Distribution storage user identification method based on two-stage multi-feature

Abstract

The invention discloses a two-stage multi-feature-based distribution and storage user identification method, which overcomes the problems of high cost, low efficiency and poor identification precision of distribution and storage user identification in the prior art, and comprises the following steps: clustering users by using a kmeans algorithm and determining a potential distribution and storage user group in combination with a custom rule; carrying out multi-dimensional typical distribution and storage user load feature and energy consumption feature extraction on the potential distribution and storage users based on the preliminary identification result; based on the extracted multiple features, a machine learning classification model is adopted to carry out accurate identification training on the distribution and storage users; and utilizing the trained machine learning classification model to identify the storage allocation user. Through a two-stage algorithm, end-to-end classification identification is directly realized based on the power grid side data, the accuracy of distribution and storage user identification is improved, a large amount of manpower troubleshooting is not needed, the manpower cost and the troubleshooting time cost are greatly reduced, and the user distribution and storage identification accuracy is high.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid optimization operation, and in particular to a storage distribution user identification method based on two-stage multi-features. Background Art

[0002] With the continuous construction and development of new power systems, the energy storage industry has developed rapidly, and more and more industrial and commercial users have begun to install customer-side energy storage. Configuring customer-side energy storage can achieve the timed storage and release of electric energy to meet the user's personalized electricity needs, such as peak-valley electricity price arbitrage, demand management, power quality improvement, emergency standby, etc.

[0003] However, customer-side energy storage is usually installed behind the user's meter, that is, it is directly connected to the user's electricity facilities. It does not have a separate meter and is generally independent of the grid. This is because customer-side energy storage mainly serves the user's own electricity optimization, rather than directly participating in the operation and regulation of the grid. This installation method allows the customer-side energy storage system to flexibly optimize the charging and discharging strategy according to the user's own electricity consumption mode and the current electricity price policy. In addition, due to the diverse system platform architectures and heterogeneous intelligent monitoring terminals adopted by energy storage suppliers, it is also difficult to securely and uniformly access the customer-side energy storage monitoring data to the grid system. There is currently no unified standard, which makes it difficult for grid operators to identify whether users are configured with customer-side energy storage. By identifying whether users are configured with user-side energy storage, grid operators can better understand the distribution of grid energy storage resources, optimize the allocation and local consumption of power resources, and improve the grid's supply guarantee capabilities. At present, manual investigation is usually used to determine whether users are configured with energy storage, which has high labor costs and time costs. Summary of the invention

[0004] The purpose of the present invention is to overcome the problems of high cost, low efficiency and poor identification accuracy of distribution and storage user identification in the prior art, and provide a distribution and storage user identification method based on two-stage multi-features. Through a two-stage algorithm, end-to-end classification and identification is realized directly based on grid-side data, thereby improving the accuracy of distribution and storage user identification. There is no need for a large amount of manpower investigation, which greatly reduces the manpower cost and investigation time cost. The user distribution and storage identification has high accuracy, and rapid identification and reasoning of whether a large number of power users are equipped with storage can be realized.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for identifying storage users based on two-stage multi-features, characterized in that it includes the following steps: S1: Use kmeans algorithm to cluster users and combine custom rules to determine potential storage user groups; S2: Based on the preliminary identification results, multi-dimensional typical load characteristics and energy consumption characteristics of potential distribution storage users are extracted; S3: Based on the extracted multiple features, a machine learning classification model is used to accurately identify and train distribution storage users; S4: Use the trained machine learning classification model to identify storage users.

[0006] In the first stage, the kmeans algorithm was used to achieve preliminary classification of massive users, and potential distribution and storage users were screened out in combination with custom rules. In the second stage, the external load curve characteristics and energy consumption characteristics that may exist in distribution and storage users were fully considered, and a machine learning classification model was introduced to achieve the training and reasoning identification of distribution and storage users. Compared with the manual investigation method, the two-stage distribution and storage user identification method proposed in the present invention directly realizes end-to-end classification and identification based on the grid-side data, without the need for a large amount of manpower investigation. This method has high popularity and fast calculation speed, and can realize the rapid identification and reasoning of whether massive power users are equipped with storage.

