A method, system and product for evaluating the collaborative regulation potential of multi-source user-side resources

Abstract

The present invention belongs to the field of power systems, and provides a method, a system and a product for evaluating the collaborative regulation potential of multiple user-side resources. The technical solution is based on the short-term prediction data and historical data of distributed photovoltaic and user-level flexible load user-side resources to evaluate the upward and downward regulation potential of a single user-side resource; then, considering the sensitivity of different user-side resources to electricity price signals, the regulation potential is corrected; finally, considering the synergistic effect of user-side resources within the aggregator, the collaborative regulation potential of user-side resources within the aggregator is evaluated and calculated. This method breaks through the problem that the traditional method for evaluating the regulation ability of multiple user-side resources relies on equipment modeling and has a large workload, and provides a new idea for evaluating the adjustable ability of multiple user-side resources.

Description

Technical Field

[0001] The present invention belongs to the field of evaluating the collaborative regulation potential of user-side resources in the power system, and particularly relates to a method, a system and a product for evaluating the collaborative regulation potential of multi-source user-side resources. Background Art

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] With the social development and industrial structure adjustment, the proportion of new energy is increasing continuously, and the peak-valley difference of the power grid is increasing continuously. Due to its flexible regulation ability, user-side resources contribute to solving many problems faced by the current power system. Therefore, evaluating and predicting the regulation potential of user-side resources is the key to grid optimal dispatching.

[0004] To ensure the effective utilization of the load regulation capabilities of multiple user sides such as industry, commerce, and residents, it is necessary to master their power characteristics. Therefore, how to accurately predict their loads is the key to studying the load regulation potential of different industries' user sides.

[0005] Current load forecasting methods are mainly divided into three categories: traditional methods, similar-day methods, and artificial intelligence model forecasting. Traditional load forecasting methods mainly analyze and statistically process historical load data and their influencing factors, and establish a mathematical model of the load change trend to predict the load data in the future time period. As the scale of the power system increases, traditional methods are increasingly difficult to meet the needs of modern large systems. The similar-day method is based on the periodic characteristics of electricity consumption loads and directly uses the load data of historical similar days as the forecasting results. However, affected by factors such as social economy and climate change, the characteristics of electricity consumption loads are becoming increasingly complex and changeable. In this context, relying solely on similar days will not provide completely accurate information. In recent years, the rise of artificial intelligence algorithms such as artificial neural networks and ensemble learning has provided new ideas for load forecasting. Artificial intelligence algorithms can make full use of the advantages of machine learning to capture the complex non-linear relationships between power loads and various influencing factors. Through continuous training and updating, these algorithms can adapt to the changes in load data and environmental conditions, and thus have broad application prospects in the context of current massive load data. However, in current artificial intelligence-based load forecasting research, usually only the influencing factors are input into the forecasting model and then the load forecasting results are directly output, without considering the differences in load characteristics under different weather scenarios, and also without making full use of the advantages of the similar-day method in determining the load change trend. In addition, a single forecasting model cannot exhibit good forecasting performance in all scenarios.

[0006] Distributed photovoltaic power prediction on the user side has been one of the research hotspots in recent years. Accurate power prediction is of great significance for evaluating its regulation potential. Traditional photovoltaic power prediction methods mainly include physical models and data-driven methods. Physical models are built based on the physical characteristics of photovoltaic systems, but they are highly dependent on meteorological data and have high model complexity. Data-driven methods train models through historical data and have strong adaptability and flexibility. In recent years, with the development of machine learning and deep learning technologies, various data-driven methods have been applied to distributed photovoltaic power prediction. For example, methods such as support vector machine (SVM), artificial neural network (ANN), and extreme learning machine (ELM) have achieved good results in photovoltaic power prediction. However, when dealing with large-scale distributed photovoltaic power plants, these methods often face problems such as large amounts of data and high computational complexity.

[0007] Currently, research has been carried out on the adjustable potential of multiple user-side resources. Some scholars have proposed an evaluation method for the adjustable potential of power system loads based on interruptibility, transferability, and product characteristics, and evaluated the adjustable potential of industrial users based on the correlation weighting method. Some scholars have constructed a load response model for residential communities and predicted the regulation potential of residential community loads using the probability distribution method. There are also scholars who have constructed an evaluation index system for adjustable potential containing multiple features by exploring the adjustable potential of different types of industrial users, and carried out industrial load potential evaluation using the VMD-TCN potential analysis method. However, most of the current evaluations of the adjustable potential of multiple user-side resources start from the perspective of equipment models. This method requires the establishment of a large number of equipment models, and the types of parameters required for equipment models such as equipment types, quantities, and operating characteristics are numerous and not easy to obtain, which has become a major problem in evaluating the adjustable potential. Summary of the Invention

