Energy output prediction method, system and medium for clustered energy storage system

Abstract

The present invention discloses a method, system and medium for predicting energy output of a cluster-dispersive energy storage system. The method of the present invention includes extracting energy storage device feature vectors of energy output of each energy storage device type for energy storage device types that affect the energy output of the cluster-dispersive energy storage system; using a gated recurrent unit GRU to extract time series features based on the energy output data of the cluster-dispersive energy storage system on a typical historical day to obtain a historical feature vector at the current time t; combining the energy storage device feature vector and the historical feature vector at the current time t, and using a Bayesian neural network BNN pre-trained by an AO Sky Eagle optimization algorithm to predict the probability distribution of the energy output of the cluster-dispersive energy storage system. The present invention aims to solve the problems of low prediction accuracy, poor stability and difficulty in hyperparameter tuning in the energy output prediction method of the traditional energy output prediction model for the cluster-dispersive energy storage system.

Description

Technical Field

[0001] The present invention belongs to the field of energy technology, and in particular relates to an energy output prediction method, system and medium for a clustered energy storage system. Background Art

[0002] With the increasing popularity and increase of renewable energy (such as wind and solar energy) in energy supply, its volatility and intermittency bring challenges to grid operation and energy management. Energy storage technology, as a key means to solve the problem of renewable energy volatility, continues to grow and develop. By combining a variety of energy storage technologies, the clustered energy storage system provides a more flexible and efficient energy management solution. Compared with traditional energy storage systems, the clustered energy storage system can optimize energy management strategies, improve energy utilization efficiency, and reduce energy waste by accurately predicting energy output. Accurately predicting the energy output of the clustered energy storage system in the region is of great significance to improving the overall implementation effect of the clustered energy storage system and promoting the consumption of new energy. Therefore, the energy output prediction method of the clustered energy storage system has become one of the key technologies that must be mastered in the operational decision-making of the power market players.

[0003] A lot of research has been done on the energy output prediction methods of energy storage systems. Statistical methods are based on statistical analysis of historical time series data to predict the energy output of energy storage systems. Auto-regressive and moving average (ARMA) and auto-regressive integrated moving average (ARIMA) are typical statistical methods. ARMA requires a stable time series, so the input data must be fixed; ARIMA introduces the difference method based on the research of ARMA, which can handle non-stationary time series data. In addition, statistical methods also include exponential smoothing, Markov chain model, regression analysis, Kalman filtering, etc. Compared with physical methods, statistical methods have a relatively simple modeling process. Statistical models have better prediction effects on the energy output of clustered energy storage systems with a prediction time scale of 1~6h. However, as the prediction time scale increases, the prediction accuracy of statistical models decreases.

[0004] The energy output prediction methods of clustered energy storage systems based on machine learning mainly include support vector machine (SVM), ensemble learning and extreme learning machine (ELM). The basic idea of ​​SVM is to find the best separation hyperplane between two or more classes and maximize the margin around the separation hyperplane. In order to break the limitations of traditional support vector machines in processing data, the model effect is improved by kernel function optimization, and the training data is mapped to a higher-dimensional space. SVM can greatly reduce the complexity of prediction while ensuring prediction accuracy. However, the prediction performance of SVM is closely related to the selection of parameters. For example, if the penalty factor is too large or too small, the generalization ability of SVM will be weakened. Ensemble learning is a supervised learning algorithm that uses multiple learners to improve the prediction ability of weak models. Typical ensemble learning models include: Boosting of serialized individual learners, Bagging of parallelized individual learners, and random forest (RF). In the prediction of photovoltaic power generation, ensemble learning has become one of the most commonly used algorithms because it can reduce data noise and uncertainty of model errors. However, it is difficult to determine the number of weak learners and the model combination strategy. The ELM model has fast learning speed and strong generalization ability, but it cannot handle real-time data input and has poor stability.

