Non-intrusive adjustable power load monitoring method, electronic equipment and medium

Abstract

The invention discloses a non-intrusive adjustable power load monitoring method, electronic equipment and a medium, and the method comprises the steps: obtaining current and voltage data of a household load in a to-be-monitored region, and carrying out the preprocessing of the current and voltage data, and obtaining feature data; screening out a multi-dimensional feature vector from the feature data; and inputting the multi-dimensional feature vector into a pre-trained LightGBM model optimized based on an ant colony algorithm to obtain a power load identification result of the to-be-monitored area. According to the method, the ant colony algorithm is adopted to carry out iterative optimization on the hyper-parameters of the LightGBM model, the optimal parameter combination of the LightGBM model is rapidly and accurately determined, the LightGBM model optimized based on the ant colony algorithm is utilized to realize power load classification identification, and the identification accuracy is improved.

Description

Technical Field

[0001] This application belongs to the field of power technology, specifically relating to a non-intrusive adjustable power load monitoring method, electronic equipment, and medium. Background Technology

[0002] Non-intrusive power load monitoring identifies individual power-consuming units and their operating status by analyzing and processing data collected from regional total load metering. It can be widely applied in smart grids, smart buildings, and other fields. In recent years, with the increasing penetration of new energy sources, the importance of non-intrusive monitoring of adjustable power loads has been growing, becoming a crucial step in the identification of adjustable power loads.

[0003] Driven by continuous breakthroughs in machine learning algorithms, load identification technology has achieved multi-dimensional integrated applications. Techniques such as Support Vector Machines (SVM), Adaboost, and LightGBM have been widely used for load identification in non-intrusive load monitoring. However, traditional modeling methods mainly rely on classic machine learning algorithms such as Support Vector Machines (SVM) and decision trees. These methods have inherent limitations when dealing with complex power load characteristics. Secondly, although adjusting parameters can improve model accuracy, this also significantly increases computational and memory costs. Among numerous machine learning algorithms, Gradient Boosting Decision Trees (GBDT) have been widely used in the field of load forecasting due to their excellent performance.

[0004] LightGBM, as a highly efficient improvement framework to Gradient Boosting Decision Tree (GBDT), integrates several innovative technologies and boasts advantages due to its low memory consumption and computational complexity. Firstly, it employs a histogram approximation algorithm to accelerate the feature splitting process; secondly, it introduces One-Sided Gradient Sampling (GOSS) to optimize data sample weights; and thirdly, it reduces feature dimensionality through the Exclusive Feature Bundling (EFB) method. This technological system significantly improves model training efficiency and prediction accuracy. However, it still faces several shortcomings when dealing with high-dimensional data and complex feature spaces: 1. Its leaf-first growth strategy may lead to overfitting on small datasets or with few features; 2. LightGBM is highly sensitive to hyperparameter selection and requires careful adjustment of multiple parameters (such as learning rate, tree depth, and number of leaves) to achieve optimal performance. Although traditional methods provide various hyperparameters for fine-tuning model performance, selecting the optimal values ​​is challenging and requires significant time and experimentation to adjust various parameters; 3. Most current algorithm designs aim to prevent LightGBM from getting trapped in local optima, and its application in parameter selection before model construction is limited. Therefore, there is an urgent need for a non-intrusive adjustable power load monitoring method that is adaptable to initial conditions and can be used for global optimization to meet the current needs of adjustable load monitoring. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this application proposes a non-intrusive adjustable power load monitoring method, electronic device, and medium, which introduces the ant colony algorithm from the optimization algorithm to optimize the parameters of LightGBM, thereby reducing the computation time and improving the recognition efficiency.

[0006] This application is achieved through the following technical solution:

[0007] A non-intrusive adjustable power load monitoring method, comprising:

[0008] Acquire the current and voltage data of household loads in the area to be monitored and preprocess them to obtain characteristic data;

[0009] Multidimensional feature vectors are selected from the feature data;

[0010] The multidimensional feature vector is input into a pre-trained LightGBM model optimized based on the ant colony algorithm to obtain the power load identification results of the area to be monitored.

