Electrochemical energy storage power station planning operation collaborative optimization method considering uncertainty of new energy power generation and load demand

Abstract

The invention provides an electrochemical energy storage power station planning operation collaborative optimization method considering new energy power generation and load demand uncertainty, and relates to the field of electrochemical energy storage planning and operation. Performing abnormal data processing on the measured value; calculating a relative error value between the standardized source load data and the predicted value, and optimizing mixed parameters of a pre-configured Gaussian mixture model in combination with an adaptive momentum regularization expectation maximization algorithm; constructing an electrochemical energy storage power station configuration optimization model in combination with the source load uncertainty model; and based on an electrochemical energy storage power station configuration optimization model result, combining an electrochemical energy storage power station power balanced distribution strategy of a health degree attenuation effect to realize energy storage power station power distribution. According to the method, the source-load probability distribution is converted into a deterministic power system flexibility requirement by setting the confidence level, a premise is provided for the optimization configuration of a subsequent electrochemical energy storage power station, and the planning reasonability of the energy storage power station is ensured.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage planning and operation, and more specifically, to a collaborative optimization method for the planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand. Background Technology

[0002] Currently, the scale of new energy generating units connected to the power system continues to climb. Simultaneously, rapid socio-economic development has led to increasingly diverse load types and continuously growing load demand. This results in significant uncertainty and mismatch between power output and consumption at both the power source and load ends, making the need for flexible regulation resources extremely urgent. Energy storage technology, especially electrochemical energy storage, is a key means to improve grid flexibility and optimize the grid-friendliness of new energy connections. Its advantages, such as fast response, flexible regulation, and lack of geographical limitations, are increasingly highlighting its importance in the power system.

[0003] However, existing energy storage planning and operation still have significant limitations. On the one hand, there is insufficient uncertainty modeling. Stochastic programming and robust optimization methods struggle to coordinate the multidimensional stochasticity of wind and solar power output and load demand; robust optimization is overly conservative, while stochastic programming suffers from high computational complexity. On the other hand, there is a disconnect between planning and operation. Traditional methods employ deterministic models or single-scenario optimization, failing to consider the long-term uncertainties of renewable energy and load, leading to capacity mismatch (such as over-investment or insufficient regulation capacity).

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In view of this, the present invention provides a collaborative optimization method for the planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand, in order to solve the aforementioned problems.

[0006] To solve the above problems, the specific technical solution adopted by the present invention is as follows:

[0007] A collaborative optimization method for the planning and operation of electrochemical energy storage power stations, taking into account the uncertainties of new energy power generation and load demand, includes the following steps:

[0008] S1. Obtain the measured values ​​of source and load data in the regional power system as actual values, and perform abnormal data processing on the measured values ​​based on the 3-sigma principle, physical rules, and spatiotemporal feature map structure to obtain standardized source and load data.

[0009] S2. Obtain the predicted values ​​corresponding to the source load data in the regional power system, and calculate the relative error between the standardized source load data and the predicted values. Based on the relative error, optimize the mixing parameters of the pre-configured Gaussian mixture model using the adaptive momentum regularization expectation maximization algorithm to obtain the source load uncertain model.

[0010] S3. Construct a full life cycle cost model for electrochemical energy storage power stations based on the unit combination framework, and combine it with the source-load uncertainty model to construct an electrochemical energy storage power station configuration optimization model;

[0011] S4. Construct a health estimation model for electrochemical energy storage power stations based on physical information fusion, graph convolution, and long short-term memory neural networks. Based on the results of the configuration optimization model for electrochemical energy storage power stations, combine the power equalization allocation strategy of electrochemical energy storage power stations with the health decay effect to realize the power allocation of energy storage power stations.

[0012] Preferably, the process of obtaining standardized source-load data by using measured values ​​of source-load data from the regional power system as actual values ​​and performing anomaly data processing on the measured values ​​based on the 3-sigma principle, physical rules, and a multi-level anomaly data identification mechanism using spatiotemporal feature map structures includes the following steps:

[0013] S11. Perform data cleaning and missing value detection on the measured values ​​of source and load data in the regional power system, and fill in the missing values ​​using the mean imputation method.

[0014] S12. Using the 3-sigma principle, perform preliminary screening of outliers on the measurement values ​​of the source payload data after data cleaning.

[0015] S13. Based on physical rules, perform a second screening process on the outlier values ​​of the source load data after the initial screening process.

[0016] S14. Based on the spatiotemporal graph feature structure, construct a Laplace graph, and use the Laplace graph to perform outlier screening on the measured values ​​of the source load data after secondary screening to obtain standardized source load data.

[0017] Preferably, the secondary outlier screening process for the source load data measurements after the initial screening, based on physical rules, includes the following steps:

[0018] S131. Based on the characteristics of photovoltaic and wind power, construct physical models for photovoltaic power generation and wind power generation respectively;

[0019] S132. Based on the physical models of photovoltaic power generation and wind power generation, the theoretical output of photovoltaic power and the theoretical output of wind power are calculated by acquiring environmental meteorological data of photovoltaic power stations and wind power stations respectively.

[0020] S133. Based on the theoretical output of photovoltaic and wind power, use upper and lower limit thresholds to identify outliers in the measured values ​​of source and load data after preliminary screening and processing, and remove the identified outliers.

[0021] Preferably, the step of constructing a Laplace graph based on the spatiotemporal graph feature structure, and performing a third outlier screening on the source load data after the second screening using the Laplace graph, includes the following steps:

[0022] S141. Based on the spatiotemporal relationship between the power grid structure and photovoltaic and wind farms in the regional power system, construct a Laplace diagram;

[0023] S142. By identifying outliers in the measured values ​​of the source load data after secondary screening, the mutation of the Laplace plot structure is used to identify outliers.

[0024] The expression for identifying outliers in the source load data after secondary screening by analyzing mutations in the Laplace diagram's graph structure is as follows:

[0025]

[0026] In the formula, s i This represents outliers, and ΔL represents the rate of change of the Laplace plot over time. H represents the set of adjacent nodes of node i. ij Let σ represent the edge features between node i and node j, and let Λ represent the activation function. τ Represents a time-dynamic filter, Θ τ The spatiotemporal graph features are represented by τ, where τ represents the index of the time dimension, and H represents the feature parameters. (l) H represents the edge feature vector of the l-th layer. (l+1) This represents the edge feature vector of the (l+1)th layer;

[0027] S143. Remove the identified outliers.

[0028] Preferably, the steps of obtaining the predicted values ​​corresponding to the source load data in the regional power system, calculating the relative error between the standardized source load data and the predicted values, and optimizing the mixing parameters of the pre-configured Gaussian mixture model based on the relative error and the adaptive momentum regularized expectation maximization algorithm to obtain the source load uncertainty model include the following steps:

[0029] S21. Initialize the conditions for the adaptive momentum regularized expectation maximization algorithm, including the number of sub-distributions, batch size, distribution parameters, optimizer initial momentum, variance, and number of iterations.

[0030] S22. Based on the predicted values ​​corresponding to the source load data in the regional power system, calculate the error values ​​between the standardized source load data and the predicted values ​​to obtain a dataset of relative prediction error values. Based on the batch size, divide the error value dataset to obtain several sub-batch sets.

[0031] S23. For each data point in a subset, the expectation step is used to calculate the posterior probability of each data point belonging to each Gaussian distribution.

[0032] The formula for calculating the posterior probability is:

[0033]

[0034] In the formula, λ ik This represents the posterior probability that a data point belongs to the k-th Gaussian distribution. Let π represent the probability density function of the k-th Gaussian distribution. k Let represent the mixing coefficient of the k-th Gaussian distribution, K represent the number of subdistributions, and j represent the index value of the number of subdistributions. Let represent the probability density function of the j-th Gaussian distribution;

[0035] S24. Construct the likelihood function, and update the parameters of the Gaussian mixture model and the distribution parameters of the mixture components with the goal of maximizing the likelihood function, to obtain the source load uncertainty model.

