A method for evaluating the peak-load shifting and valley-filling potential of power users

Through the dual-stream residual convolutional neural network combined with the daily matrix and the weekly matrix, the extensive and subjective problems of the potential assessment of peak-to-valve valley filling in power users in the existing technology are solved, and a multi-dimensional and objective assessment of the potential of power users is realized. Improves the accuracy and objectivity of the evaluation.

CN114418374BActive Publication Date: 2024-11-01国网福建省电力有限公司营销服务中心 +1
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202210040348.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-14
Publication Date
2024-11-01
Estimated Expiration
2042-01-14

AI Technical Summary

Technical Problem

The existing method of peak-to-valve potential assessment of electricity users has extensive and subjective problems. The clustering algorithm cannot accurately evaluate the potential of each household. The regulation modeling method has subjectiveness in weight allocation, which affects the evaluation results. objectivity.

Method used

The dual-current residual convolutional neural network is used to combine the daily matrix and the weekly matrix. By constructing a multi-dimensional index matrix, it reduces manual participation and allocates the weight of neural unit by itself, and discovers the potential of peak shifting and valley filling from power big data, combining with industry electricity use Comprehensive analysis of the situation, customer credit and environmental pollution factors.

Benefits of technology

A multi-dimensional and objective assessment of the potential of power users to move peaks and fill valleys has been achieved, reducing the influence of subjective factors, accurately analyzing the potential of each household, and improving the accuracy and objectivity of the evaluation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114418374B_ABST
    Figure CN114418374B_ABST
Patent Text Reader

Abstract

The present invention provides a method for evaluating the peak-shifting and valley-filling potential of power users, comprising the following steps: Step 1, constructing a two-dimensional array matrix, including a daily matrix and a weekly matrix; the daily matrix is ​​a matrix X containing the daily data of the user nm , n Indicates n Data indicators, a total of 10 indicators; m Indicates m time points, which takes values ​​at 15-minute intervals for 24 hours, and represents the time in sequence 00:00, 00:15, 00:30, ... 23:45, so the value range in a day is 1 to 96; the weekly matrix is ​​the matrix Z containing the user's weekly data jk , j Indicates j Data indicators, a total of 10 indicators; k Indicates k Week, the value range in one year is 1 to 52; step two, construction of a two-stream residual convolutional neural network; step three, network training and identification; the application of this technical solution can realize the discovery of users' peak shifting and valley filling potential from power big data, reduce manual participation, and be less affected by subjective factors.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of rationally regulating power grid supply and demand relations, and in particular to a method for evaluating peak-load shifting and valley-filling potential of power users. Background Art

[0002] As the domestic and international situation becomes more and more complicated, coal prices have risen beyond expectations. In addition, the proportion of distributed power sources such as photovoltaics and wind power in the power grid has increased rapidly. The peak load regulation on the power generation side alone cannot meet the needs of power grid stability. Therefore, some areas may experience sudden and local power shortages. At present, the Chinese government generally considers economic, technical, and electricity demand aspects to adopt peak load avoidance and power supply methods, and regulates the user's electricity consumption time and total amount in accordance with the law to ensure the supply and demand relationship of the power grid. Therefore, selecting users with great potential for peak load shifting and valley filling for orderly electricity use can better maintain the balance of supply and demand and ensure the safety of power grid operation. Obviously, it is very necessary to study a potential assessment of the peak load shifting and valley filling capabilities of power users.

[0003] With the continuous development of data collection and communication technology, smart energy meters with HPLC modules are becoming more and more popular. The success rate of collecting power consumption information of power users is getting higher and higher, the data granularity is getting finer and finer, and the indicators considered by the peak shifting and valley filling evaluation method are becoming more and more three-dimensional and complete. At present, the potential assessment of peak shifting and valley filling mainly includes combining clustering methods and scheduling modeling.

[0004] (1) Clustering-based method: This method clusters and classifies the collected user load curves, and groups users with similar load characteristics into one category, with each category representing a power consumption pattern. The load curves of each category are then analyzed using indicators such as peak-to-valley difference and peak ratio, to obtain the peak-to-valley shifting potential of each category of users.

[0005] (2) Method based on regulation modeling: This method comprehensively analyzes the user's regulation value system through the user's historical load, customer credit, output value and other aspects, and then uses weight calculation methods such as expert method and hierarchical analysis method to assign weights to historical load, customer willingness, output value and other aspects to complete the evaluation of the customer's potential for peak shifting and valley filling.