[0005] Compared with machine learning methods, the energy output prediction method of the clustered energy storage system based on deep learning has the characteristics of high accuracy and strong robustness. Deep neural networks can automatically learn any complex mapping from input to output and support multiple inputs and outputs. Compared with traditional fully connected neural networks, CNN reduces the number of parameters that need to be estimated due to the use of weight sharing, but the hyperparameter tuning is more difficult; RNN variants LSTM and GRU perform well in the prediction of renewable energy power generation output and other time series; DBN is prone to fall into local minima during fine-tuning due to the limitations of the traditional BP algorithm. Summary of the invention

[0006] Technical problem to be solved by the present invention: In view of the above-mentioned problems in the prior art, a method and system for predicting energy output of a clustered-dispersed energy storage system are provided. The present invention aims to solve the problems of low prediction accuracy, poor stability and difficulty in hyperparameter tuning in the traditional energy output prediction method of the energy output prediction model for the clustered-dispersed energy storage system.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for predicting energy output of a clustered energy storage system comprises the following steps:

[0009] S101, extracting energy storage device feature vectors of energy output of each energy storage device type in a clustered energy storage system for energy storage device types that affect the energy output of the clustered energy storage system; using a gated recurrent unit GRU to extract time series features based on energy output data of the clustered energy storage system on a typical historical day to obtain a historical feature vector at the current time t;

[0010] S102, integrating the energy storage device feature vector and the historical feature vector at the current time t, and using the Bayesian neural network BNN pre-trained by the AO Sky Eagle optimization algorithm to predict the probability distribution of the energy output of the clustered energy storage system.

[0011] Optionally, before step S101, the step of analyzing and obtaining the type of energy storage equipment in the clustered-dispersed energy storage system that affects the energy output of the clustered-dispersed energy storage system is also included: respectively calculating the correlation coefficients between various types of energy storage equipment in the clustered-dispersed energy storage system and the energy output of the clustered-dispersed energy storage system, determining the type of energy storage equipment whose correlation coefficient exceeds a preset threshold and generating energy storage equipment characteristics from its output data.

[0012] Optionally, the calculation function expression of the correlation coefficient is:

[0013] ,

[0014] In the above formula, For the The correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The direct correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The indirect correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, where:

[0015] ,

[0016] ,

[0017] In the above formula, is the partial regression coefficient, T is the number of hours per day, For the The moment Output data of energy storage devices, For the The average output data of the energy storage device at T times, To provide energy for the clustered energy storage system. is the average energy output of the cluster-dispersive energy storage system at T moments, For the Class and The correlation coefficient between the output data of similar numerical energy storage devices, For the k The direct correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The moment Output data of similar energy storage devices.

[0018] Optionally, the function expression of the partial regression coefficient is:

[0019] ,

[0020] In the above formula, For the The moment Output data of energy storage devices, For the The average output data of the energy storage device at T times, To provide energy for the clustered energy storage system. is the average energy output of the clustered energy storage system at T moments, where T is the number of moments per day.

[0021] Optionally, in step S101, using a gated recurrent unit GRU to extract time series features based on the energy output data of the clustered energy storage system on a typical historical day refers to using a cascaded L The gated recurrent unit GRU extracts time series features based on the energy output data of the clustered energy storage system on typical historical days, and any l The function expression of the input vector of the gated recurrent unit GRU is:

[0022]

[0023] In the above formula, For the l The input vector of the gated recurrent unit GRU, concat is the concatenation operation, For the The energy storage device feature vector at the moment, For the The energy output of the clustered energy storage system at any moment; l GRU calculates the reset gate , Update Gate and the hidden state of the hidden layer and candidate hidden states The function expression is:

[0024] ,

[0025] ,

[0026] ,

[0027] ,

[0028] In the above formula, is the sigmoid activation function, is the summary of the current input and the previous hidden layer state, For the The output of the hidden layer of gated recurrent units, is the hyperbolic tangent activation function, , and Reset Gate , Update Gate and the weight matrix of the hidden layer; and L The hidden state of the hidden layer of the GRU As L The historical feature vector at the current time t output by the gated recurrent unit GRU .

