Load regulation potential evaluation method and system for power system and medium

Abstract

The invention discloses a load regulation potential evaluation method and system for a power system and a medium, mainly relates to the technical field of load regulation, and is used for solving the problems that an existing method is difficult to accurately describe complex power consumption behaviors of power loads and is difficult to popularize and apply under the condition of insufficient data or strict privacy protection. Comprising the steps that a trained nonlinear autoregressive neural network is utilized, and based on historical data of a preset time period and external environment data of the day, a predicted load curve of the day is obtained through fitting; on the basis of Euclidean distances between key features in the prediction load curve and the typical daily clustering centers, determining the class with the closest distance from H * T typical daily clustering centers as the typical daily clustering center corresponding to the prediction load curve on the current day; and calculating to obtain the valley filling capability and the peak clipping capability of the evaluation subject.

Description

Technical Field

[0001] The present application relates to the field of load regulation technology, and in particular to a method, system and medium for evaluating load regulation potential of a power system. Background Art

[0002] In order to achieve the low-carbon transformation of the power system, the role of demand-side response in balancing power supply and demand and optimizing resource allocation is becoming increasingly significant. To this end, accurately assessing the regulation potential of demand-side loads is of guiding significance for formulating effective strategies for power peak shaving and valley filling and building a flexible, low-carbon, and reliable power system. After accurately assessing the load regulation capacity, time-of-use electricity prices, demand response incentives and other means can be used to effectively guide load electricity consumption behavior, thereby alleviating the imbalance between power supply and demand.

[0003] The existing load regulation potential assessment technologies are roughly divided into the following two categories: 1. Model-driven load potential assessment: Establish a load model that can be transferred, interrupted, and cut, and evaluate the load regulation potential in combination with the user's actual power consumption curve. 2. Data-driven load potential assessment: Based on a large amount of historical load data, use machine learning and other data analysis methods to evaluate the load regulation potential.

[0004] However, both of the above traditional methods have limitations. 1. Due to the lack of data support, the credibility of the evaluation results of the model-driven method is low, and it is difficult to accurately describe the complex power consumption behavior of the power load, especially in the industrial industry. 2. The data-driven method relies on a large amount of historical data and needs to consider multi-dimensional external influencing factors, such as time, temperature, price, etc. It is difficult to promote and apply when there is insufficient data or strict privacy protection. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present application provides a load regulation potential assessment system, method and medium for a power system, so as to solve the problem that the existing methods are difficult to accurately characterize the complex power consumption behavior of the power load and are difficult to promote and apply in the case of insufficient data or strict privacy protection.

[0006] In a first aspect, the present application provides a method for evaluating load regulation potential of a power system, the method comprising: Obtain the daily load curve of any assessment subject within a preset historical time period, and then extract the key features in the daily load curve as daily load clustering data, normalize the daily load clustering data, and obtain curve sample data; wherein, the daily load curve is a curve composed of key features collected at preset collection time points, and the key features at least include load power; use a preset clustering algorithm and a preset first clustering number H to cluster the curve sample data into a preset number of typical daily clustering centers; use K-means Clustering, preset the second clustering number T, based on the characteristics of three dimensions of daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time, perform secondary clustering on the curve sample data to obtain H*T typical daily clustering centers; use the trained nonlinear autoregressive neural network to fit the daily forecast load curve based on the historical data of the preset time period and the external environment data of the day; wherein the historical data consists of daily load power and external environment data within the preset time period; based on the Euclidean distance between the key features in the forecast load curve and the typical daily clustering center, determine the closest one from the H*T typical daily clustering centers as the typical daily clustering center corresponding to the forecast load curve of the day; based on the forecast load curve, the maximum value curve, minimum value curve and reserved percentage of the load power in the typical daily clustering center corresponding to the forecast load curve, calculate and obtain the valley filling capacity and peak shaving capacity of the evaluation subject.

