding to the diagonal element introducing maximum approximation and the other is to store and process the diagonal and all the off-diagonal elements in each row introducing zero approximation. And there are many situations in between the two extremes stated in the above to be determined experimentally depending on the nature of the problem for optimal use of computational resources (computer time and computer storage). In electrical circuits (networks), one situation is to store off-diagonal elements only corresponding to directly connected nodes (level-1 connectivity) to a given node, which is the same sparsity of the matrix complex [C] or real [C]. For a given node, other situations are to store off-diagonal elements corresponding to directly connected nodes (level-1), and directly connected nodes to level-1 nodes (level-2 nodes), and directly connected nodes to level-2 nodes (level-3 nodes), and so on. For a given node, the level of outward connectivity is to be determined experimentally to determine number of off-diagonal element required to be stored and processed in the complex [Z]-matrix and the real [Z]-matrix for efficient and reliable computation.
In equations (69) and (70): vectors [V] and [I] are of complex voltage and complex current element components respectively. Vectors [ΔV] and [ΔI] are composed of complex voltage correction and complex current mismatch components respectively. Voltage and current quantities appear in electrical circuits. Equation (69) corresponds to equation (59), Equation (70) corresponds to equation (58), and relevant quantities are defined in equations from (60) to (65), wherein complex [Z] becomes complex [C]−1.
Equation (69) corresponds to two super decoupled equations (36) and (37), and equation (70) corresponds to two super decoupled equations (32) and (34). Relevant quantities defined in equations (38) to (50), and (51) to (53). It should be noted that all quantities involved are real and not complex, wherein real [Z] becomes equivalent to two super decoupled real matrices [Yf]−1 and [Ye]−1.
Application of Newton-Raphson approach to solution of simultaneous non-linear algebraic equations involves calculation of correction vector in each iteration and requires updating as in equations (33) and (35) in case of decoupled models.
All the computation models and their solution methods developed in this application are for electrical power network. However, similar computation models and their solution methods can be developed using techniques developed in this application for all possible areas of study and application that requires solution of linear or non-linear simultaneous algebraic equations. Computation models and their solution methods could be for a system, a circuit, a machine, an apparatus, a device, a material etc.
It should be noted that ZPL or SZPL are embarrassingly parallel problems, and readily amenable to parallel processing. This inventor believes, an approach outlined in the above is likely to work.
If it works subject to verification by this inventor, it can produce grand simplifications in the sense that no need for specialized triangulation and back-substitution or factorization software, and no need for storing indexing and addressing information required for processing elements of factorized matrix. It appears the next numerical wonder is brewing. This inventor is humbled listening words Guardians of Galaxy chanting: “Mr. Patel, You are the one, chosen”.
The model constitutes {(69) or (70)}, {(71) or (72)} and {(73s) or (74s)}. It involves one solution of {(73) or (74)}. However, {(73a) or (73(c) or (74a) or (74c)} constitutes one equation for each node except the Slack-node, and equations for all the nodes can be solved in parallel, just like Gauss numerical method with self-iteration for each-node to handle local non-linearity. Self-iterations were introduced by this inventor first-time in the year 2005 in his patent # U.S. Pat. No. 7,788,051 and Canadian Patent #2548096 issued Jan. 5, 2011. This is a grand Gaussification of all the possible classical numerical methods. Equation {(73b) or (73d) or (74b) or (74d)} constitute one equation for each node except the slack-node, and equations for all the nodes can be solved in sequence like Gauss-Seidel numerical method with self-iteration for each-node to handle local non-linearity. This is a grand Gauss-seidelization of all the possible classical numerical methods. Gauss numerical method is ambarrassingly parallel. However, the best approach seems to solve nodal equation for each node and nodal equations of its directly connected nodes in sequence like Gauss-Seidel numerical method with self-iteration for each-node to handle local non-linearity on separate processor simultaneously in parallel by the technique introduced by this inventor in his patent # U.S. Pat. No. 7,788,051 and Canadian Patent #2548096 issued Jan. 5, 2011. The parallelizaton technique of patent # U.S. Pat. No. 7,788,051 has produced 10-times speed-up in Ybus formulation of Gauss-Seidel loadflow method involving self-iterations. The same parallelization technique applied to models (69) and (70), could potentially produce 20-to-40 times speed-ups. It looks like a revolution (Patelution) in numerical computation. The complex matrix [Z] in equations (69) and (70) can also be created by using building algorithm to create complex inverted coefficient matrix [C] or complex inverted admittance matrix [Y] and its different variations.