[0029] Optionally, in step S102, the probability distribution of energy output of the clustered energy storage system is predicted by using the Bayesian neural network BNN pre-trained by the AO Eagle optimization algorithm, including:

[0030] S201, parameters for the probability layer of the Bayesian neural network Set the prior distribution to a proportional mixed Gaussian distribution:

[0031] ,

[0032] In the above formula, is the parameter of the probability layer The prior distribution of is the number of parameters; is the mixing ratio; is at a mean of 0 and a variance of Under the normal distribution of The probability density at ; is at a mean of 0 and a variance of Under the normal distribution of The probability density at ;

[0033] S202: The energy storage device characteristic vector at the current time t is calculated according to the following formula: and the historical feature vector Concatenate to get the input feature vector of the Bayesian neural network at the current time t :

[0034] ,

[0035] In the above formula, concat is a concatenation operation;

[0036] S203, treat the input feature vector as a random variable , the energy output of the clustered energy storage system is considered as a random variable , record the probability layer parameters The posterior probability is , introducing the approximate distribution ,in is a learnable parameter, ,in ~ For the 1st~ The mean of the parameters, ~ For the 1st~ Parameter mapping values ​​of parameters, approximate distribution is a Gaussian distribution with a mean of 0 and independent components:

[0037] ,

[0038] In the above formula, For the mean , the variance is Under the normal distribution of The probability density at , and For the The mean of the parameters; For the The variance of the parameter, The variance of the parameters With The mapping relationship between the mapping values ​​of the parameters is:

[0039] ;

[0040] S204, sampling from a standard normal distribution Group Dimensional data , respectively, according to the following formula to calculate the The parameter Sample value :

[0041] ,

[0042] In the above formula, For the Group Dimensional data; It represents the Hadamard product, which means the multiplication of the corresponding elements of the two matrices before and after;

[0043] S205, in Parameters for group sampling Next, input the feature vector To Parameters for group sampling The energy output sampling value of the clustering energy storage system at the current time t is obtained by forward propagation in the Bayesian neural network with parameters as follows: :

[0044] ,

[0045] In the above formula, For the first Parameters for group sampling is the parameter of the Bayesian neural network, the input feature vector at the current time t is the forward propagation of the input, and the energy output sampling value of the clustering energy storage system at the current time t is As the probability distribution output of the energy output of the clustered energy storage system predicted by the Bayesian neural network BNN.

[0046] Optionally, before step S101, the method also includes using the AO Eagle optimization algorithm to train the Bayesian neural network BNN for predicting the probability distribution of the energy output of the clustered energy storage system based on the characteristic vector of the energy storage device and the historical characteristic vector at the current time t, and the training of the Bayesian neural network BNN includes using the AO Eagle optimization algorithm to optimize the parameters of the Bayesian neural network BNN, including the learning rate, the number of hidden nodes and the training batch size.

[0047] In addition, the present invention also provides an energy output prediction system for a cluster-dispersive energy storage system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the energy output prediction method for the cluster-dispersive energy storage system.

[0048] In addition, the present invention also provides a computer-readable storage medium, in which a computer program / instruction is stored. The computer program / instruction is programmed or configured to execute the energy output prediction method for a clustered energy storage system through a processor.

[0049] In addition, the present invention also provides a computer program product, including a computer program / instruction, which is programmed or configured to execute the energy output prediction method for a clustered energy storage system through a processor.

[0050] Compared with the prior art, the present invention mainly has the following advantages: the energy output prediction method for the cluster-dispersed energy storage system of the present invention includes extracting the energy storage device feature vector of the energy output of each energy storage device type according to the energy storage device type that affects the energy output of the cluster-dispersed energy storage system in the cluster-dispersed energy storage system; using the gated recurrent unit GRU to extract the time series features according to the energy output data of the cluster-dispersed energy storage system on a typical historical day to obtain the historical feature vector at the current time t; combining the energy storage device feature vector and the historical feature vector at the current time t, using the Bayesian neural network BNN pre-trained by the AO Sky Eagle optimization algorithm to predict the probability distribution of the energy output of the cluster-dispersed energy storage system, which is obtained by the energy storage device The selection of types and the extraction of time series features are combined with the use of the Bayesian neural network BNN pre-trained by the AO Tianying optimization algorithm to realize the probability distribution prediction of the energy output of the clustering and decentralized energy storage system. The combination of the AO Tianying optimization algorithm, the gated recurrent unit GRU and the Bayesian neural network BNN is realized, which can solve the problems of low prediction accuracy, poor stability and difficulty in hyperparameter tuning in the traditional energy output prediction model used for the clustering and decentralized energy storage system. It can automatically optimize the hyperparameters of the energy output prediction model of the clustering and decentralized energy storage system, and can greatly improve the accuracy and stability of the prediction, thereby providing a strong guarantee for the flexible storage and scheduling of the clustering and decentralized energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of the basic flow of the method of the embodiment of the present invention.

