Adjustable resource cluster multi-time node response potential evaluation method based on ae-lstm

Abstract

The application discloses a response potential evaluation method of adjustable resource cluster based on an AE-LSTM algorithm, and comprises the following steps: 1, collecting original power data and excitation intensity data; 2, decomposing the power consumption power sequence of the adjustable resource individual by using a discrete wavelet transform; 3, classifying the decomposed sequence of the adjustable resource cluster by using K-medoids clustering based on a DTW algorithm; 4, extracting the characteristic value of the center point data of each sequence of the clustered adjustable resource cluster by using the AE-LSTM algorithm, obtaining the parameters of the potential evaluation probability distribution, and calculating the response potential probability distribution of the adjustable resource individual. The application can effectively improve the characteristic value extraction efficiency of the AE-LSTM neural network, solve the problem that individual differences are neglected in adjustable resource potential evaluation, and thus can realize effective evaluation on the scheduling potential of the adjustable object in the adjustable cluster.

Description

Technical Field

[0001] This invention relates to the field of power system adjustable resource response potential assessment, and specifically to a multi-time-node potential assessment method for adjustable resource clusters considering different operating characteristics. Background Technology

[0002] Integrating demand-side response into the grid energy management system can improve load curves by encouraging active consumer participation, thereby achieving economic benefits and reducing carbon emissions from the power system. Demand response can not only effectively reduce peak loads and installed capacity of traditional generating units, but also help improve the reliability and robustness of the grid. Accurate assessment of the response potential of adjustable resources can effectively promote the participation of various adjustable resources in power system ancillary services, which is of great significance for power system dispatching and planning.

[0003] Existing methods for assessing adjustable resource potential generally estimate it directly based on user capacity ratios. However, users with different operating characteristics vary in terms of incentive intensity and adjustable range, failing to consider the impact of individual differences on the response range of adjustable resources and not fully exploring the coordinated and optimized utilization of capacity resources in the distribution network through demand response. Furthermore, the accuracy of adjustable resource response potential assessment is affected by external uncertainties such as meteorological conditions and response behavior, resulting in insufficient systematicity and effectiveness in the assessment. These inadequate considerations lead to poor practical results in distribution network collaborative control optimization, reducing the broad applicability of demand response. Summary of the Invention

[0004] To overcome the shortcomings of the existing technology, this invention proposes a method for evaluating the multi-time-node response potential of adjustable resource clusters based on AE-LSTM. This method aims to comprehensively consider the differences in individual response intensity and working characteristics of adjustable resources, thereby providing effective guidance for flexible load security control, source-load optimization scheduling, and power grid planning and construction.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] The present invention provides a method for evaluating the response potential of an adjustable resource cluster based on the AE-LSTM algorithm, characterized by the following steps:

[0007] Step 1: Sample raw power data and excitation intensity data at T time points daily from an adjustable resource cluster containing M adjustable resource individuals, thereby obtaining a raw power data set A = {A1, A2, ..., A...} for N days. m ,...,A M} and the set of excitation intensity data δ={δ1,δ2,...,δm ,...,δ M}, where A m Let A represent the set of original power data for the m-th adjustable resource individual, and A m ={A m,1 A m,2 ,...,A m,n ,...,A m,N}, A m,n Let represent the original power data sequence of the m-th adjustable resource individual on day n, and Let represent the raw power data of the m-th adjustable resource individual at time t on day n. Then, the raw power data of the m-th adjustable resource individual at time t on day N is denoted as . δ m Let represent the set of incentive intensity data for the m-th adjustable resource individual, and δ m ={δ m,1 ,δ m,2 ,...,δ m,t ,...,δ m,T}, δ m,t This represents the incentive intensity data for the m-th adjustable resource individual at the t-th time point each day;

[0008] Step 2: Decompose the raw power data of individual adjustable resources using discrete wavelet transform:

[0009] Using the db6 wavelet function to Decompose into K a This allows us to obtain a high-frequency signal from the m-th adjustable resource individual at the t-th time point on day N. And K of the m-th adjustable resource individual at the t-th time point on day N. a Low-frequency signal of layer in, This represents the k-th time point of the m-th adjustable resource individual at the t-th time point on day N. a Low-frequency signals in the layer, and This represents the k-th time point of the m-th adjustable resource individual on day n at time t. a Low-frequency raw power data; thus obtaining high-frequency signals of M adjustable resource individuals at time point t on day N. And M adjustable resource individuals at time point t on day N. a The low-frequency signal of the layer is denoted as in, This represents the k-th time point of M adjustable resource individuals at time t on day N. a Low-frequency signals in the layer, and in, This represents the high-frequency raw power data of the m-th adjustable resource individual at time t on day n.

