Patent Application
20140081704
The present disclosure pertains to power and particularly to stabilization of power grids. More particularly, the disclosure pertains to buying and selling power.
The disclosure reveals a system for purchasing and selling power that fairly accommodates sellers and buyers. For instance, a submarket may be formed between a utility company or retailer and its consumer or customer. The utility or retailer may eliminate differences between generated or purchased power based on day-ahead predictions and demanded power in a given day. Mechanisms used for elimination of power differences may incorporate purchasing or selling power on the spot market, and affecting a demand for power with demand response programs. A difference between purchased power and demanded power may be minimized by forming an optimal power stack having a mix of power of the demand response program, power at the spot market and/or power of ancillary services. A transmission and system operator (TSO) may operate a distribution grid and maintain grid stability through a use of ancillary services. The utility may pay a fee for elimination of its eventual power imbalance to the TSO. An optimization sequence may be implemented to minimize the difference between the purchased power and demanded power, and to maximize profit.
In
A utility company may be responsible for stabilization of the power grid and for this purpose can use several stabilization mechanisms. The utility company or companies may have made an effort to reduce usage of the ancillary services because of their high prices. A demand response (DR) program may be another option to ensure a stability of the grid by influencing the demand. With an application of the program, the utility company may change customers' loads if a change is beneficial. Utility companies may offer various demand response programs to their customers and each customer can participate in a DR program in its own way. DR programs may be divided into programs with discrete decisions and real-time pricing.
However, DR programs may have considerable drawbacks for both sides—consumers and utility companies. First, the utility company may face a rather difficult decision. In a case of programs with discrete decisions, a DR adjustment is not necessarily smooth and may represent a complex combinatory issue. In case of real-time pricing, it may be difficult to determine an appropriate price as well as the reactions of the consumers that are of a stochastic nature. “Stochastic” may be of or pertain to a process involving a randomly determined sequence of observations each of which is considered as a sample of one element from a probability distribution.
On the other hand, the consumers may have to consider their reactions to a DR event with respect to changing prices (in the case of real-time pricing). Alternatively, the customers may face more or less discrete decisions.
Demand bidding programs may just exploit fixed incentives (e.g., 0.50 cents/kW in day-ahead mode and 0.60 cents/kW in day-of mode). The participants may then just decide whether they should submit their bids and determine what amount of power they are willing to curtail. Bids may be gathered by a utility or at a demand response automation server (DRAS) and evaluated when the time for bid-sending is over. Such an approach may have some disadvantages. First, the fixed discount rate may not necessarily be always beneficial because of its inherent inability to react on current conditions (e.g., real time price, actual demand, and so on). It is simply not necessarily a result of continuous trading but may be rather of an apparent long-term over-designed estimate. Second, the programs may count just with the demand reduction on the participants' side. However, when a utility is facing a power surplus, it may be beneficial for the utility to provide an incentive payment to a customer who commits to move some power required operation (i.e., a re-schedulable load) to a time interval with a surplus.
The present approach may provide a business model for utilities and their consumers that copes with above-mentioned issues, and also be a related decision support tool for utilities for bringing in significant savings.
One goal may be to create a virtual submarket between a utility company (retailer) and its customer. A customer may actively participate in a DR program and supply bids for a load increase or reduction to the virtual submarket (located on utility side).
Customers may evaluate and submit bids that consist of an energy amount and a corresponding price. A price may depend on the particular case and can be categorized as revenue in the case of load reduction, or as a discount price for an additional load in the case of a load increase.
A utility company may need to eliminate differences between generated (purchased) and demanded power. Three kinds of mechanisms may be utilized for elimination of a power difference. The mechanisms may 1) use ancillary services, 2) purchase or sell power at the spot market, or 3) influence the demand via DR program events, respectively. Each mechanism may have its advantages and disadvantages.
