User load clustering method and system based on multiple adjustment characteristic indexes, terminal equipment and storage medium

Abstract

The invention discloses a user load clustering method and system based on multiple adjustment characteristic indexes, terminal equipment and a storage medium, and belongs to the field of user load clustering. The method constructs multiple types of adjustment characteristic indexes and a comprehensive clustering score objective function, and uses a Bayesian optimization algorithm to adaptively adjust AP clustering parameters, so that the user load clustering efficiency is improved; according to the method, intelligent clustering analysis of user load characteristics is realized, and the clustering number and the initial center do not need to be manually preset, so that the autonomy and adaptability of the clustering process are ensured, the subjective influence of manual parameter setting is avoided, the accuracy and reliability of complex load data clustering are improved, and the user experience is improved. The problems that in the prior art, due to the fact that a static clustering algorithm is adopted, clustering parameters depend on manual setting and cannot be automatically adjusted according to data distribution, the clustering adaptability is insufficient and the precision is low when multi-type and high-volatility load data are processed are solved.

Description

Technical Field

[0001] The present invention relates to the field of user load clustering, and in particular to a user load clustering method, system, terminal equipment and storage medium based on multiple adjustment characteristic indicators. Background Art

[0002] With the large-scale access of new energy, electric vehicles, distributed energy storage and other flexible resources to the power system, user loads have shown complex characteristics of multi-type, high volatility and strong randomness. This makes user load clustering, as the basis for analyzing and predicting load data and making power system dispatch decisions, face the challenge of significantly increased data complexity.

[0003] Existing technologies mostly use distance-based static clustering algorithms (such as K-means), which achieve clustering by optimizing the objective function of minimizing the intra-cluster distance and maximizing the inter-cluster distance. However, this type of method relies on manually setting clustering parameters (such as the initial center and the number of clusters) and cannot automatically adjust according to the data distribution. In particular, when processing load data sets with irregular cluster shapes and uneven distribution, improper parameter settings can easily lead to low clustering accuracy, making it difficult to effectively reveal the inherent structure of the load data.

[0004] Therefore, the existing technology uses static clustering algorithms such as K-means based on distance measurement, which relies on manual setting of clustering parameters and cannot be automatically adjusted according to data distribution, resulting in insufficient clustering adaptability and low accuracy when processing multi-type and highly volatile load data. Summary of the Invention

[0005] The present invention provides a user load clustering method, system, terminal device and storage medium based on multiple adjustment characteristic indicators, which can solve the problem that the existing technology relies on manually set clustering parameters due to the use of static clustering algorithms and cannot be automatically adjusted according to data distribution, resulting in insufficient clustering adaptability and low accuracy when processing multi-type and highly volatile load data.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a user load clustering method based on multiple adjustment characteristic indicators, including:

[0007] Obtain load data for several users;

[0008] Calculate the regulation characteristic index of each user based on the load data of each user;

[0009] According to the search domain of the clustering parameters of the AP clustering algorithm, several different clustering parameter combinations are randomly generated. Then, users are clustered according to the different clustering parameter combinations and the adjustment characteristic indicators of each user, and the user load clustering results under several different clustering parameter combinations are obtained.

[0010] According to the user load clustering results under several different clustering parameter combinations and the constructed clustering score objective function, the clustering score value of the user load clustering result under each clustering parameter combination is calculated, and the clustering score values ​​of several different clustering parameter combinations are obtained;

[0011] Inputting several different clustering parameter combinations and corresponding clustering score values ​​into the FNN model, training the FNN model, and obtaining a trained FNN model;

[0012] According to the trained FNN model prediction and Bayesian optimization algorithm, the clustering parameter combination is iteratively optimized in the clustering parameter search domain to obtain the clustering parameter combination with the optimal clustering score value;

[0013] According to the clustering parameter combination with the optimal clustering score value and the adjustment characteristic index of each user, user load clustering is performed to obtain the final user load clustering result.

[0014] Furthermore, the regulation characteristic indicators include: primary and secondary reduction load change rate, peak flexibility load rate, day-ahead peak load rate, intraday peak-to-valley difference rate and load fluctuation rate;

[0015] The calculation expression of the primary and secondary load reduction change rate is:

[0016] γ1(i)=P peak (i)-min(P peak-1 (i), P peak+1 (i));

[0017] Where γ1(i) is the primary and secondary load reduction rate of the i-th user; P peak (i) is the peak value of the load curve of the i-th user in a typical day; P peak-1 (i),P peak+1 (i) is the load value in the time period adjacent to the peak

[0018] The calculation expression of the peak flexibility load rate is:

[0019]

[0020] Where, γ2(i) is the peak flexibility load rate of the i-th user; is the average load of the i-th user in a typical day;

[0021] The calculation expression of the day-ahead peak load rate is:

[0022]

[0023] Where, γ3(i) is the peak load rate of the i-th user during the day; is the electricity consumption of the i-th user throughout the day;

[0024] The calculation expression of the intraday peak-to-valley difference rate is:

[0025]

[0026] Where, γ4(i) is the peak-to-valley difference rate of the i-th user within a day; P valley (i) is the minimum value of the load curve of the i-th user in a typical day;

[0027] The calculation expression of the load fluctuation rate is:

[0028]

[0029] Where, γ5(i) is the load fluctuation rate of the i-th user; σ III (i) and μ III (i) are the load variance and mean of the i-th user on a typical day.