[0036] The expression for the likelihood function is:

[0037]

[0038] In the formula, Q(θ|θ) (t) ) represents the likelihood function, θ (t) Z represents the parameter value in the t-th iteration; Z represents the latent variable. Let denot represent the class classification of a sample in a Gaussian mixture model, and p(X,Z|θ) represent the joint probability distribution.

[0039] Preferably, the construction of a full life-cycle cost model for an electrochemical energy storage power station based on a unit combination framework, and the construction of a configuration optimization model for the electrochemical energy storage power station in conjunction with a source-load uncertainty model, includes the following steps:

[0040] S31. Based on the charging and discharging data of the electrochemical energy storage power station, estimate the equivalent operating life of the electrochemical energy storage power station;

[0041] S32. Based on the equivalent operating years of the electrochemical energy storage power station and in combination with the impact of rated capacity on the total life cycle cost of the electrochemical energy storage power station, calculate the investment cost and annual operation and maintenance cost of the electrochemical energy storage power station.

[0042] S33. Based on the investment cost and annual operation and maintenance cost of electrochemical energy storage power stations, construct a full life cycle cost model for electrochemical energy storage power stations;

[0043] S34. Transform the source-load uncertainty model into a power system flexibility constraint, and calculate the boundary range of the source-load uncertainty model at the significance level;

[0044] S35. Construct an optimization model for the configuration of electrochemical energy storage power stations based on power system flexibility constraints and the full life cycle cost model of electrochemical energy storage power stations.

[0045] Preferably, the construction of the electrochemical energy storage power station configuration optimization model based on power system flexibility constraints and the full life cycle cost model of the electrochemical energy storage power station includes the following steps:

[0046] S351. Based on the full life cycle cost model of electrochemical energy storage power stations, determine the objective function of the configuration optimization model for electrochemical energy storage power stations;

[0047] The objective function expression of the electrochemical energy storage power station configuration optimization model is:

[0048]

[0049] In the formula, J represents the number of conventional thermal power units; T represents the total number of time periods; This represents the start-up cost of a conventional thermal power unit j; This represents the power generation cost of a conventional thermal power unit (j). This indicates the output of a conventional thermal power unit j during time period t; u j,t This indicates the on / off status of a conventional thermal power unit j during time period t;

[0050] S352. Based on the constraints of the power system unit combination model, and combined with the power system flexibility constraints and the operation constraints of the electrochemical energy storage power station, determine the constraints of the configuration optimization model of the electrochemical energy storage power station; the constraints of the power system unit combination model include unit output constraints, unit ramping constraints, unit minimum start-up and shutdown time constraints, and power balance constraints.

[0051] Preferably, the construction of the electrochemical energy storage power station health estimation model based on physical information fusion-graph convolution-long short-term memory neural network, and the realization of power allocation of the energy storage power station based on the configuration optimization model results of the electrochemical energy storage power station, combined with the power equalization allocation strategy of the electrochemical energy storage power station based on the health decay effect, includes the following steps:

[0052] S41. Construct an aging dynamic model for energy storage batteries, and combine it with a physical information encoder, graph convolutional network, and long short-term neural network to construct a power station health estimation model.

[0053] S42. Estimate the energy storage health of the electrochemical energy storage power station based on the power station health estimation model;

[0054] S43. For energy storage units within an electrochemical energy storage power station, update the actual capacity of each energy storage unit using the energy storage health status of the electrochemical energy storage power station.

[0055] S44. Obtain the dispatch instructions issued by the dispatching agency, and allocate the power of the electrochemical energy storage power station in conjunction with the configuration optimization model of the electrochemical energy storage power station.

[0056] Preferably, the construction of the energy storage battery aging dynamics model, combined with a physical information encoder, graph convolutional network, and long short-term neural network, to construct a power plant health estimation model includes the following steps:

[0057] S411. Based on the aging dynamics model of the energy storage battery, obtain the mechanism information of the energy storage battery, and use a physical information encoder to embed the mechanism information of the energy storage battery into a neural network to obtain physical coding features.

[0058] S412. Based on physical coding features, the electrical and thermal coupling relationships inside the electrochemical energy storage power station are obtained through a graph convolutional neural network, resulting in the graph convolution feature matrix.

[0059] S413. Input the feature matrix of the graph convolution into the long short-term neural network, and use the output of the long short-term neural network as the input of the fully connected layer to obtain the power plant health estimation model of the chemical energy storage power plant.

[0060] Preferably, the step of obtaining the dispatch instructions issued by the dispatching agency and allocating the power of the electrochemical energy storage power station in conjunction with the configuration optimization model of the electrochemical energy storage power station includes the following steps:

[0061] S441. Obtain the dispatch instructions issued by the dispatching agency and determine the status of the dispatch instructions. Based on the results of the electrochemical energy storage power station configuration optimization model, determine the working mode that the electrochemical energy storage power station needs to adjust.

[0062] S442. Determine the current operating mode of the electrochemical energy storage power station, estimate the available power based on the state of each energy storage unit, and calculate the total available power of the electrochemical energy storage power station based on the available power of each energy storage unit.

[0063] S443. Based on the dispatch instructions of the electrochemical energy storage power station and the total available power of the electrochemical energy storage power station, calculate the adaptive power allocation factor of the energy storage unit and allocate the power of each energy storage unit.

[0064] The formula for calculating the adaptive power allocation factor of the energy storage unit and allocating the power of each energy storage unit is as follows:

[0065]

[0066] In the formula, F represents the adaptive power allocation factor of the electrochemical energy storage power station, and P i P represents the power allocated to the i-th energy storage unit. Dispatch Indicates a scheduling instruction, P i,ava This represents the available power of the i-th energy storage unit.

[0067] The beneficial effects of this invention are as follows:

[0068] 1. This invention employs a three-level anomaly data identification mechanism to accurately screen abnormal data in new energy power generation, maximizing the consistency between data distribution and actual distribution. Furthermore, it proposes an adaptive momentum regularization expectation maximization algorithm to overcome the slow convergence speed, susceptibility to local optima, and covariance matrix degradation problems inherent in traditional expectation maximization algorithms. A GMM model of new energy power generation and load demand is established, and by setting confidence levels, the probability distribution of source and load is transformed into deterministic power system flexibility requirements, providing a prerequisite for the subsequent optimized configuration of electrochemical energy storage power stations and ensuring the rationality of energy storage power station planning.

[0069] 2. This invention uses an electrochemical energy storage power station health estimation model based on physical information fusion-graph convolution-long short-term memory neural network to estimate the SOH of the energy storage power station. GCN can effectively extract spatial features, and LSTM can effectively extract temporal features. This improvement can extract high-dimensional complex features from the operation data of the energy storage power station and has high accuracy in assessing the remaining capacity of energy storage.