[0008] To solve at least one of the technical problems existing in the above background technology, the present invention provides a method, system, and product for evaluating the collaborative regulation potential of multiple user-side resources, which breaks through the problem that the traditional evaluation method of the regulation ability of multiple user-side resources depends on equipment modeling and has a large workload, and efficiently completes the evaluation of the adjustable ability of multiple user-side resources.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] The first aspect of the present invention provides a method for evaluating the collaborative regulation potential of multiple user-side resources, including the following steps:

[0011] Based on the obtained power consumption load characteristic data, similar-day load data of various users, and the constructed integrated learning load prediction model for multiple users, obtain the load prediction results for each type of user;

[0012] Cluster distributed photovoltaic power stations, determine the source domain and target domain based on the clustering results, establish and train a short-term photovoltaic power prediction model based on the source domain data; transfer the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain to obtain a short-term photovoltaic power prediction model applicable to the target domain, and further predict to obtain distributed photovoltaic power data.

[0013] Combine multi-source user load data and distributed photovoltaic data to establish the regulation potential of a single user-side resource. Considering the sensitivity of different user-side resources to electricity price signals, correct the regulation potential, and consider the synergistic effect of user-side resources within the aggregator to calculate the coordinated regulation potential of user-side resources within the aggregator.

[0014] Furthermore, the load prediction results of each type of user are obtained based on the acquired electricity load characteristic data of various users, similar-day load data, and the constructed multi-source user integrated learning load prediction model, including:

[0015] Construct a training set and a prediction set based on the acquired electricity load characteristic data of various users and similar-day load data. Train multiple base learners based on the training set, and then predict the trained base learners based on the prediction set to obtain a prediction result dataset.

[0016] Use the prediction result dataset as a secondary training dataset and a prediction dataset. Train the meta-learner based on the secondary training dataset, and predict the load prediction results of various users based on the prediction dataset and the trained meta-learner.

[0017] Furthermore, the clustering of distributed photovoltaic power stations and the determination of the source domain and target domain based on the clustering results include using the AP clustering algorithm to cluster distributed photovoltaic power stations, and defining the clustering center of each cluster as the source domain according to the clustering results, and classifying the remaining power stations as the target domain.

[0018] Furthermore, the short-term photovoltaic power prediction model uses a Transformer neural network.

[0019] Furthermore, when transferring the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain, the model parameters are adjusted with the goal of minimizing the RMSE and MAPE.

[0020] Furthermore, combining multi-source user load data and distributed photovoltaic data to establish the regulation potential of a single user-side resource includes:

[0021] Calculate the baseline output of each user-side resource in different seasons;

[0022] Take the maximum value that satisfies the principle and the minimum value as t the maximum baseline output and the minimum baseline output at the moment;

[0023] Based on the future t Based on the predicted values of different types of user-side resources and the maximum baseline output at the moment, the flexibility up-regulation potential of different types of user-side resources is calculated. Based on the predicted values of different types of user-side resources and the minimum baseline output at the future t moment, the flexibility down-regulation potential of different types of user-side resources is obtained.

[0024] Further, the adjusted regulation potential after correcting the regulation potential is:

[0025] ,

[0026] ,

[0027] wherein, and are the adjusted up-regulation potential and down-regulation potential, and are the t flexibility up-regulation potential and down-regulation potential of different types of user-side resources at the moment before correction, is the load demand at time t before demand response, is the change in load demand at time t before and after participating in demand response.

[0028] Further, the coordinated regulation potential is:

[0029] ,

[0030] ,

[0031] wherein, is the user-side resource i adjusted up-regulation potential, is the user-side resource i adjusted down-regulation potential, is the distributed photovoltaic, is the load.

[0032] The second aspect of the present invention provides a multi-user-side resource coordinated regulation potential evaluation system, including:

[0033] A load prediction module, which is used to obtain the load prediction result of each type of user based on the acquired power consumption load characteristic data of various users, similar-day load data, and the constructed multi-user integrated learning load prediction model;

[0034] A photovoltaic power prediction module, which is used to cluster distributed photovoltaic power stations, determine the source domain and the target domain based on the clustering results, establish and train a short-term photovoltaic power prediction model based on the source domain data; transfer the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain to obtain a short-term photovoltaic power prediction model suitable for the target domain, and further predict to obtain distributed photovoltaic power data;

[0035] An adjustment potential evaluation module, which is used to establish the adjustment potential of a single user-side resource by combining multi-user load data and distributed photovoltaic data, correct the adjustment potential considering the sensitivity of different user-side resources to electricity price signals, and calculate the coordinated adjustment potential of the user-side resources within the aggregator considering the synergistic effect of the user-side resources within the aggregator.