1) Ancillary services may represent an ample power source with deterministic prices, but these prices can be high. 2) On the spot market, the power may be sold or bought under market prices which are of a stochastic nature and unknown until trading time. 3) In the present DR mechanism, prices for DR may be given by customers and the prices may be nondeterministic, but known. The utility may have full control of acceptance of the customer's bids (DR power). The first two mechanisms may affect the supply side and the last mechanism (DR) may affect the demand side. The utility may make a decision about an optimal structure of a power stack used for elimination of a power difference. The power may be considered as a mix of DR power, power bought on spot market, and ancillary services power.
A present decision support system may be provided in a form of, e.g., a web service residing on a cloud, on an automated demand response server and help to find an optimal ratio of power mixing. The system may use a scenario approach for overcoming the uncertainty that is included in customer's loads and in final spot market prices. Advanced optimization algorithms (i.e., mixed integer stochastic optimization) may be employed for optimal power mixing and optimal customer bids selection. Probabilistic measures may be exploited for an evaluation of risk. A level of risk may be specified by the utility (e.g., conservative behavior versus aggressive behavior).
There may be basically two major features of the present approach. 1) The customers may have the opportunity to influence the final incentives (which are fixed in a current demand bidding program (DBP)) as they are allowed to send the bids that consist not only of a power reduction/increase amount but also an expected price for each particular amount. Furthermore, not only may load reduction bids be requested but also bids may be made for load increases. The utility may then decide whether it is economically beneficial to exploit these bids or, e.g., sell the power surplus back to the market. 2) A present decision support tool may help the utility to make the most beneficial decisions in every step of an operation. For example, the tool may suggest an optimum amount of power to be taken from accepted bids (besides the decision about which bids should be accepted), an optimum amount of power that should be traded at the market, and so forth (e.g., from ancillary services). The decisions may be generated by the optimization tool that considers the stochastic nature of involved variables by using a so-called scenario approach (i.e., a significant principle from stochastic optimization).
There may be devices placed on the customer site that can communicate with a DR server. Each customer may be allowed to submit bids that consist of a provided amount of demand reduction/increase, time interval and price offer. The bid may also have a form of a function (price=function(power increase or reduction)). Bids may be generated either manually or automatically whenever they are requested by the DRAS (demand response automation server) hosting the virtual sub-market application. Similarly with respect to demand bidding programs, which may already be run by most of the utilities supporting a demand response, the bids may be requested the day before a particular event (i.e., day-ahead mode) or directly during the event day (i.e., day-of mode, near future or real time DR).
When an electricity demand forecaster, running at the DRAS, indicates that, on the next day (a day-ahead mode), there may be a high probability of a mismatch between purchased/generated power and a forecasted demand, the request for bids may be sent to virtually all DR participants. The participants may be requested to send their bids up to some fixed deadline on the day before event. The participants of the particular DR program may then submit their bids. The utility may evaluate the most profitable composition of a power stack needed for overcoming the purchased/generated power versus demand discrepancy. Nearly, the same mechanism may be utilized for the day-of events. The events may be generated when more accurate (forecast horizon in order of hours) predictions are available. The participants may then have, of course, less time to submit their bids; however, they can expect higher payments as the final price should be more influenced by the spot market and the ancillary services price. In both modes (the day-ahead and day-of), the virtual sub-market application running at DRAS may be responsible for generating recommendations for a utility in how to compound the power from different sources (e.g., the market, ancillary services, DR bids, and so on) in the final corrective action that matches the demand and supply with each other.
A decision support system may be based on a virtual energy market (VEM). Accomplishments may incorporate establishing a new mechanism for demand responses, creating a virtual submarket between a retailer (utility) and its customers, and bringing benefits to the retailer (i.e., higher profit) and to consumers (i.e., more savings).
The graph may be plot RPenalty (revenue—penalty) versus PAS (power—ancillary services) as indicated by lines 27.
Particular bids may consist of an amount of energy, duration time and incentives. The incentives may depend of the particular case which may provide revenue in the event of load reduction and a discount of price for additional loads in the event of a load increase.
A retailer (utility) may eliminate differences between purchased and load power or energy. The difference may be eliminated via DR, spot market or ancillary power. The retailer may make a decision about accepting bids from customers.
