A power system load curve clustering analysis method

Abstract

The application discloses a power system load curve clustering analysis method, determines the target and requirement of power system load curve clustering analysis, carries out data preprocessing, uses a self-organizing feature mapping neural network to construct a feature extraction model, uses an MAML algorithm to carry out multi-task learning, carries out rapid adjustment through meta learning, applies a Gaussian mixture model to cluster the extracted features, uses indexes to evaluate the clustering effect, carries out clustering result analysis, determines whether the predetermined performance standard is met, and further carries out iterative optimization, so that the self-organizing feature mapping neural network, meta learning and the Gaussian mixture model are combined, and greater value is brought to power system operation and energy management.

Description

Technical Field

[0001] This invention relates to the field of power system engineering technology, and in particular to a method for cluster analysis of power system load curves. Background Technology

[0002] In the field of power systems, load curve clustering analysis is an important problem. When discussing existing techniques for load curve clustering analysis, the following methods are typically involved.

[0003] K-Means Clustering: K-Means is a common clustering method that divides data points into K clusters, such that each data point belongs to the center of the nearest cluster. In load curve clustering, K-Means can group similar load curves, but the number of clusters K needs to be specified in advance.

[0004] Hierarchical clustering: Hierarchical clustering organizes data points hierarchically by constructing a hierarchical tree, thus forming a clustering hierarchy. In load curve clustering, hierarchical clustering can help discover patterns and similarities at different levels.

[0005] Density clustering: Density clustering methods (such as DBSCAN) can identify data points with similar densities, thus enabling the discovery of irregularly shaped clusters. In load curve clustering, density clustering can handle situations where data points are unevenly distributed or have a lot of noise.

[0006] Fuzzy clustering: Fuzzy clustering allows data points to belong to multiple clusters, represented by a certain degree of membership. In load curve clustering, fuzzy clustering can handle situations where data points do not completely belong to a single cluster.

[0007] Deep learning methods: In recent years, deep learning methods have been increasingly used in load curve clustering analysis. For example, convolutional neural networks (CNNs) can capture local patterns in load curves, and recurrent neural networks (RNNs) can consider temporal relationships.

[0008] In summary, while existing load curve clustering analysis techniques may perform well in some cases, they are limited by issues such as diverse data distributions, scarce labeled data, and adaptive learning. Therefore, a new method is needed to improve the accuracy of load curve clustering analysis to meet the requirements of load classification and forecasting in power systems. Summary of the Invention

[0009] This application provides a method for clustering analysis of power system load curves, which improves the accuracy, efficiency, and adaptability of power system load curve clustering analysis.

[0010] This application provides a method for clustering analysis of load curves in a power system, including the following steps:

[0011] A method for clustering analysis of power system load curves, characterized by the following steps: Step 1, determining the objectives and requirements of power system load curve clustering analysis;

[0012] Step 2: Data preprocessing, collecting relevant power system load data, organizing the data, and removing data noise;

[0013] Step 3: Use a self-organizing feature map neural network to build a feature extraction model, select appropriate hyperparameters for training, and extract key features of the power system load curve.

[0014] Step 4: Use the MAML algorithm for multi-task learning to enable the model to quickly adapt to different clustering tasks. Through rapid adjustment via meta-learning, further optimize and accurately extract features.

[0015] Step 5: Apply Gaussian mixture model to cluster the extracted features, determine the number of clusters and the parameters of the Gaussian components, and classify and segment the power system load curves.

[0016] Step 6, Results Analysis: Use indicators to evaluate the clustering effect and analyze the clustering results to determine whether the predetermined performance standards are met.

[0017] Step 7: Iterative optimization. Steps 3 to 6 can be iterated repeatedly as needed until the conditions for a satisfactory result are met.

[0018] Preferably, step 2 specifically includes:

[0019] Step 21: Load the power system load curve data;

[0020] Step 22: Fill in or delete missing values ​​in the data;

[0021] Step 23: Standardize the data.

[0022] Preferably, step 3 specifically includes:

[0023] Step 31: Create the SOM model and define parameters such as the grid size, learning rate, and neighborhood function of the SOM;

[0024] Step 32: Initialize the weights, either randomly or using PCA.

[0025] Step 33, train the SOM using a random or batch training algorithm;

[0026] Step 34: Visualize the results and use the visualized SOM to understand the feature mapping.

[0027] Preferably, step 4 specifically includes:

[0028] Step 41: Extract SOM features. Extract features from SOM and use them as input for meta-learning.