[0036] The third aspect of the present invention provides a program product.

[0037] A program product, which is a computer program product, includes a computer program, and when the computer program is executed by a processor, it implements the steps in a method for evaluating the coordinated adjustment potential of multi-user-side resources as described above.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] 1. Based on the short-term prediction curves and historical data of user-side resources such as distributed photovoltaics and user-level flexible loads, the present invention evaluates the upward and downward adjustment potentials of a single user-side resource; then, considering the sensitivity of different user-side resources to electricity price signals, the adjustment potential is corrected; finally, considering the synergistic effect of the user-side resources within the aggregator, the coordinated adjustment potential of the user-side resources within the aggregator is evaluated and calculated. This method breaks through the problem that the traditional method for evaluating the adjustment ability of multi-user-side resources depends on equipment modeling and has a large workload, and provides a new idea for evaluating the adjustable ability of multi-user-side resources.

[0040] 2. Considering that a single model cannot achieve good prediction performance in various prediction scenarios, the present invention separately establishes an ensemble learning load prediction model for each type of user, including industrial, commercial, and residential users, and fuses multiple base learners to obtain a strong learning model, thereby improving the prediction accuracy.

[0041] 3. Aiming at the problem that the separate prediction calculation requirement for each site of the distributed photovoltaic power prediction model is large, the present invention proposes a distributed photovoltaic power prediction method based on AP clustering transfer learning, constructs a short-term power prediction model based on the Transformer neural network, and avoids redundant training costs through transfer learning based on AP clustering, effectively improving the prediction accuracy of distributed photovoltaic power prediction in complex prediction scenarios.

[0042] Advantages of additional aspects of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not unduly limit the present invention.

[0044] Figure 1 is a flowchart of a method for evaluating the collaborative regulation potential of multi-user side resources provided by an embodiment of the present invention;

[0045] Figure 2 is a flowchart of load forecasting based on multi-model Stacking ensemble learning provided by an embodiment of the present invention;

[0046] Figure 3 is a flowchart of transfer learning provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further descriptions of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0049] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0050] As mentioned in the background art, most of the current adjustable potential evaluations for multi-user side resources start from the perspective of device models, and this method requires the establishment of a large number of device models. The types of parameters required for device models, such as device types, quantities, and operating characteristics, are numerous and not easily obtained, which becomes a major problem when evaluating the adjustable potential. Based on the short-term prediction curves and historical data of user-side resources such as distributed photovoltaics and user-level flexible loads, the present invention evaluates the upward and downward adjustment potentials of a single user-side resource; then, considering the sensitivity of different user-side resources to electricity price signals, the adjustment potential is corrected; finally, considering the synergistic effect of user-side resources within the aggregator, the collaborative adjustment potential of user-side resources within the aggregator is evaluated and calculated. This method breaks through the problem that the traditional evaluation method of the adjustable capacity of multi-user side resources relies on device modeling and has a large workload, and provides a new idea for the evaluation of the adjustable capacity of multi-user side resources.

[0051] Embodiment 1

[0052] As Figure 1 shown, this embodiment provides a method for evaluating the collaborative adjustment potential of multi-user side resources, including the following steps:

[0053] Step 1: Obtain the load prediction results of each type of user based on the acquired electricity load characteristic data and the constructed multi-user integrated learning load prediction model;

[0054] As Figure 2 shown, it specifically includes the following steps:

[0055] Step 101: Obtain the electricity load characteristic data of various users and the load data of similar days

[0056] In this embodiment, the electricity load characteristic data includes meteorological characteristic data such as apparent temperature, humidity, irradiance, instantaneous precipitation, and wind speed that affect the electricity load;

[0057] Considering the periodic characteristics of the electricity load, the input features also include time tags and the day-of-week numbers of the current day. In addition, since the electricity loads of adjacent dates are relatively similar, when predicting the load of the th day, the load data from the th to the th day are also used as input features to better predict the change trend of the daily load and thus improve the accuracy.

[0058] Step 102: Construct a training set and a prediction set based on the acquired electricity load characteristic data of various users and the load data of similar days, train multiple base learners constructed based on the training set, and then predict the trained base learners based on the prediction set to obtain a prediction result dataset;

[0059] In this embodiment, the base learners adopt Support Vector Machine (SVM) and Deep Neural Network (DNN).