[0010] Step 3: Classify the high-frequency and low-frequency signals after decomposition of the adjustable resource cluster using K-medoids clustering based on the DTW algorithm:

[0011] Step 3.1, for B t Clustering is performed on the high-frequency signals to obtain the final updated N at time point t. k Individuals with high-frequency cluster centers in, This represents the m-th update at time point t. b A high-frequency cluster center individual;

[0012] Step 3.2, from Randomly select N k An individual with adjustable resources at time point k on day t. a The low-frequency signal of layer k is used as the kth layer a If the low-frequency cluster center individuals are in the layer, then MN remain. k An individual with adjustable resources at time point k on day t. a The low-frequency signal of layer k is used as the kth a The low-frequency non-clustering center individuals are identified; then, following steps 3.1.2-3.1.5, the final updated k-th time point at time t is obtained. a Low-frequency cluster center individuals in, This represents the k-th update at time point t. a Layer m b 1. 2. 3. 4. 5. 5. 6. 7. 8. 9. ... a Low-frequency cluster center individuals

[0013] Step 3.3, will and After merging, the high-frequency signal and K at time point t are obtained. a Cluster center set of low-frequency signals in, This indicates the k-th time point at time t. e Individual cluster centers;

[0014] Step 4: Use the AE-LSTM algorithm to extract feature values ​​from the cluster center data after clustering, calculate the general and random parameters of the potential assessment model, and calculate the probability distribution of the response potential of individual adjustable resources.

[0015] Step 4.1: Constructing the AE-LSTM neural network includes: an input layer, an encoder LSTM network layer, and a decoder LSTM network layer; used to extract the cluster center set. At time point t, the k-th time... e Individual cluster centers eigenvalues in, Represents the optimal characteristic matrix The eigenvalue of the i-th row and j-th column;

[0016] Step 4.2: Collect data separately The basic threshold for the incentive intensity response generated by the individual with adjustable resources. The upper threshold of the individual's response to incentive intensity for the available adjustable resources Maximum response rate of the individual with available adjustable resources

[0017] Step 4.3: Based on the least squares method... and respectively with By performing fitting, the values ​​at time point t are obtained. The kth e Individual cluster centers The basic threshold fitting function of the excitation intensity response At time point t The kth e Individual cluster centers Upper threshold fitting function of the excitation intensity response At time point t The kth e individual Maximum response rate fitting function

[0018] Step 4.4, in Then, reselect a cluster center at time point t and repeat steps 4.1-4.3 until all high-frequency signals and K at time point t are obtained. a Fitting function for cluster centers of low-frequency signals in layers;

[0019] Step 4.5: Decompose the high-frequency data of the raw power at time point t over N days and K... a After calculating the eigenvectors of the low-frequency data in each layer, they are substituted into the basic threshold fitting function of the cluster center to which the data belongs. The results are then superimposed to obtain the basic threshold of the m-th adjustable resource individual at time t.

[0020] The high-frequency data of the raw power decomposition at time point t on N days and Ka After calculating the feature vectors of the low-frequency data in each layer, they are substituted into the upper limit threshold fitting function of the cluster to which the data belongs, and the results are superimposed to obtain the upper limit threshold of the m-th adjustable resource individual at time t.

[0021] The high-frequency data of the raw power decomposition at time point t on N days and K a After calculating the eigenvectors of the low-frequency data in each layer, they are substituted into the maximum response rate fitting function of the cluster to which the data belongs. The results are then superimposed to obtain the maximum response rate of the m-th adjustable resource individual at time t.