[0011] In some embodiments, the preprocessing includes:

[0012] Resampling, noise filtering, steady-state period segmentation, amplitude normalization, and feature extraction.

[0013] In some implementations, the step of filtering multidimensional feature vectors from the feature data to form an optimal dataset includes:

[0014] Calculate the segmentation equity for each feature, sort the features by importance based on the segmentation equity, and remove low-importance features whose cumulative contribution is below a threshold;

[0015] Principal component analysis is used to compress the feature dimensions, forming a multidimensional feature vector.

[0016] In some implementations, the training process of the LightGBM model optimized based on the ant colony algorithm includes:

[0017] Extract typical residential load data from the benchmark dataset and preprocess it;

[0018] Select the optimal dataset based on the preprocessed data;

[0019] The optimal parameter combination of the LightGBM model is determined by iterative optimization using the ant colony algorithm.

[0020] The LightGBM model takes a multidimensional feature vector as input and outputs a device identification label.

[0021] In some implementations, the method of using ant colony optimization to iteratively optimize and determine the optimal parameter combination of the LightGBM model includes:

[0022] Initialize the ant colony algorithm parameters and the model parameter search space;

[0023] In each iteration, the ant selects a set of parameter combinations, trains the LightGBM model using the optimal dataset, and calculates the accuracy as the fitness. Each ant searches the model parameter search space and selects parameter values ​​based on pheromones and heuristic information.

[0024] Update the pheromone based on the quality of the parameter combination chosen by the ant;

[0025] Determine whether the number of iterations has reached the upper limit or whether the accuracy has converged. If so, the currently selected parameter combination is taken as the optimal parameter combination of the LightGBM model; otherwise, continue to the next round of iteration.

[0026] In some implementations, selecting parameter values ​​based on pheromones and heuristic information includes:

[0027] A dynamic adaptive heuristic factor is adopted, that is, the weights are dynamically adjusted according to the parameter contributions during the iteration process;

[0028] A hybrid global and local update algorithm is used to update pheromones during the iteration process.

[0029] In some implementations, the extraction and preprocessing of typical residential load data from the benchmark dataset includes:

[0030] Extract current and voltage data, as well as device tags, of typical residential loads from the benchmark dataset;

[0031] Preprocessing of current and voltage data of typical household loads yields multidimensional feature data;

[0032] The preprocessing mentioned above includes:

[0033] Resampling, noise filtering, steady-state period segmentation, amplitude normalization, and feature extraction.

[0034] In some implementations, selecting the optimal dataset based on the preprocessed data includes:

[0035] Calculate the segmentation equity for each feature, sort the features by importance based on the segmentation equity, and propose low-importance features with a cumulative contribution below a threshold;

[0036] Principal component analysis is used to compress feature dimensions and generate an optimal dataset, which includes multidimensional feature vectors and their labels.

[0037] Secondly, this application proposes an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-described non-intrusive adjustable power load monitoring methods.

[0038] Thirdly, this application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described embodiments of the non-intrusive adjustable power load monitoring method.

[0039] This application proposes a non-intrusive adjustable power load monitoring method, which uses an ant colony algorithm to iteratively optimize the hyperparameters of the LightGBM model, quickly and accurately determining the optimal parameter combination of the LightGBM model. This ant colony-optimized LightGBM model is then used to classify and identify power loads, improving the identification accuracy. Cross-validation shows that the classification and identification algorithm used in this application demonstrates good model generalization and load identification capabilities in practical engineering applications. Furthermore, during the parameter iterative optimization process, this method also employs an adaptive dynamic heuristic factor and a global-local hybrid update method, ensuring that even with a powerful heuristic function, the ants do not completely lose their exploration ability, avoiding being misled by the heuristic function itself and falling into local optima.