[0060] Among them, the Support Vector Machine (SVM) maps the non - linear problem in the low - dimensional space to the high - dimensional space by introducing the kernel method, so that a linear method can be used to solve it. The optimization goal of SVM is to find a function such that the prediction error is minimized within the range of , where is the non - linear mapping function, and the RBF kernel function is adopted in this embodiment.

[0061] The optimization goal and constraint conditions of SVM are shown as follows:

[0062] ,

[0063] In the formula, and b represent the weight vector and the bias term respectively, and are slack variables, which are used to handle the errors outside the - insensitive loss, C is the penalty parameter, is the i th input feature, is the i th prediction result;

[0064] When training the SVM, the SVM adopts the - insensitive loss function, the purpose of which is not to calculate the loss within the error range of to improve the robustness of the model. The expression of the

[0065] ,

[0066] This loss function means that when the error between the predicted value and the actual value y is less than or equal to , the loss is not counted, and the loss is only counted when it is greater than .

[0067] Among them, the Deep Neural Network (DNN) includes an input layer, multiple hidden layers and an output layer. After the input layer receives the feature vector composed of input features, the hidden layer passes through the weight matrix and the bias vector Process the input and calculate the result of the linear transformation :

[0068] ,

[0069] ,

[0070] wherein is the activation function; is the output of the l th layer, representing the result after the activation function acts.

[0071] The output layer generates the final prediction result. For a regression problem, the output layer does not use an activation function but directly outputs a continuous real value:

[0072] ,

[0073] wherein is the predicted value, and are the weight matrix and bias vector of the output layer respectively, is the output of the penultimate layer.

[0074] During training, the DNN uses the gradient descent algorithm to optimize the weights and biases to minimize the loss function. The above units work together to enable the DNN to learn the complex mapping relationship between the input features and the target output, thereby achieving accurate prediction.

[0075] Step 103: Use the prediction result data set as the secondary training data set and the prediction data set, train the meta-learner based on the secondary training data set, and predict the load prediction results of various users based on the prediction data set and the trained meta-learner;

[0076] In this embodiment, the meta-learner adopts an extreme gradient boosting (XGBoost). Its basic idea is to gradually add new decision trees to correct the errors of the previous round of learners, thereby improving the prediction performance of the overall model.

[0077] The objective function of XGBoost consists of a loss function and a regularization term:

[0078] ,

[0079] wherein is used to represent the parameter set; is the loss function, is the regularization term, and the two can be obtained from the following two formulas:

[0080] ,

[0081] ,

[0082] where M is the number of leaf nodes of the tree, is the weight of the j th leaf node, the table is the t th tree; and are regularization parameters, , are the i th true value and predicted value respectively.

[0083] The predicted values of each tree are weighted to obtain the final predicted value:

[0084] ,

[0085] where T is the total number of trees.

[0086] Using XGBoost as the final meta-model has the following advantages: First, it is good at handling large-scale data sets, and can significantly improve the training speed and efficiency by using parallel processing and cache optimization; Second, by introducing regularization terms, it can effectively prevent overfitting and improve the generalization ability of the model; Finally, XGBoost supports undersampling and multiple loss functions, enhancing its flexibility and adaptability in different tasks and data sets.

[0087] Considering that a single model cannot achieve good prediction performance in various prediction scenarios, the above solution builds an ensemble learning load prediction model for each type of user, namely industrial, commercial, and residential users, and fuses multiple base learners to obtain a strong learning model, thereby improving the prediction accuracy.

[0088] Step 2: Cluster the distributed photovoltaic power stations, define the clustering center of each cluster as the source domain according to the clustering results, and classify the remaining power stations into the target domain; establish and train a short-term photovoltaic power prediction model based on the source domain data; transfer the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain to obtain a short-term photovoltaic power prediction model applicable to the target domain;

[0089] Aiming at the problem of large computational requirements for individual prediction of each site in the distributed photovoltaic power prediction model, a short-term photovoltaic power prediction model based on the Transformer neural network is constructed, and redundant training costs are avoided through transfer learning based on AP clustering, effectively improving the prediction accuracy of distributed photovoltaic power prediction in complex prediction scenarios.

