User side energy storage charging and discharging strategy based on electric power spot price prediction model

Abstract

The invention provides a user side energy storage charging and discharging strategy based on an electric power spot price prediction model, and belongs to the technical field of electric power system automation, and the strategy comprises the steps: collecting and preprocessing electricity price sequence data and related market feature data, guaranteeing the data integrity and consistency, and carrying out the data cleaning and standardization processing; constructing an inner-layer framework of the double-layer prediction model, constructing a multi-model prediction algorithm, and realizing preferential prediction; an outer layer frame of the double-layer prediction model is constructed, an outer layer algorithm is calibration window integrated prediction based on Bayesian model averaging, and an interval prediction result of the electric power spot price is obtained; constructing a deterministic optimization scheduling problem model based on complete information; a two-stage minimum and maximum regret value optimization scheduling problem based on uncertainty factors is constructed, a system scheduling objective function and constraint conditions are included, and an efficient energy storage charging and discharging plan is formulated, so that the income is improved, and the reliability of the system is enhanced at the same time.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automation, and particularly to a charging and discharging strategy for user-side energy storage based on a power spot price prediction model. Background Art

[0002] With the gradual maturity and popularization of the power spot market, the power trading mechanism has become more flexible and market-oriented. The power spot market reflects the changes in the supply-demand relationship through real-time electricity prices, promoting the optimal allocation of power resources. However, the power spot price has high volatility and uncertainty, which brings new challenges to the charging and discharging strategies of user-side energy storage systems. User-side energy storage systems play an increasingly important role in the power system, being able to effectively balance power supply and demand, relieve the grid pressure, and improve energy utilization efficiency. However, to fully utilize the advantages of energy storage systems, the key lies in formulating reasonable charging and discharging strategies, enabling them to charge when the electricity price is low and discharge when the electricity price is high, so as to maximize economic benefits. The background of this research is to address the challenges brought by the fluctuations of power spot prices, and propose a charging and discharging strategy for user-side energy storage based on a power spot price prediction model. This strategy collects and preprocesses electricity price sequence data and related market characteristic data, constructs a two-layer prediction model to achieve accurate prediction of power spot prices, and combines deterministic and uncertainty optimal scheduling models to formulate an efficient energy storage charging and discharging plan, thereby reducing the electricity cost while improving the reliability of the system. Summary of the Invention

[0003] The main purpose of the embodiments of the present disclosure is to propose a charging and discharging strategy for user-side energy storage based on a power spot price prediction model. This strategy collects and preprocesses electricity price sequence data and related market characteristic data, constructs a two-layer prediction model to achieve accurate prediction of power spot prices, and combines deterministic and uncertainty optimal scheduling models to formulate an efficient energy storage charging and discharging plan, thereby reducing the electricity cost while improving the reliability of the system.

[0004] To achieve the above purpose, the embodiments of the present disclosure propose a charging and discharging strategy for user-side energy storage based on a power spot price prediction model, which is characterized in that it is applied to a user-side energy storage system, and the strategy includes:

[0005] Collect and preprocess electricity price and market characteristic data;

[0006] Construct an inner-layer prediction framework based on variational mode decomposition;

[0007] Construct an outer-layer prediction framework based on Bayesian model averaging for calibration window ensemble prediction;

[0008] Construct a deterministic user-side energy storage charging and discharging strategy model under complete information;

[0009] Construct a two-stage minimax regret optimization scheduling problem based on uncertainty factors;

[0010] According to the collected and preprocessed electricity price and market characteristic data, the constructed inner-layer prediction framework based on variational mode decomposition, the constructed outer-layer prediction framework of calibration window ensemble prediction based on Bayesian model averaging, the constructed deterministic user-side energy storage charge and discharge strategy model under complete information, and the constructed two-stage minimax regret optimization scheduling problem based on uncertainty factors, construct a user-side energy storage charge and discharge strategy based on an electricity spot price prediction model.

[0011] In some embodiments, the collecting and preprocessing of electricity price and market characteristic data includes:

[0012] Collect electricity price sequence data and related market characteristic data to ensure the integrity and consistency of the data;

[0013] Data cleaning and normalization processing, filling in missing values, and eliminating outliers;

[0014] According to the collected electricity price sequence data and related market characteristic data, and the data cleaning and normalization processing, filling in missing values and eliminating outliers, collect and preprocess electricity price and market characteristic data.