[0006] The potential assessment of peak shifting and valley filling established by clustering, regulation modeling and other methods has the following main shortcomings:

[0007] (1) The clustering algorithm is relatively rough. It evaluates the potential of users by category and does not accurately assess the potential of each household.

[0008] (2) The evaluation indicators of clustering algorithms are relatively simple and do not comprehensively consider various factors such as economy and industry characteristics.

[0009] (3) Although the regulatory modeling method takes into account various influencing factors, there is still a certain degree of subjectivity in the weight allocation of influencing factors. Weight allocation is prone to errors, which will affect the final evaluation results.

[0010] (4) When calculating economic, environmental and other influencing factors, the regulatory modeling method contains many subjective and fuzzy descriptions to illustrate the numerical value. For example, when describing the user's contribution to the regional economy, [0.1, 0.3, 0.7] is used to represent low, average, and high. The numerical value is highly subjective, and different people may have different subjective judgments.

[0011] The existing methods for assessing users' peak-to-valley shifting potential have some problems such as being extensive and highly subjective. Although the extensive assessment method is objective, it takes fewer factors into consideration. The control modeling method considers many comprehensive methods, but has problems such as relatively subjective weight allocation. Summary of the invention

[0012] In view of this, the purpose of the present invention is to provide a method for evaluating the peak-load shifting and valley-filling potential of power users, so as to discover the peak-load shifting and valley-filling potential of users from power big data, reduce manual participation, and be less affected by subjective factors.

[0013] To achieve the above object, the present invention adopts the following technical solution: a method for evaluating the peak load shifting and valley load filling potential of an electricity user, comprising the following steps:

[0014] Step 1: construct a two-dimensional array matrix, including a daily matrix and a weekly matrix;

[0015] Among them, the daily matrix is ​​the matrix X containing the user's daily data nm , n represents the nth data indicator, a total of 10 indicators; m represents the mth time point;

[0016] Daily matrix X nm Focus on the user's load rate and load growth rate information, and compare it with the overall situation of the user's industry or city, including 10 indicators:

[0017] The larger the user load value, the greater the potential for peak-load shifting and valley-loading. Therefore, the first indicator X 1m is the load size of the user at the mth time point;

[0018] If the user load is much higher than the average load, the user's peak load shifting and valley load filling potential may be greater; therefore, the second indicator X 2m and the third indicator X 3m The difference between the average load of all users in the city and industry where the user is located and the load of the user;

[0019] The greater the load curve change rate, the more obvious the curve fluctuations, and the greater the peak-to-valley difference of the curve may be. Therefore, the fourth indicator X4m is the load growth rate of the user at the mth time point, and its formula is:

[0020]

[0021] Considering the similarity between the user's load curve and the whole network's load curve, the more similar the growth rate is, the more consistent the change trend is, and the more likely it is that the peak shifting and valley filling will be effective; therefore, the fifth indicator X 5m It is the difference between the load growth rate of the whole network at the mth time point and the load growth rate of the user at the mth time point;

[0022] Similarly, the sixth indicator X 6m and the seventh indicator X 7m are the difference between the load growth rate of the user's city and industry at the mth time point and the load growth rate of the user at the mth time point;

[0023] Power data as the eighth indicator X 8m , is the power consumption of the user from the m-1th time point to the mth time point; the power consumption is obtained by collecting the frozen indication curve through the user information collection system and subtracting the previous and next two time periods;

[0024] The ninth indicator X 9m The power consumption from the m-1th time point to the mth time point divided by the user's full-day power consumption;

[0025] The tenth indicator X 10m The amount of electricity from the m-1th time point to the mth time point multiplied by the time-of-use electricity price of the period;

[0026] At this point, the matrix is ​​constructed;

[0027] Weekly matrix is ​​the matrix Z containing the user's weekly data jk ,j represents the jth data indicator, with a total of 10 indicators; k represents the kth week, and the value range in a year is 1 to 52;

[0028] Zhou Matrix Z jk Focusing on the user's long-term electricity consumption information and unit electricity output value information, and comparing it with the overall situation of the user's industry or city, it includes 10 indicators; for long-term information, the relationship between seasonal changes and user electricity consumption can be seen. If the user's electricity consumption is higher during the peak periods of summer and winter, it means that the potential for regulation is greater; therefore, the first indicator Z 1k The power consumption of the user in week k-1 and week k;

[0029] Second Z 2k and the third indicator Z 3k The difference between the average weekly electricity consumption of all users in the user's city and industry and the weekly electricity consumption of the user;

[0030] The fourth indicator Z 4k The fifth indicator Z is the growth rate of the user's electricity consumption in the kth week. 5k The difference between the power growth rate of the entire network in the kth week and the power growth rate of the user in the kth week;