[0090] As Figure 3As shown in the figure, it specifically includes the following steps:

[0091] Step 201: Cluster the distributed photovoltaic power stations, and define the clustering center of each cluster as the source domain according to the clustering results, and classify the remaining power stations into the target domain;

[0092] In this embodiment, the AP clustering algorithm is used to cluster the distributed photovoltaic power stations; AP clustering is an unsupervised clustering algorithm based on information transmission for cluster division, which can adapt to various types of data distributions, is not affected by the selection of the initial center, and can effectively identify "consistent output" power stations when facing complex and non-linear distributed photovoltaic power station data in reality, and the clustering results can be more stable. The core idea of the AP clustering algorithm is that there is no need to pre-specify the number of clusters, all data points are used as potential clustering centers, and each data point sends two types of information to other points: attraction information and belonging information, and the clustering center is determined by iterating these two types of information. The attraction information represents the degree to which a power station is suitable as a power station relative to other candidate clustering centers, and the information is transmitted from the power station to the candidate clustering center power station ; the belonging information represents the degree to which a power station chooses the power station as the clustering center, and is sent from the candidate clustering center power station to the power station . For each data point , its clustering center is the point that makes the largest.

[0093] Specifically, it includes the following steps:

[0094] Step 2011: After preprocessing the historical power data of the existing distributed photovoltaic power stations, calculate the initial similarity between the power stations , and obtain the similarity matrix . In this embodiment, the DTW distance is selected as the initial similarity between two photovoltaic power stations.

[0095] Step 2012: Set the initial reference degree and the number of iterations. The diagonal elements of the matrix characterize the reference degree for a power station to become a clustering center. The larger this value is, the higher the possibility of this data point being a clustering center, and the number of final clustering centers will also increase. Calculated according to the DTW distance, its value is 0, but in the AP clustering algorithm, the degree of a power station as a clustering center cannot be 0. Therefore, before the iteration starts, it is assumed that all points have the same ability to become clustering centers, and the reference degree is generally set to the minimum value or median of all values in the similarity matrix.

[0096] Step 2013: Initially, the algorithm needs to initialize the attraction matrix and the membership matrix as zero matrices, and then calculate the elements of the attraction matrix and the elements of the membership matrix . The calculation formulas are as follows:

[0097] ,

[0098] ,

[0099] ,

[0100] In the formula, is all elements except the th column in the th row of the similarity matrix; represents all elements except the th column in the th row of the membership matrix; represents the sum of all elements greater than zero except the th row in the th column of the attraction matrix; The self-membership is equal to the sum of the positive attractions obtained from other points, is the element in the k th row and k th column of the attraction matrix, is the element in the i th row and k th column of the similarity matrix;

[0101] Step 2014: Update and . In the iterative process, in order to ensure the convergence speed and its stability, a damping coefficient needs to be introduced for adjustment, which is expressed by the formula:

[0102] ,

[0103] where represents the number of iterations.

[0104] Step 2015: After each update of the attraction information and the membership information, select the cluster center for each sample point. Repeat the update of the attraction matrix and the membership matrix until the matrix change tends to be stable or reaches the predetermined number of iterations. Calculate the silhouette coefficient according to the clustering result. Change the reference degree, recalculate, and confirm the best clustering result to complete the division of the photovoltaic power station group.

[0105] Step 2016: To determine the best clustering result for the division of the photovoltaic power station and ensure the effectiveness of the clustering result, the Silhouette coefficient (SC) is selected as the evaluation function for the clustering result, and its formula is as follows:

[0106] ,

[0107] In the formula, is the silhouette coefficient of the power station sample points after clustering; is the average distance between the power station sample points and other power station sample points in the same cluster, which is called the cohesion degree and indicates the tightness of this point with this cluster; is the average distance between the power station sample points and all samples in other clusters, which is called the separation degree and indicates the degree of alienation of this point from other clusters. Each sample point corresponds to a silhouette coefficient value, and the average value of the silhouette coefficients of all sample points is calculated as the total silhouette coefficient, and the value range is [-1, 1]. The closer its value is to 1, the better the clustering effect. By comparing the silhouette coefficients under different numbers of clusters, the best clustering result of the photovoltaic power station is selected. The station groups divided according to this method can basically be equivalent to small-scale photovoltaic power station groups with consistent power generation. Step 202: Establish and train a short-term photovoltaic power prediction model based on source domain data;

[0108] In this embodiment, the short-term photovoltaic power prediction model adopts a Transformer neural network;

[0109] The Transformer completely abandons network structures such as RNN and CNN, and only adopts the Attention mechanism (i.e., the attention mechanism) for machine translation tasks and achieves excellent results.

[0110] The Attention mechanism enables the neural network model to assign different degrees of attention to the input data through weight settings, so as to capture the time series dependence characteristics of the input data. The Attention mechanism can usually be expressed as follows: mapping the query (Q) and key-value pairs

[0111] to the output, where the query, each key, and each value are vectors, and the output is the weighted sum of all values in , where the weights are calculated from the query and each key.