[0052] Figure 2 Schematic diagram of the process of the AO-GRU-BNN model in an embodiment of the present invention.

[0053] Figure 3 It is a schematic diagram of the optimization process of the AO Eagle optimization algorithm in an embodiment of the present invention.

[0054] Figure 4 The figure is a comparison chart of the prediction results of the method according to the embodiment of the present invention and the existing method.

[0055] Figure 5 1 and 2 are power prediction intervals given by three probability prediction methods at different confidence levels in the embodiments of the present invention. DETAILED DESCRIPTION

[0056] like Figure 1 and Figure 2 As shown, the energy output prediction method for the clustered energy storage system of this embodiment includes the following steps:

[0057] S101, extracting energy storage device feature vectors of energy output of each energy storage device type in a clustered energy storage system for energy storage device types that affect the energy output of the clustered energy storage system; using a gated recurrent unit GRU to extract time series features based on energy output data of the clustered energy storage system on a typical historical day to obtain a historical feature vector at the current time t;

[0058] S102, integrating the energy storage device feature vector and the historical feature vector at the current time t, and using the Bayesian neural network BNN pre-trained by the AO Aquila Optimizer to predict the probability distribution of the energy output of the clustered energy storage system.

[0059] The method of this embodiment combines the AO Tianying optimization algorithm, the gated recurrent unit GRU and the Bayesian neural network BNN, so it is named AO-GRU-BNN model. Figure 2 As shown, it can solve the problems of low prediction accuracy, poor stability and difficulty in hyperparameter tuning in the traditional energy output prediction model for clustered and decentralized energy storage systems, realize automatic optimization of hyperparameters for the energy output prediction model of clustered and decentralized energy storage systems, and can greatly improve the prediction accuracy and stability, thereby providing strong guarantee for the flexible storage and scheduling of clustered and decentralized energy storage systems.

[0060] Before step S101, this embodiment also includes the step of analyzing and obtaining the type of energy storage device in the clustered energy storage system that affects the energy output of the clustered energy storage system: respectively calculating the correlation coefficients between various types of energy storage devices in the clustered energy storage system (such as lithium iron phosphate battery energy storage, supercapacitors and thermal storage tanks, etc.) and the energy output of the clustered energy storage system, determining the type of energy storage device whose correlation coefficient exceeds a preset threshold and generating energy storage device characteristics from its output data.

[0061] In this embodiment, the calculation function expression of the correlation coefficient is:

[0062] ,

[0063] In the above formula, For the The correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The direct correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The indirect correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, where:

[0064] ,

[0065] ,

[0066] In the above formula, is the partial regression coefficient, T is the number of hours per day, For the The moment Output data of energy storage devices, For the The average output data of the energy storage device at T times, To provide energy for the clustered energy storage system. is the average energy output of the cluster-dispersive energy storage system at T moments, For the Class and The correlation coefficient between the output data of similar numerical energy storage devices, For the k The direct correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The moment Output data of similar energy storage devices.

[0067] In this embodiment, the functional expression of the partial regression coefficient is:

[0068] ,

[0069] In the above formula, For the The moment Output data of energy storage devices, For the The average output data of the energy storage device at T times, To provide energy for the clustered energy storage system. is the average energy output of the clustered energy storage system at T moments, where T is the number of moments per day. Specifically, in this embodiment, the energy output of the clustered energy storage system and the output data of the energy storage equipment in a certain place in Hunan are used for correlation analysis, and factors with correlation coefficients greater than 0.2 (lithium iron phosphate battery energy storage, supercapacitors, thermal storage tanks, etc.) are selected. factors, the vector composed of these energy storage device factor data That is The energy storage device feature vector at the moment.