[0029] Step 42: Perform power system load curve clustering analysis using MAML;

[0030] Step 43, train MAML.

[0031] Preferably, step 5 specifically includes:

[0032] Step 51: Create the GMM by initializing it by determining the number of clusters;

[0033] Step 52, train GMM, using SOM features to train GMM.

[0034] Preferably, step 5, which determines the number of clusters, specifically involves selecting the cluster with the largest silhouette coefficient.

[0035] Preferably, step 5, the method for classifying and segmenting the power system load curve, is as follows:

[0036] If the goal is to divide the load curve into different time periods, then based on the clustering results, time periods with similar characteristics can be grouped into one category.

[0037] If the goal is to distinguish between different types of load curves, load curves with similar load patterns are grouped into one category based on the clustering results.

[0038] Preferably, in step 6, the silhouette coefficient, Davis-Bourdin index, and variance ratio criterion are used to evaluate the clustering effect.

[0039] This application presents a powerful, flexible, and accurate clustering analysis method for power system load curves, employing self-organizing feature mapping, model-independent meta-learning, and Gaussian mixture models. Its numerous advantages make it widely applicable in power system operation, optimization, and analysis. Compared to existing clustering and analysis techniques, it offers a more accurate, flexible, and interpretable solution, contributing to the intelligent and efficient operation of power systems.

[0040] 1. By employing SOM, the feature extraction capability is enhanced. Using SOM, this application is able to capture more complex and deeper features. Compared with traditional feature extraction methods, SOM can better represent the intrinsic structure and patterns of power load curves.

[0041] 2. Through the meta-learning techniques of MAML, this approach can adapt to new tasks more quickly and works well even with limited labeled data. This is a significant advantage compared to many traditional machine learning methods.

[0042] 3. Based on the robustness and accuracy of GMM in cluster analysis, this patent can more accurately classify different categories or patterns of power load curves by using Gaussian mixture models for cluster analysis.

[0043] 4. Through automated evaluation and iterative optimization processes, this approach can continuously improve clustering quality. This ensures the quality and consistency of clustering results and allows the model to be fine-tuned as needed.

[0044] 5. This scheme is designed to allow for transfer learning between different power systems and different types of load data. This means that applying the model in new or different environments may be easier and more effective.

[0045] 6. Through more accurate load curve clustering analysis, power companies can better understand load patterns, thereby optimizing resource allocation and improving power supply reliability and efficiency.

[0046] 7. Both SOM and GMM have a degree of interpretability, which can help power system operators understand the key characteristics of load patterns. This helps to analyze the system not only numerically, but also intuitively understand how it works. Attached Figure Description

[0047] Figure 1 This is a flowchart of the method in this application;

[0048] Figure 2 The image shows the effect of clustering 100 points in this application. Detailed Implementation

[0049] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0050] like Figure 1 As shown, this application begins with data preprocessing, where power load data is organized and formatted. Then, key features are extracted using Self-Organizing Feature Map (SOM) training, and Model-Independent Meta-Learning (MAML) is used for multi-task learning to adapt to different clustering tasks. Subsequently, a Gaussian Mixture Model (GMM) is used to accurately cluster the extracted features. The subsequent results analysis evaluates the clustering effectiveness. If the results are unsatisfactory, the process returns to SOM training for iterative optimization; if the conditions are met, the entire analysis process ends. Specifically, the following steps are included:

[0051] Step 1: Determine the objectives and requirements of the power system load curve clustering analysis.

[0052] Identify load patterns: Through analysis and clustering, identify different load patterns existing in the power system and reveal the inherent laws and characteristics of power load.

[0053] Improve system efficiency: By identifying and analyzing different load patterns, optimize the operation strategy of the power system to improve the system's operating efficiency and economy.

[0054] Optimize resource allocation: Based on different load patterns, optimize the allocation of power resources to ensure the stable and reliable operation of the power system.

[0055] Improve forecast accuracy: Optimize the load forecasting model based on clustering results to improve the accuracy of load forecasting.

[0056] Step 2: Data preprocessing. Collect relevant power system load data, organize the data, and remove data noise.

[0057] Step 21: Load power system load curve data

[0058] import pandas as pd

[0059] data_path='path_to_your_data.csv'

[0060] data = pd.read_csv(data_path)

[0061] Step 22: Fill in or delete missing values ​​in the data.

[0062] data.fillna(method='ffill',inplace=True)

[0063] Step 23: Standardize the data.