[0007] The online learning method provided in the embodiment of the present application uses system cluster analysis to take the daily load data (daily load curve) within the preset historical time period of the evaluation subject as a feature, and performs a primary clustering of the historical daily load data; secondly, the K-means clustering method is used to select characteristic vectors such as daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time for secondary clustering analysis, in order to capture the user's complex electricity consumption behavior and habits; then, based on the cluster analysis, a nonlinear autoregressive neural network is used for prediction, mainly using historical load data and external environmental data (such as electricity prices, weather, production plans, etc.) for neural network training; finally, in order to accurately evaluate the load regulation potential, the optimal match between the predicted day and the cluster center is calculated, and the difference between the upper and lower boundaries of the cluster center and the predicted load is calculated based on the matching result, and then the load regulation potential (valley filling capacity and peak shaving capacity) is calculated.

[0008] In one implementation of the present application, the daily load curve is a 96-point daily load curve.

[0009] In one implementation of the present application, a preset clustering algorithm and a preset first clustering number H are used to cluster the curve sample data into a preset number of typical day clustering centers, specifically including: S1. Based on the formula: , Calculate the Euclidean distance between two curve sample data ;in, represents one of the two curve sample data, Represents the other of the two curve sample data; t represents the time point, M = 24 hours / time interval between time points; S2, merge the two classes with the smallest Euclidean distance into a new class, and perform the K=K-1 operation; where the initial value of K is the total number of days of the curve sample data; S3. Repeat S1-S2 until k=H.

[0010] In one implementation of the present application, before fitting and obtaining the predicted load curve for the day based on the historical data of the preset time period and the external environment data of the day by using the trained nonlinear autoregressive neural network, the method further includes: Get the key parameters of the nonlinear autoregressive neural network: training step size, transfer function g(), output layer expected value Y, hidden layer expected value H, and learning rate ; By formula: , Calculate the hidden layer output prediction value ;in, is the weight matrix from the input layer to the hidden layer, is the threshold of the hidden layer; Represents historical data of input; By formula: ; Calculate the hidden layer error , and then use the formula: ,renew ; Among them, Δ represents the derivative of the transfer function; By formula: , Calculate the output layer output prediction value ;in, is the weight matrix from the hidden layer to the output layer, is the threshold of the output layer; By formula: ; Calculate the output layer error , and then use the formula: ,renew ; The nonlinear autoregressive neural network iterates until and The preset error threshold is reached.

[0011] In one implementation of the present application, K-means clustering is used, a second clustering number T is preset, and the curve sample data is secondary clustered based on the characteristics of three dimensions: daily load rate, daily peak-to-valley difference rate, and daily maximum load utilization time, specifically including: S1. Extract daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time from curve sample data; Then construct several three-dimensional vector samples: ; Where i represents the i-th sample; i∈[0,L], L represents the total number of samples; represents the daily load rate of the i-th sample, represents the daily peak-to-valley difference rate of the i-th sample, represents the daily maximum load utilization time of the i-th sample; S2, randomly select T samples as the initial cluster centers; S3, using the formula: , Calculate the distance from each sample to the cluster center, and then Assign to the nearest cluster center; in, represents the jth cluster center, j∈[0,T]; S4, through the formula: , Recalculate each cluster center; where, Indicates the number of samples contained in the current j-th cluster center; S5. Repeat S3-S4 until the cluster center no longer changes or the preset number of iterations is reached.

[0012] In one implementation of the present application, based on the predicted load curve, the maximum value curve, the minimum value curve, and the reserved percentage of the load power in the typical day cluster center corresponding to the predicted load curve, the valley filling capacity and peak shaving capacity of the evaluation subject are calculated, specifically including: The maximum value curve is adjusted upward by the reserved percentage to obtain the final maximum value curve; wherein the maximum value curve is a curve composed of the maximum values ​​of each acquisition time point in the cluster center of a typical day; The minimum value curve is adjusted down by the reserved percentage to obtain the final minimum value curve; wherein the maximum value curve is a curve composed of the minimum values ​​of each acquisition time point in the cluster center of a typical day; By formula: , calculate and obtain the valley filling capacity at each sampling time point on the predicted day ;in, represents the maximum value at the corresponding acquisition time point, represents the predicted value of the predicted load curve at the corresponding acquisition time point; By formula: , calculate and obtain the predicted peak shaving capacity at each collection time point on the day ;in, Indicates the minimum value at the corresponding acquisition time point.