[0070] In step S101 of this embodiment, the gated recurrent unit GRU (Gated Recurrent Unit) is used to extract the time series features according to the energy output data of the clustered energy storage system on a typical historical day. LThe gated recurrent unit GRU extracts time series features based on the energy output data of the clustered energy storage system on typical historical days, and any l The function expression of the input vector of the gated recurrent unit GRU is:

[0071]

[0072] In the above formula, For the l The input vector of the gated recurrent unit GRU, concat is the concatenation operation, For the The energy storage device feature vector at the moment, For the The energy output of the clustered energy storage system at any moment; l GRU calculates the reset gate , Update Gate and the hidden state of the hidden layer and candidate hidden states The function expression is:

[0073] ,

[0074] ,

[0075] ,

[0076] ,

[0077] In the above formula, is the sigmoid activation function, is the summary of the current input and the previous hidden layer state, For the The output of the hidden layer of gated recurrent units, is the hyperbolic tangent activation function, , and Reset Gate , Update Gate and the weight matrix of the hidden layer; and L The hidden state of the hidden layer of the GRU As L The historical feature vector at the current time t output by the gated recurrent unit GRU .

[0078] In step S102 of this embodiment, the probability distribution of energy output of the clustered energy storage system is predicted by using the Bayesian neural network BNN pre-trained by the AO Eagle optimization algorithm, including:

[0079] S201, parameters for the probability layer of the Bayesian neural network Set the prior distribution to a proportional mixed Gaussian distribution:

[0080] ,

[0081] In the above formula, is the parameter of the probability layer The prior distribution of is the number of parameters; is the mixing ratio; is at a mean of 0 and a variance of Under the normal distribution of The probability density at ; is at a mean of 0 and a variance of Under the normal distribution of The probability density at ;

[0082] S202: The energy storage device characteristic vector at the current time t is calculated according to the following formula: and the historical feature vector Concatenate to get the input feature vector of the Bayesian neural network at the current time t :

[0083] ,

[0084] In the above formula, concat is a concatenation operation;

[0085] S203, treat the input feature vector as a random variable , the energy output of the clustered energy storage system is considered as a random variable , record the probability layer parameters The posterior probability is , introducing the approximate distribution ,in is a learnable parameter, ,in ~ For the 1st~ The mean of the parameters, ~ For the 1st~ Parameter mapping values ​​of parameters, approximate distribution is a Gaussian distribution with a mean of 0 and independent components:

[0086] ,

[0087] In the above formula, For the mean , the variance is Under the normal distribution of The probability density at , and For the The mean of the parameters; For the The variance of the parameter, The variance of the parameters With The mapping relationship between the mapping values ​​of the parameters is:

[0088] ;

[0089] S204, sampling from a standard normal distribution Group Dimensional data , respectively, according to the following formula to calculate the The parameter Sample value :

[0090] ,

[0091] In the above formula, For the Group Dimensional data; It represents the Hadamard product, which means the multiplication of the corresponding elements of the two matrices before and after;

[0092] S205, in Parameters for group sampling Next, input the feature vector To Parameters for group sampling The energy output sampling value of the clustering energy storage system at the current time t is obtained by forward propagation in the Bayesian neural network with parameters as follows: :

[0093] ,

[0094] In the above formula, For the first Parameters for group sampling is the parameter of the Bayesian neural network, the input feature vector at the current time t is the forward propagation of the input, and the energy output sampling value of the clustering energy storage system at the current time t is As the probability distribution output of the energy output of the clustered energy storage system predicted by the Bayesian neural network BNN.

[0095] Before step S101 of this embodiment, the method further includes using the AO Eagle optimization algorithm to train the Bayesian neural network BNN for predicting the probability distribution of the energy output of the clustering and dispersing energy storage system based on the energy storage device feature vector and the historical feature vector at the current time t, and the training of the Bayesian neural network BNN includes using the AO Eagle optimization algorithm to optimize the parameters of the Bayesian neural network BNN, including the learning rate, the number of hidden nodes, and the training batch size. The AO Eagle optimization algorithm is an existing well-known method and can be divided into the following five stages: initialization, expansion search stage, reduction search stage, expansion development stage, reduction development stage, such as Figure 3 As shown, it mainly includes:

[0096] S301, in the initialization stage, the initialization population needs to randomly initialize the search individual position within the search range, and its function expression is:

