This disclosure relates generally to operation of distributed energy resources.
The use of distributed energy resources (DERs) is becoming common. Examples of residential distributed energy resources are batteries, solar photovoltaic panels, small wind turbines, natural-gas-fired fuel cells, and emergency backup generators, usually fueled by natural gas, gas oline or diesel fuel. Examples of commercial and industrial distributed energy resources are batteries and other storage, combined heat and power systems, solar photovoltaic panels, wind, hydropower, biomass combustion or cofiring, municipal solid waste incineration, fuel cells fired by natural gas or biomass and reciprocating combustion engines, including backup generators, which may be fueled by oil. Any of these options may be used by the electrical grid operator, as well as residential, commercial or industrial third parties.
Installation and operation of any of the DERs is an expense proposition, so choice of location and operating protocols are important when options are available.
For illustration, there are shown in the drawings certain examples described in the present disclosure. In the drawings, like numerals indicate like elements throughout. The full scope of the inventions disclosed herein are not limited to the precise arrangements, dimensions, and instruments shown. In the drawings:
Optimal siting of a DER and operation of the DER depends on many factors: the properties of the DER itself, the electrical environment into which the DER is placed, the economic environment into which the DER is placed, and others. Further, the DER itself changes the environments once it is installed, so analysis should also take that into consideration.
Referring to
While the above is an overall summary, provided here are details on the various steps. Besides locations at a zip code granularity, the details of the DER inputs are in different categories of DER capital and physical characteristics, operating limits and preferences, and available wholesale market revenue streams. Examples include but are not limited to:
Historical nodal prices of electricity market products and services for each location are retrieved from a database containing those details.
A neural network 200 as shown in
In one embodiment, the neural network 200 is constructed of five linear layers with four LeakyReLU (Leaky Rectified Linear Unit) layers 206, 210, 214, 218 in between them, the final linear layer 220 serving as the layer to output price prediction. The linear layers 204, 208, 212, 216, 220 utilized have respective output feature sizes of 4096, 2048, 1024, 1024, and 1 in that order. For the purposes of optimization, the AdamW optimizer is used for the training process for its ability to better generalize to unseen data than the traditional Adam optimizer regarding many datasets. Once trained, upon receiving initial projections about future generation mix, load growth and weather, obtained from various sources, the output is the initial price prediction for a desired period of time. Preferably all calculations and predictions are done using 15 minute or hourly data. In some embodiments, the price predictions for given locations are performed prior to and separate from the remainder of the processing, so that the initial pricing prediction is a database lookup, rather than real time computations.
It is understood that this is an exemplary neural network and other neural network configurations can be used after proper training.
The price prediction outputs are then fed into the optimization engine. In one embodiment, the optimization is an MILP optimization considering detailed DER performance model, operational constraints and ISO/RTO specific market rules. The objective function of this optimization problem is to maximize resulting market revenues minus operational cost of the DER. In addition to financial calculations, the optimization output provides hourly operational behavior for the decided DER.
For example, dispatching an energy storage DER is a non-linear optimization problem that the optimization operation solves using linear programming transformation techniques and Mixed Integer Linear Programming (MILP) optimization.
The optimal decision maximizes the profit while it satisfies all the physical and operational constraints. In one embodiment, one important physical characteristic of an energy storage DER is the state-of-charge (SoC) which shows how much energy is stored in the DER. If the energy storage DER participates in Energy and Ancillary Services (AS) markets, the optimization solver decides when to charge the DER from the grid, when to discharge the DER into the grid to participate in the Energy market, and when to keep the capacity and energy available to participate in the AS market. The optimization solver predicts the DER SoC after participation in the Energy and AS markets to accurately decide about the next interval dispatch while satisfying the SoC constraint.
The initial prices are predicted based on the traditional behavior of the system generation units and electricity demand. However, by increasing the penetration of the DER assets in the system, the traditional behavior will be affected. For this reason, in a separate process, future DERs penetration in the system has been assessed in terms of types, sizes, and locations. The optimization engine determines DERs behavior based on the predicted prices by optimizing their profit considering all physical and operational constraints.
The hourly dispatch profiles are then taken back to the future pricing datasets to serve as a modification to initial hourly projections about the system supply and demand. Consequently, the data features regarding generation mix and load for the future are changed impactfully at an hourly level in regards to the added DERs and their locations. This updated data is then fed through the same neural network as before, thus outputting final price predictions reflective of the DERs' impact on the grid and the market.
The final price predictions are then used as inputs to the optimization process for the specific DER. The final outputs of the optimization process are then the final economics and hourly dispatch schedule for each location and DER of interest. These outputs are then provided for review, with the dispatch schedule forming the basis for operating the DER.
After some period of operation, the price prediction process can be repeated while updating the training set of the neural network model and considering any ISO/RTO regulatory changes.
The server 302 includes a processor 308, RAM 310 used to store programs and data during operation of the system 300, a network interface card (NIC) 312 to connect to the network 304, and non-volatile program storage 314. Programs contained in the storage 314 are an operating system 316; an overall DER location and operation program 318 which executes the flowchart 100; a neural network 320, such as the neural network 200, and linear programming transformation and MILP optimization program 322. Storage 314 further includes a database containing the data needed to operate the neural network and the optimization, including location data, DER specifications, nodal prices, tariffs, existing generation data, existing load data, weather, and market rules.
It is understood that this is a highly simplified illustration of the system, 300 and an actual system may be configured in numerous different ways to perform the operations.
By performing a second pass based on load and other changes due to the impact of the DERs, better predictions and an improved dispatch or operational schedule are obtained.
The various examples described are provided by way of illustration and should not be construed to limit the scope of the disclosure. Various modifications and changes can be made to the principles and examples described herein without departing from the scope of the disclosure and without departing from the claims which follow.
This application claims priority to U.S. Provisional Application Ser. No. 63/363,209, filed Apr. 19, 2022, the contents of which are incorporated herein in their entirety by reference.
Number | Date | Country | |
---|---|---|---|
63363209 | Apr 2022 | US |