[0015] In some embodiments, the constructing of the inner-layer prediction framework based on variational mode decomposition includes:

[0016] Use the variational mode decomposition method to decompose the electricity price sequence into multiple subsequences;

[0017] Construct a multi-model prediction method;

[0018] Perform model selection and integration;

[0019] According to the use of the variational mode decomposition method to decompose the electricity price sequence into multiple subsequences, the constructed multi-model prediction method, and the performing of model selection and integration, construct an inner-layer prediction framework based on variational mode decomposition.

[0020] In some embodiments, the constructing of the multi-model prediction method includes:

[0021] Construct a time series interval prediction model based on the fusion of quantile regression convolutional bidirectional long short-term memory neural network and attention mechanism;

[0022] Construct an interval prediction algorithm based on XGBoost;

[0023] Construct a multi - model prediction method based on the constructed time - series interval prediction model that fuses the attention mechanism with a quantile regression convolutional bidirectional long - short - term memory neural network and the constructed interval prediction algorithm based on XGBoost.

[0024] In some embodiments, the construction of the outer - layer prediction framework for calibration window ensemble prediction based on Bayesian model averaging includes:

[0025] Set a dynamic calibration window;

[0026] Train the inner - layer prediction model;

[0027] Perform Bayesian model averaging optimization;

[0028] Perform error correction and multi - model integration on the model;

[0029] Construct the outer - layer prediction framework for calibration window ensemble prediction based on Bayesian model averaging according to the setting of the dynamic calibration window, the training of the inner - layer prediction model, the performance of Bayesian model averaging optimization, and the performance of error correction and multi - model integration on the model.

[0030] In some embodiments, the construction of the deterministic optimization scheduling problem model based on complete information includes:

[0031] Construct the objective function of the deterministic optimization scheduling problem based on complete information;

[0032] Construct the constraint conditions of the deterministic optimization scheduling problem based on complete information;

[0033] Construct the deterministic optimization scheduling problem model based on complete information according to the constructed objective function of the deterministic optimization scheduling problem based on complete information and the constructed constraint conditions of the deterministic optimization scheduling problem based on complete information.

[0034] In some embodiments, the construction of the objective function of the deterministic optimization scheduling problem based on complete information includes:

[0035] Construct an energy storage system operation aging model;

[0036] Construct a cost model for participating in the electricity market transaction;

[0037] Construct the objective function of the deterministic optimization scheduling problem based on complete information according to the constructed energy storage system operation aging model and the constructed cost model for participating in the electricity market transaction.

[0038] In some embodiments, the construction of the constraint conditions of the deterministic optimization scheduling problem based on complete information includes:

[0039] Construct a constraint condition model for the operation of the energy storage system;

[0040] Construct a model for the constraints of participating in the electricity market transaction;

[0041] Construct a model for the system power balance constraint;

[0042] According to the constructed operation constraint model of the energy storage system, the constructed model for the constraints of participating in the electricity market transaction, and the constructed model for the system power balance constraint, construct an objective function for the deterministic optimization scheduling problem under complete information.

[0043] In some embodiments, the construction of the two-stage min-max regret optimization scheduling problem based on uncertainty factors includes:

[0044] Construct an objective function for the two-stage min-max regret optimization scheduling problem;

[0045] Construct constraints for the two-stage min-max regret optimization scheduling problem;

[0046] According to the constructed objective function for the two-stage min-max regret optimization scheduling problem and the constructed constraints for the two-stage min-max regret optimization scheduling problem, construct a two-stage min-max regret optimization scheduling problem based on uncertainty factors. Brief Description of the Drawings

[0047] Figure 1 It is a flowchart of the charging and discharging strategy of the user-side energy storage based on the electricity spot price prediction model provided by the embodiments of the present disclosure. Detailed Description of the Embodiments

[0048] In order to make the objectives, technical solutions and advantages of the present invention clearer, the following further describes the present application in detail with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] With the gradual maturity and popularization of the electricity spot market, the electricity trading mechanism has become more flexible and market-oriented. The electricity spot market reflects the changes in the supply and demand relationship through real-time electricity prices, promoting the optimal allocation of electricity resources. However, the electricity spot price has high volatility and uncertainty, which brings new challenges to the charging and discharging strategies of the user-side energy storage system. The user-side energy storage system plays an increasingly important role in the power system, and can effectively balance the electricity supply and demand, relieve the grid pressure and improve the energy utilization efficiency. However, to fully utilize the advantages of the energy storage system, the key lies in formulating reasonable charging and discharging strategies, enabling it to charge when the electricity price is low and discharge when the electricity price is high, so as to maximize the economic benefits.