[0097] ,

[0098] In the above formula, To search for population size; The dimension of the problem; is a random vector with values ​​between 0 and 1; For the given problem a lower bound; For the given problem The population is composed of the learning rate, the number of hidden nodes, and the training batch size;

[0099] S302, in the expanded search phase, the Skyhawk increases its vertical height to obtain a larger search range, so as to better hunt prey during flight and determine the optimal hunting range. The mathematical description of this process is as follows:

[0100] ,

[0101] In the above formula, To use the search method In the The search results obtained during the round, For the front The best solution obtained in the round of iterations, is the maximum search round, Same as above, a random value between 0 and 1. For the The average value of the solutions in this round of iteration is obtained by the following formula:

[0102] ,

[0103] In the above formula, N is the number of solutions in this round.

[0104] S303, in the narrowing search phase, the Skyhawk will hover within the range of the prey and prepare to attack. The mathematical description of this process is as follows:

[0105] ,

[0106] In the above formula, To use the search method ; in the The search results obtained during the round, is the Levy flight distribution strategy used to describe the flight of the Skyhawk.

[0107] S304, during the expansion development phase, the Skyhawk will be ready to land and attack, and will use a low altitude slow descent attack method to conduct a preliminary attack to determine the reaction of the prey. The mathematical description of this process is as follows:

[0108] ,

[0109] In the above formula, To use the search method In the The search results obtained during the round, and To develop the adjustment parameters, the values ​​can be taken according to actual needs, for example, 0.1 is taken in this embodiment.

[0110] S305, in the shrinking development stage, the Sky Eagle will launch a random attack on the prey on the ground. The mathematical description of this process is as follows:

[0111] ,

[0112] In the above formula, To use the search method In the The search results obtained during the round; is the quality function value used to balance the search strategy; for the various movements made by the eagle in the pursuit of its prey; is a linearly decreasing flight slope value, and:

[0113] ,

[0114] ,

[0115] ,

[0116] S306, determine whether the maximum number of iterations has been reached, if not, jump to step S302; otherwise, output the optimal parameters, end and exit.

[0117] In order to verify the energy output prediction method for the clustered energy storage system in this embodiment, the energy output of the clustered energy storage system in a Hunan industrial park and the output data of related energy storage equipment are used as data sets. At the same time, in order to verify the effectiveness of the proposed model (AO-GRU-BNN) for the energy output prediction method for the clustered energy storage system in this embodiment, the normal distribution method (ND) and the kernel density estimation method (KDE) are used for comparative experiments. The prediction results of the energy output probability distribution of the above-mentioned clustered energy storage system are compared. Figure 4 As shown. Figure 4 It can be seen that, compared with the normal distribution method (ND) and the kernel density estimation method (KDE), the peak value of the model (AO-GRU-BNN) proposed in the energy output prediction method for the clustered energy storage system in this embodiment is closer to the actual value of the energy output of the clustered energy storage system at that moment; at the same time, the probability density distribution of the model (AO-GRU-BNN) proposed in the energy output prediction method for the clustered energy storage system in this embodiment is more concentrated and has a smaller variance, which better describes the distribution of the energy output of the clustered energy storage system near the actual value, while the probability density distribution of the normal distribution method (ND) and the kernel density estimation method (KDE) is wider, the characterization of uncertainty is more vague, and the accuracy is not high.

[0118] Figure 5 The power prediction intervals given by the three probability prediction methods at different confidence levels in this embodiment are shown in Table 1. The average interval widths of the three different probability prediction methods are shown in Table 1.

[0119] Table 1 Comparison of prediction results of average interval width ( / kW) of three probability prediction methods

[0120] Confidence interval / % ND KDE AO-GRU-BNN 50 1.563 0.979 1.089 70 2.099 1.898 1.674 90 3.386 3.658 2.317

[0121] Depend on Figure 5 It can be seen from Table 1 that at different confidence levels, when the overall prediction accuracy is low, the average interval width of KDE is small. However, as the prediction accuracy increases, the interval width of KDE increases rapidly. When the prediction accuracy is required to be 90%, the average interval width of the model (AO-GRU-BNN) proposed in the energy output prediction method for the cluster-dispersed energy storage system in this embodiment is much smaller than that of the other two methods, which proves the superiority of the model (AO-GRU-BNN) proposed in the energy output prediction method for the cluster-dispersed energy storage system in this embodiment in the probability prediction of energy output of the cluster-dispersed energy storage system.