[0013] In a second aspect, the present application provides a load regulation potential assessment system for a power system, the system comprising: The acquisition module is used to obtain the daily load curve of any evaluation subject within a preset historical time period, and then extract the key features in the daily load curve as daily load clustering data, normalize the daily load clustering data, and obtain curve sample data; wherein, the daily load curve is a curve composed of key features collected at preset collection time points, and the key features include at least load power; the clustering module is used to cluster the curve sample data into a preset number of typical daily clustering centers using a preset clustering algorithm and a preset first clustering number H; and use K-means Clustering, preset the second clustering number T, based on the characteristics of three dimensions of daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time, perform secondary clustering on the curve sample data to obtain H*T typical daily clustering centers; a fitting module is used to use the trained nonlinear autoregressive neural network to fit the daily forecast load curve based on the historical data of the preset time period and the external environment data of the day; wherein the historical data consists of daily load power and external environment data within the preset time period; a determination module is used to determine the closest type from the H*T typical daily clustering centers as the typical daily clustering center corresponding to the daily forecast load curve based on the Euclidean distance between the key features in the forecast load curve and the typical daily clustering center; a calculation module is used to calculate the valley filling capacity and peak shaving capacity of the evaluation subject based on the forecast load curve, the maximum value curve, the minimum value curve and the reserved percentage of the load power in the typical daily clustering center corresponding to the forecast load curve.

[0014] In one implementation of the present application, the fitting module includes a training unit, Used to obtain the key parameters of nonlinear autoregressive neural networks: training step size, transfer function g(), output layer expected value Y, hidden layer expected value H, and learning rate ; By formula: , Calculate the hidden layer output prediction value ;in, is the weight matrix from the input layer to the hidden layer, is the threshold of the hidden layer; Represents historical data of input; By formula: ; Calculate the hidden layer error , and then use the formula: ,renew ; Among them, Δ represents the derivative of the transfer function; By formula: , Calculate the output layer output prediction value ;in, is the weight matrix from the hidden layer to the output layer, is the threshold of the output layer; By formula: ; Calculate the output layer error , and then use the formula: ,renew ; The nonlinear autoregressive neural network iterates until and The preset error threshold is reached.

[0015] In one implementation of the present application, the computing module includes a computing unit, The maximum value curve is adjusted upward by the reserved percentage to obtain the final maximum value curve; wherein the maximum value curve is a curve composed of the maximum values ​​of each acquisition time point in the cluster center of a typical day; The minimum value curve is adjusted down by the reserved percentage to obtain the final minimum value curve; wherein the maximum value curve is a curve composed of the minimum values ​​of each acquisition time point in the cluster center of a typical day; By formula: , calculate and obtain the valley filling capacity at each sampling time point on the predicted day ;in, represents the maximum value at the corresponding acquisition time point, represents the predicted value of the predicted load curve at the corresponding acquisition time point; By formula: , calculate and obtain the predicted peak shaving capacity at each collection time point on the day ;in, Indicates the minimum value at the corresponding acquisition time point.

[0016] In a third aspect, the present application provides a non-volatile computer storage medium having computer instructions stored thereon, which when executed implement a method for evaluating load regulation potential for a power system as described above.

[0017] Those skilled in the art can understand that the present application has at least the following beneficial effects: 1. Accurate load boundary characterization: This application constructs a more detailed load power consumption behavior through two clustering analyses, effectively capturing the fluctuation and uncertainty of the load. Therefore, compared with the first clustering, the second clustering further enhances the accuracy of the boundary calculation.

[0018] 2. Accurate regulation potential prediction: The power prediction model based on nonlinear autoregressive neural network generates a high-precision predicted daily load curve through historical load data and external influencing factors (electricity price, weather, time, etc.), and combines the clustering boundaries obtained by secondary clustering to accurately evaluate the user's peak shaving and valley filling capabilities.

[0019] 3. Strong model robustness: Since historical load data already includes a variety of complex influencing factors such as changes in electricity prices and seasonal differences, the system can maintain a stable assessment of regulation potential in scenarios such as changes in the external environment or a high degree of confidentiality of production time series data, thereby improving the robustness and applicability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Some embodiments of the present disclosure are described below with reference to the accompanying drawings, in which: Figure 1 It is a flow chart of a method for evaluating load regulation potential of a power system provided in an embodiment of the present application.