[0030] Furthermore, according to the search domain of the clustering parameters of the AP clustering algorithm, several different clustering parameter combinations are randomly generated, and users are clustered according to the several different clustering parameter combinations and the adjustment characteristic indicators of each user, to obtain user load clustering results under the several different clustering parameter combinations, including:

[0031] According to the search domain of the clustering parameters of the AP clustering algorithm, uniform sampling is performed using the Latin hypercube sampling method to generate several different clustering parameter combinations; wherein the clustering parameters are the preference value and the damping coefficient;

[0032] For each different clustering parameter combination, the AP clustering algorithm is called to cluster the users according to the adjustment characteristic indicators of each user, and the user load clustering results under each different clustering parameter combination are obtained.

[0033] Furthermore, the calling of the AP clustering algorithm to cluster users includes:

[0034] Calculate the Euclidean distance between users based on the adjustment characteristic index of each user, and determine the similarity matrix based on the Euclidean distance;

[0035] Initialize the attraction matrix and the attribution matrix, and iteratively update the attraction matrix and the attribution matrix until the maximum number of iterations is reached to obtain the final attraction matrix and the attribution matrix;

[0036] The final attraction matrix and belonging matrix are added together to obtain the judgment matrix. The cluster center is determined according to the judgment matrix, and each user sample is assigned to the corresponding cluster to obtain the user load clustering result.

[0037] Furthermore, the expression of the constructed clustering score objective function is:

[0038]

[0039] Where F1 is the clustering score; a(i) is the average distance between user load data sample i and other samples in the same cluster; b(i) is the minimum average distance between sample i and samples in all other clusters; s(i) represents the silhouette coefficient; B k is the weighted square sum of the distances between the cluster centroid and the global centroid; W k is the sum of squares of the distances from the sample to the centroid of the cluster to which it belongs; n is the number of clusters; v pen is the penalty term for the number of clusters; imb is the penalty term for imbalance in the number of users within a cluster; CH is the clustering quality evaluation index, which is used to evaluate the clustering quality by the ratio of the inter-cluster dispersion to the intra-cluster dispersion; k is the number of all clusters.

[0040] Furthermore, the clustering parameter combination is iteratively optimized in the clustering parameter search domain based on the trained FNN model prediction and Bayesian optimization algorithm to obtain the clustering parameter combination with the optimal clustering score value, including:

[0041] For each candidate clustering parameter combination in the search domain, the clustering score of each candidate clustering parameter combination is predicted by the trained FNN model, and the expected value of each candidate parameter combination is calculated by the expected improvement function according to the clustering score, and the clustering parameter combination with the highest expected value is selected as the new clustering parameter combination;

[0042] According to the new clustering parameter combination and the adjustment characteristic index of each user, the AP clustering algorithm is used to cluster the users and obtain the real clustering score value of the new clustering parameter combination;

[0043] The new clustering parameter combination and the corresponding true clustering score value are used as the training set of the FNN model, and the FNN model is retrained;

[0044] Repeat the iteration until the preset number of iterations is reached, and select the clustering parameter combination with the best clustering score value from all iteration results.

[0045] Based on the above method embodiment, the present invention provides a corresponding system embodiment;

[0046] An embodiment of the present invention provides a user load clustering system based on multiple regulation characteristic indicators, comprising: a load data acquisition module, a regulation characteristic indicator calculation module, a clustering parameter combination sampling module, a clustering score objective function calculation module, an FNN model training module, an optimal clustering parameter combination determination module, and a user load clustering module;

[0047] The load data acquisition module is used to acquire load data of several users;

[0048] The regulation characteristic index calculation module is used to calculate the regulation characteristic index of each user based on the load data of each user;

[0049] The clustering parameter combination sampling module is used to randomly generate a number of different clustering parameter combinations based on the search domain of the clustering parameters of the AP clustering algorithm, and cluster users based on the different clustering parameter combinations and the adjustment characteristic indicators of each user to obtain user load clustering results under the different clustering parameter combinations;

[0050] The clustering score objective function calculation module is used to calculate the clustering score value of the user load clustering result under each clustering parameter combination based on the user load clustering results under several different clustering parameter combinations and the constructed clustering score objective function, and obtain the clustering score values ​​of several different clustering parameter combinations;

[0051] The FNN model training module is used to input a plurality of different clustering parameter combinations and corresponding clustering score values ​​into the FNN model, train the FNN model, and obtain a trained FNN model;

[0052] The optimal clustering parameter combination determination module is used to iteratively optimize the clustering parameter combination within the clustering parameter search domain based on the trained FNN model prediction and Bayesian optimization algorithm to obtain the clustering parameter combination with the optimal clustering score value;

[0053] The user load clustering module is used to perform user load clustering based on the clustering parameter combination with the optimal clustering score value and the adjustment characteristic index of each user to obtain a final user load clustering result.

