Power load prediction method and system based on privacy protection distributed clustering

Abstract

The invention discloses a power load prediction method and system based on privacy protection distributed clustering, and the method comprises the steps: firstly extracting load features to form a local data set, then executing an iteration clustering process, executing local clustering distribution by each power user in each iteration process to generate a plurality of clusters, and carrying out the clustering of the plurality of clusters; calculating a secret fragment according to the local sum of the load features in each cluster and the number of the load features, updating a clustering center according to the secret fragment, re-executing local clustering distribution until a clustering termination condition is reached, generating a final cluster, constructing a corresponding LSTM load prediction model for each cluster, and performing power load prediction. According to the method, the privacy of the power load data of the users is protected, meanwhile, high-quality user grouping and accurate load prediction are achieved, even if part of the users do not participate in calculation, the system can operate normally, and the method is suitable for electricity selling companies to carry out user group division and refined load prediction in the power market environment.

Description

Technical Field

[0001] The present invention relates to the fields of power market, load forecasting and privacy computing, and in particular to a power load forecasting method and system based on privacy-preserving distributed clustering. Background Art

[0002] Cluster analysis, as an important preprocessing step for load forecasting, can group users with similar electricity consumption characteristics together, thereby improving the accuracy of subsequent forecasting models. However, traditional clustering methods require direct access to users' detailed electricity consumption data, such as time-of-day electricity consumption and load curve characteristics. These data directly reflect users' production and operation activities and electricity consumption behavior patterns and are considered core business secrets. Against the backdrop of increasingly fierce competition in the electricity market, users' demand for data privacy protection is becoming increasingly strong, making it difficult for power sales companies to collect original electricity consumption data. Therefore, there is an urgent need for a method that can provide high-precision clustering results while protecting the privacy of users' electricity load data, laying the foundation for subsequent deep learning-based load forecasting. Summary of the Invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a power load forecasting method and system based on privacy-preserving distributed clustering, which can accurately group large users without obtaining the users' original electricity consumption data, and provide support for the LSTM-based load forecasting model.

[0004] Technical solution: The power load forecasting method based on privacy-preserving distributed clustering described in the present invention includes the following steps:

[0005] The power users and power sales companies participating in the cluster calculation are identified as central coordinators. Each power user extracts load characteristics based on its own historical power load data to form a local data set.

[0006] Initial cluster centers are determined and distributed to power users. An iterative clustering process is performed based on local data sets. In each iteration, each power user performs local cluster allocation to generate several clusters. A secret shard is calculated based on the local sum of load characteristics in each cluster and the number of load characteristics. The cluster center is updated based on the secret shard, and local cluster allocation is re-executed until the cluster termination condition is reached to generate the final cluster.

[0007] Add calibration noise to the cluster centers in the final clustering;

[0008] A corresponding LSTM load forecasting model is constructed for each cluster after adding calibration noise to perform power load forecasting.

[0009] Furthermore, after updating the cluster center according to the secret shards, it also includes power load anomaly detection and empty cluster processing;

[0010] The method for detecting power load anomalies includes: calculating the distance variance between the load characteristics in each cluster and the cluster center, identifying abnormal load characteristics based on the distance variance and its corresponding threshold, and constructing a corresponding LSTM load prediction model for each abnormal load characteristic;

[0011] The method for processing empty clusters includes: using the strategy of splitting the largest cluster to create a new cluster center.

[0012] Furthermore, the clustering termination condition includes: the maximum change of the cluster center is less than its corresponding threshold, or the maximum number of iterations is reached.

[0013] Furthermore, the method for determining the initial cluster centers includes: selecting the initial cluster centers based on the K-means++ principle, and adding random perturbations to each initial cluster center.

[0014] Furthermore, the method of selecting the initial cluster center based on the K-means++ principle includes:

[0015] Divide the grid according to the numerical range of the load characteristics and calculate the load characteristic density distribution within each grid;

[0016] Randomly select a point from the high-density grid as the first initial cluster center;

[0017] The minimum sum of squared distances from the load characteristics in each grid to the first initial cluster center is calculated, and subsequent cluster centers are selected with probability proportional to the sum of squared distances.