A retailer may purchase an energy or power profile for the next day in the day-a-head market. The purchase may depend on a load forecast for the next day. Generally, power differences may occur because of load uncertainties. A retailer may want to eliminate the differences in an economically optimal way.
Difference elimination possibilities may incorporate: 1) Letting a system operator making use of ancillary services to eliminate the difference (i.e., an expensive way); 2) Selling or purchasing, for instance, electricity on the market (i.e., price is given by market—stochastic approach); and 3) Making a demand response action (i.e., price is given by relationship of a trader-consumer—deterministic approach).
One may note an optimization at a one time instance of a ΔP elimination with a set 31 of equations indicated in
One may note an optimization of a ΔP elimination with a set 32 of equations indicated in
R(ΔP)=RLoad(P)−RDR(α·ΔP)−RSpot(β·ΔP)−RAS(χ·ΔP)−RPurchased.
RLoad(P) may indicate a deterministic price (known) and load reflection. RDR(α·ΔP) may indicate a stochastic price (known) and limited power. RSpot(β·ΔP) may indicate a stochastic price (unknown) and partially limited power. RAS(χ·ΔP) may indicate a deterministic price (penalty) and “unlimited” power. RPurchased may indicate purchased power and is not necessarily important for optimization.
For a load greater than supply, a set 34 of equations, as shown in
R(ΔP)=RLoad(P)−RDR(α·ΔP)+RSpot(β·ΔP)−RAS(χ·ΔP)−RPurchased.
RLoad(P) may indicate a deterministic price (known) and load reflection. RDR(α·ΔP) may indicate a stochastic price (known) and limited power. RSpot(β·ΔP) may indicate a stochastic price (unknown) and partially limited power. RAS(χ·ΔP) may indicate a deterministic price (penalty) and “unlimited” power. RPurchased may indicate purchased power and is not necessarily important for optimization.
In
where α+β+χ=1. A scenario approach may be used to overcome an uncertainty. Solving the optimization task may be done with a presently selected approach. Probabilistic measures may lead to a determination of risk.
An optimization sequence may incorporate: 1) Reading historical data from a database; 2) Constructing one or more load forecasting models; 3) Retrieving external information about prices and weather trends; 4) Retrieving bids from consumers; 5) Using the models, generating scenarios and considered parameter combinations (α, β, χ) (e.g., 0.6, 0.3, 0.1); 6) For each parameter combination (α, β, χ), a) evaluating a cost function for particular settings (α, β, χ) over virtually all scenarios, and b) using probabilistic measures, e.g., a combination the brings a highest revenue at a given risk level (such as revenue achieved with 95 percent probability), as an aggregation function for virtually all scenarios in determination of a final value of the cost function for the combination (α, β, χ); 7) Finding optimal values of parameters α*, β*, χ* from aggregated values; 8) Informing consumers about acceptance; and 9) Measuring a real operation and saving the operation to a database.
As to step 2, concerning model construction, models may be needed for: a) Distributions P(Weather), P(Prices), P(Behavior|Weather) and/or P(Behavior); b) Mapping Consumption(Weather, Behavior, Acceptance); and c) Mapping Profit(Consumption, Acceptance, Prices). One may note profit as revenue and cost but as also involving accepted and fulfilled incentives.
The models may be obtained or construed from historical data (i.e., a black box), possibly with use of: a) Some apparent relationships such as summing up the total consumption of particular consumers; b) External information such as weather forecasts, public holidays, and so forth; and c) Behavior to be modeled as a function of time explaining modeling residuals of black box models of the consumption conditioned by weather and acceptance.
As to step 5, concerning application of a scenario approach, a scenario may represent uncertain information in the system. Knowing the scenario and making a decision, a next evolution of the system may be determined. In a case of DR programs, scenarios may involve: a) Weather (temperature, humidity, solar radiation) which may not necessarily depend on decisions; b) Consumer behavior patterns (daily, weekly, yearly trends) which may be affected by acceptance of demand response bids; c) Spot market (prices); and d) Impact of DR (e.g., in the hope that the consumer is able to fulfill the bid).