[0064] from sklearn.preprocessing import StandardScaler

[0065] scaler = StandardScaler()

[0066] data_scaled=scaler.fit_transform(data)

[0067] Step 3: Use a self-organizing feature map neural network (SOM) to build a feature extraction model, select appropriate hyperparameters for training, and extract key features of the power system load curve.

[0068] Step 31: Create the SOM model and define the grid size, learning rate, neighborhood function, and number of iterations for the SOM.

[0069] SOM is used to transform the input space into a mesh structure that is typically lower in dimension, while preserving the topology of the input space.

[0070] import MiniSom from minisom

[0071] som_shape = (20, 20)

[0072] som=MiniSom(som_shape[0], som_shape[1], data_scaled.shape[1], sigma=1.0, learning_rate=0.5)

[0073] Grid Size: The grid size determines the number and layout of neurons in the competing layer. Common layouts include rectangles and hexagons. For example, a 10x10 rectangular grid will have 100 neurons.

[0074] Learning Rate: The learning rate determines the magnitude of weight updates. A high learning rate may lead to rapid learning but may miss the optimal solution, while a low learning rate may lead to more accurate learning but requires more iterations.

[0075] Neighborhood Function:

[0076] The neighborhood function defines which neurons will be updated in each iteration. The Gaussian function is a commonly used neighborhood function.

[0077] Number of iterations: The number of iterations determines the total number of times the model is trained. Each iteration updates the neuron weights.

[0078] Here is a Python code example of how to set these parameters and create an SOM model:

[0079] import MiniSom from minisom

[0080] #Grid Size

[0081] x_size = 10

[0082] y_size = 10

[0083] #Learning Rate

[0084] learning_rate = 0.5

[0085] #Number of iterations

[0086] iterations = 10000

[0087] #Initialize the SOM model

[0088] som=MiniSom(x=x_size,y=y_size,input_len=data.shape[1],sigma=1.0,

[0089] learning_rate=learning_rate)

[0090] #Initialize weights

[0091] som.random_weights_init(data)

[0092] #Training the model

[0093] som.train_random(data,iterations)

[0094] Step 32: Initialize the weights, either randomly or using PCA.

[0095] som.pca_weights_init(data_scaled)

[0096] Step 33, train the SOM using a random or batch training algorithm.

[0097] som.train_batch(data_scaled,1000)#

[0098] Step 34: Visualize the results and use the visualized SOM to understand the feature mapping.

[0099] from pylab import bone,pcolor,colorbar,plot,show

[0100] bone()

[0101] pcolor(som.distance_map().T)

[0102] colorbar()

[0103] show()

[0104] Step 4: Use the MAML algorithm for multi-task learning, enabling the model to quickly adapt to different clustering tasks. Through rapid adjustment via meta-learning, the features extracted are further optimized and accurately refined.

[0105] Meta-learning is a strategy for training models that can quickly adapt to different tasks.

[0106] Step 41: Extract SOM features. Extract features from SOM and use them as input for meta-learning.

[0107] som_features=[som.winner(x)for x in data_scaled]

[0108] Step 42: Perform power system load curve clustering analysis using Model Independent Meta-Learning (MAML).

[0109] import learn2learn as l2l

[0110] import torch

[0111] model = YOUR_MODEL # Define the model structure based on SOM features

[0112] maml=l2l.algorithms.MAML(model,lr=0.01)

[0113] Step 43, train MAML.

[0114] for iteration in range(NUM_ITERATIONS):

[0115] #Meta-learning training is performed here, with iterative training process.

[0116] pass

[0117] Step 5: Apply Gaussian Mixture Model (GMM) to cluster the extracted features, determine the number of clusters and the parameters of the Gaussian components, and classify and segment the power system load curves.

[0118] GMM is a soft clustering algorithm that uses a probabilistic model to describe each cluster.

[0119] Step 51: Create GMM by initializing the GMM by determining the number of clusters.

[0120] from sklearn.mixture import GaussianMixture

[0121] gmm=GaussianMixture(n_components=number_of_clusters)

[0122] Step 52, train GMM, using SOM features to train GMM.

[0123] gmm.fit(som_features)

[0124] labels=gmm.predict(som_features)

[0125] Gaussian Mixture Models (GMMs) are used to cluster the extracted features. The following explains how to determine the number of clusters, the parameters of the Gaussian components, and the basis and methods for classification and segmentation when applying GMMs.