[0018] Furthermore, the method for each power user to perform local cluster allocation to generate several clusters includes: calculating the weighted Euclidean distance between the load characteristics and the cluster center, and adding controllable noise to the weighted Euclidean distance; and allocating the load characteristics to the category corresponding to the cluster center with the smallest weighted Euclidean distance.

[0019] Furthermore, a secret shard is calculated based on the local sum of the load characteristics in each cluster and the number of load characteristics. Updating the cluster center based on the secret shard includes: updating the cluster center using the Shamir secret sharing scheme, including four stages of secret distribution, secret verification, secret aggregation and secret reconstruction, including the following steps:

[0020] The local sum of the shared load characteristics and the polynomial of the number of load characteristics are constructed respectively, and the secret shards are calculated. The secret shards are distributed to the power users. After receiving the secret shards, each power user performs secret verification and secret aggregation, reconstructs the secret by interpolation according to the aggregation structure, and calculates the new cluster center.

[0021] The power load forecasting system based on privacy-preserving distributed clustering according to the present invention includes:

[0022] The load feature extraction unit is used to determine the power users and power sales companies participating in the cluster calculation as the central coordinator. Each power user extracts load features based on its own historical power load data to form a local data set;

[0023] A cluster allocation unit is configured to determine initial cluster centers and distribute them to power users, and perform an iterative clustering process based on a local data set. In each iteration, each power user performs local cluster allocation to generate a number of clusters. A secret shard is calculated based on the local sum of the load characteristics in each cluster and the number of load characteristics. The cluster center is updated based on the secret shard, and the local cluster allocation is re-executed until a cluster termination condition is met to generate a final cluster.

[0024] A differential privacy processing unit, used to add calibration noise to the cluster centers in the final clustering;

[0025] The load forecasting unit is used to build a corresponding LSTM load forecasting model for each cluster after adding calibration noise to perform power load forecasting.

[0026] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the power load forecasting method based on privacy-preserving distributed clustering is implemented.

[0027] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the power load forecasting method based on privacy-preserving distributed clustering is implemented.

[0028] Beneficial effects: Compared with the prior art, the advantages of the present invention are:

[0029] (1) By combining the K-means++ initialization strategy with secret sharing technology, the clustering quality and algorithm convergence speed are improved while protecting the privacy of user power load data, providing a more accurate user grouping basis for subsequent LSTM load forecasting;

[0030] (2) Adoption The threshold secret sharing mechanism provides fault tolerance for the system. Even if some users are offline or submit incorrect data, the system can still operate normally. This is of great significance to large users in the power system.

[0031] (3) The introduction of a zero-knowledge proof verification mechanism can effectively prevent attacks by malicious users submitting erroneous data, ensuring the reliability and security of clustering results;

[0032] (4) Through the specially designed power load feature extraction and weighted distance calculation mechanism, the sensitivity and accuracy of clustering to power load characteristics are improved;

[0033] (5) The clustering results are seamlessly integrated with LSTM deep learning predictions, achieving the goal of accurate load forecasting for different user groups while protecting user privacy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of a method for predicting power load according to an embodiment of the present invention.

[0035] Figure 2 This is a diagram of the power load forecasting system architecture according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0037] like Figure 1 As shown, the power load forecasting method based on privacy-preserving distributed clustering includes the following steps.

[0038] Step 1: System initialization. Determine the clustering The electricity users and the electricity sales company act as the central coordinator and set the number of clusters. , secret sharing threshold , maximum number of iterations and convergence threshold .

[0039] Step 2: Load data feature extraction and preparation. Each user Based on its own historical power load data, extract power consumption characteristics to form a local data set , where each eigenvector Indicates the User's The electricity consumption feature vector is dimensional real number vector, including characteristics such as daily load curve shape, peak-to-valley ratio, load rate, peak period electricity consumption ratio, etc. Indicates the users have a total of Features.

[0040] Load characteristics include time domain characteristics, frequency domain characteristics and enterprise attribute characteristics; time domain characteristics include daily load curve shape characteristics, peak-to-valley difference ratio, load rate, peak period electricity consumption proportion and load growth rate; frequency domain characteristics include power spectrum density characteristics, main frequency components, statistical characteristics, electricity consumption statistical distribution characteristics, periodic indicators and volatility indicators; enterprise attribute characteristics include industry category (after coding processing), scale indicators and energy consumption pattern coding.