As to step 5, concerning generating scenarios, items to be noted may incorporate: 1) Sampling a trajectory of weather, prices; 2) Conditioned in this trajectory, sampling a trajectory of behavior; and 3) Determining consumption as a function of acceptance. The steps may be repeated in that many scenarios are generated. Thus, each scenario “s” may produce mapping—Consumptions(Acceptance). Consumption and prices may directly determine utility profit—Profits(Acceptance).
As to step 6, an aim of optimization may be to maximize profit with ensuring an elimination of a power difference—ΔP+ΔPCORR=0 and ΔPCORR=(a·ΔP)+(β·ΔP)+(χ·ΔP), where (α·ΔP) pertains to RDR, (β·ΔP) pertains to RSPOT, and (χ·ΔP) pertains to RAS. An optimization algorithm may find an optimal combination of a spot market power 13 and an ancillary services power χ. A demand response power a may be a discrete variable determined by acceptance. An impact of individual consumers may be assumed to be reasonably independent. With respect to a search algorithm, genetic algorithms may be proposed because of a discontinuous objective function. Other heuristics may be applicable if a set of accepted bids does not depend on a simple sort.
As to step 6, concerning risk measures, the following factors may be considered. Using a selected measure may determine an aggregation function. Parameter a may represent an optimal combination of supplied bids (discretized value). A selected combination may optimize an objective function (e.g., profit) over virtually all scenarios with a consideration of risk. Possible aggregation approaches may incorporate: 1) Mean—a combination may maximize expected profit over virtually all scenarios; 2) Worst case—a combination may maximize a minimal profit over virtually all scenarios; and 3) Percentile—a combination may maximize a profit that is given by N-th percentile of the objective functions for virtually all scenarios.
Reschedulable loads 65 may incorporate, for example, loads such as 1 kW for 2 hours and 20 kW for one hour. The loads may be of any other amounts and durations. For each of such loads may be a schedule or graph 67 showing DR dollars versus time with increments of price along the time line. There may be a time deadline where the price ($DR) is stopped at a fixed level. The price at the fixed level may be, for instance, a normal price.
Before the each optimization procedure run, the algorithm should have the following items at its disposal. It may be noted that the solution described herein may just deal with static optimization, i.e., the optimization task is solved separately for each time slot (e.g., one hour) or a several time slots in row but with the no correlation between the slots being assumed. The present approach may be easily extended to solve a dynamic optimization problem (allowing inter-slot dependencies).
The items may incorporate the following. 1) A weather forecast for each DR participant (sharing weather resources may be exploited in advance). The probability distribution function, e.g., for OAT, may be estimated for a given time-slot. 2) Spot market price prediction for given time slot in a form of probability distribution function. It may be estimated based on the historical data. 3) Individual electrical energy demand models for each DR participating load. An example of global multivariate regression model (ToD=Time-of-Day could simulate the occupancy level which is not usually available) may be:
L=a
0
+a
1·OAT+a2·ToD+a3·ToD2+Accp·DR
, where Accep is a bid acceptance status (binary) and DR is a general term representing the influence of demand response action. 4) Bids (nominations) from all potential DR participants, where each bid (load [kW] reduction/increase) may be a function of time slot and incentive expected. It may mean that multiple bids for the same time slot are allowed and participants are allowed to offer their own price in contrast to current incentive politics produced exclusively by utilities.
Symbol 82 indicates generating a sufficient number of scenarios. Create set of test scenarios (say 1000). Every scenario can be described by a vector (or matrix for multi-step) of values generated based on estimated (historical data based) probability distribution functions. Following the example load model, the random variables of the 3-participants scenario vector are generated based on distributions
[P(price,P(OAT1),P(OAT2),P(OAT3),P(DR1|OAT1),P(DR2|OAT2),P(DR3|OAT3)]
where first term is the spot market price distribution, next three terms are distributions of outdoor air temperatures for given time-slot and last three terms are conditional distributions of load reduction/increase capabilities for given outdoor air temperatures.