[0050] Based on this, the main purpose of the embodiments of the present disclosure is to propose a user-side energy storage charging and discharging strategy based on an electricity spot price prediction model, which will be specifically described through the following embodiments.

[0051] Referring to Figure 1 , a user-side energy storage charging and discharging strategy based on an electricity spot price prediction model according to an embodiment of the present disclosure is applied to a user-side energy storage system. The strategy includes, but is not limited to, steps S110 to S150.

[0052] S110, collect and preprocess electricity price and market feature data;

[0053] S120, construct an inner-layer prediction framework based on variational mode decomposition;

[0054] S130, construct an outer-layer prediction framework for calibration window integration prediction based on Bayesian model averaging;

[0055] S140, construct a deterministic user-side energy storage charging and discharging strategy model under complete information;

[0056] S150, construct a two-stage min-max regret value optimization scheduling problem based on uncertainty factors.

[0057] In step S110, data preparation: collect and preprocess electricity price sequence data and related market feature data.

[0058] Data cleaning: Check for missing values, outliers, and duplicate records in the data. Use appropriate methods to fill in missing values, eliminate or correct outliers, and remove duplicate data entries. Ensure that all data entries are accurate and consistent.

[0059] Data integration: Combine data from different sources and ensure that the data formats are consistent. For data with different formats or structures, conversion should be performed to make it unified into a format suitable for subsequent analysis.

[0060] Data preprocessing: Smooth the data to improve data quality. Moving average is a commonly used smoothing technique for reducing data volatility. By calculating the average of several adjacent points in the data sequence, moving average can eliminate random fluctuations and highlight long-term trends.

[0061] In step S120, construct an inner-layer prediction framework based on variational mode decomposition. Decompose the electricity price sequence into multiple subsequences, apply multiple prediction models and evaluate their performance, select and integrate the optimal subsequence prediction results. The process of the inner-layer prediction framework is as follows: 1. Use a hybrid decomposition method of variational mode decomposition (VMD) to decompose the electricity price sequence into multiple subsequences.

[0062] Initialize the mode function uk (t) and the central frequency ω k

[0063] Define a variational formula to minimize the total bandwidth of modal components with constraints:

[0064]

[0065] Use the ADMM method to iteratively solve the variational problem:

[0066]

[0067] Iteratively update the modal function u k (t) and the central frequency ω k Until convergence;

[0068] Reconstruct all modal functions into the initial signal:

[0069]

[0070] where u k (t) is the k-th modal function, representing the signal component at time t; ω k is the central frequency of the k-th modal function, representing the main frequency component of this mode; δ(t) is the Dirac function, also known as the unit impulse function, which is infinite at t = 0 and zero elsewhere, and its integral is 1; is the estimated value of the k-th modal function at the (n + 1)-th iteration; is the estimated value of the central frequency of the k-th modal function at the (n + 1)-th iteration; x(t) is the original signal, and its value at time t; λ is the regularization parameter, used to control the smoothness of the solution.

[0071] 2. Feature extraction

[0072] Suppose there is a time series dataset of electricity spot prices Extract the following features:

[0073]

[0074] where x t represents the feature matrix; X time represents the time feature; X lag represents the lag feature; X rolling represents the moving average feature.

[0075] 3. Apply multiple prediction models to each subsequence, including but not limited to: a time series interval prediction model based on quantile regression convolutional bidirectional long short-term memory neural network fusion attention mechanism (QRC-BiLSTM-Attention), an interval prediction algorithm based on XGBoost;

[0076] 3.1 Time Series Interval Prediction Model Based on Quantile Regression Convolutional Bidirectional Long Short-Term Memory Neural Network Fusion Attention Mechanism (QRC-BiLSTM-Attention)

[0077] Algorithm Process: This model extracts features through a convolutional neural network, captures the bidirectional dependencies of time series with bidirectional LSTM, fuses the attention mechanism to improve the effectiveness of feature extraction, and finally realizes interval prediction through quantile regression.