[0021] Figure 2 It is a schematic diagram of the internal structure of a load regulation potential assessment system for a power system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] It should be understood by those skilled in the art that the embodiments described below are only preferred embodiments of the present disclosure, and do not mean that the present disclosure can only be implemented through the preferred embodiments. The preferred embodiments are only used to explain the technical principles of the present disclosure, and are not used to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work should still fall within the protection scope of the present disclosure.

[0023] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0024] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0025] The embodiment provides a method for evaluating load regulation potential of a power system. Figure 1 As shown, the method provided in the embodiment of the present application mainly includes the following steps: Step 110: Obtain the daily load curve of any assessment subject within a preset historical time period, and then extract key features in the daily load curve as daily load clustering data, normalize the daily load clustering data, and obtain curve sample data.

[0026] It should be noted that the daily load curve is a curve composed of key features collected at preset collection time points, and the collection time points can be any feasible time points. For example, the daily load curve can be a 96-point daily load curve. The key features at least include load power, and those skilled in the art can add key features according to actual needs. For example, the daily load curve in the preset historical time period can be a 96-point daily load curve for 365 days a year.

[0027] To avoid dimensionality problems, the daily load clustering data is standardized so that the load power curve data is scaled to the [0,1] interval. Those skilled in the art can remove bad points according to actual needs.

[0028] Step 120: using a preset clustering algorithm and a preset first clustering number H, cluster the curve sample data into a preset number of typical day cluster centers.

[0029] This step can be specifically as follows: S1. Based on the formula: , Calculate the Euclidean distance between two curve sample data ;in, represents one of the two curve sample data, Indicates the other of the two curve sample data; t indicates the time point, M = 24 hours / time interval between time points. When it is a 96-point daily load curve, M = 96.

[0030] S2. Merge the two classes with the smallest Euclidean distance into a new class and perform the K=K-1 operation; where the initial value of K is the total number of days of the curve sample data.

[0031] S3. Repeat S1-S2 until k=H.

[0032] It will be appreciated by those skilled in the art that the present application uses a system clustering algorithm in step 120 to implement a clustering of the load power curve. The different daily load data of the same evaluation subject are regarded as independent classes, and the daily load data with the closest distance are gradually merged according to a specific distance metric until all daily load data are merged into H complete historical load data sets. The basic principle is to select the class that minimizes the sum of square deviations within the class each time two classes are merged. In this way, daily load data with similar power curve characteristics are classified into the same class, and the power consumption characteristics between different classes remain relatively different. Since this method generates a hierarchical clustering dendrogram, the reasonable number of clusters and the samples contained in each class can be determined according to the structure of the hierarchical diagram, for example, 96 points of daily load data can be divided into typical days with high power consumption and typical days with low power consumption.

[0033] Step 130: Using K-means clustering and presetting the second cluster number T, based on the characteristics of three dimensions, namely, daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time, perform secondary clustering on the curve sample data to obtain H*T typical daily cluster centers.

[0034] It should be noted that the daily load rate is the ratio of actual power consumption to the maximum load capacity in a day, reflecting the utilization rate of power equipment. The daily peak-to-valley difference rate describes the difference between the peak and trough of load in a day, reflecting the volatility of the load. The daily maximum load utilization time indicates the time when the maximum load is reached in a day, reflecting the time distribution characteristics of the power load. By clustering the above three dimensional features, users' power consumption behaviors can be classified into different types, so as to more accurately evaluate their flexible load adjustment potential.

[0035] This step can be specifically as follows: S1. Extract daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time from curve sample data; Then construct several three-dimensional vector samples: ; Where i represents the i-th sample; i∈[0,L], L represents the total number of samples; represents the daily load rate of the i-th sample, represents the daily peak-to-valley difference rate of the i-th sample, Indicates the daily maximum load utilization time of the i-th sample. ; S2. Randomly select T samples as initial cluster centers.