[0054] Furthermore, the expression of the constructed clustering score objective function is:

[0055]

[0056] Where F1 is the clustering score; a(i) is the average distance between user load data sample i and other samples in the same cluster; b(i) is the minimum average distance between sample i and samples in all other clusters; s(i) represents the silhouette coefficient; B k is the weighted square sum of the distances between the cluster centroid and the global centroid; W k is the sum of squares of the distances from the sample to the centroid of the cluster to which it belongs; n is the number of clusters; v pen is the penalty term for the number of clusters; imb is the penalty term for imbalance in the number of users within a cluster; CH is the clustering quality evaluation index, which is used to evaluate the clustering quality by the ratio of the inter-cluster dispersion to the intra-cluster dispersion; k is the number of all clusters.

[0057] Based on the above-mentioned method embodiment, another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a user load clustering method based on multiple adjustment characteristic indicators as described in the above-mentioned embodiment.

[0058] Based on the above method embodiment, another embodiment of the present invention provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the user load clustering method based on multiple adjustment characteristic indicators described in the above embodiment.

[0059] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0060] The present invention realizes intelligent clustering analysis of user load characteristics by constructing multi-type regulation characteristic indicators and comprehensive clustering score objective functions, and using the Bayesian optimization algorithm to adaptively adjust AP clustering parameters. There is no need to manually preset the number of clusters and the initial center, thereby ensuring the autonomy and adaptability of the clustering process, and avoiding the subjective influence of manual parameter setting, improving the accuracy and reliability of complex load data clustering, and solving the problem of the existing technology relying on manual setting of clustering parameters and being unable to automatically adjust according to data distribution due to the use of static clustering algorithms, resulting in insufficient clustering adaptability and low precision when processing multi-type and highly volatile load data. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flowchart of a method for clustering user loads based on multiple adjustment characteristic indicators provided by an embodiment of the present invention;

[0062] Figure 2 Typical user adjustment characteristics of cluster category 1 after clustering provided by the embodiment of the present invention;

[0063] Figure 3 Typical user adjustment characteristics of cluster category 2 after clustering provided by the embodiment of the present invention;

[0064] Figure 4 A typical daily load curve of users in clustering category 1 after clustering provided by an embodiment of the present invention;

[0065] Figure 5 A typical daily load curve of users of clustering category 2 after clustering provided by an embodiment of the present invention;

[0066] Figure 6This is a module diagram of a user load clustering system based on multiple adjustment characteristic indicators provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0068] Example 1:

[0069] Reference Figure 1 : A flowchart of a method for clustering user loads based on multiple adjustment characteristic indicators provided by an embodiment of the present invention; to address the problem in the prior art of relying on manually set clustering parameters due to the use of static clustering algorithms and being unable to automatically adjust according to data distribution, resulting in insufficient clustering adaptability and low accuracy when processing multi-type and highly volatile load data, the method includes at least the following steps:

[0070] Step S1: Obtain load data of several users;

[0071] In this embodiment, the power load data of each user in the target area can be obtained through the smart meter collection system; the user load data includes the user's load value and power consumption in a typical day.

[0072] In this embodiment, after acquiring load data of a plurality of users, the method further includes: performing data preprocessing on the acquired user load data; wherein the data preprocessing includes missing value filling and outlier processing.

[0073] In this embodiment, the missing value filling method can be used, such as mean filling and linear interpolation. The load data is sorted in time series and the intervals are not large, so the linear interpolation method is used. The formula of the linear interpolation method is as follows:

[0074]

[0075] Where, P t+2 (i) and P t (i) are the load values ​​at time t+2 and time t respectively; P t+1 (i) is the missing load value at time t+1.

[0076] In this embodiment, the abnormal value is when the difference between the current load value and the adjacent load value in the load data set exceeds a preset value, then the data is considered abnormal and the horizontal processing method can be used to handle the abnormality. The processing method is as follows:

[0077]

[0078] Where, is the set threshold, P t (i) is the user load value at time t on a typical day.