[0031] The Sixth Z 6k and the seventh indicator Z 7k are the difference between the growth rate of electricity consumption in the city and industry where the user is located in the kth week and the growth rate of electricity consumption of the user in the kth week;

[0032] The Eighth Z 8k The indicator is the year-on-year growth rate of electricity consumption of users in the kth week;

[0033] The ninth Z 9k The indicator is the user's network loss rate in the kth week, and its formula is:

[0034]

[0035] Where D zk represents the total amount of electricity supplied by the power grid to users in week k; D yk represents the electricity consumption of the user in the kth week;

[0036] Similarly, regarding the grid balance index, there is also the assessment of the user power factor. The lower the user power factor, the more reactive power is generated, the greater the harm to the grid, and the more it needs to be included in the orderly power consumption list; therefore, the tenth Z 10k The indicator is the difference between the user's power factor in the kth week and the assessment power factor;

[0037] At this point, the weekly matrix is ​​constructed;

[0038] Interpolate and supplement the missing data. i The calculation formula is:

[0039]

[0040] In the formula, NaN represents missing values; avg() represents the average value; y represents all the data of the user under this indicator, which is a one-dimensional vector;

[0041] Step 2: Construction of a two-stream residual convolutional neural network;

[0042] Step 3: Network training and identification. Before training, build a training data set, find out users who have participated in orderly power consumption this year, and build a weekly matrix and a daily matrix on typical feature days for each user last year, and label them as having the potential to shift peaks and fill valleys. At the same time, randomly select users who have not participated in orderly power consumption from the entire network, the number of which is the same as that of users who have participated in orderly power consumption, and label them as having low potential to shift peaks and fill valleys. At this point, the data set construction is complete.

[0043] Then the entire data set is divided into a training set and a test set at a ratio of 4:1, and input into a two-stream residual network for training; the initial parameters of the model are set: the initialization parameters are set to the default values, the dimensions of the day matrix and the week matrix are set, the average pooling method is used for the pooling layer, and the softmax probability calculation and the number of classifications are set to 2; then the data set is input into the website for training;

[0044] After the training is completed, a network model of the probability that a user has the potential to shift peaks and fill valleys can be obtained; then, the daily matrix and weekly matrix of the user to be evaluated are input into the network to obtain whether the user has the potential to shift peaks and fill valleys and the corresponding probability;

[0045] If the user has the potential to shift peaks and fill valleys, the user is corrected by combining the electricity consumption of the industry, customer credit and environmental pollution factors, and a comprehensive judgment score S is given to the user. The first part of the formula is the potential probability obtained by the residual network, the second part is the electricity consumption of the industry where the user is located, and the third part is the subjective judgment part. The specific formula is:

[0046]

[0047] Where tanh is the hyperbolic tangent function, which keeps the value range of the score S between [0,1]; P is the peak shifting and valley filling potential probability of the dual-stream residual network output; Q and W are both industry electricity consumption coefficients, where Q is the annual electricity consumption of the user's industry as a percentage of the annual electricity consumption of the entire network, and W is the number of users in the user's industry who have participated in orderly electricity consumption as a percentage of the number of users in the entire network who have participated in orderly electricity consumption; C t is the user's credit status coefficient, specifically the number of defaults, overdue payments, and electricity thefts by the user in the tth month; H t is the environmental pollution coefficient of the user, specifically the number of times the user was warned and rectified by the environmental protection department in the tth month; T is the total number of months from the establishment of the power user file to the present;

[0048] Finally, the comprehensive judgment scores S of all users are sorted, and those with higher judgment scores are selected to participate in peak shifting and valley filling.

[0049] In a preferred embodiment, in the day matrix, 24 hours are taken at intervals of 15 minutes, and the times are represented in sequence as 00:00, 00:15, 00:30, ... 23:45, so the value range in a day is 1 to 96.

[0050] In a preferred embodiment, for the daily matrix residual network, the input is a 10×96 matrix. For the data information features with an interval of 15 minutes, the first layer selects a 3×3 convolution kernel, the number of convolution kernels is 8, and the step size can be set to 1×2. The first layer does not use a pooling layer to retain the indicator information; in order to ensure that the indicator dimension does not decrease during the calculation of this layer and the time dimension is consistent with the weekly matrix, a zero padding operation is used during the calculation to ensure that the output dimension is 9×52; the second layer uses a 3×3 convolution kernel, the number of convolution kernels is 16, the step size is 1×1, the pooling layer is 1×2, and zero padding is used to ensure the output dimension; the third layer uses a 3×3 convolution kernel, the number of convolution kernels is 32, the step size is 1×1, the pooling layer is 2×2, and zero padding is also used to ensure the output dimension is 32×5×13;