[0007] Preferably, the step S1 comprises: S1.1: Obtain the historical load data of each user, average the load data at the same time, and obtain the typical load curve of each user; S1.2: Take the typical load curve of each user as a sample, construct a sample set, cluster the sample set using the kmeans algorithm, and determine the number of clusters; S1.3: Calculate the cluster center load curve of each cluster, the average load Lv of the time-of-use electricity price valley period of the cluster center load curve of each cluster, and the average load La of the first hour of the valley period; S1.4: If Lv>2*La, all users in the cluster are preliminarily defined as potential storage users.

[0008] Preferably, the step S2 comprises: S2.1: Calculate the statistical characteristics of the typical load curve of the potential storage user object; S2.2: Obtain the electricity consumption data of the potential distribution and storage user group during the historical time-of-use electricity price period, and calculate the energy consumption characteristics of the time period; S2.3: Calculate the load characteristics of the typical load curve of the potential storage user object at the TOU electricity price mutation point; S2.4: Calculate the similarity characteristics between the daily load curve of the potential storage user object and the average typical user load curve; S2.5: Calculate the photovoltaic-storage association characteristics of the potential storage allocation user object.

[0009] Preferably, S2.5 includes: calculating the user's load average during the valley price period during the day and the load average during the period with the most abundant sunlight based on the typical load curve of the potential energy storage user object, and calculating the difference between the two, and judging the probability of the user configuring energy storage based on the size of the difference.

[0010] Preferably, the similarity feature is a quotient of the sum of the Pearson correlation coefficients between the daily load curve of the user within a specified date interval and the average typical user load curve and the selected comparison days.

[0011] As a preferred method, the extracted multiple features are merged, and for user i, the feature vector input into the machine learning classification model is X i ={x1,x2,…,x n}, where x n Represents the nth feature after multi-feature extraction.

[0012] As a preference, the binary cross entropy loss is used as the loss function Loss of the machine learning classification model. Among them, y is the real label of whether the user has storage. is the probability that the machine learning classification model predicts that it is allocated storage.

[0013] Preferably, the statistical characteristics include the maximum load and the minimum load in each time period, the load variance and the load average; the energy consumption characteristics include the power value in each peak and valley time period, the peak-valley power difference and the valley power difference.

[0014] As a preferred method, the elbow method is used to determine the number of clusters: For different values ​​of the number of clusters K, run the K-means algorithm to form K clusters each time and calculate the SSE value of each cluster; draw a line graph of the K value and the corresponding SSE value with the K value as the horizontal axis and the SEE value as the vertical axis; Find the point on the line graph where the SSE decrease rate decreases suddenly. The K value corresponding to this point is the optimal number of clusters.

[0015] Preferably, for each cluster, the average position of all load curves in the cluster is calculated, and the average position is the new cluster center.

[0016] Therefore, the present invention has the following beneficial effects: 1. In the first stage, the kmeans algorithm is introduced to realize the clustering of massive users and combine with custom rules to determine the potential distribution and storage user groups, which realizes the rapid preliminary screening of massive users and reduces the computing cost; in the second stage, the load characteristics and energy consumption characteristics of the distribution and storage users are fully considered, and the photovoltaic storage association characteristics when the users have distributed photovoltaics are considered, and the corresponding feature engineering is carried out to improve the accuracy of distribution and storage user identification.

[0017] 2. Introduce the machine learning CatBoost classification model to realize the training and reasoning of storage user identification. Compared with the existing manual investigation method, it adopts a data-driven approach to reduce labor and time costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The figure is a flowchart of the steps of the storage user identification method based on two-stage multiple features. DETAILED DESCRIPTION

[0019] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments: Embodiment 1: like Figure 1 In the illustrated embodiment, a method for identifying storage users based on two-stage multiple features can be seen, and its operation process is as follows: Step 1, using the kmeans algorithm to cluster users and combining custom rules to determine the potential storage user group; Step 2, based on the preliminary identification results, multi-dimensional typical storage user load characteristics and energy consumption characteristics of potential storage users are extracted; Step 3, based on the extracted multiple features, a machine learning classification model is used to perform accurate storage user identification training; Step 4, using the trained machine learning classification model to identify storage users.