[0070] 3. This invention estimates the State of Health (SOH) of an electrochemical energy storage power station based on a physical information fusion-graph convolution-long short-term memory neural network model. The SOH is then used to update the state of the energy storage units and achieve adaptive power allocation based on SOH decay, which greatly improves the service life of the energy storage power station and effectively enhances its operational performance and economic value. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0072] Figure 1 This is a flowchart of a collaborative optimization method for the planning and operation of an electrochemical energy storage power station that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0073] Figure 2This is a schematic diagram of a three-level abnormal data identification mechanism in a collaborative optimization method for the planning and operation of an electrochemical energy storage power station that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0074] Figure 3 This is a research framework for optimizing the configuration and power allocation strategies of energy storage power stations, taking into account the flexibility requirements of the distribution network, in a collaborative optimization method for the planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0075] Figure 4 This is a probability distribution diagram of the relative error of photovoltaic prediction in a collaborative optimization method for planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0076] Figure 5 This is a probability distribution diagram of the relative error of wind power prediction in a collaborative optimization method for the planning and operation of an electrochemical energy storage power station that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0077] Figure 6 This is a probability distribution diagram of the relative error of load forecast in a collaborative optimization method for planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0078] Figure 7 This is a schematic diagram of the wind-solar-load curve in a basic scenario of an electrochemical energy storage power station planning and operation collaborative optimization method that takes into account the uncertainty of new energy power generation and load demand according to an embodiment of the present invention.

[0079] Figure 8 This is a schematic diagram of the unit scheduling stacking diagram in an electrochemical energy storage power station planning and operation collaborative optimization method that takes into account the uncertainty of new energy power generation and load demand according to an embodiment of the present invention;

[0080] Figure 9 This is a schematic diagram of the basic scenario of the energy storage power station scheduling plan and the energy storage power station SOC in a collaborative optimization method for planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand according to an embodiment of the present invention.

[0081] Figure 10 This is a schematic diagram of the internal topology of an electrochemical energy storage power station in an electrochemical energy storage power station planning and operation collaborative optimization method that takes into account the uncertainty of new energy power generation and load demand according to an embodiment of the present invention.

[0082] Figure 11 This is the initial SOC result of each energy storage unit in an electrochemical energy storage power station in a collaborative optimization method for planning and operation of an electrochemical energy storage power station that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0083] Figure 12 This is the SOC result of each energy storage unit in the electrochemical energy storage power station after one operation of the proposed strategy in a collaborative optimization method for planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0084] Figure 13 This is the SOC result of each energy storage unit in the electrochemical energy storage power station after the proposed strategy has been run twice, in a collaborative optimization method for planning and operation of electrochemical energy storage power stations that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0085] Figure 14 This is the power allocation result of each energy storage unit in the electrochemical energy storage power station after one operation of the proposed strategy in an electrochemical energy storage power station planning and operation collaborative optimization method that takes into account the uncertainty of new energy power generation and load demand, according to an embodiment of the present invention.

[0086] Figure 15 This is the power allocation result of each energy storage unit in the electrochemical energy storage power station after the proposed strategy has been run twice, according to an embodiment of the present invention, which takes into account the uncertainty of new energy power generation and load demand in the planning and operation of an electrochemical energy storage power station. Detailed Implementation

[0087] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0088] According to an embodiment of the present invention, a collaborative optimization method for the planning and operation of an electrochemical energy storage power station that takes into account the uncertainty of new energy power generation and load demand is provided.

[0089] It should be noted that, as Figure 3 As shown, this invention establishes a Gaussian mixture model of the relative error of source-load prediction, then provides the capacity configuration of the energy storage power station from the scheduling level, and formulates specific energy and power allocation strategies at the energy storage power station level.

[0090] Specifically, firstly, the data on new energy power generation (source) and load demand (load) are preprocessed, specifically wind power generation data, photovoltaic power generation data, and load demand data, including error and blank data processing and normalization. Secondly, for the processed source-load data, the error value of the predicted source-load data relative to the actual value is calculated, and this relative error value is used as the input of a Gaussian mixture model (GMM), and the mixture parameters are optimized iteratively through the expected maximum value. Then, the traditional unit combination model is improved by incorporating the full life cycle cost of the electrochemical energy storage power station, adding source-load uncertainty constraints and energy storage operation constraints, and constructing an electrochemical energy storage power station configuration optimization model. Finally, an electrochemical energy storage power station health (SOH) estimation model based on a physical information fusion-graph convolution-long short-term memory (GCN-LSTM) neural network is constructed, and an adaptive power balance allocation strategy for electrochemical energy storage power stations considering the SOH decay effect is proposed to achieve reasonable power allocation of energy storage power stations.

[0091] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figures 1-2 As shown, the electrochemical energy storage power station planning and operation collaborative optimization method considering the uncertainty of new energy power generation and load demand according to an embodiment of the present invention includes the following steps:

[0092] S1. Obtain the measured values ​​of source and load data in the regional power system as actual values, and perform abnormal data processing on the measured values ​​based on the 3-sigma principle, physical rules, and spatiotemporal feature map structure to obtain standardized source and load data.

[0093] In a preferred embodiment, the measured values ​​of source-load data in the regional power system are used as actual values, and a multi-level anomaly data identification mechanism based on the 3-sigma principle, physical rules, and spatiotemporal feature map structure is used to process the measured values ​​to obtain standardized source-load data. This includes the following steps:

[0094] S11. Perform data cleaning and missing value detection on the measured values ​​of source and load data in the regional power system, and fill in the missing values ​​using the mean imputation method.

[0095] S12. Using the 3-sigma principle, perform preliminary screening of outliers on the measurement values ​​of the source payload data after data cleaning.

[0096] S13. Based on physical rules, perform a second screening process on the outlier values ​​of the source load data after the initial screening process.

[0097] As a preferred embodiment, the secondary outlier screening process for the measured values ​​of the source load data after the initial screening, based on physical rules, includes the following steps:

[0098] S131. Based on the characteristics of photovoltaic and wind power, construct physical models for photovoltaic power generation and wind power generation respectively;

[0099] S132. Based on the physical models of photovoltaic power generation and wind power generation, the theoretical output of photovoltaic power and the theoretical output of wind power are calculated by acquiring environmental meteorological data of photovoltaic power stations and wind power stations respectively.

[0100] S133. Based on the theoretical output of photovoltaic and wind power, use upper and lower limit thresholds to identify outliers in the measured values ​​of source and load data after preliminary screening and processing, and remove the identified outliers.

[0101] S14. Based on the spatiotemporal graph feature structure, construct a Laplace graph, and use the Laplace graph to perform a third outlier screening on the measurement values ​​of the source load data after the second screening process.

[0102] As a preferred embodiment, the step of constructing a Laplace graph based on the spatiotemporal graph feature structure, and performing a third outlier screening on the source load data after the second screening using the Laplace graph, includes the following steps:

[0103] S141. Based on the spatiotemporal relationship between the power grid structure and photovoltaic and wind farms in the regional power system, construct a Laplace diagram;

[0104] S142. By identifying outliers in the measured values ​​of the source load data after secondary screening, the mutation of the Laplace plot structure is used to identify outliers.

[0105] S143. Remove the identified outliers.

[0106] Specifically, load demand data generally does not fluctuate drastically, so secondary screening is unnecessary. However, for photovoltaic and wind power generation data, due to their inherent randomness and volatility, using only 3-sigma screening may inadvertently delete valid data, affecting subsequent analysis. Therefore, this invention adopts a three-level anomaly data identification mechanism for new energy power generation data, with the main steps including: preliminary screening based on the 3-sigma principle → secondary screening based on physical rules → tertiary screening based on abrupt changes in the spatiotemporal graph feature structure.

[0107] The acquired new energy power generation data and load demand data are cleaned to process missing and abnormal data, and rows or columns with many missing data are deleted.

[0108] Assuming that the renewable energy generation data and load demand data follow a Gaussian distribution, outliers are initially screened out using the 3-sigma principle. The calculation formula is as follows:

[0109]

[0110] In the formula, μ represents the mean of X, ∑ represents the standard deviation of X, and when |z|>3, it is outlier data and needs to be removed.

[0111] The physical rule-based secondary screening starts with the actual physical models of photovoltaic and wind power generation, calculates the theoretical output of photovoltaic and wind power based on environmental meteorological data, and then identifies abnormal data by superimposing certain upper and lower thresholds.