[0051] Weekly matrix residual network, the input is a 10×52 matrix. For weekly data, because the interval of the time dimension is large, the first and second layer structures are consistent with the second and third layers of the daily matrix, and the output is also 32×5×13;

[0052] After building the two networks, a two-stream residual network is established; a two-stream structure with two residual networks in parallel is used, namely the daily matrix channel and the weekly matrix channel, which respectively extract the features of daily data and weekly data; the user's daily data and the user's weekly data are input into the corresponding two channels respectively, and after identification, the features of the two channels are averaged and fused, that is, the output tensors of the daily matrix and the weekly matrix are added and numerically averaged, and finally input into the fully connected layer and use the softmax method to evaluate the user's potential for peak shifting and valley filling.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The present invention uses a modified dual-stream residual network to evaluate the user's peak-to-valley shifting potential index through multi-dimensional indicators. The network can automatically allocate the weights of neural units and discover the user's peak-to-valley shifting potential from the power big data, reducing manual participation and being less affected by subjective factors. Compared with the scheduling modeling method, it is more objective; compared with the extensive evaluation of the clustering algorithm, the present invention can perform potential analysis on each household.

[0055] 2. The daily matrix and weekly matrix established by the present invention imitate the pixel matrix form of the picture. There is a certain internal connection between the indicators adjacent to each other in the matrix, so that there is a certain relationship between the elements in the matrix, which can improve the accuracy of potential judgment.

[0056] 3. The final judgment formula for users’ peak-to-valley shifting. The double-stream residual convolutional neural network obtains the probability of users’ peak-to-valley shifting potential, combines it with factors such as the electricity consumption of the industry, customer credit, and environmental pollution, and conducts a comprehensive judgment on users based on both subjective and objective factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of a dual-stream residual network structure according to a preferred embodiment of the present invention;

[0058] Figure 2 Schematic diagram of the working process of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0060] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0061] 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 application; 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 their combinations.

[0062] A method for evaluating the peak-load shifting and valley-filling potential of an electricity user comprises the following steps:

[0063] Step 1: Data collection and processing

[0064] Based on the existing construction conditions of the power marketing business application system and the power consumption information collection system, the weekly and daily data of power users can be collected remotely with a granularity of up to 15 minutes. Then, based on the user's production level, power consumption in the industry or region, and electricity charges, the input part of the residual network is established by scientifically and fully considering multiple indicators.

[0065] Inspired by neural networks in the field of computer vision: researchers generally use a two-dimensional pixel matrix to represent images, and computers recognize image information by reading the pixel matrix. Furthermore, there is a certain connection between the values ​​of two adjacent pixels in the image. For example, there is a certain connection between two adjacent pixels of a piece of clothing in the image. The pixel value generally does not change suddenly. If there is a sudden change, it represents a special meaning, such as the edge of the clothing, the junction of the clothing pattern, etc. Therefore, the input of the present invention imitates the two-dimensional pixel matrix of the image, which is two two-dimensional array matrices, divided into a daily matrix and a weekly matrix. There is also a certain connection between the two elements in each matrix, where the daily matrix is ​​a matrix X containing the user's daily data. nm , n represents the nth data indicator, a total of 10 indicators; m represents the mth time point, which is to take values ​​at intervals of 15 minutes for 24 hours, and represent the time in sequence 00:00, 00:15, 00:30, ... 23:45, so the value range in a day is 1 to 96. The weekly matrix is ​​the matrix Z containing the user's weekly data jk ,j represents the jth data indicator, with a total of 10 indicators; k represents the kth week, and the value range in a year is 1 to 52. The specific meanings and indicators of the daily matrix and weekly matrix are introduced below:

[0066] 1. Construction of daily matrix

[0067] Daily matrix X nm Focus on the user's load rate, load growth rate and other information, and compare it with the overall situation of the user's industry or city, including 10 indicators. The following is a brief introduction to each indicator:

[0068] To determine whether a user has the possibility of shifting peak loads to fill valley loads, we first consider the user's power load curve. The larger the load value, the greater the potential for shifting peak loads to fill valley loads. Therefore, the first indicator X 1m It is the load size of the user at the mth time point.

[0069] Consider comparing the user load with the average load of all users in the city or industry where the user is located. If the user load is much higher than the average load, the user's peak load shifting and valley filling power potential may be greater. 2m and the third indicator X 3m They are the difference between the average load of all users in the user's city and industry and the user's load.