[0020] The two-stage multi-feature based storage user identification method provided in this embodiment includes preliminary identification in the first stage and feature engineering in the second stage.

[0021] In the first stage, load curve clustering is performed to find several clusters, and the cluster center of each cluster is preliminarily screened using the Huayun algorithm to find suspicious users. The kmeans algorithm is used to preliminarily classify massive users, and combined with custom rules to screen out potential storage users.

[0022] In the second stage, the external load curve characteristics and energy consumption characteristics of the distribution storage users are fully considered, and a machine learning classification model is introduced to realize the training and inference identification of distribution storage users. External load curve characteristics and energy consumption characteristics include various statistical characteristics of the load curve (maximum and minimum loads in different time periods, load curve variance, etc.); power consumption characteristics in different time periods (peaks and valleys), peak-valley differences, peak-valley differences, and valley-valley differences; mutation points (the number of signal mutation points, signal mutation points in a specific time period (for example, the difference between 9 o'clock and 8:45); continuous constant points (a flat curve in the user's typical load curve); daily variation pattern of the load curve: if the user's daily load curve shows a similar pattern over a period of time, especially during peak demand periods, this may indicate that the user uses an energy storage system to manage daily load fluctuations (which can be represented by the mean or variance of cosine similarity); the correlation between load and photovoltaic output: for users who have installed photovoltaic systems, if the load curve does not increase significantly during periods of sufficient sunlight, but is relatively stable during periods of insufficient sunlight, this may indicate that the user uses an energy storage system to balance the intermittent nature of photovoltaic output, and whether the user's load increases during the valley period when photovoltaic output is the largest (compared to the previous period).

[0023] Compared with the manual investigation method, the two-stage storage user identification method provided in this embodiment directly realizes end-to-end classification and identification based on the grid-side data, without the need for a large amount of manpower investigation. This method has high popularity and fast calculation speed, and can realize rapid identification and reasoning of whether a large number of power users are equipped with storage.

[0024] Embodiment 2: Step 1: Use the kmeans algorithm to cluster users and combine custom rules to determine the potential distribution and storage user group; Step 2: Based on the preliminary identification results, extract the multi-dimensional typical distribution and storage user load characteristics and energy consumption characteristics of potential distribution and storage users; Step 3: Based on the extracted multiple features, use the machine learning classification model to accurately identify the distribution and storage users; Step 4: Use the trained machine learning classification model to identify the distribution and storage users.

[0025] Compared with the direct dispatch of research personnel by the relevant departments of the power operator to inquire on-site users, the two-stage storage user identification method proposed in this embodiment directly realizes end-to-end classification and identification based on grid-side data, without the need for a large amount of manpower investigation. This method has high popularity and fast calculation speed, and can realize rapid identification and reasoning of whether a large number of power users are equipped with storage.

[0026] The technical scheme and technical effects of the present invention will be further described below through specific examples. The following examples are explanations of the present invention and the present invention is not limited to the following examples.

[0027] Phase 1: Preliminary identification and screening of storage users.

[0028] The number of users served by power operators is huge. In order to improve the identification accuracy of distribution and storage users, the kmeans algorithm is first used to cluster the massive users and combined with custom rules to determine the potential distribution and storage user groups, thereby achieving preliminary identification and screening of the massive users.

[0029] The specific process of the first stage is as follows: Step (1): Obtain the historical load data of each user from the current date (D-day) to 365 days ago (D-365 days), remove the load data of non-working days, and calculate the average value of the load data at the same time to obtain the typical load curve of each user (taking one day as an example). With the load sampling frequency of 15 minutes of the current data acquisition system, the typical load curve contains 96 points of data.

[0030] Step (2): Take the typical load curve of each user as a sample, combine all samples to form a sample set of the typical load curve of the input user, use the kmeans algorithm to cluster the sample set of the typical load curve of the user, and use the elbow method to determine the number of clusters.