[0112] The Transformer model adopts an encoder-decoder structure, where the encoder maps the input sequence to a continuous representation , then the decoder generates an output sequence , and outputs a result at each time step.

[0113] When training the short-term photovoltaic power prediction model, the processed key meteorological factors are used as inputs, and the neural network is trained for the first time with the historical power data at the corresponding time as the output; the test set of the source domain is used to verify the reliability of the model, and the model parameters are adjusted with the goal of minimizing RMSE and MAPE until the prediction accuracy requirements are met. At this time, the completed network model is denoted as .

[0114] Step 203: Transfer the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain to obtain a short-term photovoltaic power prediction model suitable for the target domain;

[0115] Considering that the feature extraction of the neural network is more obvious in the upper layers while the underlying information is relatively basic, the parameters of all layers are kept basically unchanged and only fine-tuned, while the parameters of the fully connected layer are retrained. The data division results of the source domain and the target domain are the same. Continuing with the goal of minimizing the RMSE and MAPE of the prediction error, a prediction model suitable for the target domain is finally constructed, denoted as ,

[0116] In the process of transfer learning, the source domain and the target domain are two core concepts. The source domain is the domain that already has a large amount of labeled data. Usually, the model is pre-trained in the source domain with sufficient data. The target domain is the domain where it is hoped that the model performs well, but usually has less labeled data. Generally, only the pre-trained model is fine-tuned to make full use of the source domain data to improve the performance of the model on the target data. The data spaces of the source domain and the target domain are respectively expressed by the formulas:

[0117] ,

[0118] ,

[0119] where and respectively represent the data spaces of the source domain and the target domain, and respectively represent the features of the data spaces of the source domain and the target domain, and respectively represent the corresponding labels.

[0120] The tasks of the source domain and the target domain are to find the optimal source domain parameters and of the corresponding mapping function in the appropriate mapping functions and the optimal target domain parameters , to make the learning task and as close as possible to the label and , while transfer learning is to fine-tune based on the optimal source domain parameters to make the target domain parameters reach the optimum as quickly as possible, and are expressed as follows:

[0121] ,

[0122] ,

[0123] Step 204: Predict the distributed photovoltaic power data according to the short-term photovoltaic power prediction model applicable to the target domain.

[0124] The present invention proposes a distributed photovoltaic power prediction method based on AP clustering transfer learning, constructs a short-term power prediction model based on the Transformer neural network, and avoids redundant training costs through transfer learning based on AP clustering, effectively improving the prediction accuracy of distributed photovoltaic power prediction in complex prediction scenarios.

[0125] Step 3: Combine the multi-source user load data, distributed photovoltaic data and historical data to calculate the collaborative regulation potential of multi-source user-side resources for the collaborative regulation potential of multi-source user-side resources.

[0126] Specifically, it includes the following steps:

[0127] Step 301: Based on the load prediction curves of different types of users such as industrial, commercial, and residential users and the power prediction curves of distributed photovoltaics, evaluate and calculate the regulation potential of different types of user-side resources;

[0128] Specifically, it includes the following steps:

[0129] Step 3011: Calculate the baseline output of each user-side resource in different seasons;

[0130] Statistically predict the output value distribution of different types of user-side resources at the same historical moment within a specified duration range before the prediction date, such as within one month, and calculate the mean and variance .

[0131] Step 3012: Take the maximum and minimum values that satisfy the principle as t the maximum baseline output and the minimum baseline output at the

[0132] Step 3013. Based on the predicted values of different types of user-side resources, the maximum baseline output, and the minimum baseline output at future t moments, calculate the flexibility regulation potential of different types of user-side resources;

[0133] Among them, the calculation formula for the upward flexibility regulation potential of different types of user-side resources is as follows:

[0134] ,

[0135] In the formula, is t the upward flexibility regulation potential of different types of user-side resources at moment is t the maximum baseline output of different types of user-side resources at moment is t the predicted value of different types of user-side resources at moment

[0136] The calculation formula for the downward flexibility regulation potential of different types of user-side resources is as follows:

[0137] ,

[0138] In the formula, is t the downward flexibility regulation potential of different types of user-side resources at moment is t the minimum baseline output of different types of user-side resources at moment

[0139] Step 302. Considering the sensitivity of different user-side resources to electricity price signals, correct the regulation potential;

[0140] Time-of-use electricity price is a system that divides daily electricity consumption into peak, valley, and normal periods according to the power supply and demand situation of the power grid and the seasonal load characteristics, and reasonably determines the time-of-use electricity price difference and the electricity price for the corresponding periods. Time-of-use electricity price provides economic incentives for users through the electricity price differences in different periods. During peak hours, the electricity price is higher, while during valley hours, the electricity price is lower. This encourages users to use electricity during valley hours, thus saving electricity bills. Time-of-use electricity price affects the regulation potential of user-side resources by influencing the load value of users.