[0126] The silhouette coefficient can be used to measure the stability and compactness of clusters. You can select the number of clusters with the highest silhouette coefficient.

[0127] Determine the parameters of the Gaussian components. In a Gaussian Model, each Gaussian component has its mean, covariance, and weights. These parameters are automatically optimized and determined during model training using maximum likelihood estimation.

[0128] Classification Basis and Methods. After Gaussian Mixture (GMM) clustering, each data point is assigned to a Gaussian component (i.e., a category) based on its characteristics. The classification is primarily based on the probability of a data point within each Gaussian component. The component to which the data point has the highest probability is the component to which it is classified.

[0129] Segmentation criteria and methods: If the goal is to divide the load curve into different time periods, the time periods with similar characteristics can be grouped together based on the clustering results.

[0130] If the goal is to distinguish between different types of load curves, load curves with similar load patterns can be grouped into one category based on the clustering results.

[0131] Below is a simple Python code example that uses GaussianMixture from sklearn to apply GMM clustering:

[0132] from sklearn.mixture import GaussianMixture

[0133] from sklearn.metrics import silhouette_score

[0134] #X extracted the feature data for you

[0135] #n_clusters_list is a list of the number of clusters you want to test, for example: [2,3,4,5]

[0136] best_score = -1

[0137] best_n_clusters = 0

[0138] for n_clusters in n_clusters_list:

[0139] gmm=GaussianMixture(n_components=n_clusters,random_state=0)

[0140] gmm.fit(X)

[0141] labels = gmm.predict(X)

[0142] score=silhouette_score(X,labels)

[0143] if score > best_score:

[0144] best_score = score

[0145] best_n_clusters=n_clusters

[0146] #Retrain the model using the optimal number of clusters

[0147] final_gmm=GaussianMixture(n_components=best_n_clusters,random_state=0)final_gmm.fit(X)

[0148] final_labels=final_gmm.predict(X)

[0149] Step 6: Results analysis. Use appropriate metrics, such as silhouette coefficient and Davis-Bourdin index, to evaluate the clustering effect and perform clustering result analysis to determine whether the predetermined performance criteria are met.

[0150] from sklearn.metrics import silhouette_score,davies_bouldin_score

[0151] silhouette_avg=silhouette_score(som_features,labels)

[0152] db_score=davies_bouldin_score(som_features,labels)

[0153] if silhouette_avg>some_threshold and db_score <another_threshold:

[0154] print("Result is acceptable")

[0155] else:

[0156] print("Result is not satisfactory, continue iteration")

[0157] #Iterative code optimization

[0158] Evaluation indicators:

[0159] Silhouette Score: The silhouette score is a commonly used indicator for evaluating clustering performance. Its value ranges from -1 to 1, and a higher value indicates a better clustering performance.

[0160] Calinski-Harabasz Index: Also known as the variance ratio criterion, the Calinski-Harabasz Index is one of the evaluation indicators of the quality of clustering. The larger the value, the better the clustering effect.

[0161] The Davies-Bouldin Index is used to evaluate the effectiveness of clustering algorithms. A lower Davies-Bouldin Index value indicates better clustering results.

[0162] Step 7: Iterative optimization. Steps 3 to 6 can be iterated repeatedly as needed until the conditions for a satisfactory result are met.

[0163] If the results analysis does not meet the requirements, return to the SOM training phase for iterative optimization, adjusting model parameters and feature selection until the performance standards are met.

[0164] Finally, the cluster analysis is completed, and all processes and results are recorded. The final clustering results and analysis report of the power system load curves are output.

[0165] The above methods enable accurate clustering and analysis of power system load curves. By combining self-organizing feature map neural networks, model-independent meta-learning, and Gaussian mixture models, complex clustering tasks can be handled in a flexible and accurate manner, ensuring the accuracy and robustness of the analysis.

[0166] like Figure 2 The image shows the clustering effect of 100 points selected from a set of power load data. The Silhouette Score is an indicator for evaluating the clustering effect and is used to measure the quality of the clustering results. The Silhouette Score ranges from -1 to 1, with a higher value indicating a better clustering effect. The clustering result in this case is 0.865, indicating a good clustering effect.