[0031] The steps to determine the number of clusters using the elbow method are as follows: a) For different values ​​of the number of clusters K (starting from 2 and gradually increasing), run the K-means algorithm, form K clusters each time, and calculate the SSE of each cluster. SSE refers to the sum of the squares of the Euclidean distances from all points in a cluster to its cluster center (centroid).

[0032] b) With K value as the horizontal axis and SEE value as the vertical axis, draw a line graph of K value and corresponding SSE value.

[0033] c) Find the point on the line graph where the SSE decrease rate decreases sharply. This point is usually called the "elbow point". The K value corresponding to the "elbow point" is usually selected as the optimal number of clusters.

[0034] Step (3): Calculate the cluster center load curve of each cluster. For each cluster, calculate the average position of all typical load curves in the cluster, and this average position becomes the new cluster center. For the cluster center of the jth cluster, its cluster center μ j It can be calculated by the following formula: Among them, C j is the sample set of typical load curves of all input users in the jth cluster, N j It is cluster C j The number of samples in x i It is cluster C j Typical load curve of user i in .

[0035] Step (4): Calculate the average load L during the valley period of the time-of-use electricity price of the cluster center load curve of each cluster v And the average load L in the first hour of the off-peak period a , if the average load during the valley period of the time-of-use electricity price is greater than twice the average load of the hour before the valley period, that is: L v >2*L a , All users in the cluster are initially defined as potential storage users for further calculation in the second phase.

[0036] The second stage: storage user identification model based on multiple features.

[0037] Based on the calculations in the first stage, the potential distribution storage user object group can be determined. The multi-feature typical distribution storage user energy consumption feature extraction is carried out for the target object (pre-distribution storage user object group), and the machine learning model is used for training and reasoning based on the extracted multi-features.

[0038] The main steps include: Step 1: Based on the preliminary identification results, extract the multi-dimensional load characteristics and energy consumption characteristics of potential storage users.

[0039] (1) Statistical load characteristics.

[0040] According to the typical load curve (daily) of the user calculated in step (1) of the first stage, various statistical characteristics of the typical load curve of the potential storage user object are calculated. The statistical characteristics mainly include the maximum load, minimum load, load variance and load average value in each period of peak, flat and valley.

[0041] (2) Electricity characteristics.

[0042] Obtain the electricity data of each time-of-use electricity price period from the current date (D-day) to 365 days ago (D-365 days) for the potential storage user group, and calculate the energy consumption characteristics of each time period. The energy consumption characteristics specifically include the peak and valley electricity values ​​in each period, the peak-valley electricity difference, the peak-valley electricity difference, and the valley-valley electricity difference.

[0043] (3) Load characteristics at the time-of-use electricity price mutation point.

[0044] Based on the typical load curve (daily) of the user calculated in step (1) in the first stage, the load characteristics of the typical load curve of the potential storage user object at the time-of-use electricity price mutation point are calculated.

[0045] For example, if the A price is implemented in the period [tN, t), and the B price is implemented in the period [t, t+N), then time t is the time-of-use price mutation point. The current time-of-use electricity prices are mostly divided into peak-valley or peak-valley prices, and different electricity prices are implemented in multiple periods of the day, so there will be multiple time-of-use price mutation points. Calculate the load difference value L of the user at each time-of-use price mutation point Δtou : L Δtou =L tou -L tou-last , Where, L tou is the load value at the time of the sudden change of the time-of-use electricity price, L tou-last It is the load value at the moment before the time-of-use electricity price mutation point (the time interval is determined by the measurement frequency of the monitoring device, which is 15 minutes in this embodiment).

[0046] (4) Similarity characteristics of daily load curves.

[0047] Based on the typical user load curve (daily) calculated in step (1) of the first stage, calculate the Pearson correlation coefficient between the daily load curve of the user in the specified date range in the past and the average typical user load curve. Assume that the load curve of a typical user j is denoted as Lj mean , the daily load curve of the user on a certain day is recorded as Lj actual , then the calculation formula of the Pearson correlation coefficient r on that day is: Where, L mi For Lj mean The i-th observation value of ai For Lj actual The i-th observation value of For Lj mean The sample mean of For Lj actual The sample mean of .