[0041] Step 3: Safely initialize the cluster center.

[0042] Step 3.1: User securely submits load statistics.

[0043] Step 3.2: The electricity sales company selects the initial cluster center based on the K-means++ principle ,in Indicates the Cluster centers, each cluster center corresponds to a category. The specific steps are as follows:

[0044] (a) Users securely aggregate statistical information about the power load feature space based on secret sharing, including the numerical range of load features in each dimension and the load feature density distribution after dividing the feature space into grids;

[0045] (b) The electricity sales company randomly selects a point as the first cluster center ;

[0046] (c) For arrive :Users securely calculate the minimum sum of squared distances from each grid's load feature point to the existing cluster center based on secret sharing, and the power sales company selects the next cluster center with a probability proportional to the sum of squared distances. .

[0047] Step 3.3: Add random perturbations to each initial cluster center to obtain the perturbed cluster centers .in , The mean is 0 and the variance is Gaussian distribution.

[0048] Step 3.4: The electricity sales company will add the initial cluster center after disturbance Safely distributed to all electricity users.

[0049] Step 4: Iterate the clustering process. In each iteration , perform the following operations.

[0050] Step 4.1: Each user performs local cluster assignment.

[0051] (a) Calculate each local load characteristic vector The weighted Euclidean distance between each cluster center:

[0052] ;

[0053] in, Represents a user No. Features and The weighted Euclidean distance between cluster centers, Represents a user No. The first eigenvector dimensional component, Indicates the The first cluster center dimensional component, represents the total number of dimensions of the feature vector, It is The weight of the dimensional feature is dynamically adjusted based on the importance of the power load feature, that is, the contribution of the feature to the load forecast accuracy is automatically calculated and meets the requirements. .

[0054] (b) To prevent accurate distance calculation results The user's sensitive information is deduced, and controllable noise is added to the distance to obtain the noise-added :

[0055] ;

[0056] in The mean is 0 and the variance is Gaussian distribution.

[0057] (c) Assign each feature vector to the category corresponding to the cluster center with the smallest distance.

[0058] (d) Calculate the Category collection The sum of the load eigenvectors and quantity :

[0059] ;

[0060] ;

[0061] in, Represents the feature vector Assigned to category , Represents a user The set of load characteristic vectors, Represents a user Belongs to category The number of load characteristic vectors, express Belongs to category The sum of all load eigenvectors of .

[0062] Step 4.2: Cluster center update based on threshold secret sharing. For each category ,implement:

[0063] Step 4.2.1: Secret distribution.

[0064] For each user :

[0065] (a) Constructed for shared load characteristics and The t-1 degree polynomial:

[0066] ;

[0067] in, is a user From a finite field The randomly selected coefficients in .

[0068] (b) Constructing the number of shared data points of Polynomial of degree:

[0069] ;

[0070] in is a user From a finite field The randomly selected coefficients in .

[0071] (c) Compute the secret shard and distribute it to all users:

[0072] For each user :calculate Shard ,calculate Shard , secure transmission of fragment pairs To the rest of the users. For users A unique identifier, usually a non-zero integer, Represents a user Load characteristic value At the point The fragment value at Represents a user Number of data points At the point The fragment value at .

[0073] Step 4.2.2: Secret Verification.

[0074] For each user :

[0075] (a) Generate a set of categories Zero-knowledge proof ,prove:

[0076] (i) ,and is the effective load characteristic and;

[0077] (ii) ,and is a non-negative integer;

[0078] (iii) and is the correctly calculated shard;

[0079] (b) will prove Sent together with the fragments;

[0080] (c) Each user verifies the received shard and certificate: if the verification passes, the shard is accepted; if the verification fails, the shard is marked as invalid and reported to the power sales company.

[0081] Step 4.2.3: Secret aggregation.

[0082] For each user :

[0083] (a) Aggregate the valid fragments received:

[0084] ;

[0085] .

[0086] in, Represents a user The collected categories The result of shard aggregation of load characteristic values, Represents a user The collected categories The result of shard aggregation is the number of data points.

[0087] (b) Sending the aggregation results to the electricity sales company .

[0088] Step 4.2.4: Secret reconstruction.