Symbol 83 indicates evaluating expected demands. For every acceptance vector (selected in the first step) the expected total demand (sum of individual participants' demands) now may be evaluated against virtually all (e.g., 1000) scenarios (scenario=vector of realizations of random variables).
Symbol 84 indicates evaluating an expected profit distribution. Having the spot market price for each scenario and known penalty politics for exploiting the ancillary services (i.e., excessive power consumption), the expected profit may be evaluated for each scenario given the acceptance vector. It may be seen as a profit distribution over all testing scenarios for given acceptance vector.
Symbol 85 indicates finding an optimum acceptance vector. Profit distributions may then be evaluated for all testing acceptance vectors. The optimization may search for such an acceptance vector that maximizes the profit with the given required level of confidence (i.e., risk level). Note the set of testing acceptance vectors was found by the search procedure based on the bid ordering or on some other search approach (genetic algorithm) in the first step.
Symbol 86 indicates exploiting the acceptance vector. The optimum acceptance vector may then support the decisions about whom to accept the DR bid and when.
Each column may have graphs 95 and 96 of P(Weather) versus Weather and of P(Occupancy) versus Occupancy, respectively. The energy market may be represented with a graph 97 of P(Price) versus Price.
A table 98 may show data for a number of scenarios for three situations with indications of Toa1, Toa2, Toa3, Occ1, Occ2 and Occ3. Price may be indicated for each scenario. In this example, weather may be determined by outside temperature Toa. If occupancy is not available, then it may be replaced by a time-of-day variable that is able to capture the occupancy profile sufficiently.
To recap, a system for optimizing a balance of power, may incorporate a first mechanism that decides about purchasing power from ancillary services, a second mechanism that purchases or sells power at a spot market, a third mechanism that purchases or sells power according to a demand response program, and a processor having a connection to the first, second and third mechanisms. The processor may process a reduction of a difference between purchased power of a supplier and demanded power of a consumer, by determining an amount of power bought and/or sold with one or more of the first, second and third mechanisms.
The difference between purchased power and demanded power may be minimized by forming an optimal power stack. The power stack may incorporate a mix having power via the demand response program, power at the spot market, and/or power from ancillary services. The processor may determine the mix of the power stack to minimize the difference between purchased power and demanded power.
The system may further incorporate an optimization sequence. The optimization sequence may incorporate maximizing profit and/or minimizing the difference between the purchased power and the demanded power.
Some terms may incorporate PLoad as demanded power, PPurchased as purchased power, and ΔP as the difference between PLoad and PPurchased. Also, there may PDR=αΔP, PSpotβΔP, and PAS=χΔP ΔPCorrection may incorporate PDR, PSpot and PAS. ΔP+ΔPCorrection=0 and α+β+χ≈1 may be applicable.
α could be a discrete variable representing a magnitude of acceptance of the consumer in the demand response program. A power difference may be ΔP=PLoad−PPurchased. For a load greater than supply, ΔP−PDR−PSpot−PAS=0 and R(ΔP)=RLoad(P)−RDR(αΔP)+RSpot(βΔP)−RAS(χΔP)−RPurchased. Also, −PDR−PSpot−PAS=ΔPCorrection, R(ΔP) may be profit, RLoad(P) may be a price of the load, and RDR(αΔP) may be a price of power determined between the utility and the consumer in a demand response relationship. Also, RSpot(βΔP) may be a price of power on an open market, and RAS(χΔP) may be a price of power from a system operator providing ancillary services at a set price. One may have α+β+χ≈1, and
for optimization.
A power difference may be ΔP=PLoad−PPurchased. For a load less than supply, there may be ΔP+PDR+PSpot+PAS=0 and R(ΔP)=RLoad(P)−RDR(αΔP)−RSpot(βΔP)−RAS(χΔP)−RPurchased. There may be PDR+PSpot+PAS=ΔPCorrection. R(ΔP) may be profit, and RLoad(P) may be a price of the load. RDR(αΔP) may be a price of power determined between the utility and the consumer in a demand response relationship. RSpot(βΔP) may be a price of power on an open market, and RAS(χΔP) may be a price of power from a system operator providing ancillary services at a set price. One may have α+β+χ≈1, and
for optimization.