[0040] Accordingly, the electronic device and computer-readable storage medium proposed in this application achieve the same technical effects as described above. Attached Figure Description

[0041] The accompanying drawings, which are included to provide a further understanding of the embodiments of this application and form part of this application, do not constitute a limitation on the embodiments of this application. In the drawings:

[0042] Figure 1 This is a flowchart of the non-intrusive adjustable power load monitoring method proposed in the embodiments of this application;

[0043] Figure 2 This is a block diagram illustrating the principle of the non-intrusive adjustable power load monitoring device proposed in this application embodiment;

[0044] Figure 3 This is a schematic diagram of the architecture of the non-intrusive adjustable power load monitoring system proposed in the embodiments of this application;

[0045] Figure 4 This is a schematic diagram of the electronic device proposed in the embodiments of this application;

[0046] Figure 5 This is a schematic diagram of a computer-readable storage medium proposed in an embodiment of this application;

[0047] Figure 6 This is a flowchart illustrating the iterative optimization process based on improved ACD in an embodiment of this application.

[0048] Figure 7 This is a comparison chart showing the optimization performance of the improved ACO-LightGBM algorithm proposed in this application and the traditional ACO-LightGBM algorithm.

[0049] Figure 8 This is a comparison of the confusion matrix of (a) the improved ACO-LightGBM algorithm and (b) the traditional KNN classification and identification algorithm proposed in the embodiments of this application under common load identification conditions;

[0050] Figure 9 Statistics on load identification of the improved ACO-LightGBM based on mathematical features.

[0051] Figure reference numerals and corresponding component names:

[0052] 200-Power load monitoring device, 201-Preprocessing unit, 202-Feature screening unit, 203-Classification and identification unit, 300-Power load monitoring system, 301-Input device, 302-Output device, 303-Processor A, 304-Memory A, 400-Electronic device, 410-Memory B, 420-Processor B, 411-Computer program A, 500-Computer-readable storage medium, 511-Computer program B. Detailed Implementation

[0053] In the following, the terms “comprising” or “may include” as used in the various embodiments of this application indicate the presence of a function, operation, or element of the invention and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in the various embodiments of this application, the terms “comprising,” “having,” and their cognates are intended only to indicate a specific feature, number, step, operation, element, component, or combination of the foregoing and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or adding one or more combinations of the foregoing.

[0054] In various embodiments of this application, the expression "or" or "at least one of A and / or B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A and / or B" may include A, may include B, or may include both A and B.

[0055] The terms used in the various embodiments of this application (such as "first," "second," etc.) may modify various constituent elements in the various embodiments, but do not limit the corresponding constituent elements. For example, the above terms do not limit the order and / or importance of the elements. The above terms are only used for the purpose of distinguishing one element from other elements. For example, a first user device and a second user device refer to different user devices, although both are user devices. For example, without departing from the scope of the various embodiments of this application, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element.

[0056] It should be noted that if a description is made of "connecting" one component to another, then the first component can be directly connected to the second component, and a third component can be "connected" between the first and second components. Conversely, when a component is "directly connected" to another component, it can be understood that there is no third component between the first and second components.

[0057] The terminology used in the various embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the various embodiments of this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. The terms (such as those defined in a generally used dictionary) are to be interpreted as having the same meaning as in the context of the relevant technical field and are not to be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0059] This application proposes a non-intrusive adjustable power load monitoring method, such as... Figure 1 As shown, the method proposed in this application includes the following steps:

[0060] Step 110: Obtain the current and voltage data of the household load in the area to be monitored and preprocess them to obtain characteristic data;

[0061] Step 120: Select multidimensional feature vectors from the preprocessed feature data;

[0062] Step 130: Input the multidimensional feature vector into the pre-trained LightGBM model optimized by the ant colony algorithm to obtain the power load identification results of the area to be monitored.

[0063] Furthermore, in step 110 of this application embodiment, the data preprocessing process includes: resampling, noise filtering, steady-state period segmentation, amplitude normalization, and feature extraction (harmonic features, power features, VI trajectory extraction, etc.).