[0089] Power sales companies implement:

[0090] (a) Collect at least Aggregate results for valid users ;

[0091] Assume that the valid user set is ,in Indicates the number of valid users, satisfying ;

[0092] (b) Reconstruct the secret using Lagrange interpolation and :

[0093] ;

[0094] ;

[0095] in is the Lagrange coefficient, which is calculated as: .

[0096] in, 、 Different indexes for valid users, , is a user A unique identifier for is a user A unique identifier for the .

[0097] (c) Calculation New cluster centers : ;if ( is the preset minimum cluster size threshold), and the cluster is marked as abnormal.

[0098] Step 4.3: Detect and handle power load anomalies.

[0099] (a) Load abnormality detection:

[0100] (i) For Category collection , calculate the internal load characteristic vector Variance of distance to cluster center :

[0101] ;

[0102] (ii) Identification distance exceeds The load characteristics of the abnormal point ( is an adjustable parameter);

[0103] (iii) Special treatment of abnormal points, such as establishing an independent abnormal load model.

[0104] (b) Empty cluster processing:

[0105] If a category set of , then use the strategy of splitting the largest cluster: find the cluster with the most members ,choose The load characteristic point farthest from the center is used as the center to create a new cluster center to replace the empty cluster.

[0106] Step 4.4: Termination condition check.

[0107] (a) Calculate the maximum change in cluster center: ;

[0108] (b) If Or the maximum number of iterations has been reached , the algorithm terminates; otherwise, update the cluster center (For all ), return to step 4.1 and continue iterating.

[0109] Step 5: Differential privacy enhancement processing.

[0110] Step 5.1, add calibration noise to the final cluster center:

[0111] ;

[0112] in: is the cluster center after adding noise, is the sensitivity, which indicates the maximum impact that a load characteristic point may have on the cluster center. It's a privacy budget. is the noise parameter;

[0113] Step 5.2, return the cluster center after differential privacy protection .

[0114] Step 6: Build an LSTM load forecasting model based on clustering results.

[0115] Step 6.1, for each cluster , the electricity sales company collects the anonymized historical load data of users in this category;

[0116] Step 6.2: Build a dedicated LSTM load forecasting model for each cluster The LSTM load forecasting model includes a data preprocessing layer for data normalization, time series feature extraction, and external factor encoding (such as temperature, date type, etc.).

[0117] The LSTM network structure includes an input layer, a multi-layer LSTM layer, an attention mechanism layer, and an output layer; the input layer is used to receive historical load sequences and external features; the multi-layer LSTM layer is used to capture the long-term and short-term dependencies of the load; the attention mechanism layer is used to focus on important historical time periods; and the output layer is used to predict the load in future time periods.

[0118] Model optimization and adjustment methods include: automatically adjusting network hyperparameters based on the load characteristics of users within the cluster; using early stopping to prevent overfitting; and regular updates to adapt to changes in power consumption patterns.

[0119] Step 6.3: When a new user joins, assign him / her to the nearest cluster based on the anonymized features he / she submits. , and use the corresponding prediction model Carry out load forecasting.

[0120] The power load forecasting method based on privacy-preserving distributed clustering in this embodiment is applied to a scenario where a power sales company clusters and forecasts power load characteristics for large users. The steps are as follows.

[0121] S101, system initialization. In this embodiment, (Large industrial users), (Classification by electrical characteristics), , , For example.

[0122] S102: Load data feature extraction and preparation. Based on their historical load data, each electricity user extracts a 30-dimensional feature vector containing the daily load curve shape characteristics, peak-to-valley ratio, load factor, peak-hour electricity consumption ratio, and load volatility indicators at 24 time points. The user retains the original load data and only uses the extracted features for subsequent calculations.

[0123] S103, securely initialize the cluster center. This step includes:

[0124] S103.1, users securely submit load statistics information through secret sharing, including the numerical range of load characteristics and grid density information. Specifically, the 30-dimensional feature space is divided into grid units, and each user calculates the distribution of local load characteristics in each grid. A threshold secret sharing scheme for securely aggregating global mesh density.