A system for managing energy, may incorporate a server, a virtual energy marketing (VEM) module connected to the server, a utility energy source connected to the VEM module, a meter data management (MDM) database connected to the VEM module, an energy consumer connected to the server and the MDM database, and an energy market source connected to the VEM module.
The VEM module may incorporate a decision engine connected to the utility energy source and the server, a scenario generator connected to the decision engine, a forecaster mechanism connected to the scenario generator and the MDM database, and a probability distribution generator connected to the forecaster, the energy market database and the weather forecast database. The system for managing energy may further incorporate a weather forecast database connected to the probability distribution generator.
The utility may provide information about power unbalance between loaded power and purchased power and/or grid status to the decision engine. Demand response signals and business information may be exchanged between the consumer and the server. The decision engine may provide optimal timing and selection of demand response resources to the server. The consumer may provide energy consumption data to the MDM database. The forecaster may receive selected relevant data from the MDM database. The energy market source may provide a market price to the probability distribution generator.
The system may further incorporate a weather forecast database connected to the probability distribution generator. The weather forecast database may provide weather parameters to the probability distribution generator. The server may be a demand response automation server.
An approach for coordinating power transactions, may incorporate finding out an amount of purchased power of a utility, finding out an amount of demanded power by a consumer, minimizing a power difference between an amount of purchased power of the utility and an amount of demanded power by the consumer, minimizing the power difference that depends on, at least in part, purchasing power from a system operator providing ancillary services at a set price, selling or purchasing power on the open market at market price, and/or selling or purchasing power at a price determined between the utility and the consumer in a demand response relationship.
The power difference may be between an amount of purchased power of the utility and an amount of demanded power by the consumer, with optimizing a combination of PSpot, PAS and PDR Goals of optimizing the combination may incorporate maximizing profit to the utility and minimizing the power difference.
αΔP, βΔP and χΔP may represent portions of the respective power that constitute the power difference between the amount of purchased power of the utility and the amount of demanded power by the consumer. Parameters α, β and χ may be determined to minimize the power difference, where α+β+χ≈1.
In the approach, the power difference may be ΔP=PLoad−PPurchased. For a load greater than supply, ΔP=PDR⊕PSpot−PAS=0 and R(ΔP)βRLoad(P)−RDR(αΔP)+RSpot(βΔP)−RAS(χΔP)−RPurchased. There may be −PDR−PSpot−PAS=ΔPCorrection. R(ΔP) may be profit, and RLoad(P) may be a price of the load. RDR(αΔP) may be a price of power determined between the utility and the consumer in a demand response relationship. RSpot(βΔP) may be a price of power on an open market. RAS(χΔP) may be a price of power from a distribution company providing ancillary services at a set price, and α+β+χ≈1.
The power difference may be ΔP=PLoad−PPurchased. For a load less than supply, ΔP+PDR+PSpot+PAS=0 and R(ΔP)=RLoad(P)−RDR(αΔP)−RSpot(βΔP)−RAS(χΔP)−RPurchased. There may be PDR+PSpot+PAS=ΔPCorrection. R(ΔP) may be profit. RLoad(P) may be a price of the load. RDR(αΔP) may be a price of power determined between the utility and the consumer in a demand response relationship. RSpot(βΔP) may be a price of power on an open market. RAS(χΔP) may be a price of power from a distribution company providing ancillary services at a set price, and α+β+χ≈1.
In the present specification, some of the matter may be of a hypothetical or prophetic nature although stated in another manner or tense.
Although the present system and/or approach has been described with respect to at least one illustrative example, many variations and modifications will become apparent to those skilled in the art upon reading the specification. It is therefore the intention that the appended claims be interpreted as broadly as possible in view of the related art to include all such variations and modifications.