[0064] Furthermore, in step 120 of this application embodiment, the multidimensional feature vector acquisition process includes: calculating the segmentation equity of each feature, sorting the features according to their importance based on the segmentation equity, removing low-importance features with a cumulative contribution rate below a threshold (e.g., 90%), and reducing redundant information; then using principal component analysis algorithm to compress the feature dimension, reduce training time, retain the main feature information, and generate a multidimensional feature vector.

[0065] Furthermore, in step 130 of this embodiment, the training process of the LightGBM model optimized based on the ant colony algorithm includes:

[0066] Step 131: Extract typical household load data from the PLAID (Plug-Load Appliance Identification Dataset) benchmark dataset and preprocess it to obtain multidimensional feature data. The PLAID benchmark dataset acquires voltage and current parameters of residential electrical equipment based on a 30kHz high-frequency acquisition protocol, covering the operating parameters and electrical characteristics of 11 typical household appliances. Each device category in the PLAID dataset contains dozens of specific model samples. Multiple sets of observation data were recorded for each device during different state transitions. This observation data was processed: through data cleaning and other steps, time windows of several seconds containing the stable operation phase and the startup transient process were extracted, and abnormal data such as noise caused by measurement errors were removed, ultimately retaining 1074 instances. Specifically, this includes current and voltage data and device labels for typical household loads such as air conditioners, fluorescent lamps, fans, refrigerators, hair dryers, heaters, incandescent lamps, computers, microwave ovens, vacuum cleaners, and washing machines. Table 1 shows examples from the PLAID benchmark dataset. The specific preprocessing steps include: resampling, noise filtering, steady-state period segmentation, amplitude normalization, and feature extraction (harmonic features, power features, and VI trajectory extraction). Resampling, noise filtering, steady-state period segmentation, amplitude normalization, harmonic features, and power feature extraction can all be implemented using existing technologies in this field. Furthermore, the extraction of the VI trajectory feature, i.e., the VI trajectory matrix, is based on the standardized voltage-current function relationship of household appliances. A mathematical model is established within one steady-state period and ultimately displayed through an image. The specific implementation steps are as follows:

[0067] Extract the load cycle voltage-current timing signal and perform dual-channel data standardization processing;

[0068] Build a A square matrix is ​​used as the feature grid for the VI trajectory, and all matrix elements are set to zero in the initial stage.

[0069] The VI timing signal is mapped to the projection transformation. In a two-dimensional matrix of order 2, the formula is as follows:

[0070]

[0071]

[0072] in, , These represent the row index and column index of the matrix, respectively. , They represent the normalized i-th Voltage and current values ​​at each sampling point; This indicates rounding up; the VI trajectory matrix can be obtained using the above formula;

[0073] Simultaneously, the extracted harmonic features are integrated with the VI trajectory matrix to form a hybrid feature matrix. The steps are as follows:

[0074] The characteristic sequence is confirmed, which includes three additional parameters, arranged in order as follows: 3rd harmonic amplitude, 5th harmonic amplitude and 7th harmonic amplitude;

[0075] Standardize the same feature from different samples to ensure the range is within... The formula is as follows:

[0076]

[0077] in, For the normalized first The first sample One eigenvalue; For the first before normalization The first sample One eigenvalue; , The first The maximum and minimum sample values ​​of each feature.

[0078] Convert the standardized eigenvalues ​​to Bit-bin encoding enables the mapping of numerical values ​​to discrete bit patterns;

[0079] generate Two-dimensional feature matrix As a supplementary feature matrix, all elements of this matrix are initialized to zero.

[0080] Fill binary values ​​into the two-dimensional feature matrix This is then horizontally concatenated with the VI trajectory matrix to form a hybrid feature matrix.