[0022] Step 4.6: Collect the load reduction rate η of the m-th adjustable resource at the t-th time point in the historical data. m,t The probability distribution is obtained using the point estimation method. Where N represents a normal distribution, μ m,t σ m,t Let the mean and standard deviation of the load reduction rate of the m-th adjustable resource individual at time point t be given respectively; and we have:

[0023]

[0024]

[0025] In equations (6) and (7), ξ t,m Let be the potential assessment parameters for the m-th adjustable resource individual at time point t. Let be the mean and standard deviation of the potential assessment parameters for the m-th adjustable resource individual at time point t;

[0026] Step 4.7: Calculate the incentive intensity of the m-th adjustable resource individual at time point t using equation (5). Response potential

[0027]

[0028] The method for evaluating the multi-time-node response potential of adjustable resource clusters based on AE-LSTM described in this invention is also characterized in that step 3.1 includes:

[0029] Step 3.1.1, in B t Randomly select N k If the high-frequency signal of an adjustable resource individual at time point t on day N is used as the high-frequency clustering center individual, then the remaining MN k The high-frequency signal of an adjustable resource individual at the t-th time point on N days is taken as the high-frequency non-clustering center individual;

[0030] Step 3.1.2: Solve for the m-th equation using equations (1) and (2). a A high-frequency non-clustering center individual and the mth b Individuals with high-frequency cluster centers DTW distance

[0031]

[0032]

[0033] In equations (1) and (2), for The Nth signal value, for The Nth signal value, for and The current distance, and for and The cumulative distance; for The (N-1)th signal value and The (N-1)th signal value The cumulative distance; for The (N-1)th signal value and The Nth signal value The cumulative distance, for The Nth signal value and the (N-1)th signal value The cumulative distance;

[0034] Step 3.1.3: Calculate the DTW distance between all high-frequency non-clustered center individuals and each high-frequency clustered center individual at time point t, and assign the high-frequency non-clustered center individual with the smallest DTW distance to the corresponding high-frequency clustered center individual, thereby obtaining the well-clustered high-frequency population at time point t.

[0035] Step 3.1.4: Calculate the high-frequency cluster centers of the clustered high-frequency population at time point t using the K-medoids algorithm, and obtain the updated high-frequency cluster center individuals;

[0036] Step 3.1.5: Submit the updated high-frequency cluster centers to step 3.1.2 sequentially until the maximum number of iterations is reached, thus obtaining the final updated N at time point t.k Individuals with high-frequency cluster centers in, This represents the m-th update at time point t. b A high-frequency cluster center individual.

[0037] Step 4.1 includes:

[0038] Step 4.1.1: Set the threshold parameter to L. mse Set the sliding window length parameter to n a ;

[0039] The input layer uses a sliding window algorithm to... Convert to Nn a +1 vector, denoted as the k-th vector at time point t. e Sliding feature matrix of an individual in, This indicates the k-th time point at time t. e Individual cluster centers The nth a vectors, and express v in the middle a One data point;

[0040] Step 4.1.2, the encoder LSTM network layer... Encode the k-th digit. e Individual cluster centers Feature matrix Among them, L a The output dimension of the LSTM neural network;

[0041] Step 4.1.3, the decoder LSTM network layer... Perform decoding to obtain the k-th... e Reduction matrix of each cluster center individual

[0042] Step 4.1.4, Construction and The mean squared error loss is calculated, and the parameters of the AE-LSTM neural network are trained using gradient descent until the mean squared error loss is less than L. mse At that time, the optimal AE-LSTM model after training is obtained, and the optimal feature matrix is ​​obtained from the encoder layer LSTM network layer in the optimal AE-LSTM model after training. Extract the k-th time point at time t e Individual feature vectors in, Represents the optimal characteristic matrix The eigenvalue of the i-th row and j-th column.

[0043] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the adjustable resource cluster multi-time node response potential assessment method, and the processor is configured to execute the program stored in the memory.

[0044] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the steps of the method for evaluating the multi-time-node response potential of a tunable resource cluster.

[0045] Compared with existing technologies, the beneficial effects of this invention are reflected in:

[0046] 1. This invention considers the decomposition and feature extraction of various adjustable resource electricity consumption behaviors to obtain adjustable potential model parameters that reflect the differences in response characteristics at different time points. It accurately assesses the response potential of large-scale adjustable resource clusters, providing effective assistance for practical regulation.

[0047] 2. This invention comprehensively considers the different working characteristics of adjustable resource clusters and proposes to use wavelet decomposition of time series and K-medoids clustering based on the DTW algorithm to improve the reliability of data prediction for adjustable resource response potential assessment and reduce the number of subsequences of adjustable resource clusters, thereby significantly improving feature extraction efficiency and reducing redundant calculations in different adjustable resource potential assessment processes.