[0070] Considering that the greater the load curve change rate, the more obvious the curve fluctuations, the greater the peak-to-valley difference of the curve may be, so the fourth indicator X 4m is the load growth rate of the user at the mth time point, and its formula is:

[0071]

[0072] Considering the similarity between the user's load curve and the whole network's load curve, the more similar the growth rate is, the more consistent the change trend is, and the more likely it is that the peak shifting and valley filling will be effective. 5m It is the difference between the load growth rate of the entire network at the mth time point and the load growth rate of the user at the mth time point.

[0073] Similarly, in order to make the evaluation more accurate, consider the sixth X 6m and the seventh indicator X 7m They are respectively the difference between the load growth rate of the user's city and industry at the mth time point and the load growth rate of the user at the mth time point.

[0074] Considering that the load index is an instantaneous quantity, it cannot well represent the characteristics of the power data in this time period. Therefore, the power data is used as the eighth index X 8m , is the power consumption of the user from the m-1th time point to the mth time point. The power consumption is collected by the user information collection system and the frozen indication curve is obtained by subtracting the two time periods before and after.

[0075] Considering the peak-to-valley ratio of user power consumption, the greater the peak-time ratio, the greater the necessity and potential of the user to shift power to fill the valley. 9m It is the power consumption from the m-1th time point to the mth time point divided by the user's full-day power consumption.

[0076] Considering the economic aspect, the more electricity a user pays, the greater their contribution to the local economy may be. Excessive regulation may cause regional economic problems. 10m The amount of electricity from the m-1th time point to the mth time point multiplied by the time-of-use electricity price for that period.

[0077] At this point, the daily matrix is ​​completed, which is a 10×96 matrix. The indicators start from the load, and then derive related indicators such as load difference and load growth rate, and then add time to get the power indicator, and then add economic considerations from the power to get the electricity cost indicator. There is a certain connection between indicators, and the specific data are objectively obtained.

[0078] 2. Construction of Weekly Matrix

[0079] Zhou Matrix Z jk Focusing on the user's long-term electricity consumption information, unit electricity output value and other information, and similarly comparing it with the overall situation of the user's industry or city, it includes 10 indicators. For long-term information, we can see the relationship between seasonal changes and user electricity consumption. If the user's electricity consumption is high during the peak periods of summer and winter, it means that there is a great potential for regulation. Therefore, the first indicator Z 1k is the user's electricity consumption in week k-1 and week k.

[0080] The difference between the user's weekly electricity consumption and the average electricity consumption of all users in the region and industry, the electricity growth rate and other information are also considered. 2k and the third indicator Z 3k They are the difference between the average weekly electricity consumption of all users in the user's city and industry and the user's weekly electricity consumption.

[0081] The fourth indicator Z 4k The fifth indicator Z is the growth rate of the user's electricity consumption in the kth week. 5k It is the difference between the power growth rate of the entire network in the kth week and the power growth rate of the user in the kth week.

[0082] The Sixth Z 6k and the seventh indicator Z 7k They are respectively the difference between the electricity growth rate of the user's city and industry in the kth week and the electricity growth rate of the user in the kth week.

[0083] Considering the year-on-year growth of user power consumption, if the user power consumption decreases year-on-year, it means that the user may have taken power-saving measures, and the potential for peak shifting and valley filling is reduced. 8k The indicator is the year-on-year growth rate of electricity consumption of users in the kth week.

[0084] In addition to the above considerations of power consumption and power growth rate, power consumption can also be used to determine the user's network loss rate. If the user's network loss rate is higher, it means that the user is more destructive to the balance of supply and demand of the regional power grid, and the more orderly power consumption is needed. Therefore, the ninth Z 9k The indicator is the user's network loss rate in the kth week, and its formula is:

[0085]

[0086] Where D zk represents the total amount of electricity supplied by the power grid to users in week k; D yk Represents the electricity consumption of the user in the kth week.

[0087] Similarly, regarding the grid balance index, there is also the assessment of the user's power factor. The lower the user's power factor, the more reactive power is generated, the greater the harm to the grid, and the more it needs to be included in the orderly power consumption list. 10k The indicator is the difference between the user's power factor in the kth week and the assessment power factor.

[0088] At this point, the weekly matrix is ​​constructed, which is a 10×52 matrix. The relationship between the indicators is similar to that of the daily matrix. The indicators start with the power quantity, derive related indicators such as the power quantity difference and power quantity growth rate, and then consider the network loss rate and power factor indicators of the degree of damage to the balance of supply and demand of the power grid. The indicator data is objectively obtained.