[0081] Table 1 Examples of Plaid benchmark datasets

[0082] Equipment type equipment Example air conditioner 14 66 fluorescent lamp 31 175 fan 23 115 refrigerator 21 38 hair dryer 32 156 heater 7 35 Incandescent light bulb 23 114 computer 34 172 Micro-wave oven 28 139 Vacuum cleaner 8 38 washing machine 9 26 total 230 1074

[0083] Step 132: Select the optimal dataset: Calculate the segmentation equity of each feature (e.g., current span, self-intersection point, change in instantaneous admittance, angle between the maximum and minimum points, distance between the maximum and minimum points, etc.). In this embodiment, a histogram optimization strategy is used to process the segmentation equity of each feature, which evaluates only for each discrete interval, significantly reducing the frequency of segmentation equity evaluation and improving computational efficiency. Based on the segmentation equity, the features are sorted by importance, and low-importance features with a cumulative contribution below a threshold (e.g., 90%) are removed to reduce redundant information. Then, principal component analysis algorithm is used to compress the feature dimensions, reduce training time, retain the main feature information, and generate the optimal dataset. This optimal dataset contains multi-dimensional feature vectors and corresponding labels (equipment type labels) for subsequent iterative training.

[0084] Step 133: The ant colony algorithm is used for iterative optimization to determine the optimal parameter combination of the LightGBM model. The input of the LightGBM model is a multi-dimensional feature vector, and the output is a device identification label. The hyperparameters include: num_leaves (number of leaves), learning_rate (learning rate), lambda_l1 (L1 regularization), lambda_l2 (L2 regularization), and n_estimators (number of base learners).

[0085] The specific iterative optimization process includes:

[0086] Initialization: Each ant represents a set of LightGBM parameter combinations, and the number of ants, number of iterations, initial pheromone, parameter search space, etc. are set;

[0087] Iterative loop: In each iteration, the ant selects a set of parameter combinations, trains the LightGBM model using the optimal dataset, and calculates the accuracy as fitness: Each ant searches in the parameter search space and selects parameter values ​​based on pheromones and heuristic information;

[0088] Pheromone Update: Based on the quality of the parameter combination chosen by the ant, the pheromone is updated: high-accuracy solutions (i.e., higher-quality solutions) will receive more pheromones, thus attracting more ants; low-accuracy solutions will have diminished pheromones.

[0089] Convergence criterion: Determine whether the number of iterations has reached the upper limit or whether the accuracy has converged. If so, the currently selected parameter combination is taken as the optimal parameter combination for the LightGBM model; otherwise, continue to the next iteration.

[0090] Furthermore, during the iteration process, the ant colony algorithm of this application adopts a dynamic adaptive heuristic factor, that is, the weights are dynamically adjusted according to the parameter contributions during the iteration process to solve the static weight bias problem. The dynamic heuristic function formula is as follows:

[0091]

[0092] in: The learning rate; For parameters The number of times it appears in the historical best solution; This represents the number of ants in the ant colony.

[0093] Simultaneously, a hybrid global and local update algorithm is employed to update pheromones during the iteration process. The core principle is as follows: global update enhances the pheromone concentration of the historically optimal path during iteration, guiding the population towards a high-quality solution region; local update, after an ant completes its path search, applies pheromone decay to its traversed path, preventing a single dominant path from prematurely monopolizing the search direction, thereby maintaining population diversity. The formula for enhancing the global pheromone in the initial stage is shown below:

[0094]

[0095]

[0096] in: For path arrive exist The pheromone concentration at any given time; For path arrive exist The pheromone concentration at any given time; The global pheromone evaporation rate; For pheromone increment; The pheromone enhancement coefficient; The objective function value of the optimal solution; This represents the ranking of the optimal solution in the iteration history.

[0097] To maintain search diversity along local paths, the formulas for pheromone volatilization and local pheromone volatilization rate are as follows:

[0098]

[0099]

[0100] in: Local pheromone evaporation rate; This is the initial value for pheromones; Based on local volatility; This represents the total number of iterations.