[0079] Step S2: Calculate the regulation characteristic index of each user based on the load data of each user;

[0080] In this embodiment, the regulation characteristic indicators include: primary and secondary reduction load change rate, peak flexibility load rate, day-ahead peak load rate, intraday peak-to-valley difference rate, and load fluctuation rate;

[0081] The calculation expression of the primary and secondary load reduction change rate is:

[0082] γ1(i)=P peak (i)-min(P peak-1 (i), P peak+1 (i));

[0083] Where γ1(i) is the primary and secondary load reduction change rate of the i-th user, which refers to the load difference between the peak load in the daily load curve of user i and the load between two adjacent time periods. The larger its value, the greater the load reduction potential of user i; P peak (i) is the peak value of the load curve of the i-th user in a typical day; P peak-1 (i),P peak+1 (i) is the load value in the time period adjacent to the peak

[0084] The calculation expression of the peak flexibility load rate is:

[0085]

[0086] Where γ2(i) is the peak flexibility load rate of the i-th user, which refers to the standard deviation of user i on a typical day. The larger the value, the stronger the user's flexibility and the greater the potential for the user to participate in real-time peak shaving. is the average load of the i-th user in a typical day;

[0087] The calculation expression of the day-ahead peak load rate is:

[0088]

[0089] Where γ3(i) is the day-ahead peak load rate of the i-th user, which refers to the peak value of the load curve of user i on a typical day divided by the total daily electricity consumption. The larger the day-ahead peak load rate, the greater the day-ahead response potential of user i. is the electricity consumption of the i-th user throughout the day;

[0090] The calculation expression of the intraday peak-to-valley difference rate is:

[0091]

[0092] Where γ4(i) is the peak-to-valley difference rate of the i-th user, which refers to the ratio of the peak-to-valley difference of the typical daily load of user i to the peak value. The larger the value, the greater the potential for translation. valley (i) is the minimum value of the load curve of the i-th user in a typical day;

[0093] The calculation expression of the load fluctuation rate is:

[0094]

[0095] Where γ5(i) is the load fluctuation rate of the i-th user, which refers to the variance of the load curve of user i in a typical day divided by the mean. The larger the load fluctuation rate, the greater the potential for user i to shift; σ III (i) and μ III (i) are the load variance and mean of the i-th user on a typical day.

[0096] In this embodiment, in order to ensure that the gradients of the five indicators are offset due to different magnitudes, the load data needs to be normalized to obtain the same weight tendency and map different variables to the value range [0, 1].

[0097] The normalization formula is as follows.

[0098]

[0099] Where, γ n (i) is the data after index normalization, γ n (i) is the initial data of the indicator, is the average value of the initial data of the indicator, γ n (i) max , γ n (i) min are the maximum and minimum values ​​of the initial data of the indicator, n∈{1, 2, 3, 4, 5}.

[0100] Step S3: Based on the search domain of the clustering parameters of the AP clustering algorithm, several different clustering parameter combinations are randomly generated. Then, based on the different clustering parameter combinations and the adjustment characteristic indicators of each user, the users are clustered to obtain the user load clustering results under the different clustering parameter combinations.

[0101] In this embodiment, the search domain of the clustering parameters of the AP clustering algorithm is used to randomly generate several different clustering parameter combinations. Users are clustered based on the several different clustering parameter combinations and the adjustment characteristic indicators of each user. The user load clustering results obtained under the several different clustering parameter combinations include:

[0102] According to the search domain of the clustering parameters of the AP clustering algorithm, uniform sampling is performed using the Latin hypercube sampling method to generate several different clustering parameter combinations; wherein the clustering parameters are the preference value and the damping coefficient;

[0103] In this embodiment, the search domain of the clustering parameter is [p min ,p max ]×[λ min ,λ max ], P is the preference value, λ is the damping coefficient;

[0104] For each different clustering parameter combination, the AP clustering algorithm is called to cluster the users according to the adjustment characteristic indicators of each user, and the user load clustering results under each different clustering parameter combination are obtained.

[0105] In this embodiment, calling the AP clustering algorithm to cluster users includes:

[0106] Calculate the Euclidean distance between users based on the adjustment characteristic index of each user, and determine the similarity matrix based on the Euclidean distance;

[0107] Initialize the attraction matrix and the attribution matrix, and iteratively update the attraction matrix and the attribution matrix until the maximum number of iterations is reached to obtain the final attraction matrix and the attribution matrix;

[0108] The final attraction matrix and belonging matrix are added together to obtain the judgment matrix. The cluster center is determined according to the judgment matrix, and each user sample is assigned to the corresponding cluster to obtain the user load clustering result.

[0109] In this embodiment, the calculation expression for calling the AP clustering algorithm to cluster users is:

[0110]

[0111] r (t) (i,j)=s(i,j)-max j′≠j{a (t-1) (i,j′)+s(i,j′)};

[0112]

[0113] Where d(i, j) is the Euclidean distance between user i and user j; s(i, j) is the similarity matrix between user i and user j; r (t) (i, j) is the attraction matrix between user i and user j in the tth iteration; a (t) (i, j) is the attribution matrix between user i and user j in the tth iteration; a (t) (j, j) is the degree matrix of user j’s belonging to himself as the cluster center at the tth iteration, which is a (t) A variant of the degree matrix of (i, j); r (t) (j, j) is the attraction matrix of user j to itself as the cluster center at the tth iteration; r (t) (i′, j) is the attraction matrix between user i′ and user j at the tth iteration, i′≠i, i′≠j; is the updated attraction matrix; is the updated attribution matrix; E(i, j) is the judgment matrix.