[0089] 3. Data Preprocessing

[0090] Although the current data collection technology is very mature, it is still possible that the user data is missing due to communication errors, metering device failures and other problems. Therefore, it is necessary to interpolate and supplement the missing data. i The calculation formula is:

[0091]

[0092] In the formula, NaN represents missing values; avg() represents the average value; y represents all the data of the user under this indicator, which is a one-dimensional vector.

[0093] Step 2: Construction of two-stream residual convolutional neural network

[0094] refer to Figure 2 Schematic diagram of the dual-stream residual network structure. Through the previous brief analysis of the residual convolutional neural network, it can be seen that the residual network has a good feature extraction and classification effect in the field of image vision, but the matrix elements involved in the present invention are small, and the number of layers of the standard residual network model is not required, and the input matrix is ​​not a block matrix, so the network needs to be modified and lightweight.

[0095] For the daily matrix residual network, the input is a 10×96 matrix. Considering that the information characteristics of data with an interval of 15 minutes may be similar, a 3×3 convolution kernel is selected in the first layer, the number of convolution kernels is 8, and the step size can be set to 1×2 to appropriately reduce the redundancy of the time dimension. Considering the small number of indicators, the first layer does not use the pooling layer to retain the indicator information. In order to ensure that the indicator dimension does not decrease during the calculation of this layer and the time dimension is consistent with the weekly matrix, zero padding is used during calculation to ensure that the output dimension is 9×52. The second layer uses a 3×3 convolution kernel, the number of convolution kernels is 16, the step size is 1×1, the pooling layer is 1×2, and zero padding is used to ensure the output dimension. The third layer uses a 3×3 convolution kernel, the number of convolution kernels is 32, the step size is 1×1, the pooling layer is 2×2, and zero padding is also used to output the dimension of 32×5×13.

[0096] The weekly matrix residual network has a 10×52 input matrix. For weekly data, because the interval of the time dimension is large, the first and second layer structures are consistent with the second and third layers of the daily matrix, and the output is also 32×5×13.

[0097] After building the two networks, the dual-stream residual network in the present invention can be established. Use two modified residual networks in parallel with a dual-stream structure, namely the daily matrix channel and the weekly matrix channel, to extract features of daily data and weekly data respectively. Input the user's daily data and the user's weekly data into the corresponding two channels respectively, and after identification, averagely fuse the features of the two channels, that is, add the output tensors of the daily matrix and the weekly matrix and perform numerical averaging, and finally input them into the fully connected layer and use the softmax method to evaluate the user's potential for peak shifting and valley filling.

[0098] Step 3: Network training and recognition

[0099] Before training, we build a training data set, find out users who have participated in orderly power consumption this year, and build a weekly matrix for each user last year and a daily matrix on a typical feature day (for example, when the network load is the highest in a quarter), and label them as having the potential to shift peaks and fill valleys. At the same time, we randomly select users who have not participated in orderly power consumption from the entire network, the number of which is the same as that of users who have participated in orderly power consumption, and label them as having low potential to shift peaks and fill valleys. At this point, the data set is built.

[0100] Then the entire data set is divided into a training set and a test set at a ratio of 4:1, and input into a two-stream residual network for training. The initial parameters of the model of the present invention are set: the initialization parameters are set to the default values, the dimensions of the day matrix and the week matrix are set, the pooling layer uses the average pooling method, and the softmax probability calculation and the number of classifications are 2. Then the data set is input into the website for training.

[0101] After the training is completed, a network model of the probability that a user has the potential to shift peaks and fill valleys can be obtained. Then, the daily matrix and weekly matrix of the user to be evaluated are input into the network to determine whether the user has the potential to shift peaks and fill valleys and the corresponding probability.

[0102] Furthermore, if the user has the potential to shift peaks and fill valleys, the user is corrected by combining the electricity consumption of the industry, customer credit, environmental pollution and other factors, and a comprehensive judgment score S is given to the user. The first part of the formula is the potential probability obtained by the residual network, the second part is the electricity consumption of the industry where the user is located, and the third part is the subjective judgment part. The specific formula is:

[0103]

[0104] Where tanh is the hyperbolic tangent function, which can keep the value range of the score between [0,1]; P is the peak shifting and valley filling potential probability of the dual-stream residual network output; Q and W are both industry electricity consumption coefficients, where Q is the annual electricity consumption of the user's industry as a percentage of the annual electricity consumption of the entire network, and W is the number of users in the user's industry who have participated in orderly electricity consumption as a percentage of the number of users in the entire network who have participated in orderly electricity consumption; Ct is the user's credit status coefficient, specifically the number of defaults, overdue payments, and electricity thefts by the user in the tth month; H t is the environmental pollution coefficient of the user, specifically the number of times the user is warned and rectified by the environmental protection department in the tth month; T is the total number of months from the establishment of the power user file to the present.