[0048] Therefore, the similarity characteristics of the daily load curve of user j can be expressed as the quotient of the sum of the Pearson correlation coefficients between the daily load curve of the user and the average typical user load curve within the specified date interval and the selected comparison days. That is: In the formula, r d is the Pearson correlation coefficient between the user j on the dth day and the average typical user load curve, and n is the number of selected comparison days.

[0049] (5) Photo-storage correlation characteristics.

[0050] Since a high proportion of industrial and commercial users install distributed photovoltaics, and most of them adopt the "self-generation for own use, surplus power to the grid" operation mode, if this type of user configures energy storage, the user's load curve will become more complicated, making it difficult to identify whether the user has configured energy storage. For example, during periods of sufficient sunlight, the output of distributed photovoltaics increases, and the user's required mains electricity will decrease, and the load curve will generally show a downward trend. However, when users configure energy storage, distributed photovoltaic power generation is not only used to meet the user's own load needs, but also needs to be used to charge the user-side energy storage, so as to achieve on-site consumption of distributed photovoltaics as much as possible. Considering that the charging and discharging power of the energy storage on the user side is generally large and is generally charged during valley hours, if the charging power is greater than the maximum output of distributed photovoltaics, mains electricity is still required for energy storage charging. Therefore, the photovoltaic storage association feature is introduced in this embodiment.

[0051] According to the typical load curve of the user calculated in step (1) in the first stage, calculate the average load value Lv of the user during the daytime valley electricity price period dayand the load mean value L during the period with the most abundant sunlight (13:00-14:00 is selected in this embodiment, and in other embodiments, it can be modified according to actual conditions) pvmax , and calculate the load mean value Lv during the daytime valley electricity price period day The average load value L during the period with the most abundant sunlight pvmax The difference L diff , that is, L diff =Lv day -L pvmax In general, if the average load value L during the period with the most abundant sunlight is pvmax Still less than the average load value Lv during the daytime valley electricity price period day , the probability of users configuring energy storage is relatively high.

[0052] Step 2: Configure the user identification model.

[0053] After completing the multi-feature extraction in the first step, based on the extracted multi-features, this embodiment uses the machine learning classification model CatBoost to perform accurate identification training for storage users.

[0054] Since the input features of the machine learning classification model need to be processed into a table form, this embodiment directly uses a merge operation to realize the combination of all the multiple features extracted in the first step. For example, for user i, its input classifier feature vector can be expressed as X i ={x1,x2,…,x n}, where x1 represents the first feature after multi-feature extraction, which is defined as the maximum load during the peak period, and x n It represents the nth feature after multi-feature extraction, and the same applies to other features.

[0055] The multi-feature-based storage user identification is a classification problem, that is, the corresponding classification "boundary" is found by inputting features. In this embodiment, CatBoost machine learning classifier is selected for training, and binary cross entropy loss is used as the loss function Loss of the machine learning classification model (storage user identification model). The loss function Loss is defined as: Where y is the real label (0 or 1) of whether the user is equipped with storage. It is the probability that the machine learning classification model predicts the positive class (allocation storage).

[0056] Step 3: After completing the machine learning classification model training, use the trained machine learning classification model to identify the storage allocation of massive users.

[0057] The two-stage multi-feature based storage user identification method provided in this embodiment has the following beneficial effects: 1. In the first stage, the kmeans algorithm is introduced to cluster massive users and combined with custom rules to determine the potential storage user group, which enables rapid preliminary screening of massive users and reduces computing costs.

[0058] 2. In the second stage, the load characteristics and energy consumption characteristics of the storage users were fully considered, and the photovoltaic storage correlation characteristics when the users had distributed photovoltaics were considered. Corresponding feature engineering was carried out to improve the accuracy of storage user identification.