[0122] In summary, the method of this embodiment uses the path analysis method to analyze the impact of the type of energy storage equipment on the energy output of the cluster-dispersed energy storage system and obtain the characteristics of the energy storage equipment; secondly, the learning rate, number of hidden nodes and training batch size of the GRU-BNN model are optimized through the AO (Aquila Optimizer) algorithm and input into the network model; then, the gated recurrent unit GRU is used to extract the time series characteristics through the energy output data of the cluster-dispersed energy storage system on typical historical days; finally, the Bayesian neural network is used to predict the probability distribution of the energy output of the cluster-dispersed energy storage system based on the comprehensive energy storage equipment and time series characteristics. The method of this embodiment aims to solve the problems of low prediction accuracy, poor stability and difficulty in hyperparameter tuning of the energy output prediction method of the traditional energy storage system. For the prediction of energy processing of the cluster-dispersed energy storage system, a combined model is adopted, which integrates two different types of neural networks: the gated recurrent unit and the Bayesian neural network. By establishing a connection between GRU and BNN, their advantages in time series feature extraction and probability prediction are fully utilized to better capture the complex relationships in time series data and improve the effectiveness of prediction; at the same time, by introducing the AO algorithm for hyperparameter tuning, this method realizes the automatic optimization of the neural network structure and parameters, avoids the tedious process of manual parameter adjustment, improves the generalization ability and robustness of the model, improves the power supply flexibility and response speed of the cluster-dispersed energy storage system, alleviates the load pressure of the power system, and provides a strong guarantee for the power supply stability and reliability of the power grid. The method of this embodiment can automatically optimize the hyperparameters and greatly improve the accuracy and stability of the prediction, thereby providing a strong guarantee for the flexible storage and scheduling of the cluster-dispersed energy storage system, and solves the problems of low prediction accuracy, poor stability, and difficulty in hyperparameter tuning in the energy output prediction method of the traditional energy storage system.

[0123] In addition, this embodiment also provides an energy output prediction system for a cluster-dispersive energy storage system, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the energy output prediction method for the cluster-dispersive energy storage system.

[0124] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program / instruction is stored. The computer program / instruction is programmed or configured to execute the energy output prediction method for the clustered energy storage system through a processor.

[0125] In addition, this embodiment also provides a computer program product, including a computer program / instruction, which is programmed or configured to execute the energy output prediction method for the clustered energy storage system through a processor.

[0126] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0127] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting energy output of a clustered energy storage system, characterized in that: The steps include: S101, extracting energy storage device feature vectors of energy output of each energy storage device type in a clustered energy storage system for energy storage device types that affect the energy output of the clustered energy storage system; using a gated recurrent unit GRU to extract time series features based on energy output data of the clustered energy storage system on a typical historical day to obtain a historical feature vector at the current time t; S102, combining the energy storage device feature vector and the historical feature vector at the current time t, and using the Bayesian neural network BNN pre-trained by the AO Tianying optimization algorithm to predict the probability distribution of the energy output of the cluster-dispersed energy storage system; Before step S101, the step of analyzing and obtaining the type of energy storage device in the clustered energy storage system that affects the energy output of the clustered energy storage system is also included: respectively calculating the correlation coefficients between various types of energy storage devices in the clustered energy storage system and the energy output of the clustered energy storage system, determining the type of energy storage device whose correlation coefficient exceeds a preset threshold and generating energy storage device characteristics from its output data, wherein the calculation function expression of the correlation coefficient is: ， In the above formula, For the The correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The direct correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The indirect correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, where: ， ， In the above formula, is the partial regression coefficient, T is the number of hours per day, For the The moment Output data of energy storage devices, For the The average output data of the energy storage device at T moments, To provide energy for the clustered energy storage system. is the average energy output of the cluster-dispersive energy storage system at T moments, For the Class and The correlation coefficient between the output data of similar numerical energy storage devices, For the k The direct correlation coefficient between the type of energy storage equipment and the energy output of the clustered energy storage system, For the The moment Output data of energy storage devices, For the k The average output data of the energy storage device at T moments.