[0105] Finally, the comprehensive judgment scores S of all users are sorted, and those with higher judgment scores are selected to participate in peak shifting and valley filling.

Claims

1. A method for evaluating the peak load shifting and valley load filling potential of power users, characterized in that , including the following steps: Step 1: construct a two-dimensional array matrix, including a daily matrix and a weekly matrix; Among them, the daily matrix is ​​the matrix X containing the user's daily data nm , n represents the nth data indicator, a total of 10 indicators; m represents the mth time point; Daily matrix X nm Focus on the user's load rate and load growth rate information, and compare it with the overall situation of the user's industry or city, including 10 indicators: The larger the user load value, the greater the potential for peak-load shifting and valley-loading. Therefore, the first indicator X 1m is the load size of the user at the mth time point; If the user load is much higher than the average load, the user's peak load shifting and valley load filling potential may be greater; therefore, the second indicator X 2m and the third indicator X 3m The difference between the average load of all users in the city and industry where the user is located and the load of the user; The greater the load curve change rate, the more obvious the curve fluctuations, and the greater the peak-to-valley difference of the curve may be. Therefore, the fourth indicator X 4m is the load growth rate of the user at the mth time point, and its formula is: Considering the similarity between the user's load curve and the whole network's load curve, the more similar the growth rate is, the more consistent the change trend is, and the more likely it is that the peak shifting and valley filling will be effective; therefore, the fifth indicator X 5m It is the difference between the load growth rate of the whole network at the mth time point and the load growth rate of the user at the mth time point; Similarly, the sixth indicator X 6m and the seventh indicator X 7m are the difference between the load growth rate of the user's city and industry at the mth time point and the load growth rate of the user at the mth time point; Power data as the eighth indicator X 8m , is the power consumption of the user from the m-1th time point to the mth time point; the power consumption is obtained by collecting the frozen indication curve through the user information collection system and subtracting the previous and next two time periods; The ninth indicator X 9m The power consumption from the m-1th time point to the mth time point divided by the user's full-day power consumption; The tenth indicator X 10m The amount of electricity from the m-1th time point to the mth time point multiplied by the time-of-use electricity price of the period; At this point, the matrix is ​​constructed; Weekly matrix is ​​the matrix Z containing the user's weekly data jk ,j represents the jth data indicator, with a total of 10 indicators; k represents the kth week, and the value range in a year is 1 to 52; Zhou Matrix Z jk Focusing on the user's long-term electricity consumption information and unit electricity output value information, and comparing it with the overall situation of the user's industry or city, it includes 10 indicators; for long-term information, the relationship between seasonal changes and user electricity consumption can be seen. If the user's electricity consumption is higher during the peak periods of summer and winter, it means that the potential for regulation is greater; therefore, the first indicator Z 1k The power consumption of the user in week k-1 and week k; Second Z 2k and the third indicator Z 3k The difference between the average weekly electricity consumption of all users in the user's city and industry and the weekly electricity consumption of the user; The fourth indicator Z 4k The fifth indicator Z is the growth rate of the user's electricity consumption in the kth week. 5k The difference between the power growth rate of the entire network in the kth week and the power growth rate of the user in the kth week; The Sixth Z 6k and the seventh indicator Z 7k are the difference between the growth rate of electricity consumption in the city and industry where the user is located in the kth week and the growth rate of electricity consumption of the user in the kth week; The Eighth Z 8k The indicator is the year-on-year growth rate of electricity consumption of users in the kth week; The ninth Z 9k The indicator is the user's network loss rate in the kth week, and its formula is: Where D zk represents the total amount of electricity supplied by the power grid to users in week k; D yk represents the electricity consumption of the user in the kth week; Similarly, regarding the grid balance index, there is also the assessment of the user power factor. The lower the user power factor, the more reactive power is generated, the greater the harm to the grid, and the more it needs to be included in the orderly power consumption list; therefore, the tenth Z 10k The indicator is the difference between the user's power factor in the kth week and the assessment power factor; At this point, the weekly matrix is ​​constructed; Interpolate and supplement the missing data. i The calculation formula is: In the formula, NaN represents missing values; avg() represents the average value; y represents all the data of the user under this indicator, which is a one-dimensional vector; Step 2: Construction of a two-stream residual convolutional neural network; Step 3: Network training and identification. Before training, build a training data set, find out users who have participated in orderly power consumption this year, and build a weekly matrix and a daily matrix on typical feature days for each user last year, and label them as having the potential to shift peaks and fill valleys. At the same time, randomly select users who have not participated in orderly power consumption from the entire network, the number of which is the same as that of users who have participated in orderly power consumption, and label them as having low potential to shift peaks and fill valleys. At this point, the data set construction is complete. Then the entire data set is divided into a training set and a test set at a ratio of 4:1, and input into a two-stream residual network for training; the initial parameters of the model are set: the initialization parameters are set to the default values, the dimensions of the day matrix and the week matrix are set, the average pooling method is used for the pooling layer, and the softmax probability calculation and the number of classifications are set to 2; then the data set is input into the website for training; After the training is completed, a network model of the probability that a user has the potential to shift peaks and fill valleys can be obtained; then, the daily matrix and weekly matrix of the user to be evaluated are input into the network to obtain whether the user has the potential to shift peaks and fill valleys and the corresponding probability; If the user has the potential to shift peaks and fill valleys, the user is corrected by combining the electricity consumption of the industry, customer credit and environmental pollution factors, and a comprehensive judgment score S is given to the user. The first part of the formula is the potential probability obtained by the residual network, the second part is the electricity consumption of the industry where the user is located, and the third part is the subjective judgment part. The specific formula is: Where tanh is the hyperbolic tangent function, which keeps the value range of the score S between [0,1]; P is the peak shifting and valley filling potential probability of the dual-stream residual network output; Q and W are both industry electricity consumption coefficients, where Q is the annual electricity consumption of the user's industry as a percentage of the annual electricity consumption of the entire network, and W is the number of users in the user's industry who have participated in orderly electricity consumption as a percentage of the number of users in the entire network who have participated in orderly electricity consumption; C t is the user's credit status coefficient, specifically the number of defaults, overdue payments, and electricity thefts by the user in the tth month; H t is the environmental pollution coefficient of the user, specifically the number of times the user was warned and rectified by the environmental protection department in the tth month; T is the total number of months from the establishment of the power user file to the present; Finally, the comprehensive judgment scores S of all users are sorted, and those with higher judgment scores are selected to participate in peak shifting and valley filling.