[0059] 3. Introduce the machine learning CatBoost classification model to realize the training and reasoning of storage user identification. Compared with the existing manual investigation method, it adopts a data-driven approach to reduce labor and time costs.

[0060] 4. By identifying whether the user has configured a customer-side energy storage device, the grid operator can better understand the performance indicators and operating conditions of the customer-side energy storage, thereby optimizing the operation and management strategies of the grid and promoting customer-side energy storage services to ensure power supply and optimize power system operation.

[0061] The two-stage multi-feature-based storage user identification system provided in this embodiment is based on the kmeans algorithm combined with custom rules to achieve preliminary identification of whether a large number of users are potential storage users in the first stage. In the second stage, based on the preliminary identification results, the multi-dimensional load characteristics and energy consumption characteristics of the storage users are extracted, and the machine learning CatBoost classification model is used based on the extracted multi-dimensional features to achieve accurate identification of storage users. Through the two-stage algorithm, end-to-end storage user identification is directly achieved based on data-driven form, which greatly reduces labor costs and troubleshooting time costs, while ensuring the accuracy and scalability of user storage identification.

[0062] Embodiment three: This embodiment provides a storage user identification system based on two stages and multiple features, including: Data collection module: collects the historical load data of each user from the current date (D-day) to 365 days ago (D-365 days) at a load sampling frequency of once every 15 minutes, and sends the collected historical load data to the data processing module.

[0063] Data processing module: pre-process the historical user load data transmitted by the data acquisition module, remove outliers, and supplement missing values. It can also perform pre-processing such as interpolation, smoothing, and normalization. Then, for the pre-processed load data, calculate the average value of the load data at each sampling time point, and draw a typical load curve with the sampling time as the horizontal coordinate and the average value of the load data corresponding to the sampling time as the vertical coordinate. Or, draw the load curve of the user every day with the sampling time as the horizontal coordinate and the load data corresponding to the sampling time as the vertical coordinate, and then fit the load curve of the user every day to obtain the typical load curve of the user.

[0064] Potential storage user group identification module: Taking the typical load curve of a single user as a sample, a sample set containing the typical load curves of all users is constructed, and the sample set is clustered using the kmeans algorithm to screen out potential storage users.

[0065] Multi-feature extraction module for allocation and storage users: extracts multi-feature typical allocation and storage user energy consumption features from potential allocation users, and sends the extracted multi-features to the machine learning classification model.

[0066] Machine learning classification model training module: Based on the extracted multiple features of typical storage users, a merging operation is performed to obtain the input feature vector of the machine learning classification model, and the machine learning classifier CatBoost is used to realize the classification and identification of user storage.

[0067] In this embodiment, the data acquisition module includes a current sensor and a voltage sensor for measuring the user's voltage, and also includes a signal acquisition module, the current sensor and the voltage sensor are both electrically connected to the signal acquisition module; wherein the current sensor is connected to the socket to measure the socket current in the user's home. The signal acquisition module is electrically connected to the data processing module to collect current and voltage signals; the data processing module is electrically connected to the signal transmission module; the signal transmission module is connected to the cloud storage; the cloud storage is connected to the operation terminal through the network to store the electricity usage information, and the operation terminal queries and stores the electricity usage information.

[0068] In this embodiment, the clustering process of the kmeans algorithm for clustering the sample set is: obtaining the cluster k value based on the natural nearest neighbor algorithm; then calculating the local density ρ', the minimum distance δ and the decision value γ of the sample; then selecting the cluster center through the automatic cluster center selection method; optimizing the coding string to achieve the optimal clustering of historical samples by using genetic operations such as crossover and mutation; and finally allocating the remaining sample points according to the sample allocation strategy based on the weighted KNN graph.

[0069] Through the two-stage multi-feature-based storage user identification system provided in this embodiment, the power grid operator can identify whether the user has configured user-side energy storage, thereby better understanding the distribution of power grid energy storage resources, optimizing power resource allocation and local consumption, and improving the power grid's supply guarantee capability. At the same time, by identifying whether the user has configured a customer-side energy storage device, the power grid operator can better understand the performance indicators and operating conditions of the user's customer-side energy storage, thereby optimizing the operation and management strategies of the power grid, and promoting customer-side energy storage services for power supply guarantee and power system operation optimization.