[0078] 1) Extract time series features using a convolutional neural network:

[0079]

[0080] where x t is the input sequence; W CNN and b CNN are the weights and biases of the CNN; * represents the convolution operation.

[0081] 2) Bidirectional Long Short-Term Memory Network (BiLSTM)

[0082] Input the extracted features into the bidirectional LSTM network:

[0083]

[0084] where and represent the hidden states of the forward and backward LSTM respectively.

[0085] 3) Fuse the attention mechanism, calculate the attention weights and sum them up weighted:

[0086]

[0087] where v T represents the transpose of the weight vector used to calculate the attention score; W att is the weight matrix of the hidden states of the forward and backward LSTM; b att is the bias term in the attention mechanism; a t is the attention score at time step t; is the context vector at time step t.

[0088] 4) Perform quantile regression on the features after attention fusion to obtain the prediction interval:

[0089]

[0090] where τ represents the quantile; is the predicted value corresponding to the quantile.

[0091] 5) Combine the prediction results of different quantiles to form a prediction interval.

[0092] 3.2 Interval prediction algorithm based on XGBoost

[0093] For each quantile τ, train an XGBoost regression model:

[0094] where x t is the input feature vector; XGBoost τ is the XGBoost model corresponding to the quantile.

[0095] The objective function of the XGBoost model is:

[0096]

[0097] fk(x) = ω q(x) (10)

[0098] where is the loss function, Ω(f k ) is the regularization term; f k (x) is the base learner in XGBoost; F is the set of all trees; q(x) is the tree structure function that maps the input features to the leaf nodes; ω is the weight of the leaf nodes.

[0099] 4. Model selection and integration

[0100] Use multiple evaluation metrics (such as root mean square error, mean absolute error, etc.) to comprehensively evaluate the prediction performance of each subsequence model. Based on the historical prediction results, use the meta - learning method to automatically select the optimal subsequence prediction model, and integrate the optimal prediction results of each subsequence to obtain the preliminary predicted electricity price.

[0101] 4.1 Suppose there are M subsequences, and the prediction result of each subsequence model is The true value is y(t).

[0102] The root mean square error (RMSE) of the i - th subsequence model:

[0103] The mean absolute error (MAE) of the i - th subsequence model:

[0104] 4.2 Based on the historical prediction results, use the meta - learning method to automatically select the optimal subsequence prediction model

[0105] The meta - learning method automatically selects the optimal subsequence prediction model by learning the historical prediction results. Suppose there are K evaluation metrics and the historical prediction results of M subsequence models.

[0106] 1) Construct the meta-feature matrix X:

[0107]

[0108] where Metric i,K represents the k-th evaluation index of the i-th subsequence model.

[0109] 2) Construct the target vector:

[0110]

[0111] where y contains the overall performance scores of all subsequence models.

[0112] 3) Train the meta-model:

[0113] Train the meta-feature matrix and the target vector using a meta-model (such as linear regression, random forest, etc.):

[0114] MetaModel(X) → y (13)

[0115] 4) Select the optimal model: Select the optimal subsequence model through the prediction results of the meta-model.

[0116]

[0117] where i * is the index of the optimal subsequence model; X i is the meta-feature of the i-th subsequence model.

[0118] 4.3 Integrate the optimal prediction results of each subsequence to obtain the preliminary predicted electricity price. Weight and integrate the prediction results of the selected optimal subsequence models to obtain the preliminary predicted electricity price as shown in the following formula:

[0119]

[0120] where ω i is the weight of the i-th subsequence model; Metric i is a certain evaluation index (such as MAE or RMSE) of the i-th subsequence model.

[0121] In step S130, the outer framework is improved: Calibration window integrated prediction: Set a dynamic calibration window and train the inner prediction model, optimize the weights using the weighted Bayesian model averaging method, perform error correction and multi-model integration to obtain the final electricity price prediction.