[0112] The physical model for photovoltaic power generation is as follows:

[0113]

[0114] In the formula, This represents the physical model of photovoltaic power generation, the theoretical output of the photovoltaic system, and how the output changes over time; η represents the conversion efficiency of the photovoltaic module; G sc θ represents the standard solar irradiance; θ represents the angle between sunlight and the normal to the photovoltaic panel; A represents the area of ​​the photovoltaic panel; k cloud C(t) represents the cloud layer influence coefficient; C(t) represents the cloud cover.

[0115] The physical model for wind power generation is as follows:

[0116]

[0117] In the formula, This represents a physical model of wind power generation, where ρ represents air density as a function of temperature / altitude; R represents blade radius; v represents wind speed at hub height; and C represents wind speed. p The wind energy utilization coefficient is represented by β; the blade pitch angle is represented by λ; and the tip speed ratio is represented by P. rated Indicates rated power, v cut-in It means, v cut-out It means, v rated express.

[0118] The general threshold is 10% of the theoretical output of photovoltaic / wind power at the current moment. Therefore, when the photovoltaic / wind power is between 0.9 and 1.1 times the theoretical output, the threshold is lower. This is considered valid data.

[0119] In addition, the three-level screening based on the structural mutation of spatiotemporal graph features considers the spatiotemporal relationship between the power grid structure and photovoltaic and wind power stations, constructs a Laplace graph, and identifies outliers through mutations in the graph structure.

[0120] The expression for identifying outliers in the source load data after secondary screening by analyzing mutations in the Laplace diagram's graph structure is as follows:

[0121]

[0122] In the formula, s i This indicates an outlier; a value greater than 0.5 is considered abnormal, while a value less than 0.5 is considered normal. ΔL represents the rate of change over time in the Laplace plot. H represents the set of adjacent nodes of node i. ij Let σ represent the edge features between node i and node j, and let Λ represent the activation function. τ Represents a time-dynamic filter, Θ τ The spatiotemporal graph features are represented by τ, where τ represents the index of the time dimension, and H represents the feature parameters. (l) H represents the edge feature vector of the l-th layer. (l+1) This represents the edge feature vector of the (l+1)th layer;

[0123] The formula for calculating the rate of change over time in the Laplace plot is as follows:

[0124]

[0125]

[0126] In the formula, ΔL represents the rate of change of the Laplace plot over time, and L represents the Laplace plot; ||·|| F D represents the Fibonacci norm; D represents the degree matrix; A represents the adjacency matrix; d geo (i,j) represents the geographical distance; σ s Indicates the spatial attenuation coefficient; d elec (i,j) represents the electrical distance; σ e Represents the electrical attenuation coefficient; v represents the electrical weighting factor; x represents the node's x-coordinate, y represents the node's y-coordinate; ΔV ij I represents the voltage difference between nodes i and j; ij Represents the current in line i and j; Z ij This represents the equivalent impedance of nodes i and j.

[0127] Secondly, for missing data or missing values ​​resulting from the deletion of outliers, the mean is used for imputation. The calculation formula is as follows:

[0128] x k =mean(X);

[0129] In the formula, x k This represents the k-th renewable energy generation data or load demand data, X represents the set of renewable energy generation or load demand data, and mean represents the mean value processing.

[0130] S2. Obtain the predicted values ​​corresponding to the source load data in the regional power system, and calculate the relative error between the standardized source load data and the predicted values. Based on the relative error, optimize the mixing parameters of the pre-configured Gaussian mixture model using the adaptive momentum regularization expectation maximization algorithm to obtain the source load uncertain model.

[0131] It should be noted that when obtaining the predicted values ​​of source and load data in a regional power system, firstly, historical data, real-time operational data, and key forecast information (especially refined weather forecasts) are extensively collected. Subsequently, these raw data undergo preprocessing, including cleaning (handling missing and outlier values), feature selection (constructing meaningful predictive factors, such as deriving holiday markers from dates and calculating temperature and humidity indices from meteorological data), and standardization, to prepare high-quality input for model training. Based on the processed data and the prediction objective, a source and load data prediction model based on an LSTM model is constructed. The model is trained using a training set and its parameters are fine-tuned using a validation set, ultimately outputting the predicted source and load data, such as... Figure 4-6 As shown, the relative error probability distributions of photovoltaic prediction, wind power prediction, and load prediction were calculated respectively.

[0132] In a preferred embodiment, the steps of obtaining the predicted values ​​corresponding to the source load data in the regional power system, calculating the relative error between the standardized source load data and the predicted values, and optimizing the mixing parameters of the pre-configured Gaussian mixture model based on the relative error and using an adaptive momentum regularized expectation-maximization algorithm to obtain the source load uncertainty model include the following steps:

[0133] S21. Initialize the conditions for the adaptive momentum regularized expectation maximization algorithm, including the number of sub-distributions, batch size, distribution parameters, optimizer initial momentum, variance, and number of iterations.

[0134] S22. Based on the predicted values ​​corresponding to the source load data in the regional power system, calculate the error values ​​between the standardized source load data and the predicted values ​​to obtain a dataset of relative prediction error values. Based on the batch size, divide the error value dataset to obtain several sub-batch sets.

[0135] S23. For each data point in a subset, the expectation step is used to calculate the posterior probability of each data point belonging to each Gaussian distribution.

[0136] S24. Construct the likelihood function, and update the parameters of the Gaussian mixture model and the distribution parameters of the mixture components with the goal of maximizing the likelihood function, to obtain the source load uncertainty model.

[0137] It should be noted that the GMM model is used to characterize source load uncertainty. By improving the expectation-maximization algorithm, an adaptive momentum-regularized expectation-maximization algorithm is obtained. This algorithm effectively overcomes the problems of slow convergence speed, susceptibility to local optima, and covariance matrix degradation inherent in traditional expectation-maximization algorithms. The expression for the GMM model is:

[0138]

[0139] In the formula, p(x) represents the GMM model, and π k Let the mixing coefficients (weights) of the k-th Gaussian distribution satisfy the following condition: Let μ represent the probability density function of the k-th Gaussian distribution, and let μ be the parameter representing the mean. k The covariance matrix Σ k .

[0140] Specifically, the steps of the adaptive momentum regularized expectation maximization algorithm are as follows:

[0141] Given the number of subdistributions K and the batch size B;

[0142] Initialize distribution parameters

[0143] Initialize the Adam optimizer with initial momentum m0 = 0, variance v0 = 0, and iteration count t = 0.

[0144] For each subset B t The posterior probability λ of a data point belonging to the k-th Gaussian distribution is calculated using the E (expectation step) step. ik The formula for calculating the posterior probability is:

[0145]

[0146] In the formula, λ ik This represents the posterior probability that a data point belongs to the k-th Gaussian distribution. Let π represent the probability density function of the k-th Gaussian distribution. k Let represent the mixing coefficient of the k-th Gaussian distribution, K represent the number of subdistributions, and j represent the index value of the number of subdistributions. Let represent the probability density function of the j-th Gaussian distribution;

[0147] The M-step (maximization step) updates the model parameters to maximize the likelihood function Q, expressed as:

[0148]

[0149] In the formula, Q(θ|θ) (t) ) represents the likelihood function, θ (t) Z represents the parameter value in the t-th iteration; Z represents the latent variable. Let denot represent the class classification of a sample in a Gaussian mixture model, and p(X,Z|θ) represent the joint probability distribution.

[0150] Among them, the updated weights are:

[0151]

[0152] In the formula, This represents the Q function with respect to the weight π of the k-th mixture component. k The partial derivative of B, where B represents the batch size. t Denotes a sub-batch set, λ ik Let N represent the posterior probability that a data point belongs to the k-th Gaussian distribution, where N represents the total number of data points and i represents the sub-batch B. t The i-th data.