[0070] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A two-stage multi-feature based storage user identification method, characterized in that: The following steps are involved: S1: Use kmeans algorithm to cluster users and combine custom rules to determine potential storage user groups; S2: Based on the preliminary identification results, extract the load characteristics and energy consumption characteristics of potential storage users in multiple dimensions; S3: Based on the extracted multiple features, a machine learning classification model is used to accurately identify storage users; S4: Use the trained machine learning classification model to identify storage users.

2. According to the two-stage multi-feature based storage user identification method of claim 1, it is characterized in that: The step S1 comprises: S1.1: Obtain the historical load data of each user, average the load data at the same time, and obtain the typical load curve of each user; S1.2: Take the typical load curve of each user as a sample, construct a sample set, cluster the sample set using the kmeans algorithm, and determine the number of clusters; S1.3: Calculate the cluster center load curve of each cluster and the average load L of the time-of-use electricity price valley period of the cluster center load curve of each cluster v And the average load L in the first hour of the off-peak period a ; S1.4: If L v >2*L a , then all users in the cluster are initially defined as potential storage users.

3. According to the two-stage multi-feature based storage user identification method of claim 1, it is characterized in that: The step S2 comprises: S2.1: Calculate the statistical characteristics of the typical load curve of the potential storage user object; S2.2: Obtain the electricity consumption data of the potential distribution and storage user group during the historical time-of-use electricity price period, and calculate the energy consumption characteristics of the time period; S2.3: Calculate the load characteristics of the typical load curve of the potential distribution storage user object at the TOU electricity price mutation point; S2.4: Calculate the similarity characteristics between the daily load curve of the potential storage user object and the average typical user load curve.

4. According to the two-stage multi-feature based storage user identification method of claim 3, it is characterized in that: The S2 also includes calculating the photovoltaic energy storage association characteristics of the potential energy storage user object: according to the typical load curve of the potential energy storage user object, the average load of the user during the valley price period during the day and the average load during the period with the most abundant sunlight are calculated, and the difference between the two is calculated, and the probability of the user configuring energy storage is determined according to the size of the difference.

5. A two-stage multi-feature based storage user identification method according to claim 3 or 4, characterized in that: The similarity feature is the quotient of the sum of the Pearson correlation coefficients between the daily load curve of the user within the specified date interval and the average typical user load curve and the selected comparison days.

6. A method for identifying storage users based on two-stage multi-features according to claim 1, characterized in that: The extracted multiple features are merged, and for user i, the feature vector input into the machine learning classification model is X i ={x1,x2,…,x n }, where x n Represents the nth feature after multi-feature extraction.

7. A two-stage multi-feature based storage user identification method according to claim 1 or 2 or 3 or 4 or 6, characterized in that: Use binary cross entropy loss as the loss function Loss of the machine learning classification model. Among them, y is the real label of whether the user has storage. is the probability that the machine learning classification model predicts that it is allocated storage.

8. A two-stage multi-feature based storage user identification method according to claim 3 or 4, characterized in that: The statistical characteristics include the maximum load and the minimum load in each time period, the load variance and the load average; the energy consumption characteristics include the power value in each peak and valley time period, the peak-valley power difference and the valley power difference.

9. The method for identifying storage users based on two-stage multi-features according to claim 2 is characterized in that: Use the elbow method to determine the number of clusters: For different values ​​of the number of clusters K, run the K-means algorithm to form K clusters each time and calculate the SSE value of each cluster; draw a line graph of the K value and the corresponding SSE value with the K value as the horizontal axis and the SEE value as the vertical axis; Find the point on the line graph where the SSE decrease rate decreases suddenly. The K value corresponding to this point is the optimal number of clusters.

10. A two-stage multi-feature based storage user identification method according to claim 2 or 9, characterized in that: For each cluster, the average position of all load curves in the cluster is calculated, and the average position is the new cluster center.

Images