[0125] S103.2, the electricity sales company selects the initial cluster center based on the K-means++ principle. First, according to the density distribution, the first cluster center is selected from the high-density area. ; Then, subsequent cluster centers are selected iteratively so that the new cluster centers are as far away from the existing centers as possible to better cover the characteristics of different types of power loads.

[0126] S103.3, add random perturbations to each initial cluster center: , obey distribution, where The value of is small enough and does not affect the clustering performance.

[0127] S103.4, the electricity sales company will disturb the initial cluster center Distributed to all electricity users.

[0128] S104, iterative clustering process. In each iteration, perform:

[0129] S104.1, each user performs local cluster assignment. Calculate weighted Euclidean distance, where the weight is set based on the contribution of the load feature to the prediction accuracy. Add controllable noise, assign data points, and calculate the local sum and number of each type of load feature. and .

[0130] S104.2, cluster center update based on threshold secret sharing. For each category ,implement:

[0131] S104.2.1, Secret Distribution. Each user constructs a polynomial and , calculate secret shards and distribute them securely.

[0132] S104.2.2, Secret Verification. Use Bulletproofs zero-knowledge proof system to generate proof And verify the validity of the received fragments.

[0133] S104.2.3, Secret Aggregation. Aggregate valid shards and .

[0134] S104.2.4, secret reconstruction. The electricity sales company uses at least The results of valid users are reconstructed by Lagrange interpolation and , calculate the new cluster center .

[0135] S104.3, Power Load Anomaly Detection and Processing. Identify abnormal load points, such as atypical power consumption caused by temporary shutdowns and equipment failures, and use a splitting strategy to handle empty clusters.

[0136] S104.4, Termination Condition Check. Calculate the maximum change ,if Or the maximum number of iterations is reached, the process terminates; otherwise, the cluster centers are updated and the iteration continues.

[0137] S105: Differential privacy enhancement processing. Calibration noise is added to the final cluster center and the processed result is returned.

[0138] S106: Build an LSTM load forecasting model based on the clustering results. A dedicated LSTM forecasting model is built for each cluster, including a data preprocessing layer, a multi-layer LSTM structure, an attention mechanism layer, and an output layer. Each model is optimized for the load characteristics of that user type, such as the cyclical patterns of industrial enterprises and the temperature sensitivity of commercial users.

[0139] like Figure 2 As shown, the privacy-preserving distributed clustering system for power load forecasting described in the present invention includes the following modules.

[0140] System initialization module 101, used to determine the set of power users participating in the calculation, the power sales company and related parameters;

[0141] The load feature extraction module 102 is used to extract feature vectors from the original power load data;

[0142] A secure cluster center initialization module 103 is used to securely select an initial cluster center based on the K-means++ principle and secret sharing technology;

[0143] a local cluster assignment module 104 for calculating weighted distances and assigning load characteristics to the nearest cluster center;

[0144] The secret distribution module 105 is used to construct a polynomial and generate secret fragments;

[0145] Secret verification module 106, for generating and verifying zero-knowledge proofs;

[0146] Secret aggregation module 107, used to aggregate received valid fragments;

[0147] A secret reconstruction module 108, configured to reconstruct the secret using Lagrange interpolation and update the cluster center;

[0148] The load anomaly detection module 109 is used to identify and process abnormal load characteristics;

[0149] A differential privacy processing module 110 is used to add calibration noise to the final result to enhance privacy protection;

[0150] The LSTM load prediction module 111 is used to build dedicated prediction models for different user groups based on clustering results.

[0151] In actual applications, modules 102 and 104 are deployed on the user side, modules 101, 103, 108, 109, 110 and 111 are deployed on the power sales company side, and modules 105, 106 and 107 are distributed among all participants, exchanging data through secure communication channels.

[0152] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the power load forecasting method based on privacy-preserving distributed clustering is implemented.

[0153] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the power load forecasting method based on privacy-preserving distributed clustering is implemented.

[0154] The computer-readable storage media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory, or any other medium that can be used to store program code in the form of instructions or data structures and that can be accessed by a computer.

[0155] The processor is configured to execute the computer program stored in the memory to implement the various steps in the method involved in the above embodiment.