[0153] Update the mean:

[0154]

[0155] In the formula, The Q-function represents the mean μ of the k-th mixture component. k The partial derivative of B, where B represents the batch size. t Denotes a sub-batch set, λ ik Let x represent the posterior probability that a data point belongs to the k-th Gaussian distribution. i This represents the i-th data.

[0156] Update the covariance matrix:

[0157]

[0158] In the formula, The Q-function represents the covariance Σ of the k-th mixture component. k The partial derivatives; Let T denote the inverse matrix of the covariance, and let T denote the matrix transpose.

[0159] Covariance matrix regularization:

[0160]

[0161] In the formula, Let represent the variance of the k-th subdistribution in the (t+1)-th iteration, ∈ represent the regularization parameter, and I represent the identity matrix.

[0162] Update the distribution parameters of the mixed components:

[0163]

[0164] In the formula, η t Indicates the adaptive learning rate; m t The gradient first moment estimate is represented by β; the momentum decay coefficient is represented by η; the basic learning rate is represented by v. t θ represents the second moment estimate of the gradient; γ represents the variance decay coefficient; ε represents a small constant to prevent division by zero; θ (t+1) Let θ represent the updated distribution parameters. (t) This represents the distribution parameters before the update. This represents the gradient of the objective function Q.

[0165] Furthermore, since the GMM model requires the number of Gaussian distributions to be manually specified, it is highly subjective. This claim aims to optimize the GMM solution method by nesting an optimization layer outside the E and M steps to obtain the optimal number of Gaussian distributions and the parameters of each distribution. A more practical solution method is proposed: a GMM model with 2 to 8 Gaussian distributions can perfectly fit most application scenarios. Therefore, a loop layer is nested outside the adaptive momentum regularized expectation maximization algorithm steps, iterating through K = 2 to 8.

[0166] S3. Construct a full life cycle cost model for electrochemical energy storage power stations based on the unit combination framework, and combine it with the source-load uncertainty model to construct an electrochemical energy storage power station configuration optimization model;

[0167] As a preferred embodiment, the construction of a full life-cycle cost model for an electrochemical energy storage power station based on a unit combination framework, and the construction of an electrochemical energy storage power station configuration optimization model in conjunction with a source-load uncertainty model, includes the following steps:

[0168] S31. Based on the charging and discharging data of the electrochemical energy storage power station, estimate the equivalent operating life of the electrochemical energy storage power station;

[0169] S32. Based on the equivalent operating years of the electrochemical energy storage power station and in combination with the impact of rated capacity on the total life cycle cost of the electrochemical energy storage power station, calculate the investment cost and annual operation and maintenance cost of the electrochemical energy storage power station.

[0170] S33. Based on the investment cost and annual operation and maintenance cost of electrochemical energy storage power stations, construct a full life cycle cost model for electrochemical energy storage power stations;

[0171] S34. Transform the source-load uncertainty model into a power system flexibility constraint, and calculate the boundary range of the source-load uncertainty model at the significance level;

[0172] S35. Construct an optimization model for the configuration of electrochemical energy storage power stations based on power system flexibility constraints and the full life cycle cost model of electrochemical energy storage power stations.

[0173] As a preferred embodiment, the construction of the electrochemical energy storage power station configuration optimization model based on power system flexibility constraints and the full life cycle cost model of the electrochemical energy storage power station includes the following steps:

[0174] S351. Based on the full life cycle cost model of electrochemical energy storage power stations, determine the objective function of the configuration optimization model for electrochemical energy storage power stations;

[0175] S352. Based on the constraints of the power system unit combination model, and combined with the power system flexibility constraints and the operation constraints of the electrochemical energy storage power station, determine the constraints of the configuration optimization model of the electrochemical energy storage power station; the constraints of the power system unit combination model include unit output constraints, unit ramping constraints, unit minimum start-up and shutdown time constraints, and power balance constraints.

[0176] It should be noted that the total lifecycle cost of an energy storage power station is divided into one-time investment cost and annual operation and maintenance cost. The lifecycle cost model for an energy storage power station is constructed as follows:

[0177]

[0178] In the formula, This represents the daily investment and operation / maintenance costs of an energy storage power station ($); C c (r,T life ) represents the capital recovery factor; r represents the annual interest rate; E N This indicates the rated capacity (MW·h) of the energy storage power station; C E Indicates the investment cost of an energy storage power station ($); C OM This represents the unit operating and maintenance cost ($ / MW·h).

[0179] To more precisely measure the impact of rated power on the cost of electrochemical energy storage power stations, this invention adds the impact of rated capacity on the total lifecycle cost of electrochemical energy storage power stations:

[0180]

[0181] In the formula, C c (r,T life α represents the capital recovery factor. s This represents the unit price per MW of energy storage ($ / MW); β s This represents the unit price per MW·h of energy storage capacity ($); P N This indicates the rated charge / discharge power (MW) of the energy storage power station; T life C represents the equivalent operating life of an electrochemical energy storage power station. E This represents the investment cost ($) of an energy storage power station.

[0182] The formula for estimating the equivalent operating life of an electrochemical energy storage power station is as follows:

[0183]

[0184] In the formula, T life N represents the equivalent operating life of an electrochemical energy storage power station. B Indicates the number of charge / discharge cycles in a year; L CYC,iNi represents the number of cycles of the energy storage power station at the i-th charge-discharge depth; N0 represents the number of cycles when the energy storage is charged and discharged at 100% discharge depth; Di represents the number of cycles of the energy storage power station at the i-th charge-discharge depth. i Indicates depth of charge / discharge (%), k p This represents the parameters obtained from the fitting.

[0185] Furthermore, this invention transforms the source-load uncertainty model, converting uncertain source-load fluctuations into deterministic power system flexibility constraints. The expression for the power system flexibility constraints is as follows:

[0186]

[0187] In the formula, and These represent the upper and lower reserve power provided by conventional (traditional) thermal power unit j during time period t, respectively. and These represent the upper and lower reserve power provided by the electrochemical energy storage power station during time period t; Let represent the predicted power of the r-th uncertain source in time period t; Indicates the first One new energy generating unit and one system load; Let represent the lower and upper boundaries of the r-th uncertain source / load at significance level α, and J represent the total amount of conventional (traditional) thermal power units.

[0188] For the lower boundary of uncertain sources / loads involved in the power system flexibility constraints at a significance level α. and upper boundary The calculation method is as follows:

[0189]

[0190] In the formula, p r (x) represents the probability distribution of the r-th uncertain source / charge; α represents the given significance level; Let represent the confidence interval of the r-th uncertainty source / charge at the significance level α.

[0191] Specifically, the configuration optimization model for electrochemical energy storage power stations includes two parts: an objective function and constraints. The objective function adds a daily discounted cost item for the total life cycle cost of the electrochemical energy storage power station.

[0192] The objective function expression for the electrochemical energy storage power station configuration optimization model is as follows:

[0193]

[0194] In the formula, J represents the number of conventional thermal power units; T represents the total number of time periods; This represents the start-up cost of a conventional thermal power unit j; This represents the power generation cost of a conventional thermal power unit (j). This indicates the output of a conventional thermal power unit j during time period t; u j,t This represents the on / off state of a conventional thermal power unit j during time period t, a 0-1 variable;

[0195] In addition, this invention adds power system flexibility reserve constraints and electrochemical energy storage power station operation constraints to the original power system unit combination model constraints, such as unit output constraints, unit ramping constraints, unit minimum start-up and shutdown time constraints, and power balance constraints.

[0196] The unit output constraints are as follows:

[0197]

[0198] In the formula, This indicates the maximum output of a conventional thermal power unit (j). This represents the minimum output of a conventional thermal power unit j. This represents the output of a conventional thermal power unit j during time period t.