[0036] S3, using the formula: , Calculate the distance from each sample to the cluster center, and then Assign to the nearest cluster center; in, represents the jth cluster center, j∈[0,T].

[0037] S4, through the formula: , Recalculate each cluster center; where, Indicates the number of samples contained in the current j-th cluster center.

[0038] S5. Repeat S3-S4 until the cluster center no longer changes or the preset number of iterations is reached.

[0039] Those skilled in the art can understand that, in the secondary clustering, the present application uses K-means clustering to extract three key dimensional features from the user's daily load data: daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time, and performs secondary clustering analysis based on these three features.

[0040] Step 140: Using the trained nonlinear autoregressive neural network, based on the historical data of the preset time period and the external environmental data of the day, a predicted load curve for the day is obtained by fitting.

[0041] It should be noted that nonlinear autoregressive neural networks, such as NARX neural networks, are neural networks with memory and dynamic feedback functions, which can memorize and store historical power load data and combine the trained model with real-time data to predict future load data. The historical data consists of daily load power and external environment data within a preset time period.

[0042] Before fitting and obtaining the current day's forecast load curve using the trained nonlinear autoregressive neural network based on historical data of a preset time period and external environmental data of the current day, the method further includes: Get the key parameters of the nonlinear autoregressive neural network: training step size, transfer function g(), output layer expected value Y, hidden layer expected value H, and learning rate ; By formula: , Calculate the hidden layer output prediction value ;in, is the weight matrix from the input layer to the hidden layer, is the threshold of the hidden layer; Represents historical data of the input.

[0043] By formula: ; Calculate the hidden layer error , and then use the formula: ,renew ; Among them, Δ represents the derivative of the transfer function; By formula: , Calculate the output layer output prediction value ;in, is the weight matrix from the hidden layer to the output layer, is the threshold of the output layer; By formula: ; Calculate the output layer error , and then use the formula: ,renew ; The nonlinear autoregressive neural network iterates until and The preset error threshold is reached.

[0044] It should be noted that the input of the nonlinear autoregressive neural network is historical data, where the external environmental data can at least include external factors such as the adjustable characteristics of the load, temperature, humidity, product prices, etc. The output of the model is the predicted daily load curve. By learning the user's load power and external environmental data, the nonlinear autoregressive neural network can capture the load change law and provide load forecasts for a specific time period in the future.

[0045] In addition, those skilled in the art can understand that the calculation process of the above data is: forward calculation ; ; Reverse calculation ; Reverse calculation ; Reverse calculation ; .

[0046] Step 150: Based on the Euclidean distance between the key features in the forecast load curve and the typical day cluster center, determine the closest one from the H*T typical day cluster centers as the typical day cluster center corresponding to the forecast load curve of the day.

[0047] It should be noted that the H cluster centers obtained in the first clustering and the T cluster centers obtained in the second clustering can be intersected into H*T typical daily cluster centers.

[0048] Step 160: Based on the predicted load curve, the maximum value curve, the minimum value curve, and the reserved percentage of the load power in the typical day cluster center corresponding to the predicted load curve, the valley filling capacity and the peak shaving capacity of the evaluation subject are calculated.

[0049] It should be noted that the reserved percentage may be 5%.

[0050] This step can be specifically as follows: The maximum value curve is adjusted upward by the reserved percentage to obtain the final maximum value curve, where the maximum value curve is a curve composed of the maximum values ​​of each collection time point in the cluster center of a typical day.

[0051] The minimum value curve is adjusted down by the reserved percentage to obtain the final minimum value curve, where the maximum value curve is a curve composed of the minimum values ​​of each collection time point in the typical daily cluster center.

[0052] By formula: , calculate and obtain the valley filling capacity at each sampling time point on the predicted day ;in, represents the maximum value at the corresponding acquisition time point, Represents the predicted value of the predicted load curve at the corresponding acquisition time point.

[0053] By formula: , calculate and obtain the predicted peak shaving capacity at each collection time point on the day ;in, Indicates the minimum value at the corresponding acquisition time point.