[0167] By combining Self-Organizing Eigenmaps (SOM), Model-Independent Meta-Learning (MAML), and Gaussian Mixture Models (GMM) for cluster analysis of power system load curves, a powerful, flexible, and accurate cluster analysis method can be achieved. The fusion and combination of these three technologies can be considered the unique feature of this patent. Specifically, the following aspects can be considered as unique combination methods:

[0168] 1. Step-by-step fusion:

[0169] This patent employs a step-by-step fusion strategy. First, it uses SOM to extract features and reduce dimensionality of the data. Then, it utilizes MAML for meta-learning. Finally, it applies GMM for cluster analysis. Each step is closely linked, providing the necessary input and support for the next step.

[0170] 2. The interaction between feature extraction and meta-learning:

[0171] Features extracted through SOM can be better adapted to the meta-learning model, enabling MAML to find optimized model parameters faster and more accurately. This interaction between feature extraction and meta-learning can be considered a unique fusion method.

[0172] 3. Flexible optimization strategies:

[0173] This method allows users to flexibly adjust the parameters of each model according to actual needs, thereby optimizing the entire cluster analysis process. This flexible optimization strategy is also part of the unique fusion method of this patent.

[0174] 4. Multilevel analysis:

[0175] From feature extraction in SOM to meta-learning in MAML, and then to clustering in GMM, these three methods are combined to form a multi-level, multi-dimensional analysis method that can comprehensively and deeply reveal the inherent laws of power system load curves.

[0176] 5. Iterative optimization:

[0177] The results are analyzed to determine whether the predetermined performance criteria are met. If not, SOM training and meta-learning are repeated, thus realizing a dynamic and iteratively optimized combination method.

[0178] These unique combination and fusion methods are all aimed at better solving the clustering analysis problem of power system load curves, improving the accuracy and robustness of the model, and achieving more efficient and reliable power system management and control.

[0179] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

[0180] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for clustering analysis of load curves of an electric power system, characterized in that, Includes the following steps: Step 1: Determine the objectives and requirements of the power system load curve clustering analysis; Step 2: Data preprocessing, collecting relevant power system load data, organizing the data, and removing data noise; Step 3: Use a self-organizing feature map neural network to build a feature extraction model, select hyperparameters for training, and extract key features of the power system load curve; Step 4: Use the MAML algorithm for multi-task learning to enable the model to quickly adapt to different clustering tasks. Through rapid adjustment via meta-learning, further optimize and accurately extract features. Step 5: Apply Gaussian mixture model to cluster the extracted features, determine the number of clusters and the parameters of the Gaussian components, and classify and segment the power system load curves. Step 6, Results Analysis: Use indicators to evaluate the clustering effect and analyze the clustering results to determine whether the predetermined performance standards are met. Step 7: Iterative optimization. Repeat steps 3 to 6 until the conditions for a satisfactory result are met.

2. The power system load curve clustering analysis method as described in claim 1, characterized in that, Step 2 specifically includes: Step 21: Load the power system load curve data; Step 22: Fill in or delete missing values ​​in the data; Step 23: Standardize the data.

3. The power system load curve clustering analysis method as described in claim 1, characterized in that, Step 3 specifically includes: Step 31: Create the SOM model and define the grid size, learning rate, and neighborhood function parameters of the SOM; Step 32: Initialize the weights, either randomly or using PCA. Step 33, train the SOM using a random or batch training algorithm; Step 34: Visualize the results and use the visualized SOM to understand the feature mapping.

4. The power system load curve clustering analysis method as described in claim 1, characterized in that, Step 4 specifically includes: Step 41: Extract SOM features. Extract features from SOM and use them as input for meta-learning. Step 42: Perform power system load curve clustering analysis using MAML; Step 43, train MAML.

5. The power system load curve clustering analysis method as described in claim 1, characterized in that, Step 5 specifically includes: Step 51: Create the GMM by initializing it by determining the number of clusters; Step 52, train GMM using SOM features processed by the MAML algorithm.

6. The power system load curve clustering analysis method as described in claim 1, characterized in that, Step 5 determines the number of clusters, specifically by selecting the clusters with the largest silhouette coefficients.

7. The power system load curve clustering analysis method as described in claim 1, characterized in that, Step 5, the method for classifying and segmenting the power system load curve, is as follows: If the goal is to divide the load curve into different time periods, then based on the clustering results, time periods with similar characteristics can be grouped into one category. If the goal is to distinguish between different types of load curves, load curves with similar load patterns are grouped into one category based on the clustering results.

8. The power system load curve clustering analysis method as described in claim 1, characterized in that, Step 6 uses the silhouette coefficient, Davis-Bourdin index, and variance ratio criterion to evaluate the clustering effect.

Images