[0048] 3. This invention comprehensively considers the temporal characteristics of the electricity consumption behavior of adjustable resource clusters and proposes a feature extraction method based on the AE-LSTM algorithm. It has a good processing effect on data with continuous time characteristics and can effectively extract feature values ​​of the response potential of adjustable resource clusters, thus helping to classify and regulate various types of adjustable resources in the region. Attached Figure Description

[0049] Figure 1 This is a flowchart of the adjustable resource potential assessment process of the present invention. Detailed Implementation

[0050] In this embodiment, a method for assessing the multi-time-node response potential of adjustable resource clusters based on an autoencoder-long short-term memory (AE-LSTM) network predicts the response potential of different adjustable resources based on feature extraction of historical power consumption behavior of the adjustable resource clusters. The method includes: 1. Collecting raw power data and excitation intensity data; 2. Decomposing the power consumption sequence of individual adjustable resources using discrete wavelet transform; 3. Classifying the decomposed sequences of the adjustable resource clusters using K-medoids clustering based on the DTW (Dynamic Time Warping) algorithm; 4. Extracting feature values ​​from the centroid data of each sequence of the clustered adjustable resource clusters using the AE-LSTM algorithm, calculating the general and random parameters of the potential assessment model, and calculating the probability distribution of the response potential of individual adjustable resources. Specifically, as... Figure 1 As shown, the procedure is as follows:

[0051] Step 1: Sample raw power data and excitation intensity data at T time points daily from an adjustable resource cluster containing M adjustable resource individuals, thereby obtaining a raw power data set A = {A1, A2, ..., A...} for N days. m ,...,A M} and the set of excitation intensity data δ={δ1,δ2,...,δ m ,...,δ M}, where A m Let A represent the set of original power data for the m-th adjustable resource individual, and A m ={A m,1 A m,2 ,...,A m,n ,...,A m,N}, A m,n Let represent the original power data sequence of the m-th adjustable resource individual on day n, and Let represent the raw power data of the m-th adjustable resource individual at time t on day n. Then, the raw power data of the m-th adjustable resource individual at time t on day N is denoted as . δ m Let represent the set of incentive intensity data for the m-th adjustable resource individual, and δ m ={δ m,1 ,δ m,2 ,...,δ m,t ,...,δ m,T}, δ m,t This represents the incentive intensity data for the m-th adjustable resource individual at the t-th time point each day;

[0052] Step 2: Decompose the raw power data of individual adjustable resources using discrete wavelet transform. Wavelet decomposition effectively reduces noise and interference. Smoothing the high-frequency components of the original sequence further enhances the accuracy and reliability of the data.

[0053] Using the db6 wavelet function to Decompose into K a This allows us to obtain a high-frequency signal from the m-th adjustable resource individual at the t-th time point on day N. And K of the m-th adjustable resource individual at the t-th time point on day N. a Low-frequency signal of layer in, This represents the k-th time point of the m-th adjustable resource individual at the t-th time point on day N. a Low-frequency signals in the layer, and This represents the k-th time point of the m-th adjustable resource individual on day n at time t. a Low-frequency raw power data; thus obtaining high-frequency signals of M adjustable resource individuals at time point t on day N. And M adjustable resource individuals at time point t on day N. a The low-frequency signal of the layer is denoted as in, This represents the k-th time point of M adjustable resource individuals at time t on day N. a Low-frequency signals in the layer, and in, This represents the high-frequency raw power data of the m-th adjustable resource individual at time t on day n.

[0054] Step 3: Classify the high-frequency and low-frequency signals after decomposition of the adjustable resource cluster using K-medoids clustering based on the DTW algorithm. DTW minimizes the total distance between two time series observations. It achieves this by locally stretching or compressing the sequences to make them as similar as possible. Since the time alignment between sequences is not fixed, the DTW algorithm can detect periodic changes in the lead-lag correlation between time series, thus providing accurate dynamic analysis.