2. A method for evaluating the peak load shifting and valley filling potential of an electricity user according to claim 1, characterized in that ,In the day matrix, it takes values ​​at 15-minute intervals for 24 hours, and represents the time in sequence 00:00, 00:15, 00:30, … 23:45, so the value range in a day is 1 to 96.

3. A method for evaluating the peak load shifting and valley filling potential of an electricity user according to claim 2, characterized in that For the daily matrix residual network, the input is a 10×96 matrix. For the data information features with an interval of 15 minutes, the first layer selects a 3×3 convolution kernel, the number of convolution kernels is 8, and the step size can be set to 1×2. The first layer does not use a pooling layer to retain the indicator information; in order to ensure that the indicator dimension does not decrease during the calculation of this layer and the time dimension is consistent with the weekly matrix, the calculation uses zero padding to ensure that the output dimension is 9×52; the second layer uses a 3×3 convolution kernel, the number of convolution kernels is 16, the step size is 1×1, the pooling layer is 1×2, and zero padding is used to ensure the output dimension; the third layer uses a 3×3 convolution kernel, the number of convolution kernels is 32, the step size is 1×1, the pooling layer is 2×2, and zero padding is also used to ensure the output dimension is 32×5×13; Weekly matrix residual network, the input is a 10×52 matrix. For weekly data, because the interval of the time dimension is large, the first and second layer structures are consistent with the second and third layers of the daily matrix, and the output is also 32×5×13; After building the two networks, a two-stream residual network is established; A dual-stream structure using two parallel residual networks, one for the daily matrix channel and the other for the weekly matrix channel, is used to extract features of daily data and weekly data respectively. The user's daily data and weekly data are input into the corresponding two channels respectively. After identification, the features of the two channels are averaged and fused. That is, the output tensors of the daily matrix and the weekly matrix are added and averaged respectively. Finally, they are input into the fully connected layer and the softmax method is used to evaluate the user's potential for peak shifting and valley filling.

Citation Information

Patent Citations

  • Novel distribution network battery energy storage system optimal?allocation?algorithm

    CN103824123A

  • Orderly power utilization decision method of user on the basis of fuzzy analytic hierarchy process

    CN105243609A

  • Cloud energy storage charging and discharging strategy forming method based on GRU multi-step prediction technology

    CN112217195A

  • Distributed power supply planning method based on capacity limitation of power supply equipment

    CN113128844A