[0054] In addition, this application Figure 2 The present invention provides a load regulation potential assessment system for a power system. Figure 2 As shown, the system provided in the embodiment of the present application mainly includes: The acquisition module 210 is used to obtain the daily load curve of any evaluation subject within a preset historical time period, and then extract the key features in the daily load curve as daily load clustering data, normalize the daily load clustering data, and obtain curve sample data; wherein, the daily load curve is a curve composed of key features collected at preset collection time points, and the key features include at least load power.

[0055] The clustering module 220 is used to cluster the curve sample data into a preset number of typical daily cluster centers using a preset clustering algorithm and a preset first cluster number H; and to perform secondary clustering on the curve sample data based on the characteristics of three dimensions, namely, daily load rate, daily peak-to-valley difference rate, and daily maximum load utilization time, by using K-means clustering and a preset second cluster number T, to obtain H*T typical daily cluster centers.

[0056] The fitting module 230 is used to use the trained nonlinear autoregressive neural network to fit the predicted load curve for the day based on the historical data of the preset time period and the external environmental data of the day; wherein the historical data consists of the daily load power and external environmental data within the preset time period.

[0057] The fitting module 230 includes a training unit for obtaining key parameters of the nonlinear autoregressive neural network: training step size, transfer function g(), output layer expected value Y, hidden layer expected value H, and learning rate .

[0058] By formula: , Calculate the hidden layer output prediction value ;in, is the weight matrix from the input layer to the hidden layer, is the threshold of the hidden layer; Represents historical data of the input.

[0059] By formula: ; Calculate the hidden layer error , and then use the formula: ,renew ; Among them, Δ represents the derivative of the transfer function; By formula: , Calculate the output layer output prediction value ;in, is the weight matrix from the hidden layer to the output layer, is the threshold of the output layer.

[0060] By formula: ; Calculate the output layer error , and then use the formula: ,renew ; The nonlinear autoregressive neural network iterates until and The preset error threshold is reached.

[0061] The determination module 240 is used to determine the typical day cluster center corresponding to the forecast load curve of the day with the closest distance from the H*T typical day cluster centers based on the Euclidean distance between the key features in the forecast load curve and the typical day cluster center.

[0062] The calculation module 250 is used to calculate the valley filling capacity and peak shaving capacity of the evaluation subject based on the predicted load curve, the maximum value curve, the minimum value curve, and the reserved percentage of the load power in the typical day cluster center corresponding to the predicted load curve.

[0063] The calculation module 250 includes a calculation unit, which adjusts the maximum value curve upward by a reserved percentage to obtain a final maximum value curve; wherein the maximum value curve is a curve composed of the maximum values ​​of each acquisition time point in the typical daily cluster center; and adjusts the minimum value curve downward by a reserved percentage to obtain a final minimum value curve; wherein the maximum value curve is a curve composed of the minimum values ​​of each acquisition time point in the typical daily cluster center.

[0064] By formula: , calculate and obtain the valley filling capacity at each sampling time point on the predicted day ;in, represents the maximum value at the corresponding acquisition time point, Represents the predicted value of the predicted load curve at the corresponding acquisition time point.

[0065] By formula: , calculate and obtain the predicted peak shaving capacity at each collection time point on the day ;in, Indicates the minimum value at the corresponding acquisition time point.

[0066] In addition, an embodiment of the present application further provides a non-volatile computer storage medium having executable instructions stored thereon, and when the executable instructions are executed, a load regulation potential assessment method for a power system as described above is implemented.

[0067] So far, the technical solutions of the present disclosure have been described in combination with the above multiple embodiments, but it is easy for those skilled in the art to understand that the protection scope of the present disclosure is not limited to these specific embodiments. Without departing from the technical principles of the present disclosure, those skilled in the art can split and combine the technical solutions in the above-mentioned various embodiments, and can also make equivalent changes or replacements to the relevant technical features. Any changes, equivalent replacements, improvements, etc. made within the technical concept and / or technical principle of the present disclosure will fall within the protection scope of the present disclosure.