[0141] Price-based DR is achieved by leveraging demand elasticity. Define the price elasticity of electricity consumption:

[0142] ,

[0143] In the formula, represents the price elasticity of period s to period t ; and are respectively the DR pre-periods Electric load and time period t Electricity price; and are the load change amount and time period s after DR respectively t Price change amount.

[0144] The load change amount of users participating in price-based DR is:

[0145] ,

[0146] The load demand of users after participating in demand response becomes

[0147] ,

[0148] In the formula, is the load demand at time t before demand response.

[0149] Considering the impact of time-of-use electricity price on the regulation potential of user-side resources, the calculated up and down regulation potentials are corrected, and the corrected regulation potential is:

[0150] ,

[0151] ,

[0152] Among them, is the load demand change amount at time t before and after participating in demand response.

[0153] Step 303: Based on the corrected regulation potential, considering the synergistic effect of user-side resources within the aggregator, evaluate and calculate the synergistic regulation potential of user-side resources within the aggregator to obtain the potential evaluation calculation result;

[0154] Specifically, based on the prediction of the adjustable potential of single user-side resources such as flexible load and distributed photovoltaic, considering the synergistic effect of multiple user-side resources, evaluate and calculate the synergistic regulation potential of users aggregating multiple user-side resources:

[0155] ,

[0156] ,

[0157] Among them, is the user-side resource i Corrected upward regulation potential, is the user-side resource i Corrected downward regulation potential, is distributed photovoltaic, is the load.

[0158] The above solution breaks through the problem that the traditional evaluation method of the adjustable capacity of multi - user side resources depends on equipment modeling and has a large workload, providing a new idea for the evaluation of the adjustable capacity of multi - user side resources.

[0159] Embodiment 2

[0160] This embodiment provides a system for evaluating the collaborative regulation potential of multi - user side resources, including:

[0161] A load forecasting module, which is used to obtain the load forecasting results of each type of user based on the obtained power consumption load characteristic data, similar - day load data of various users, and the constructed multi - user integrated learning load forecasting model;

[0162] A photovoltaic power forecasting module, which is used to cluster distributed photovoltaic power stations, determine the source domain and target domain based on the clustering results, establish and train a short - term photovoltaic power forecasting model based on source - domain data; transfer the features learned by the short - term photovoltaic power forecasting model based on source - domain data to the target domain to obtain a short - term photovoltaic power forecasting model suitable for the target domain, and further forecast to obtain distributed photovoltaic power data;

[0163] A regulation potential evaluation module, which is used to establish the regulation potential of a single user - side resource by combining multi - user load data and distributed photovoltaic data, correct the regulation potential considering the sensitivity of different user - side resources to electricity price signals, and calculate the collaborative regulation potential of the user - side resources within the aggregator considering the collaborative effect of the user - side resources within the aggregator.

[0164] It should be noted that the specific implementation method of the system for evaluating the collaborative regulation potential of multi - user side resources in this embodiment of the present invention is similar to the specific implementation method of the method for evaluating the collaborative regulation potential of multi - user side resources in Embodiment 1 of the present invention. For details, please refer to the description in the method part. To reduce redundancy, it will not be elaborated here.

[0165] Embodiment 3

[0166] This embodiment provides a computer - readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the method for evaluating the collaborative regulation potential of multi - user side resources as described above.

[0167] Embodiment 4

[0168] This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for evaluating the collaborative regulation potential of multi - user side resources as described above.

[0169] Embodiment 5

[0170] This embodiment provides a program product, which is a computer program product and includes a computer program. When the computer program is executed by a processor, it implements the steps in a method for evaluating the collaborative regulation potential of multi-user side resources as described above.