[0199] The unit's ramp-up constraints are as follows:

[0200]

[0201] In the formula, This represents the output of a conventional thermal power unit j during time period t. RU represents the output of conventional thermal power unit j in time period t-1. j RD represents the upward ramp limit of unit j per unit time; j This represents the downhill ramp limit for unit j per unit time.

[0202] The minimum start-up and shutdown time constraints for the unit are as follows:

[0203]

[0204] In the formula, This represents the minimum operating time and minimum downtime of unit j.

[0205] The power balance constraints are as follows:

[0206]

[0207] In the formula, This represents the output of a conventional thermal power unit j during time period t. This indicates the discharge and charging states of the energy storage power station during time period t. This represents the discharge power and charging power of the energy storage power station during time period t; This represents the load power during time period t; R represents the output of the i-th renewable energy unit during time period t; R represents the total number of renewable energy units.

[0208] The operational constraints for newly added electrochemical energy storage power stations are as follows:

[0209]

[0210] In the formula, SOC t This represents the State of Charge (SOC) of the energy storage power station during time period t; SOC init SOC T This represents the State of Charge (SOC) of the energy storage power station during the initial period and the Tth period; SOC min SOC max Indicates the lower and upper limits of the State of Charge (SOC) of an energy storage power station; η c η d The charging and discharging efficiency of the energy storage power station is represented by Δt; the charging and discharging time is represented by Δt. This indicates the maximum discharge power and maximum charging power of the energy storage power station; This indicates the discharge and charging states of the energy storage power station during time period t.

[0211] S4. Construct a health estimation model for electrochemical energy storage power stations based on physical information fusion, graph convolution, and long short-term memory neural networks. Based on the results of the configuration optimization model for electrochemical energy storage power stations, combine the power equalization allocation strategy of electrochemical energy storage power stations with the health decay effect to realize the power allocation of energy storage power stations.

[0212] As a preferred embodiment, the construction of an electrochemical energy storage power station health estimation model based on physical information fusion-graph convolution-long short-term memory neural network, and the realization of power allocation of the energy storage power station based on the results of the electrochemical energy storage power station configuration optimization model, combined with the power equalization allocation strategy of the electrochemical energy storage power station based on the health decay effect, includes the following steps:

[0213] S41. Construct an aging dynamic model for energy storage batteries, and combine it with a physical information encoder, graph convolutional network, and long short-term neural network to construct a power station health estimation model.

[0214] As a preferred embodiment, the construction of an energy storage battery aging dynamics model, combined with a physical information encoder, graph convolutional network, and long short-term neural network, to construct a power plant health estimation model includes the following steps:

[0215] S411. Based on the aging dynamics model of the energy storage battery, obtain the mechanism information of the energy storage battery, and use a physical information encoder to embed the mechanism information of the energy storage battery into a neural network to obtain physical coding features.

[0216] S412. Based on physical coding features, the electrical and thermal coupling relationships inside the electrochemical energy storage power station are obtained through a graph convolutional neural network, resulting in the graph convolution feature matrix.

[0217] S413. Input the feature matrix of the graph convolution into the long short-term neural network, and use the output of the long short-term neural network as the input of the fully connected layer to obtain the power plant health estimation model of the chemical energy storage power plant.

[0218] S42. Estimate the energy storage health of the electrochemical energy storage power station based on the power station health estimation model;

[0219] S43. For energy storage units within an electrochemical energy storage power station, update the actual capacity of each energy storage unit using the energy storage health status of the electrochemical energy storage power station.

[0220] S44. Obtain the dispatch instructions issued by the dispatching agency, and allocate the power of the electrochemical energy storage power station in conjunction with the configuration optimization model of the electrochemical energy storage power station.

[0221] In a preferred embodiment, the step of obtaining the dispatching instructions issued by the dispatching agency and allocating the power of the electrochemical energy storage power station in conjunction with the configuration optimization model of the electrochemical energy storage power station includes the following steps:

[0222] S441. Obtain the dispatch instructions issued by the dispatching agency and determine the status of the dispatch instructions. Based on the results of the electrochemical energy storage power station configuration optimization model, determine the working mode that the electrochemical energy storage power station needs to adjust.

[0223] S442. Determine the current operating mode of the electrochemical energy storage power station, estimate the available power based on the state of each energy storage unit, and calculate the total available power of the electrochemical energy storage power station based on the available power of each energy storage unit.

[0224] S443. Based on the dispatch instructions of the electrochemical energy storage power station and the total available power of the electrochemical energy storage power station, calculate the adaptive power allocation factor of the energy storage unit and allocate the power of each energy storage unit.

[0225] It should be noted that, as Figure 7-15 As shown, a graph convolutional memory neural network based on physical information fusion is used to estimate the state of energy (SOH) of an energy storage power station. This method integrates electrochemical mechanisms, electrical response characteristics, and artificial intelligence technologies, overcoming the limitations of traditional data-driven models. By embedding physical knowledge into the neural network structure, it achieves collaborative modeling of "mechanism-data".

[0226] The aging kinetic model expression for energy storage batteries is as follows:

[0227]

[0228] In the formula, Si V represents the state vector of the i-th energy storage battery; i T represents the voltage of the i-th energy storage battery; i Z represents the temperature of the i-th energy storage battery; i F represents the impedance of the i-th energy storage battery; i This represents the internal stress of the i-th energy storage battery; Represents the lattice phase angle of the i-th energy storage battery; I represents the operating current; E a Indicates the cell activation energy; k c β represents the lithium-ion diffusion coefficient; A represents the Arrhenius pre-exponential factor; R represents the gas constant; and c represents the lithium-ion concentration. This represents noise, which follows a normal distribution.

[0229] Then, a physical information encoder is used to embed the energy storage battery mechanism information into a neural network, as follows:

[0230]

[0231] In the formula, W represents the physical coding characteristics of the i-th energy storage battery; p b represents the weight matrix; p Γ(S) represents the bias vector; i V represents the feature extraction operator for the i-th energy storage battery. i Let represent the voltage of the i-th energy storage battery; σ represent the activation function; t represent time; and T represent the vector transpose.

[0232] Next, a graph convolutional neural network is used to further refine the electrical and thermal coupling relationships within the electrochemical energy storage power station, as follows:

[0233]

[0234] In the formula, H G (l+1) D represents the feature matrix of the (l+1)th layer; D represents the degree matrix; A represents the adjacency matrix; A ij This represents the coupling strength between the i-th and j-th energy storage batteries; Indicates the convolution weights of the l-th layer graph; σ represents the voltage gradient between the i-th and j-th energy storage cells; V Represents the voltage attenuation coefficient; ||r i -r j ||2 represents the spatial distance between the i-th and j-th energy storage batteries; σ T κ represents the temperature decay coefficient; κ represents the thermal coupling strength constant.

[0235] Subsequently, the feature matrix H of the graph convolution is... GThe model is as follows, used as input to the LSTM network:

[0236]

[0237] c t =f t c t-1 +i t σ(W cx H Gt +W ch h t-1 +b c );

[0238] h t =o t σ(c t );

[0239] In the formula, W fx W ix W ox W fh W ih W oh W cx W ch These represent the weight matrices corresponding to the inputs; b f b i b o b c Both are bias vectors; f t h t c t Indicates the output of the LSTM gate unit and the cell state, i t Let h represent the input gate vector at time step t, σ represent the activation function, and h represent the input gate vector at time step t. t-1 This represents the hidden layer state vector at time step t-1.