[0114] Step S4: Based on the user load clustering results under several different clustering parameter combinations and the constructed clustering score objective function, the clustering score value of the user load clustering result under each clustering parameter combination is calculated to obtain the clustering score values ​​of several different clustering parameter combinations;

[0115] In this embodiment, the expression of the constructed cluster score objective function is:

[0116]

[0117] Where F1 is the clustering score; a(i) is the average distance between user load data sample i and other samples in the same cluster; b(i) is the minimum average distance between sample i and samples in all other clusters; s(i) represents the silhouette coefficient; B k is the weighted square sum of the distances between the cluster centroid and the global centroid; W k is the sum of squares of the distances from the sample to the centroid of the cluster to which it belongs; n is the number of clusters; v pen is the penalty term for the number of clusters; imb is the penalty term for imbalance in the number of users within a cluster; CH is the clustering quality evaluation index, which is used to evaluate the clustering quality by the ratio of the inter-cluster dispersion to the intra-cluster dispersion; k is the number of all clusters.

[0118] Step S5: inputting several different clustering parameter combinations and corresponding clustering score values ​​into the FNN model, training the FNN model, and obtaining a trained FNN model;

[0119] In this embodiment, the several different clustering parameter combinations are used as input values ​​during FNN model training, and the clustering score values ​​corresponding to the different clustering parameter combinations are used as actual clustering score values ​​during FNN model training; the several different clustering parameter combinations are input into the FNN model to be trained, so that the FNN model to be trained predicts the clustering score values ​​according to the several different clustering parameter combinations, generates clustering score value prediction values ​​for each clustering parameter combination, calculates the prediction error based on the clustering score value prediction values ​​of each clustering parameter combination and the actual clustering score values ​​of each clustering parameter combination, and optimizes and adjusts the network parameters of the model based on the prediction error using the gradient descent method until the model training reaches a preset number of training times to obtain a trained FNN model.

[0120] Step S6: Based on the trained FNN model prediction and the Bayesian optimization algorithm, the clustering parameter combination is iteratively optimized within the clustering parameter search domain to obtain the clustering parameter combination with the optimal clustering score value;

[0121] In this embodiment, the clustering parameter combination is iteratively optimized within the clustering parameter search domain based on the trained FNN model prediction and the Bayesian optimization algorithm to obtain the clustering parameter combination with the optimal clustering score value, including:

[0122] For each candidate clustering parameter combination in the search domain, the clustering score of each candidate clustering parameter combination is predicted by the trained FNN model, and the expected value of each candidate parameter combination is calculated by the expected improvement function according to the clustering score, and the clustering parameter combination with the highest expected value is selected as the new clustering parameter combination;

[0123] According to the new clustering parameter combination and the adjustment characteristic index of each user, the AP clustering algorithm is used to cluster the users and obtain the real clustering score value of the new clustering parameter combination;

[0124] The new clustering parameter combination and the corresponding true clustering score value are used as the training set of the FNN model, and the FNN model is retrained;

[0125] Repeat the iteration until the preset number of iterations is reached, and select the clustering parameter combination with the best clustering score value from all iteration results.

[0126] In this embodiment, the calculation expression of the clustering parameter combination with the highest expected value is: next ,λ next )argmax (p，λ)EI(p, λ); where EI(p, λ) is the expected value of the candidate parameter combination.

[0127] Step S7: Perform user load clustering based on the clustering parameter combination with the optimal clustering score and the adjustment characteristic index of each user to obtain the final user load clustering result.

[0128] In this embodiment, the AP clustering algorithm is used to cluster user loads according to the clustering parameter combination with the optimal clustering score and the regulation characteristic index of each user, and a final user load clustering result is obtained.

[0129] Example 2:

[0130] In this example, data from 200 users in multiple regions of the United States on winter weekdays was used as the research object. Data sampling was set to 1 hour per point, the maximum number of BO algorithm iterations was 300, the initial number of samples was 3000, the proxy model training cycle was 500, and the damping coefficient of the AP clustering algorithm was [0.5-0.99]. The final score of the clustering objective function was 0.6807, the preference value was -40.41, the damping coefficient was 0.95, and the SIL coefficient, CH index, DBI index, and Dunn index were 0.7687, 279.2502, 0.4577, and 0.1166, respectively.