[0122] 1. Define the calibration window: Define L calibration windows, and the lengths of each calibration window are W1, W2,..., WL 。

[0123] 2. Error correction mechanism: The prediction result is corrected according to the historical prediction error.

[0124]

[0125] Among them, is the corrected prediction value of the L-th calibration window; represents the prediction value of the calibration window W l ; ε (l) (t) is the historical prediction error of the l-th calibration window at time t; k is the window size used to calculate the moving average.

[0126] 3. Multi-model integration: Suppose there are M models, and the prediction result of each model is interval prediction and are the interval prediction result, lower bound prediction value and upper bound prediction value of the m-th model at time t, respectively.

[0127] 4. Calculate the model weights within each calibration window: For each calibration window W l , use Bayesian Model Averaging (BMA) to calculate the model weights. Suppose there are M models, and the weight of model m on calibration window W l is expressed as

[0128]

[0129] Among them, P(D l |M m ) represents the likelihood of the dataset D l given model M l on calibration window W m ; P(M m ) represents the prior probability of model M m .

[0130] 5. Weighted integration of the prediction results of each calibration window: For each calibration window W l , the prediction value of this window is obtained by weighted integration of the prediction results of all models.

[0131]

[0132] Among them, and are the lower bound prediction value and upper bound prediction value of calibration window W l , respectively.

[0133] 6. Calculate the prediction weights for each calibration window: Calculate the weight for each calibration window based on the prediction error of each calibration window. Assume that the error metric is the mean squared error (MSE).

[0134]

[0135] where ω (l) is the weight of calibration window W l ; MSE (l) is the mean squared error of calibration window W l .

[0136] 7. Integrate the interval prediction results of the calibration windows:

[0137]

[0138] where and represent the final lower bound prediction value and upper bound prediction value respectively.

[0139] In step S140, a deterministic user-side energy storage charge and discharge strategy model based on complete information is constructed.

[0140] The degradation cost of the electrical energy storage unit is linearized and expressed as follows:

[0141]

[0142] where a, b, c, d are the coefficients of the battery degradation cost.

[0143] The constraints that the energy storage system needs to satisfy during operation include:

[0144]

[0145] where and represent the upper limits of the charging power and discharging power of the electrical energy storage unit respectively; SOC bat (t) represents the state of charge of the electrical energy storage unit, and SOC min and SOC max are the upper and lower limits of the state of charge respectively; η bat,c and η bat,d are the charging and discharging energy efficiencies of the electrical energy storage unit respectively; E bat represents the capacity of the energy storage system.

[0146] Therefore, the deterministic optimization scheduling problem model based on complete information is as follows:

[0147]

[0148] Among them, P Bid (t) represents the regulation amount participating in the market; and represent the real-time electricity purchase and sales volume generated due to power imbalance during the intraday stage; ρ1 and ρ2 are penalty cost coefficients; represents the electricity spot price.

[0149] In step S150, a two-stage min-max regret value optimization scheduling problem based on uncertainty factors is constructed:

[0150]

[0151] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. Any equivalent modification or change made by those of ordinary skill in the art according to the disclosure of the present invention shall be included in the protection scope recorded in the claims.

Claims

1. A charging and discharging strategy for user-side energy storage based on an electricity spot price prediction model, characterized in that, Applied to the user-side energy storage system, the strategy includes: Collect and preprocess electricity price and market characteristic data; Construct an inner-layer prediction framework based on variational mode decomposition; Construct an outer-layer prediction framework based on Bayesian model averaging for calibration window ensemble prediction; Construct a deterministic user-side energy storage charging and discharging strategy model under complete information; Construct a two-stage min-max regret optimization scheduling problem based on uncertainty factors; According to the collected and preprocessed electricity price and market characteristic data, the constructed inner-layer prediction framework based on variational mode decomposition, the constructed outer-layer prediction framework based on Bayesian model averaging for calibration window ensemble prediction, the constructed deterministic user-side energy storage charging and discharging strategy model under complete information, and the constructed two-stage min-max regret optimization scheduling problem based on uncertainty factors, construct a user-side energy storage charging and discharging strategy based on the electricity spot price prediction model.