[0240] Finally, the output of the LSTM is used as the input to the fully connected layer to ultimately output the health model of the electrochemical energy storage power station. The model is as follows:

[0241] SOH=σ(W fc h t +b i );

[0242] In the formula, SOH represents the energy storage health value output by the model, and h t W represents the final output of the LSTM layer. fc σ represents the weights of the fully connected layer, and σ represents the activation function.

[0243] In addition, for each energy storage unit within the energy storage power station, the State of Health (SOH) is estimated using the health model of the electrochemical energy storage power station, and the actual capacity of each energy storage unit is updated. The formula for calculating the actual capacity of each energy storage unit is as follows:

[0244] E′ i =E i *SOH i ;

[0245] In the formula, E′ i E represents the updated actual capacity of the i-th energy storage unit. i SOH represents the initial capacity of the i-th energy storage unit. i The SOH of the i-th energy storage unit in the health model of an electrochemical energy storage power station is represented.

[0246] According to the dispatching instruction P issued by the dispatching agency Dispatch The positive and negative values ​​adjust the charging / discharging operating mode of the energy storage power station. When P Dispatch >0 indicates that the energy storage power station is operating in discharge mode, and vice versa.

[0247] When the electrochemical energy storage power station operates in discharge mode, the available discharge power is estimated based on the state of each energy storage unit, and the calculation formula is as follows:

[0248]

[0249] In the formula, P i,ava SOC represents the available power of the i-th energy storage unit. i This represents the SOC (State of Charge) of the i-th energy storage unit. i,min η represents the lower limit of the SOC of the i-th energy storage unit. d P represents the discharge efficiency of the energy storage unit, Δt represents the time interval, and P represents the discharge efficiency of the energy storage unit. i,max This represents the rated discharge power of the i-th energy storage unit.

[0250] When the electrochemical energy storage power station is operating in charging mode, the available charging power is estimated based on the state of each energy storage unit, and the calculation formula is as follows:

[0251]

[0252] In the formula, P i,ava SOC represents the available power of the i-th energy storage unit. i,amx η represents the upper limit of the SOC of the i-th energy storage unit. c P represents the discharge efficiency of the energy storage unit. i,min This represents the rated charging power of the i-th energy storage unit.

[0253] Based on the dispatch instructions and the total available power of the electrochemical energy storage power station, the adaptive power allocation factor of the energy storage unit is calculated and the power of each energy storage unit is allocated. The calculation formula is as follows:

[0254]

[0255] In the formula, F represents the adaptive power allocation factor of the electrochemical energy storage power station, and P i P represents the power allocated to the i-th energy storage unit. Dispatch Indicates a scheduling instruction, P i,ava This represents the available power of the i-th energy storage unit.

[0256] As shown in Tables 1, 2 and 3, these are the index parameters of the energy storage power station under sequential scheduling (Strategy 1: allocating charging and discharging tasks according to the physical number of the battery packs), dynamic priority scheduling (Strategy 2: dynamically allocating the power of each battery pack according to the real-time state of charge (SOC) of the battery, and the strategy proposed in this scheme. These parameters include the state of charge (SOH) after 3 years of operation, the number of cycles after 3 years of operation, and the time of the first battery swap.

[0257] Table 1 SOH after 3 years of operation

[0258]

[0259] Table 2 Number of cycles after 3 years of operation

[0260]

[0261]

[0262] Table 3 First Battery Swap Time

[0263]

[0264] In summary, by utilizing the above-mentioned technical solutions of this invention, a three-level anomaly data identification mechanism is employed to accurately screen abnormal data in new energy power generation, maximizing the consistency between the data distribution and the actual distribution. Furthermore, an adaptive momentum regularized expectation-maximization algorithm is proposed to overcome the slow convergence speed, susceptibility to local optima, and covariance matrix degradation problems inherent in traditional expectation-maximization algorithms. A GMM model of new energy power generation and load demand is established, and by setting confidence levels, the probability distribution of source and load is transformed into deterministic power system flexibility requirements, providing a prerequisite for the subsequent optimized configuration of electrochemical energy storage power stations and ensuring the rationality of energy storage power station planning. This invention employs a health estimation model for electrochemical energy storage power stations based on physical information fusion-graph convolution-long short-term memory neural networks for SOH estimation. GCN effectively extracts spatial features, and LSTM effectively extracts temporal features. This improvement can extract high-dimensional complex features from energy storage power station operating data, achieving high accuracy in assessing remaining energy storage capacity. This invention estimates the State of Health (SOH) of an electrochemical energy storage power station based on a physical information fusion-graph convolution-long short-term memory neural network model. The SOH is then used to update the state of the energy storage units and achieve adaptive power allocation based on SOH decay, which greatly improves the service life of the energy storage power station and effectively enhances its operational performance and economic value.

[0265] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0266] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A collaborative optimization method for the planning and operation of electrochemical energy storage power stations, taking into account the uncertainty of new energy power generation and load demand, characterized in that, Includes the following steps: S1. Obtain the measured values ​​of source and load data in the regional power system as actual values, and perform abnormal data processing on the measured values ​​based on the 3-sigma principle, physical rules, and spatiotemporal feature map structure to obtain standardized source and load data. S2. Obtain the predicted values ​​corresponding to the source load data in the regional power system, and calculate the relative error between the standardized source load data and the predicted values. Based on the relative error, optimize the mixing parameters of the pre-configured Gaussian mixture model using the adaptive momentum regularization expectation maximization algorithm to obtain the source load uncertain model. S3. Construct a full life cycle cost model for electrochemical energy storage power stations based on the unit combination framework, and combine it with the source-load uncertainty model to construct an electrochemical energy storage power station configuration optimization model; S4. Construct a health estimation model for electrochemical energy storage power stations based on physical information fusion, graph convolution, and long short-term memory neural networks. Based on the results of the configuration optimization model for electrochemical energy storage power stations, combine the power equalization allocation strategy of electrochemical energy storage power stations with the health decay effect to realize the power allocation of energy storage power stations.

2. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 1, is characterized in that... The process of acquiring measured values ​​of source-load data from the regional power system as actual values, and performing anomaly data processing on the measured values ​​based on the 3-sigma principle, physical rules, and a multi-level anomaly data identification mechanism using spatiotemporal feature map structures to obtain standardized source-load data includes the following steps: S11. Perform data cleaning and missing value detection on the measured values ​​of source and load data in the regional power system, and fill in the missing values ​​using the mean imputation method. S12. Using the 3-sigma principle, perform preliminary screening of outliers on the measurement values ​​of the source payload data after data cleaning. S13. Based on physical rules, perform a second screening process on the outlier values ​​of the source load data after the initial screening process. S14. Based on the spatiotemporal graph feature structure, construct a Laplace graph, and use the Laplace graph to perform outlier screening on the measured values ​​of the source load data after secondary screening to obtain standardized source load data.

3. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 2, is characterized in that... The secondary outlier screening process for the source load data measurements after the initial screening, based on physical rules, includes the following steps: S131. Based on the characteristics of photovoltaic and wind power, construct physical models for photovoltaic power generation and wind power generation respectively; S132. Based on the physical models of photovoltaic power generation and wind power generation, the theoretical output of photovoltaic power and the theoretical output of wind power are calculated by acquiring environmental meteorological data of photovoltaic power stations and wind power stations respectively. S133. Based on the theoretical output of photovoltaic and wind power, use upper and lower limit thresholds to identify outliers in the measured values ​​of source and load data after preliminary screening and processing, and remove the identified outliers.

4. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 2, is characterized in that... The process of constructing a Laplace graph based on the spatiotemporal graph feature structure, and then using the Laplace graph to perform a third outlier screening on the source load data after the second screening, includes the following steps: S141. Based on the spatiotemporal relationship between the power grid structure and photovoltaic and wind farms in the regional power system, construct a Laplace diagram; S142. By identifying outliers in the measured values ​​of the source load data after secondary screening, the mutation of the Laplace plot structure is used to identify outliers. The expression for identifying outliers in the source load data after secondary screening by analyzing mutations in the Laplace diagram's graph structure is as follows: In the formula, s i This represents outliers, and ΔL represents the rate of change of the Laplace plot over time. H represents the set of adjacent nodes of node i. ij Let σ represent the edge features between node i and node j, and let Λ represent the activation function. τ Represents a time-dynamic filter, Θ τ The spatiotemporal graph feature parameters are represented by τ, where τ represents the index of the time dimension, and H represents the index of the time dimension. (l) H represents the edge feature vector of the l-th layer. (l+1) This represents the edge feature vector of the (l+1)th layer; S143. Remove the identified outliers.

5. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 1, is characterized in that... The process of obtaining the predicted values ​​corresponding to the source load data in the regional power system, calculating the relative error between the standardized source load data and the predicted values, and optimizing the mixing parameters of the pre-configured Gaussian mixture model based on the relative error and the adaptive momentum regularized expectation maximization algorithm to obtain the source load uncertainty model includes the following steps: S21. Initialize the conditions for the adaptive momentum regularized expectation maximization algorithm, including the number of sub-distributions, batch size, distribution parameters, optimizer initial momentum, variance, and number of iterations. S22. Based on the predicted values ​​corresponding to the source load data in the regional power system, calculate the error values ​​between the standardized source load data and the predicted values ​​to obtain a dataset of relative prediction error values. Based on the batch size, divide the error value dataset to obtain several sub-batch sets. S23. For each data point in a subset, the expectation step is used to calculate the posterior probability of each data point belonging to each Gaussian distribution. The formula for calculating the posterior probability is: In the formula, λ ik This represents the posterior probability that a data point belongs to the k-th Gaussian distribution. Let π represent the probability density function of the k-th Gaussian distribution. k Let represent the mixing coefficient of the k-th Gaussian distribution, K represent the number of subdistributions, and j represent the index value of the number of subdistributions. Let represent the probability density function of the j-th Gaussian distribution; S24. Construct the likelihood function, and update the parameters of the Gaussian mixture model and the distribution parameters of the mixture components with the goal of maximizing the likelihood function, to obtain the source load uncertainty model. The expression for the likelihood function is: In the formula, Q(θ|θ) (t) ) represents the likelihood function, θ (t) Z represents the parameter value in the t-th iteration; Z represents the latent variable. Let denot represent the class classification of a sample in a Gaussian mixture model, and p(X,Z|θ) represent the joint probability distribution.

6. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 1, is characterized in that... The steps for constructing a full life-cycle cost model for electrochemical energy storage power stations based on a unit combination framework, and for building an electrochemical energy storage power station configuration optimization model by combining a source-load uncertainty model, include the following: S31. Based on the charging and discharging data of the electrochemical energy storage power station, estimate the equivalent operating life of the electrochemical energy storage power station; S32. Based on the equivalent operating years of the electrochemical energy storage power station and in combination with the impact of rated capacity on the total life cycle cost of the electrochemical energy storage power station, calculate the investment cost and annual operation and maintenance cost of the electrochemical energy storage power station. S33. Based on the investment cost and annual operation and maintenance cost of electrochemical energy storage power stations, construct a full life cycle cost model for electrochemical energy storage power stations; S34. Transform the source-load uncertainty model into a power system flexibility constraint, and calculate the boundary range of the source-load uncertainty model at the significance level; S35. Construct an optimization model for the configuration of electrochemical energy storage power stations based on power system flexibility constraints and the full life cycle cost model of electrochemical energy storage power stations.

7. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 6, is characterized in that... The construction of the electrochemical energy storage power station configuration optimization model based on power system flexibility constraints and the full life cycle cost model of electrochemical energy storage power stations includes the following steps: S351. Based on the full life cycle cost model of electrochemical energy storage power stations, determine the objective function of the configuration optimization model for electrochemical energy storage power stations; The objective function expression of the electrochemical energy storage power station configuration optimization model is: In the formula, J represents the number of conventional thermal power units; T represents the total number of time periods; This represents the start-up cost of a conventional thermal power unit j; This represents the power generation cost of a conventional thermal power unit (j). This indicates the output of a conventional thermal power unit j during time period t; u j,t This indicates the on / off status of a conventional thermal power unit j during time period t; S352. Based on the constraints of the power system unit combination model, and combined with the power system flexibility constraints and the operation constraints of the electrochemical energy storage power station, determine the constraints of the configuration optimization model of the electrochemical energy storage power station; the constraints of the power system unit combination model include unit output constraints, unit ramping constraints, unit minimum start-up and shutdown time constraints, and power balance constraints.

8. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 1, is characterized in that... The construction of a health estimation model for electrochemical energy storage power stations based on physical information fusion, graph convolution, and long short-term memory neural networks, and the implementation of power allocation for energy storage power stations based on the results of the configuration optimization model and the power balancing allocation strategy for electrochemical energy storage power stations considering health decay effects, includes the following steps: S41. Construct an aging dynamic model for energy storage batteries, and combine it with a physical information encoder, graph convolutional network, and long short-term neural network to construct a power station health estimation model. S42. Based on the power station health estimation model, estimate the energy storage health of the electrochemical energy storage power station; S43. For energy storage units within an electrochemical energy storage power station, update the actual capacity of each energy storage unit using the energy storage health status of the electrochemical energy storage power station. S44. Obtain the dispatch instructions issued by the dispatching agency, and allocate the power of the electrochemical energy storage power station in conjunction with the configuration optimization model of the electrochemical energy storage power station.

9. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 8, is characterized in that... The construction of an aging dynamics model for energy storage batteries, combined with a physical information encoder, graph convolutional network, and long short-term neural network, to build a power plant health estimation model includes the following steps: S411. Based on the aging dynamics model of the energy storage battery, obtain the mechanism information of the energy storage battery, and use a physical information encoder to embed the mechanism information of the energy storage battery into a neural network to obtain physical coding features. S412. Based on physical coding features, the electrical and thermal coupling relationships inside the electrochemical energy storage power station are obtained through graph convolutional neural networks, resulting in the graph convolution feature matrix. S413. Input the feature matrix of the graph convolution into the long short-term neural network, and use the output of the long short-term neural network as the input of the fully connected layer to obtain the power plant health estimation model of the chemical energy storage power plant.

10. The method for coordinated optimization of planning and operation of electrochemical energy storage power stations considering the uncertainty of new energy power generation and load demand, as described in claim 8, is characterized in that... The process of obtaining dispatch instructions from the dispatching agency and allocating power to the electrochemical energy storage power station in conjunction with the electrochemical energy storage power station configuration optimization model includes the following steps: S441. Obtain the dispatch instructions issued by the dispatching agency and determine the status of the dispatch instructions. Based on the results of the electrochemical energy storage power station configuration optimization model, determine the working mode that the electrochemical energy storage power station needs to adjust. S442. Determine the current operating mode of the electrochemical energy storage power station, estimate the available power based on the state of each energy storage unit, and calculate the total available power of the electrochemical energy storage power station based on the available power of each energy storage unit. S443. Based on the dispatch instructions of the electrochemical energy storage power station and the total available power of the electrochemical energy storage power station, calculate the adaptive power allocation factor of the energy storage unit and allocate the power of each energy storage unit. The formula for calculating the adaptive power allocation factor of the energy storage unit and allocating the power of each energy storage unit is as follows: In the formula, F represents the adaptive power allocation factor of the electrochemical energy storage power station, and P i P represents the power allocated to the i-th energy storage unit. Dispatch Indicates a scheduling instruction, P i,ava This represents the available power of the i-th energy storage unit.

Images