[0055] Step 3.1: Cluster the high-frequency signals:

[0056] Step 3.1.1, in B t Randomly select N k If the high-frequency signal of an adjustable resource individual at time point t on day N is used as the high-frequency clustering center individual, then the remaining MN k The high-frequency signal of an adjustable resource individual at the t-th time point on N days is taken as the high-frequency non-clustering center individual;

[0057] Step 3.1.2: Solve for the m-th equation using equations (1) and (2). a A high-frequency non-clustering center individual and the mth b Individuals with high-frequency cluster centers DTW distance

[0058]

[0059]

[0060] In equations (1) and (2), for The Nth signal value, for The Nth signal value, for and The current distance, and for and The cumulative distance; for The (N-1)th signal value and The (N-1)th signal value The cumulative distance; for The (N-1)th signal value and The Nth signal value The cumulative distance, for The Nth signal value and the (N-1)th signal value The cumulative distance;

[0061] Step 3.1.3: Calculate the DTW distance between all high-frequency non-clustered center individuals and each high-frequency clustered center individual at time point t, and assign the high-frequency non-clustered center individual with the smallest DTW distance to the corresponding high-frequency clustered center individual, thereby obtaining the well-clustered high-frequency population at time point t.

[0062] Step 3.1.4: Calculate the high-frequency cluster centers of the clustered high-frequency population at time point t using the K-medoids algorithm, and obtain the updated high-frequency cluster center individuals;

[0063] Step 3.1.5: Submit the updated high-frequency cluster centers to step 3.1.2 sequentially until the maximum number of iterations is reached, thus obtaining the final updated N at time point t.k Individuals with high-frequency cluster centers in, This represents the m-th update at time point t. b A high-frequency cluster center individual;

[0064] Step 3.2, from Randomly select N k An individual with adjustable resources at time point k on day t. a The low-frequency signal of layer k is used as the kth layer a If the low-frequency cluster center individuals are in the layer, then MN remain. k An individual with adjustable resources at time point k on day t. a The low-frequency signal of layer k is used as the kth layer a The low-frequency non-clustering center individuals are identified; then, following steps 3.1.2-3.1.5, the final updated k-th time point at time t is obtained. a Low-frequency cluster center individuals in, This represents the k-th update at time point t. a Layer m b 1. 2. 3. 4. 5. 5. 6. 7. 8. 9. ... a Low-frequency cluster center individuals

[0065] Step 3.3 will and After merging, the high-frequency signal and K at time point t are obtained. a Cluster center set of low-frequency signals in, This indicates the k-th time point at time t. e Individual cluster centers;

[0066] Step 4: Extract feature values ​​from the cluster center data using the AE-LSTM algorithm, calculate the general and random parameters of the potential assessment model, and compute the probability distribution of the response potential of adjustable resource individuals. The AE-LSTM algorithm model can effectively handle high-dimensional, nonlinear, and sparsity problems. Due to the encoder-decoder architecture, unsupervised learning can be achieved through this model, and the original sequence can be reconstructed based on the extracted information. A relatively low-dimensional representation of the time series can be learned well from the stacked LSTM modules located behind the encoder.

[0067] Step 4.1: Constructing the AE-LSTM neural network includes: an input layer, an encoder LSTM network layer, and a decoder LSTM network layer; used to extract the cluster center set. At time point t, the k-th time... e Individual cluster centers eigenvalues;

[0068] Step 4.1.1: Set the threshold parameter to L. mse Set the sliding window length parameter to n a ;

[0069] The input layer uses a sliding window algorithm to... Convert to Nn a +1 vector, denoted as the k-th vector at time point t. e Sliding feature matrix of an individual in, This indicates the k-th time point at time t. e Individual cluster centers The nth a vectors, and express v in the middle a One data point;

[0070] Step 4.1.2, Encoder LSTM network layer pair Encode the k-th digit. e Individual cluster centers Feature matrix Among them, L a The output dimension of the LSTM neural network;

[0071] Step 4.1.3, Decoder LSTM network layer pair Perform decoding to obtain the k-th... e Reduction matrix of each cluster center individual

[0072] Step 4.1.4, Construction and The mean squared error loss is calculated, and the parameters of the AE-LSTM neural network are trained using gradient descent until the mean squared error loss is less than L. mse At that time, the optimal AE-LSTM model after training is obtained, and the optimal feature matrix is ​​obtained from the encoder layer LSTM network layer in the optimal AE-LSTM model after training. Extract the k-th time point at time t e Individual feature vectors in, express The eigenvalue of the i-th row and j-th column;