Claims

1. A method for evaluating load regulation potential of a power system, characterized in that: The method comprises: Obtain the daily load curve of any assessment subject within a preset historical time period, and then extract the key features in the daily load curve as daily load clustering data, normalize the daily load clustering data, and obtain curve sample data; wherein the daily load curve is a curve composed of key features collected at preset collection time points, and the key features at least include load power; Using a preset clustering algorithm and a preset first clustering number H, the curve sample data is clustered into a preset number of typical day clustering centers; Using K-means clustering and presetting the second clustering number T, based on the characteristics of three dimensions: daily load rate, daily peak-to-valley difference rate, and daily maximum load utilization time, the curve sample data is clustered twice to obtain H*T typical daily clustering centers; Using the trained nonlinear autoregressive neural network, based on the historical data of the preset time period and the external environmental data of the day, the predicted load curve of the day is fitted; wherein the historical data consists of the daily load power and external environmental data within the preset time period; Based on the Euclidean distance between the key features in the forecast load curve and the typical day cluster center, the closest one among H*T typical day cluster centers is determined as the typical day cluster center corresponding to the forecast load curve of the day; Based on the predicted load curve, the maximum value curve, minimum value curve and reserved percentage of the load power in the typical day cluster center corresponding to the predicted load curve, the valley filling capacity and peak shaving capacity of the evaluation subject are calculated.

2. The load regulation potential evaluation method for a power system according to claim 1, characterized in that: The daily load curve is a 96-point daily load curve.

3. The load regulation potential evaluation method for a power system according to claim 1, characterized in that: Using a preset clustering algorithm and a preset first clustering number H, the curve sample data is clustered into a preset number of typical day clustering centers, specifically including: S1. Based on the formula: ， Calculate the Euclidean distance between two curve sample data ;in, represents one of the two curve sample data, Represents the other of the two curve sample data; t represents the time point, M = 24 hours / time interval between time points; S2, merge the two classes with the smallest Euclidean distance into a new class, and perform the K=K-1 operation; where the initial value of K is the total number of days of the curve sample data; S3. Repeat S1-S2 until k=H.

4. The load regulation potential evaluation method for a power system according to claim 1, characterized in that: Before fitting and obtaining the current day's forecast load curve based on historical data of a preset time period and the current day's external environmental data using the trained nonlinear autoregressive neural network, the method further includes: Get the key parameters of the nonlinear autoregressive neural network: training step size, transfer function g(), output layer expected value Y, hidden layer expected value H, and learning rate ; By formula: ， Calculate the hidden layer output prediction value ;in, is the weight matrix from the input layer to the hidden layer, is the threshold of the hidden layer; Represents historical data of input; By formula: ； Calculate the hidden layer error , and then use the formula: ,renew ; Among them, Δ represents the derivative of the transfer function; By formula: ， Calculate the output layer output prediction value ;in, is the weight matrix from the hidden layer to the output layer, is the threshold of the output layer; By formula: ； Calculate the output layer error , and then use the formula: ,renew ; The nonlinear autoregressive neural network iterates until and The preset error threshold is reached.

5. The load regulation potential assessment method for a power system according to claim 1, characterized in that: Using K-means clustering and presetting the second cluster number T, the curve sample data is clustered twice based on the characteristics of three dimensions: daily load rate, daily peak-to-valley difference rate, and daily maximum load utilization time. Specifically, the following are included: S1. Extract daily load rate, daily peak-to-valley difference rate and daily maximum load utilization time from curve sample data; Then construct several three-dimensional vector samples: ; Where i represents the i-th sample; i∈[0,L], L represents the total number of samples; represents the daily load rate of the i-th sample, represents the daily peak-to-valley difference rate of the i-th sample, represents the daily maximum load utilization time of the i-th sample; S2, randomly select T samples as the initial cluster centers; S3, using the formula: ， Calculate the distance from each sample to the cluster center, and then Assign to the nearest cluster center; in, represents the jth cluster center, j∈[0,T]; S4, through the formula: ， Recalculate each cluster center; where, Indicates the number of samples contained in the current j-th cluster center; S5. Repeat S3-S4 until the cluster center no longer changes or the preset number of iterations is reached.