[V]=[Z] {[ΔI] or [I]} OR (69)
[ΔV]=[Z][ΔI] (70)
Wherein, though it is possible to write equations (69) and (70) in complex form or real form in terms of real and imaginary components, of involved variables/parameters relevant to problem being solved, development in the following is given only for complex versions of equations (69) and (70) involving variables/parameter (voltage, current, and admittance) relevant to an electrical circuit or a network where,
components of vectors [V], [I], [ΔV], [ΔI], and special Symbols are defined in the following:
Matrix [Z] can also be made-up of real or complex components, and it is an inverse of coefficient matrix of linear and non-linear equations organized in different possible ways including in super-decoupled form, or an inverse of the Jacobian [J]−1 or its different constant or approximated variations including decoupled or super decoupled versions. It should be noted that equations (72a) and (72b) are the same as (60a) and (60) respectively, and (60s) and (63s) are different variations of (60).
The steps of loadflow calculation by ZPL or SZPL method are shown in the flowchart of
The complex conjugate power injected into the node-p of a power network is given by the following equation (76) and its other alternative organizations.
Where, Re means “real part of” and Im means “imaginary part of”. The equation (76) can also be written for complex power injected into the node-p, instead of complex conjugate power injected into the node-p for the purpose of the following development of a Gauss-Seidel-Patel numerical and Loadflow method. However, detailed generalized propounding statement of the Gauss-Seidel-Patel numerical method will be provided in the proposed book writing project.
The Gauss-Seidel-Patel (GSP) numerical method is for solving a set of simultaneous nonlinear algebraic equations iteratively. The GSPL-method calculates complex node voltage for any node-p as given in equation (76).
Iterations start with the experienced/reasonable/logical guess for the solution. The reference node also referred to as the slack-node voltage being specified, starting voltage guess is made for the remaining (n−1)-nodes in n-node network. Node voltage value is immediately updated with its newly calculated value in the iteration process in which one node voltage is calculated at a time using latest updated other node voltage values. A node voltage value calculation at a time process is iterated over (n−1)-nodes in an n-node network, the reference node voltage being specified not required to be calculated.
Now, for the iteration-(r+1), the complex voltage calculation at node-p equation (76) and reactive power calculation at node-p equation (79), becomes:
The well-known limitation of the Gauss-Seidel numerical method to be not able to converge to the high accuracy solution, was resolved by the introduction of the concept of self-iteration of each calculated variable until convergence before proceeding to calculate the next. This is achieved by replacing equation (80) by equation (82) stated in the following where self-iteration-(sr+1) over a node variable itself within the global iteration-(r+1) over (n−1) nodes in the n-node network is depicted. During the self-iteration process only Vp and its real and imaginary components change without affecting any of the terms involving Vq. At the start of the self-iteration Vpsr=Vpr, and at the convergence of the self-iteration Vp(r+1)=Vp(sr+1)
The self-iteration process for a node is carried out until changes in the real and imaginary parts of the node-p voltage calculated in two consecutive self-iterations are less than the specified tolerance. It has been possible to establish a relationship between the tolerance specification for self-convergence and the tolerance specification for global-convergence. It is found sufficient for the self-convergence tolerance specification to be ten times the global-convergence tolerance specification.
|Δfp(sr+1)=|fp(sr+1)−fpsr|<10ε (83)
|Δep(sr+1)|=|ep(sr+1)−epsr|<10ε (84)
For the global-convergence tolerance specification of 0.000001, it has been found sufficient to have the self-convergence tolerance specification of 0.00001 in order to have the maximum real and reactive power mismatches of 0.0001 in the converged solution. However, for small networks under not difficult to solve conditions they respectively could be 0.00001 and 0.0001 or 0.000001 and 0.0001, and for large networks under difficult to solve conditions they sometimes need to be respectively 0.0000001 and 0.000001.