[0131] Reference Figure 2 and Figure 3 These are the typical user adjustment characteristics of cluster category 1 and cluster category 2 after clustering provided by the embodiment of the present invention. The colored lines in the figure represent each user. Based on the response characteristic indicators [γ1, γ2, γ3, γ4, γ5] of 200 residential users, they are automatically clustered into two categories using the BO-AP clustering algorithm. Different colors in each sub-graph represent different user response characteristics. From the clustering results, it can be seen that, except for the load reduction indicator, the numerical differences of users in the same cluster on the other four indicators are relatively small. Figure 2 The main differences between the two typical users in the sub-graph are reflected in the following two aspects: First, the overall response index value of users in category 1 is significantly higher than that of users in category 2, indicating that users in category 1 have higher load response potential; second, focusing on specific indicators, especially in the load reduction rate and fluctuation rate, the values ​​of users in category 1 are significantly higher than those in category 2. Since these two indicators reflect the overall fluctuation of the user's typical daily load curve, please refer to Figure 4 and Figure 5Figure 1 shows the average load curves of typical daily users in cluster categories 1 and 2, respectively. The colored lines in the figure represent individual users. Overall, the load curves of users in category 1 exhibit more pronounced peaks and valleys than those in category 2, making them more suitable for participating in the peak-shaving market. This is not only because the electricity consumption curves of category 1 users fluctuate more, but also because their overall power consumption is generally higher than that of category 2 users. Category 2 users have a more stable daily load curve, with power consumption generally remaining around 2kW across all periods. In contrast, category 1 users generally have load power above 2kW even during the off-peak period (9:00 AM to 4:00 PM) for residents. This characteristic also reflects that category 1 residential users are mostly working families, with many family members working outside during the day. Category 2 users may prioritize power stability, resulting in a more limited response to the market.

[0132] The examples also compare other clustering algorithms, including the adaptive BO-AP algorithm, the adaptive Harris Hawks Optimization (HHO) algorithm, the adaptive Grey Wolf Optimizer (GWO) algorithm, the K-means method with cluster types set to 2 and 4, and AP clustering. Four common clustering performance evaluation metrics, the SIL coefficient, the Davies-Bouldin index (DBI), the CH index, and the Dunn index, are used to compare the clustering results of the above methods from multiple perspectives. Compared with clustering using the AP algorithm alone, the adaptive BO-AP algorithm only outperforms the SIL coefficient. This is because the DBI index, the CH index, and the Dunn index focus more on inter-cluster and intra-cluster relationships, while the AP algorithm alone clusters 41 typical user categories. Furthermore, many of the 41 categories contain only one user, making the AP algorithm alone more effective when evaluated from the perspective of these three metrics. However, this goes against the original intention of clustering in this section. Therefore, it can be concluded that the AP algorithm alone cannot accurately identify and cluster users.

[0133] By comparing with the two K-means comparison groups, a simple calculation shows that when the final cluster number is set to 2 relative to the cluster number 4, the four indicators such as SIL coefficient and DBI index are increased by nearly 90.2%, 34.3%, 1.0%, and 404.9%, respectively. This reflects to a certain extent that the appropriate number of clusters has a great impact on the clustering effect of the K-means algorithm, while the AP algorithm does not need to consider this factor. At the same time, by comparing the results of the BO-AP algorithm with the K-means algorithm with the cluster number set to 2, it can be seen that the BO-AP algorithm has certain advantages in the four clustering effect indicators, which are increased by nearly 1.0%, 21.1%, 10.0%, and 48.6%, respectively. It can be seen that the latter has a certain improvement in the average distance from sample i to the centroid and the ratio of the minimum inter-cluster distance to the maximum intra-cluster diameter, that is, each type of user is closer.

[0134] Compared with the adaptive GWO-AP algorithm and HHO-AP algorithm, the latter two are prone to fall into local optimality. Although the BO algorithm has high requirements for the initial value, this example ensures that the initial sampling points are evenly distributed, which makes up for this shortcoming and indirectly improves the clustering accuracy.

[0135] In this implementation, the comparison results of clustering effect indicators of various clustering algorithms are shown in Table 1 below:

[0136] Table 1 Comparison of clustering effect indicators

[0137]

[0138]

[0139] Example 3:

[0140] Reference Figure 4 This is a block diagram of a user load clustering system based on multiple adjustment characteristic indicators, provided by an embodiment of the present invention. To address the problem of existing technologies that rely on manually set clustering parameters due to the use of static clustering algorithms and cannot automatically adjust according to data distribution, resulting in insufficient clustering adaptability and low accuracy when processing multi-type and highly volatile load data, the system includes at least the following modules: a load data acquisition module, an adjustment characteristic indicator calculation module, a clustering parameter combination sampling module, a clustering score objective function calculation module, an FNN model training module, an optimal clustering parameter combination determination module, and a user load clustering module.

[0141] The load data acquisition module is used to acquire load data of several users;

[0142] The regulation characteristic index calculation module is used to calculate the regulation characteristic index of each user based on the load data of each user;

[0143] The clustering parameter combination sampling module is used to randomly generate a number of different clustering parameter combinations based on the search domain of the clustering parameters of the AP clustering algorithm, and cluster users based on the different clustering parameter combinations and the adjustment characteristic indicators of each user to obtain user load clustering results under the different clustering parameter combinations;

[0144] The clustering score objective function calculation module is used to calculate the clustering score value of the user load clustering result under each clustering parameter combination based on the user load clustering results under several different clustering parameter combinations and the constructed clustering score objective function, and obtain the clustering score values ​​of several different clustering parameter combinations;

[0145] The FNN model training module is used to input a plurality of different clustering parameter combinations and corresponding clustering score values ​​into the FNN model, train the FNN model, and obtain a trained FNN model;

[0146] The optimal clustering parameter combination determination module is used to iteratively optimize the clustering parameter combination within the clustering parameter search domain based on the trained FNN model prediction and Bayesian optimization algorithm to obtain the clustering parameter combination with the optimal clustering score value;

[0147] The user load clustering module is used to perform user load clustering based on the clustering parameter combination with the optimal clustering score value and the adjustment characteristic index of each user to obtain a final user load clustering result.