[0171] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating the collaborative regulation potential of multi-user side resources, characterized in that It includes the following steps: Based on the obtained power consumption load characteristic data of various types of users, similar-day load data, and the constructed multi-user integrated learning load prediction model, obtain the load prediction results for each type of user; Cluster the distributed photovoltaic power stations, determine the source domain and target domain based on the clustering results, establish and train a short-term photovoltaic power prediction model based on the source domain data; transfer the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain to obtain a short-term photovoltaic power prediction model applicable to the target domain, and further predict to obtain distributed photovoltaic power data; Combine the multi-user load data and distributed photovoltaic data to establish the adjustment potential of a single user-side resource. Considering the sensitivity of different user-side resources to the electricity price signal, correct the adjustment potential, and consider the collaborative effect of the user-side resources within the aggregator to calculate the collaborative adjustment potential of the user-side resources within the aggregator; The adjusted adjustment potential after correcting the adjustment potential is: ， ， Among them, and are the adjusted upward potential and downward potential, and are the upward potential and downward potential of the flexibility of user-side resources of different categories at different times before correction, t is the load demand at time t before demand response, is the change in load demand at time t before and after participating in demand response;​ The collaborative adjustment potential is: ， ， Among them, is the user-side resource i The revised upward potential is the user-side resource i The revised downward potential is the distributed PV is the load 2. The method for evaluating the collaborative regulation potential of multi-user side resources according to claim 1, characterized in that The obtaining of the load prediction results for each type of user based on the obtained power consumption load characteristic data of various types of users, similar-day load data, and the constructed multi-user integrated learning load prediction model includes: Based on the obtained power consumption load characteristic data of various types of users and similar-day load data, construct a training set and a prediction set. Based on the training set, train multiple base learners, and then based on the prediction set, predict the trained base learners to obtain a prediction result data set; Use the prediction result data set as a secondary training data set and a prediction data set. Based on the secondary training data set, train the meta-learner, and based on the prediction data set and the trained meta-learner, predict to obtain the load prediction results for various types of users.

3. The method for evaluating the collaborative regulation potential of multi-user side resources according to claim 1, characterized in that The clustering of the distributed photovoltaic power stations and determining the source domain and target domain based on the clustering results includes clustering the distributed photovoltaic power stations using the AP clustering algorithm, and defining the clustering center of each cluster as the source domain, and classifying the remaining power stations as the target domain.

4. A method for evaluating the collaborative regulation potential of multi-user side resources according to claim 1, characterized in that, The short-term photovoltaic power prediction model uses a Transformer neural network.

5. A method for evaluating the collaborative regulation potential of multi-user side resources according to claim 1, characterized in that, When transferring the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain, adjust the model parameters with the minimum RMSE and MAPE as the goals.

6. The method for evaluating the collaborative regulation potential of multi-user side resources according to claim 1, characterized in that, Combining the multi-user load data and distributed photovoltaic data to establish the adjustment potential of a single user-side resource includes: Calculate the baseline output of each user-side resource in different seasons; Take the maximum value satisfying the principle and the minimum value as t the maximum baseline output and the minimum baseline output at the moment; Based on the future t The flexibility up-regulation potential of different types of user-side resources is calculated based on the predicted values of different types of user-side resources and the maximum baseline output at a future t moment, and the flexibility down-regulation potential of different types of user-side resources is obtained based on the predicted values of different types of user-side resources and the minimum baseline output at a future moment.

7. A multi - user side resource collaborative regulation potential evaluation system, characterized in that, It includes: A load prediction module, which is used to obtain the load prediction results for each type of user based on the obtained power consumption load characteristic data of various types of users, similar-day load data, and the constructed multi-user integrated learning load prediction model; A photovoltaic power prediction module, which is used to cluster the distributed photovoltaic power stations, determine the source domain and target domain based on the clustering results, establish and train a short-term photovoltaic power prediction model based on the source domain data; transfer the features learned by the short-term photovoltaic power prediction model based on the source domain data to the target domain to obtain a short-term photovoltaic power prediction model applicable to the target domain, and further predict to obtain distributed photovoltaic power data; A regulation potential evaluation module, which is used to establish the regulation potential of a single user-side resource by combining multi-source user load data and distributed photovoltaic data, correct the regulation potential considering the sensitivity of different user-side resources to electricity price signals, and calculate the collaborative regulation potential of the user-side resources within the aggregator considering the synergy of the user-side resources within the aggregator to obtain the collaborative regulation potential; The regulation potential after correcting the regulation potential is: ， ， Among them, and are the adjusted upward potential and downward potential, and are the upward potential and downward potential of the flexibility of user-side resources of different categories at different times before correction, t The load demand at time t before demand response, is the load demand at time t before demand response, is the change in load demand at time t before and after participating in demand response; The collaborative regulation potential is: ， ， Among them, is the user-side resource i the revised upward potential, is the user-side resource i the revised downward potential, is the distributed photovoltaic, is the load.

8. A program product, the program product being a computer program product, including a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps in a method for evaluating the collaborative regulation potential of multi-source user-side resources according to any one of claims 1-6.

Images