[0073] Step 4.2: Collect data separately The basic threshold for the incentive intensity response generated by the individual adjustable resource. The upper threshold of the individual's response to incentive intensity for the available adjustable resources Maximum response rate of the individual with available adjustable resources

[0074] Step 4.3: Based on the least squares method... and respectively with By performing fitting, the values ​​at time point t are obtained. The kth e Individual cluster centers The basic threshold fitting function of the excitation intensity response At time point t The kth e Individual cluster centers Upper threshold fitting function of the excitation intensity response At time point t The kth e individual Maximum response rate fitting function

[0075] Step 4.4, in Reselect Repeat steps 4.1-4.3 until all high-frequency signals and K at time point t are obtained. a Fitting function for cluster centers of low-frequency signals in layers;

[0076] Step 4.4: Decompose the high-frequency data of the raw power at time point t over N days and K... a The feature vectors obtained from the low-frequency data of each layer are substituted into the basic threshold fitting function of the cluster center to which they belong, and the corresponding results are superimposed to obtain the basic threshold of the m-th adjustable resource individual at the t-th time point using equation (3).

[0077] The high-frequency data of the raw power decomposition at time point t on N days and K a The feature vectors obtained from the low-frequency data of each layer are substituted into the upper limit threshold fitting function of the cluster center to which they belong, and the final results are superimposed to obtain the upper limit threshold of the m-th adjustable resource individual at the t-th time point using equation (3).

[0078] The high-frequency data of the raw power decomposition at time point t on N days and K a The feature vectors obtained from the low-frequency data are substituted into the maximum response rate fitting function of the cluster center to which they belong, and the final results are superimposed to obtain the maximum response rate of the m-th adjustable resource individual at the t-th time point using equation (5).

[0079]

[0080]

[0081]

[0082] In equations (3)-(5), It is the cluster center to which the high-frequency signal extracted by the m-th adjustable resource individual at time t belongs after clustering. It is the cluster center to which the low-frequency signal of the z-th layer of the m-th adjustable resource individual at time t belongs after clustering. Let represent the feature value of the high-frequency signal extracted by the m-th adjustable resource individual at time t. This represents the feature value of the low-frequency signal of the z-th layer extracted by the m-th adjustable resource individual at time t.

[0083] Step 4.6: Different types of users respond differently to electricity price fluctuations and their sensitivity to electricity prices, resulting in varying electricity consumption behaviors. According to consumer psychology models, changes in electricity consumption behavior during demand response also follow certain patterns. The relationship between the load reduction rate caused by electricity consumers' participation in demand response and the corresponding incentive intensity can be described as a piecewise linear function. The load reduction rate η of the m-th adjustable resource individual at time point t in the collected historical data is calculated. m,t The probability distribution is obtained using the point estimation method. Where N represents a normal distribution, μ m,t σ m,t Let the mean and standard deviation of the load reduction rate of the m-th adjustable resource individual at time point t be given respectively; and we have:

[0084]

[0085]

[0086] In equations (6) and (7), ξ t,m Let be the potential assessment parameters for the m-th adjustable resource individual at time point t. Let be the mean and standard deviation of the potential assessment parameters for the m-th adjustable resource individual at time point t;

[0087] Step 4.7: Calculate the incentive intensity of the m-th adjustable resource individual at time point t using equation (5). Response potential

[0088]