[0148] In this embodiment, the expression of the constructed cluster score objective function is:

[0149]

[0150] Where F1 is the clustering score; a(i) is the average distance between user load data sample i and other samples in the same cluster; b(i) is the minimum average distance between sample i and samples in all other clusters; s(i) represents the silhouette coefficient; B k is the weighted square sum of the distances between the cluster centroid and the global centroid; W k is the sum of squares of the distances from the sample to the centroid of the cluster to which it belongs; n is the number of clusters; v pen is the penalty term for the number of clusters; imb is the penalty term for imbalance in the number of users within a cluster; CH is the clustering quality evaluation index, which is used to evaluate the clustering quality by the ratio of the inter-cluster dispersion to the intra-cluster dispersion; k is the number of all clusters.

[0151] Another embodiment of the present invention provides a terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the user load clustering method based on multiple adjustment characteristic indicators as described in the above embodiment. The terminal device can be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The terminal device can include, but is not limited to, a processor and a memory.

[0152] The processor may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0153] The memory can be used to store the computer program, and the processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created based on the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device or other volatile solid-state storage device.

[0154] Another embodiment of the present invention provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the user load clustering method based on multiple adjustment characteristic indicators described in the above embodiment.

[0155] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. The computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium.

[0156] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A user load clustering method based on multiple regulation characteristic indicators, characterized in that: include: Obtain load data for several users; Calculate the regulation characteristic index of each user based on the load data of each user; According to the search domain of the clustering parameters of the AP clustering algorithm, several different clustering parameter combinations are randomly generated. Then, users are clustered according to the different clustering parameter combinations and the adjustment characteristic indicators of each user, and the user load clustering results under several different clustering parameter combinations are obtained. According to the user load clustering results under several different clustering parameter combinations and the constructed clustering score objective function, the clustering score value of the user load clustering result under each clustering parameter combination is calculated, and the clustering score values ​​of several different clustering parameter combinations are obtained; Inputting several different clustering parameter combinations and corresponding clustering score values ​​into the FNN model, training the FNN model, and obtaining a trained FNN model; According to the trained FNN model prediction and Bayesian optimization algorithm, the clustering parameter combination is iteratively optimized in the clustering parameter search domain to obtain the clustering parameter combination with the optimal clustering score value; According to the clustering parameter combination with the optimal clustering score value and the adjustment characteristic index of each user, user load clustering is performed to obtain the final user load clustering result.

2. The user load clustering method based on multiple regulation characteristic indicators according to claim 1 is characterized in that: The regulation characteristic indicators include: primary and secondary load reduction change rate, peak flexibility load rate, day-ahead peak load rate, intraday peak-to-valley difference rate, and load fluctuation rate; The calculation expression of the primary and secondary load reduction change rate is: γ1(i)=P peak (i)-min(P peak-1 (i),P peak+1 (i)); Where γ1(i) is the primary and secondary load reduction rate of the i-th user; P peak (i) is the peak value of the load curve of the i-th user in a typical day; P peak-1 (i),P peak+1 (i) is the load value in the time period adjacent to the peak The calculation expression of the peak flexibility load rate is: Where, γ2(i) is the peak flexibility load rate of the i-th user; is the average load of the i-th user in a typical day; The calculation expression of the day-ahead peak load rate is: Where, γ3(i) is the peak load rate of the i-th user during the day; is the electricity consumption of the i-th user throughout the day; The calculation expression of the intraday peak-to-valley difference rate is: Where, γ4(i) is the peak-to-valley difference rate of the i-th user within a day; P valley (i) is the minimum value of the load curve of the i-th user in a typical day; The calculation expression of the load fluctuation rate is: Where, γ5(i) is the load fluctuation rate of the i-th user; σ III (i) and μ III (i) are the load variance and mean of the i-th user on a typical day.

3. The user load clustering method based on multiple regulation characteristic indicators according to claim 2 is characterized in that: According to the search domain of the clustering parameters of the AP clustering algorithm, several different clustering parameter combinations are randomly generated, and users are clustered according to the several different clustering parameter combinations and the adjustment characteristic indicators of each user, and the user load clustering results under the several different clustering parameter combinations are obtained, including: According to the search domain of the clustering parameters of the AP clustering algorithm, uniform sampling is performed using the Latin hypercube sampling method to generate several different clustering parameter combinations; wherein the clustering parameters are the preference value and the damping coefficient; For each different clustering parameter combination, the AP clustering algorithm is called to cluster the users according to the adjustment characteristic indicators of each user, and the user load clustering results under each different clustering parameter combination are obtained.