2. The user-side energy storage charging and discharging strategy based on the electricity spot price prediction model according to claim 1, wherein, The collection and preprocessing of electricity price and market characteristic data include: Collect electricity price sequence data and related market characteristic data to ensure data integrity and consistency; Data cleaning and normalization processing, filling in missing values and eliminating outliers; According to the collected electricity price sequence data and related market characteristic data, and the data cleaning and normalization processing, filling in missing values and eliminating outliers, collect and preprocess electricity price and market characteristic data.

3. The user-side energy storage charging and discharging strategy based on the electricity spot price prediction model according to claim 1, characterized in that, The construction of the inner-layer prediction framework based on variational mode decomposition includes: Use the variational mode decomposition method to decompose the electricity price sequence into multiple subsequences; Construct a multi-model prediction method; Perform model selection and integration; According to the use of the variational mode decomposition method to decompose the electricity price sequence into multiple subsequences, the construction of the multi-model prediction method, and the performance of model selection and integration, construct an inner-layer prediction framework based on variational mode decomposition.

4. The inner layer prediction framework based on variational mode decomposition according to claim 3, characterized in that, The construction of the multi-model prediction method includes: Construct a time series interval prediction model based on the fusion of quantile regression convolutional bidirectional long short-term memory neural network and attention mechanism; Construct an interval prediction algorithm based on XGBoost; According to the constructed time series interval prediction model based on the fusion of quantile regression convolutional bidirectional long short-term memory neural network and attention mechanism, and the constructed interval prediction algorithm based on XGBoost, construct a multi-model prediction method.

5. The user-side energy storage charging and discharging strategy based on the electricity spot price prediction model according to claim 1, wherein The construction of the outer-layer prediction framework based on Bayesian model averaging for calibration window ensemble prediction includes: Set a dynamic calibration window; Train the inner-layer prediction model; Perform Bayesian model averaging optimization; Perform error correction and multi-model integration on the model; According to the setting of the dynamic calibration window, the training of the inner-layer prediction model, the performance of Bayesian model averaging optimization, and the performance of error correction and multi-model integration on the model, construct an outer-layer prediction framework based on Bayesian model averaging for calibration window ensemble prediction.

6. The user-side energy storage charging and discharging strategy based on the electricity spot price prediction model according to claim 1, characterized in that The construction of the deterministic optimization scheduling problem model under complete information includes: Construct a deterministic optimization scheduling problem objective function under complete information; Construct a deterministic optimization scheduling problem constraint condition under complete information; Construct a deterministic optimal scheduling problem model based on complete information according to the construction of the objective function of the deterministic optimal scheduling problem based on complete information and the construction of the constraint conditions of the deterministic optimal scheduling problem based on complete information.

7. The user-side energy storage charging and discharging strategy based on the electricity spot price prediction model according to claim 3, characterized in that, The construction of the objective function of the deterministic optimal scheduling problem based on complete information includes: Construct an aging model for the operation of the energy storage system; Construct a cost model for participating in the electricity market transaction; Construct an objective function of the deterministic optimal scheduling problem based on complete information according to the constructed aging model for the operation of the energy storage system and the constructed cost model for participating in the electricity market transaction.

8. The user-side energy storage charging and discharging strategy based on the electricity spot price prediction model according to claim 3, wherein, The construction of the constraint conditions of the deterministic optimal scheduling problem based on complete information includes: Construct a constraint condition model for the operation of the energy storage system; Construct a constraint condition model for participating in the electricity market transaction; Construct a system power balance constraint condition model; Construct an objective function of the deterministic optimal scheduling problem based on complete information according to the constructed constraint condition model for the operation of the energy storage system, the constructed constraint condition model for participating in the electricity market transaction, and the constructed system power balance constraint condition model.

9. The user-side energy storage charging and discharging strategy based on the electricity spot price prediction model according to claim 1, characterized in that, The construction of the two-stage min-max regret optimal scheduling problem based on uncertainty factors includes: Construct an objective function of the two-stage min-max regret optimal scheduling problem; Construct constraint conditions of the two-stage min-max regret optimal scheduling problem; Construct a two-stage min-max regret optimal scheduling problem based on uncertainty factors according to the constructed objective function of the two-stage min-max regret optimal scheduling problem and the constructed constraint conditions of the two-stage min-max regret optimal scheduling problem.

Images