[0089] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0090] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A method for evaluating the response potential of an adjustable resource cluster based on the AE-LSTM algorithm, characterized in that, The procedure is as follows: Step 1: Sample raw power data and excitation intensity data at T time points daily from an adjustable resource cluster containing M adjustable resource individuals, thereby obtaining a raw power data set A = {A1, A2, ..., A...} for N days. m ,...,A M } and the set of excitation intensity data δ={δ1,δ2,...,δ m ,...,δ M }, where A m Let A represent the set of original power data for the m-th adjustable resource individual, and A m ={A m,1 A m,2 ,...,A m,n ,...,A m,N }, A m,n Let represent the original power data sequence of the m-th adjustable resource individual on day n, and Let represent the raw power data of the m-th adjustable resource individual at time t on day n. Then, the raw power data of the m-th adjustable resource individual at time t on day N is denoted as . δ m Let represent the set of incentive intensity data for the m-th adjustable resource individual, and δ m ={δ m,1 ,δ m,2 ,...,δ m,t ,...,δ m,T }, δ m,t This represents the incentive intensity data for the m-th adjustable resource individual at the t-th time point each day; Step 2: Decompose the raw power data of individual adjustable resources using discrete wavelet transform: Using the db6 wavelet function to Decompose into K a This allows us to obtain a high-frequency signal from the m-th adjustable resource individual at the t-th time point on day N. And K of the m-th adjustable resource individual at the t-th time point on day N. a Low-frequency signal of layer in, This represents the k-th time point of the m-th adjustable resource individual at the t-th time point on day N. a Low-frequency signals in the layer, and This represents the k-th time point of the m-th adjustable resource individual on day n at time t. a Low-frequency raw power data; thus obtaining high-frequency signals of M adjustable resource individuals at time point t on day N. And M adjustable resource individuals at time point t on day N. a The low-frequency signal of the layer is denoted as in, This represents the k-th time point of M adjustable resource individuals at time t on day N. a Low-frequency signals in the layer, and in, This represents the high-frequency raw power data of the m-th adjustable resource individual at time t on day n. Step 3: Classify the high-frequency and low-frequency signals after decomposition of the adjustable resource cluster using K-medoids clustering based on the DTW algorithm: Step 3.1, for B t Clustering is performed on the high-frequency signals to obtain the final updated N at time point t. k Individuals with high-frequency cluster centers in, This represents the m-th update at time point t. b A high-frequency cluster center individual; Step 3.2, from Randomly select N k An individual with adjustable resources at time point k on day t. a The low-frequency signal of layer k is used as the kth layer a If the low-frequency cluster center individuals are in the layer, then MN remain. k An individual with adjustable resources at time point k on day t. a The low-frequency signal of layer k is used as the kth layer a The low-frequency non-clustering center individuals are identified; then, following steps 3.1.2-3.1.5, the final updated k-th time point at time t is obtained. a Low-frequency cluster center individuals in, This represents the k-th update at time point t. a Layer m b 1.

2.

3.

4.

5.

5.

6.

7.

8.