4. The user load clustering method based on multiple regulation characteristic indicators according to claim 3 is characterized in that: The calling of the AP clustering algorithm to cluster users includes: Calculate the Euclidean distance between users based on the adjustment characteristic index of each user, and determine the similarity matrix based on the Euclidean distance; Initialize the attraction matrix and the attribution matrix, and iteratively update the attraction matrix and the attribution matrix until the maximum number of iterations is reached to obtain the final attraction matrix and the attribution matrix; The final attraction matrix and belonging matrix are added together to obtain the judgment matrix. The cluster center is determined according to the judgment matrix, and each user sample is assigned to the corresponding cluster to obtain the user load clustering result.

5. The user load clustering method based on multiple regulation characteristic indicators according to claim 4 is characterized in that: The expression of the constructed clustering score objective function is: Where F1 is the clustering score; a(i) is the average distance between user load data sample i and other samples in the same cluster; b(i) is the minimum average distance between sample i and samples in all other clusters; s(i) represents the silhouette coefficient; B k is the weighted square sum of the distances between the cluster centroid and the global centroid; W k is the sum of squares of the distances from the sample to the centroid of the cluster to which it belongs; n is the number of clusters; υ pen is the penalty term for the number of clusters; imb is the penalty term for imbalance in the number of users within a cluster; CH is the clustering quality evaluation index, which is used to evaluate the clustering quality by the ratio of the inter-cluster dispersion to the intra-cluster dispersion; k is the number of all clusters.

6. The user load clustering method based on multiple regulation characteristic indicators according to claim 5 is characterized in that: The clustering parameter combination is iteratively optimized in the clustering parameter search domain based on the trained FNN model prediction and Bayesian optimization algorithm to obtain the clustering parameter combination with the optimal clustering score value, including: For each candidate clustering parameter combination in the search domain, the clustering score of each candidate clustering parameter combination is predicted by the trained FNN model, and the expected value of each candidate parameter combination is calculated by the expected improvement function according to the clustering score, and the clustering parameter combination with the highest expected value is selected as the new clustering parameter combination; According to the new clustering parameter combination and the adjustment characteristic index of each user, the AP clustering algorithm is used to cluster the users and obtain the real clustering score value of the new clustering parameter combination; The new clustering parameter combination and the corresponding true clustering score value are used as the training set of the FNN model, and the FNN model is retrained; Repeat the iteration until the preset number of iterations is reached, and select the clustering parameter combination with the best clustering score value from all iteration results.

7. A user load clustering system based on multiple regulation characteristic indicators, characterized in that: include: Load data acquisition module, regulation characteristic index calculation module, clustering parameter combination sampling module, clustering score objective function calculation module, FNN model training module, optimal clustering parameter combination determination module and user load clustering module; The load data acquisition module is used to acquire load data of several users; The regulation characteristic index calculation module is used to calculate the regulation characteristic index of each user based on the load data of each user; The clustering parameter combination sampling module is used to randomly generate a number of different clustering parameter combinations based on the search domain of the clustering parameters of the AP clustering algorithm, and cluster users based on the different clustering parameter combinations and the adjustment characteristic indicators of each user to obtain user load clustering results under the different clustering parameter combinations; The clustering score objective function calculation module is used to calculate the clustering score value of the user load clustering result under each clustering parameter combination based on the user load clustering results under several different clustering parameter combinations and the constructed clustering score objective function, and obtain the clustering score values ​​of several different clustering parameter combinations; The FNN model training module is used to input a plurality of different clustering parameter combinations and corresponding clustering score values ​​into the FNN model, train the FNN model, and obtain a trained FNN model; The optimal clustering parameter combination determination module is used to iteratively optimize the clustering parameter combination within the clustering parameter search domain based on the trained FNN model prediction and Bayesian optimization algorithm to obtain the clustering parameter combination with the optimal clustering score value; The user load clustering module is used to perform user load clustering based on the clustering parameter combination with the optimal clustering score value and the adjustment characteristic index of each user to obtain a final user load clustering result.

8. The user load clustering system based on multiple regulation characteristic indicators according to claim 7 is characterized in that: The expression of the constructed clustering score objective function is: Where F1 is the clustering score; a(i) is the average distance between user load data sample i and other samples in the same cluster; b(i) is the minimum average distance between sample i and samples in all other clusters; s(i) represents the silhouette coefficient; B k is the weighted square sum of the distances between the cluster centroid and the global centroid; W k is the sum of squares of the distances from the sample to the centroid of the cluster to which it belongs; n is the number of clusters; υ pen is the penalty term for the number of clusters; imb is the penalty term for imbalance in the number of users within a cluster; CH is the clustering quality evaluation index, which is used to evaluate the clustering quality by the ratio of the inter-cluster dispersion to the intra-cluster dispersion; k is the number of all clusters.

9. A terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements a user load clustering method based on multiple adjustment characteristic indicators as described in any one of claims 1 to 6.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the user load clustering method based on multiple adjustment characteristic indicators according to any one of claims 1 to 6.

Images