The iteration process is carried out until changes in the real and imaginary parts of the set of (n−1)-node voltages calculated in two consecutive iterations are all less than the specified tolerance −ε, as shown in equations (85) and (86). The lower the value of the specified tolerance for convergence check, the greater the solution accuracy.
|Δfp(r+1)|=|fp(r+1)−fpr|<ε (85)
|Δep(r+1)=|ep(r+1)−epr|<ε (86)
The GSP-method being inherently slow to converge, it is characterized by the use of an acceleration factor applied to the difference in calculated node voltage between two consecutive iterations to speed-up the iterative solution process. The accelerated value of node-p voltage at iteration-(r+1) is given by
V
p
(r+1)(accelerated)=Vpr+β(Vp(r+1)−Vpr) (87)
Where, β is the real number called acceleration factor, the value of which for the best possible convergence for any given network can be determined by trial solutions. The GSP-method is very sensitive to the choice of β, causing very slow convergence and even divergence for the wrong choice.
Of the four variables, real power PSHp and voltage magnitude VSHp are scheduled/specified/set at a PV-node. If the reactive power calculated using VSHp at the PV-node is within the upper and lower generation capability limits of a PV-node generator, it is capable of holding the specified voltage at its terminal. Therefore the complex voltage calculated by equation (80) or (82) by using actually calculated reactive power Qp in place of QSHp is adjusted to specified voltage magnitude by equation (88). However, in case of violation of upper or lower generation capability limits of a PV-node generator, a violated limit value is used for QSHp in (80) and (82), meaning a PV-node generator is no longer capable of holding its terminal voltage at its scheduled voltage magnitude VSHp, and the PV-node is switched to a PQ-node type.
V
p
(r+1)=(VSHpVp(r+1))/|Vp(r+1)| (88)
The steps of loadflow calculation by GSPL method are shown in the flowchart of
Equations (3) and (4) can be organized in matrix form as per Patel Numerical Method:
[IR′]=[−Y][f] (90)
[II′]=[−Y][e] (91)
where,
IR
p′=(epPSHp′+fpQSHp′)/(ep2+fp2) (92)
II
p′=(epQSHp′−fpPSHp′)/(ep2+fp2) (93)
This is the model where elements of equations (90) and (91) are defined by following equations.
[−Y]=[−B]+[G][−B]−1[G] (94)
[IR′]=[IR]−[G][−B]−1[II] (95)
[II′]=[II]+[G][−B]−1[RI] (96)
Regular loadflow models can also be obtained by differentiating on both sides of equations (89), (90) and (91).
A linear system of equations Ax=b can be written for any equation-p as equations (98) and (97). They can also be written in alternative forms like equation (76) including factor Lp of (77).
A nonlinear system of equations f(x)=y can be written for any equation-p as equations (82), which is specifically a nonlinear power flow equation of a power network involving complex variables and constant parameters.
Equations (98) and (82) are defining equations of Generalized Gauss-Seidel-Patel numerical method involving self-iterations. It should be noted that self-iterations within global iterations are analogous to the earth rotating on its own axis while making rounds around the Sun. This generalized approach for solution of both linear and nonlinear system of simultaneous algebraic equations could potentially be amenable to acceleration factors greater than 2 unlike original Gauss-Seidel numerical method subject to experimental numerical verification. Further verbal elaborations about the Generalized Gauss-Seidel-Patel numerical method will be provided as part of the proposed book writing project.
The system stores a representation of the reactive capability characteristic of each machine and these characteristics act as constraints on the reactive power, which can be calculated for each machine.
While the description above refers to particular embodiments of the present invention, it will be understood that many modifications may be made without departing from the spirit thereof. The accompanying claims are intended to cover such modifications as would fall within the true scope and spirit of the present invention.
The presently disclosed embodiments are therefore to be considered in all respect as illustrative and not restrictive, the scope of the invention being indicated by the appended claims in addition to the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2968813 | May 2017 | CA | national |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 15797649 | Oct 2017 | US |
| Child | 16431952 | US | |
| Parent | 14492428 | Sep 2014 | US |
| Child | 15797649 | US |