9. ... a Low-frequency cluster center individuals Step 3.3, will and After merging, the high-frequency signal and K at time point t are obtained. a Cluster center set of low-frequency signals in, This indicates the k-th time point at time t. e Individual cluster centers; Step 4: Use the AE-LSTM algorithm to extract feature values ​​from the cluster center data after clustering, calculate the general and random parameters of the potential assessment model, and calculate the probability distribution of the response potential of individual adjustable resources. Step 4.1: Constructing the AE-LSTM neural network includes: an input layer, an encoder LSTM network layer, and a decoder LSTM network layer; used to extract the cluster center set. At time point t, the k-th time... e Individual cluster centers eigenvalues in, Represents the optimal characteristic matrix The eigenvalue of the i-th row and j-th column; Step 4.2: Collect data separately The basic threshold for the incentive intensity response generated by the individual adjustable resource. The upper threshold of the individual's response to incentive intensity for the available adjustable resources Maximum response rate of the individual with available adjustable resources Step 4.3: Based on the least squares method... and respectively with By performing fitting, the values ​​at time point t are obtained. The kth e Individual cluster centers The basic threshold fitting function of the excitation intensity response At time point t The kth e Individual cluster centers Upper threshold fitting function of the excitation intensity response At time point t The kth e individual Maximum response rate fitting function Step 4.4, in Then, reselect a cluster center at time point t and repeat steps 4.1-4.3 until all high-frequency signals and K at time point t are obtained. a Fitting function for cluster centers of low-frequency signals in layers; Step 4.5: Decompose the high-frequency data of the raw power at time point t over N days and K... a After calculating the eigenvectors of the low-frequency data in each layer, they are substituted into the basic threshold fitting function of the cluster center to which the data belongs. The results are then superimposed to obtain the basic threshold of the m-th adjustable resource individual at time t. The high-frequency data of the raw power decomposition at time point t on N days and K a After calculating the feature vectors of the low-frequency data in each layer, they are substituted into the upper limit threshold fitting function of the cluster to which the data belongs, and the results are superimposed to obtain the upper limit threshold of the m-th adjustable resource individual at time t. The high-frequency data of the raw power decomposition at time point t on N days and K a After calculating the eigenvectors of the low-frequency data in each layer, they are substituted into the maximum response rate fitting function of the cluster to which the data belongs. The results are then superimposed to obtain the maximum response rate of the m-th adjustable resource individual at time t. Step 4.6: Collect the load reduction rate η of the m-th adjustable resource at the t-th time point in the historical data. m,t The probability distribution is obtained using the point estimation method. Where N represents a normal distribution, μ m,t σ m,t Let the mean and standard deviation of the load reduction rate of the m-th adjustable resource individual at time point t be given respectively; and we have: In equations (6) and (7), ξ t,m Let be the potential assessment parameters for the m-th adjustable resource individual at time point t. Let be the mean and standard deviation of the potential assessment parameters for the m-th adjustable resource individual at time point t; Step 4.7: Calculate the incentive intensity of the m-th adjustable resource individual at time point t using equation (5). Response potential 2. The method for evaluating the multi-time-node response potential of an adjustable resource cluster based on AE-LSTM according to claim 1, characterized in that, Step 3.1 includes: Step 3.1.1, in B t Randomly select N k If the high-frequency signal of an adjustable resource individual at time point t on day N is used as the high-frequency clustering center individual, then the remaining MN k The high-frequency signal of an adjustable resource individual at the t-th time point on N days is taken as the high-frequency non-clustering center individual; Step 3.1.2: Solve for the m-th equation using equations (1) and (2). a A high-frequency non-clustering center individual and the mth b Individuals with high-frequency cluster centers DTW distance In equations (1) and (2), for The Nth signal value, for The Nth signal value, for and The current distance, and for and The cumulative distance; for The (N-1)th signal value and The (N-1)th signal value The cumulative distance; for The (N-1)th signal value and The Nth signal value The cumulative distance, for The Nth signal value and the (N-1)th signal value The cumulative distance; Step 3.1.3: Calculate the DTW distance between all high-frequency non-clustered center individuals and each high-frequency clustered center individual at time point t, and assign the high-frequency non-clustered center individual with the smallest DTW distance to the corresponding high-frequency clustered center individual, thereby obtaining the well-clustered high-frequency population at time point t. Step 3.1.4: Calculate the high-frequency cluster centers of the clustered high-frequency population at time point t using the K-medoids algorithm, and obtain the updated high-frequency cluster center individuals; Step 3.1.5: Submit the updated high-frequency cluster centers to step 3.1.2 sequentially until the maximum number of iterations is reached, thus obtaining the final updated N at time point t. k Individuals with high-frequency cluster centers in, This represents the m-th update at time point t. b A high-frequency cluster center individual.

3. The method for evaluating the multi-time-node response potential of an adjustable resource cluster based on AE-LSTM according to claim 2, characterized in that, Step 4.1 includes: Step 4.1.1: Set the threshold parameter to L. mse Set the sliding window length parameter to n a ; The input layer uses a sliding window algorithm to... Convert to Nn a +1 vector, denoted as the k-th vector at time point t. e Sliding feature matrix of an individual in, This indicates the k-th time point at time t. e Individual cluster centers The nth a vectors, and express v in the middle a One data point; Step 4.1.2, the encoder LSTM network layer... Encode the k-th digit. e Individual cluster centers Feature matrix Among them, L a The output dimension of the LSTM neural network; Step 4.1.3, the decoder LSTM network layer pair Perform decoding to obtain the k-th... e Reduction matrix of each cluster center individual Step 4.1.4, Construction and The mean squared error loss is calculated, and the parameters of the AE-LSTM neural network are trained using gradient descent until the mean squared error loss is less than L. mse At that time, the optimal AE-LSTM model after training is obtained, and the optimal feature matrix is ​​obtained from the encoder layer LSTM network layer in the optimal AE-LSTM model after training. Extract the k-th time point at time t e Individual feature vectors in, Represents the optimal characteristic matrix The eigenvalue of the i-th row and j-th column.

4. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store programs that support the processor in executing the multi-time-node response potential assessment method for any of claims 1-3, and the processor is configured to execute the programs stored in the memory.

5. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by a processor, performs the steps of the method for assessing the multi-time-node response potential of an adjustable resource cluster as described in any of claims 1-3.

Images