2. The energy output prediction method for a clustered energy storage system according to claim 1, characterized in that: The functional expression of the partial regression coefficient is: 。 3. The energy output prediction method for a clustered energy storage system according to claim 1, characterized in that: In step S101, the gated recurrent unit GRU is used to extract the time series features based on the energy output data of the clustered energy storage system on a typical historical day. L The gated recurrent unit GRU extracts time series features based on the energy output data of the clustered energy storage system on typical historical days, and any l The function expression of the input vector of the gated recurrent unit GRU is: ， In the above formula, For the l The input vector of the gated recurrent unit GRU, concat is the concatenation operation, For the The energy storage device feature vector at the moment, For the The energy output of the clustered energy storage system at each moment; Any l GRU calculates the reset gate , Update Gate and the hidden state of the hidden layer and candidate hidden states The function expression is: ， ， ， ， In the above formula, is the sigmoid activation function, is the summary of the current input and the previous hidden layer state, For the The output of the hidden layer of gated recurrent units, is the hyperbolic tangent activation function, , and Reset Gate , Update Gate and the weight matrix of the hidden layer; and L The hidden state of the hidden layer of the GRU As L The historical feature vector at the current time t output by the gated recurrent unit GRU .

4. The energy output prediction method for a clustered energy storage system according to claim 1, characterized in that: In step S102, the probability distribution of energy output of the clustered energy storage system is predicted by using the Bayesian neural network BNN pre-trained by the AO Eagle optimization algorithm, including: S201, parameters for the probability layer of the Bayesian neural network Set the prior distribution to a proportional mixed Gaussian distribution: ， In the above formula, is the parameter of the probability layer The prior distribution of is the number of parameters; is the mixing ratio; is at a mean of 0 and a variance of Under the normal distribution of The probability density at ; is at a mean of 0 and a variance of Under the normal distribution of The probability density at ; S202: The energy storage device characteristic vector at the current time t is calculated according to the following formula: and the historical feature vector Concatenate to get the input feature vector of the Bayesian neural network at the current time t : ， In the above formula, concat is a concatenation operation; S203, treat the input feature vector as a random variable , the energy output of the clustered energy storage system is considered as a random variable , record the probability layer parameters The posterior probability is , introducing the approximate distribution ,in is a learnable parameter, ,in ~ For the 1st~ The mean of the parameters, ~ For the 1st~ Parameter mapping values ​​of parameters, approximate distribution is a Gaussian distribution with a mean of 0 and independent components: ， In the above formula, For the mean , the variance is Under the normal distribution of The probability density at , and For the The mean of the parameters; For the The variance of the parameter, The variance of the parameters With The mapping relationship between the mapping values ​​of the parameters is: ； S204, sampling from a standard normal distribution Group Dimensional data , respectively, according to the following formula to calculate the The parameter Sample value : ， In the above formula, For the Group Dimensional data; It represents the Hadamard product, which means the multiplication of the corresponding elements of the two matrices before and after; S205, in Parameters for group sampling Next, input the feature vector To Parameters for group sampling The energy output sampling value of the clustering energy storage system at the current time t is obtained by forward propagation in the Bayesian neural network with parameters as follows: : ， In the above formula, For the first Parameters for group sampling is the parameter of the Bayesian neural network, the input feature vector at the current time t is the forward propagation of the input, and the energy output sampling value of the clustering energy storage system at the current time t is As the probability distribution output of the energy output of the clustered energy storage system predicted by the Bayesian neural network BNN.

5. The energy output prediction method for a clustered energy storage system according to claim 1, characterized in that: Before step S101, the method also includes using the AO Eagle optimization algorithm to train the parameters of the Bayesian neural network BNN, including the learning rate, the number of hidden nodes, and the training batch size.

6. An energy output prediction system for a clustered energy storage system, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the energy output prediction method for a clustered energy storage system as described in any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program / instruction stored therein, characterized in that: The computer program / instruction is programmed or configured to execute the energy output prediction method for a clustered energy storage system as described in any one of claims 1 to 5 through a processor.

8. A computer program product comprising a computer program / instructions, characterized in that The computer program / instruction is programmed or configured to execute the energy output prediction method for a clustered energy storage system as described in any one of claims 1 to 5 through a processor.

Images