Claims

1. A power load forecasting method based on privacy-preserving distributed clustering, characterized in that: The steps include: The power users and power sales companies participating in the cluster calculation are identified as central coordinators. Each power user extracts load characteristics based on its own historical power load data to form a local data set. Initial cluster centers are determined and distributed to power users. An iterative clustering process is performed based on local data sets. In each iteration, each power user performs local cluster allocation to generate several clusters. A secret shard is calculated based on the local sum of load characteristics in each cluster and the number of load characteristics. The cluster center is updated based on the secret shard, and local cluster allocation is re-executed until the cluster termination condition is reached to generate the final cluster. Add calibration noise to the cluster centers in the final clustering; A corresponding LSTM load forecasting model is constructed for each cluster after adding calibration noise to perform power load forecasting.

2. The power load forecasting method based on privacy-preserving distributed clustering according to claim 1 is characterized in that: After updating the cluster center according to the secret shards, it also includes power load anomaly detection and empty cluster processing; The method for detecting power load anomalies includes: calculating the distance variance between the load characteristics in each cluster and the cluster center, identifying abnormal load characteristics based on the distance variance and its corresponding threshold, and constructing a corresponding LSTM load prediction model for each abnormal load characteristic; The method for processing empty clusters includes: using the strategy of splitting the largest cluster to create a new cluster center.

3. The power load forecasting method based on privacy-preserving distributed clustering according to claim 1, characterized in that: The clustering termination conditions include: the maximum change of the cluster center is less than its corresponding threshold, or the maximum number of iterations is reached.

4. The power load forecasting method based on privacy-preserving distributed clustering according to claim 1, characterized in that: The method for determining the initial cluster center includes: The initial cluster centers are selected based on the K-means++ principle, and random perturbations are added to each initial cluster center.

5. The power load forecasting method based on privacy-preserving distributed clustering according to claim 4 is characterized in that: Methods for selecting initial cluster centers based on the K-means++ principle include: Divide the grid according to the numerical range of the load characteristics and calculate the load characteristic density distribution within each grid; Randomly select a point from the high-density grid as the first initial cluster center; The minimum sum of squared distances from the load characteristics in each grid to the first initial cluster center is calculated, and subsequent cluster centers are selected with probability proportional to the sum of squared distances.

6. The power load forecasting method based on privacy-preserving distributed clustering according to claim 1, characterized in that: The method for each power user to perform local cluster allocation to generate a plurality of clusters includes: The weighted Euclidean distance between the load feature and the cluster center is calculated, and controllable noise is added to the weighted Euclidean distance; the load feature is assigned to the category corresponding to the cluster center with the smallest weighted Euclidean distance.

7. The power load forecasting method based on privacy-preserving distributed clustering according to claim 1, characterized in that: The secret shards are calculated based on the local sum of the load characteristics in each cluster and the number of load characteristics. The cluster center is updated based on the secret shards, including: using the Shamir secret sharing scheme to update the cluster center, including four stages of secret distribution, secret verification, secret aggregation and secret reconstruction, including the following steps: The local sum of the shared load characteristics and the polynomial of the number of load characteristics are constructed respectively, and the secret shards are calculated. The secret shards are distributed to the power users. After receiving the secret shards, each power user performs secret verification and secret aggregation, reconstructs the secret by interpolation according to the aggregation structure, and calculates the new cluster center.

8. A power load forecasting system based on privacy-preserving distributed clustering, characterized in that: include: The load feature extraction unit is used to determine the power users and power sales companies participating in the cluster calculation as the central coordinator. Each power user extracts load features based on its own historical power load data to form a local data set; A cluster allocation unit is configured to determine initial cluster centers and distribute them to power users, and perform an iterative clustering process based on a local data set. In each iteration, each power user performs local cluster allocation to generate a number of clusters. A secret shard is calculated based on the local sum of the load characteristics in each cluster and the number of load characteristics. The cluster center is updated based on the secret shard, and the local cluster allocation is re-executed until a cluster termination condition is met to generate a final cluster. A differential privacy processing unit, used to add calibration noise to the cluster centers in the final clustering; The load forecasting unit is used to build a corresponding LSTM load forecasting model for each cluster after adding calibration noise to perform power load forecasting.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is loaded into a processor, the method for power load forecasting based on privacy-preserving distributed clustering according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the power load forecasting method based on privacy-preserving distributed clustering according to any one of claims 1 to 7 is implemented.

Images