6. The load regulation potential evaluation method for a power system according to claim 1, characterized in that: Based on the predicted load curve, the maximum value curve, minimum value curve, and reserved percentage of the load power in the typical day cluster center corresponding to the predicted load curve, the valley filling capacity and peak shaving capacity of the evaluation subject are calculated, including: The maximum value curve is adjusted upward by the reserved percentage to obtain the final maximum value curve; wherein the maximum value curve is a curve composed of the maximum values ​​of each acquisition time point in the cluster center of a typical day; The minimum value curve is adjusted down by the reserved percentage to obtain the final minimum value curve; wherein the maximum value curve is a curve composed of the minimum values ​​of each acquisition time point in the cluster center of a typical day; By formula: , calculate and obtain the valley filling capacity at each sampling time point on the predicted day ;in, represents the maximum value at the corresponding acquisition time point, represents the predicted value of the predicted load curve at the corresponding acquisition time point; By formula: , calculate and obtain the predicted peak shaving capacity at each collection time point on the day ;in, Indicates the minimum value at the corresponding acquisition time point.

7. A load regulation potential assessment system for a power system, characterized in that: The system comprises: The acquisition module is used to obtain the daily load curve of any evaluation subject within a preset historical time period, and then extract the key features in the daily load curve as daily load clustering data, normalize the daily load clustering data, and obtain curve sample data; wherein the daily load curve is a curve composed of key features collected at preset collection time points, and the key features include at least load power; The clustering module is used to cluster the curve sample data into a preset number of typical daily cluster centers using a preset clustering algorithm and a preset first cluster number H; perform secondary clustering on the curve sample data based on the characteristics of three dimensions, namely, daily load rate, daily peak-to-valley difference rate, and daily maximum load utilization time, using K-means clustering and a preset second cluster number T to obtain H*T typical daily cluster centers; A fitting module is used to use the trained nonlinear autoregressive neural network to fit the current day's predicted load curve based on the historical data of a preset time period and the external environmental data of the current day; wherein the historical data consists of the daily load power and external environmental data within the preset time period; A determination module is used to determine the closest type of typical day cluster center corresponding to the forecast load curve of the day from H*T typical day cluster centers based on the Euclidean distance between the key features in the forecast load curve and the typical day cluster center; The calculation module is used to calculate the valley filling capacity and peak shaving capacity of the evaluation subject based on the predicted load curve, the maximum value curve, the minimum value curve and the reserved percentage of the load power in the typical day cluster center corresponding to the predicted load curve.

8. The load regulation potential evaluation system for a power system according to claim 7, characterized in that: The fitting module includes a training unit, Used to obtain the key parameters of nonlinear autoregressive neural networks: training step size, transfer function g(), output layer expected value Y, hidden layer expected value H, and learning rate ; By formula: ， Calculate the hidden layer output prediction value ;in, is the weight matrix from the input layer to the hidden layer, is the threshold of the hidden layer; Represents historical data of input; By formula: ； Calculate the hidden layer error , and then use the formula: ,renew ; Among them, Δ represents the derivative of the transfer function; By formula: ， Calculate the output layer output prediction value ;in, is the weight matrix from the hidden layer to the output layer, is the threshold of the output layer; By formula: ； Calculate the output layer error , and then use the formula: ,renew ; The nonlinear autoregressive neural network iterates until and The preset error threshold is reached.

9. The load regulation potential evaluation system for a power system according to claim 7, characterized in that: The computing module includes a computing unit, The maximum value curve is adjusted upward by the reserved percentage to obtain the final maximum value curve; wherein the maximum value curve is a curve composed of the maximum values ​​of each acquisition time point in the cluster center of a typical day; The minimum value curve is adjusted down by the reserved percentage to obtain the final minimum value curve; wherein the maximum value curve is a curve composed of the minimum values ​​of each acquisition time point in the cluster center of a typical day; By formula: , calculate and obtain the valley filling capacity at each sampling time point on the predicted day ;in, represents the maximum value at the corresponding acquisition time point, represents the predicted value of the predicted load curve at the corresponding acquisition time point; By formula: , calculate and obtain the predicted peak shaving capacity at each collection time point on the day ;in, Indicates the minimum value at the corresponding acquisition time point.

10. A non-volatile computer storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, a method for evaluating load regulation potential for a power system as described in any one of claims 1 to 6 is implemented.

Images