[0101] During the iteration process, dynamic heuristic functions can guide ants to make more informed choices in the initial stages, laying a better foundation for subsequent searches (including local and global update guidance). The iterative solution process described above is essentially an iterative optimization based on a Markov decision process: the algorithm constructs the solution space through the distributed exploration of individual ants and gradually converges to the optimal solution domain under the guidance of a pheromone positive feedback mechanism. The specific process is as follows... Figure 6 As shown, the core parameters of the algorithm are first initialized: the initial number of iteration cycles. Maximum number of iteration cycles Group size parameters Then, the first round of optimization process is started, the first ant is deployed to perform a global search and generate candidate solutions, and the ant number is incremented sequentially until the group completes the first round of iteration and obtains the initial solution set; the convergence of the solution set is evaluated, and if the termination condition is met, the optimal solution is output; otherwise, the iteration counter is incremented and subsequent rounds of group search are performed until the maximum iteration threshold is reached.

[0102] Based on the above iterative optimization process, the LightGBM model parameters optimized using the ant colony algorithm are shown in Table 2, and the following are obtained: Figure 7 The image shows a comparison of the iterative optimization performance using the improved ACO-LightGBM algorithm and the traditional ACO-LightGBM algorithm. Figure 7 As can be seen, the improved ACO-LightGBM algorithm used in the embodiments of this application can quickly iterate and converge to the target result.

[0103] Table 2. Parameters of the LightGBM model after ant colony optimization.

[0104] parameter value boosting_type gbdt objective regression learning_rate 0.32 lambda_l1 0.747747747 lambda_l2 0.490490490 num_leaves 23 max_depth 15.368 min_data_in_leaf 25.1675858 min_sum_hessian_in_leaf 1.84 min_gain_to_split 0.098 bagging_fraction 0.1999 bagging_freq 3.976 feature_fraction 0.03 verbose -1

[0105] In Table 2, boosting_type represents the boosting type; objective represents the objective; learning_rate represents the learning rate; lambda_l1 represents L1 regularization; lambda_l2 represents L2 regularization; num_leaves represents the number of leaves; max_depth represents the maximum depth; min_data_in_leaf represents the minimum number of samples in a single leaf node; min_sum_hessian_in_leaf represents the minimum sum of the sample weights in a leaf node; min_gain_to_split represents the minimum gain criterion used to define split nodes; bagging_fraction represents the proportion of data selected in each iteration; bagging_freq represents the frequency of randomly selected data; feature_fraction represents the proportion of randomly selected features in each iteration; and verbose represents the information complexity.

[0106] Based on the same technical concept described above, this application also proposes a non-intrusive adjustable power load monitoring device, such as... Figure 2 As shown, the power load monitoring device 200 includes:

[0107] The preprocessing unit 201 is configured to acquire the current and voltage data of the household load in the area to be detected and preprocess them to obtain feature data. The specific preprocessing process is as described in step 110 above, and will not be repeated here.

[0108] Feature filtering unit 202 is configured to filter out multi-dimensional feature vectors from the preprocessed feature data. The specific process for obtaining the optimal dataset is as described in step 120 above, and will not be repeated here.

[0109] Furthermore, the classification and identification unit 203 is configured to input the optimal dataset into a pre-trained LightGBM model optimized based on the ant colony algorithm to obtain the power load identification results of the area to be monitored. The training process of the LightGBM model optimized based on the ant colony algorithm is as described in step 130 above, and will not be repeated here.

[0110] Based on the same technical concept described above, this application also proposes a non-intrusive adjustable power load monitoring system, such as... Figure 3 As shown, the power load monitoring system 300 proposed in this application embodiment includes:

[0111] The system comprises an input device 301, an output device 302, a processor A303, and a memory A304; wherein the number of processors A303 and memory A304 can be one or more. Figure 3 The following description uses a processor A303 and a memory A304 as an example. The input device 301, output device 302, processor A303, and memory A304 can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0112] Specifically, by calling the operation instructions stored in memory A304, processor A303 executes the following steps:

[0113] Acquire the current and voltage data of household loads in the area to be monitored and preprocess them to obtain characteristic data;

[0114] Multidimensional feature vectors are selected from the preprocessed feature data;

[0115] The optimal dataset is input into a pre-trained LightGBM model optimized by the ant colony algorithm to obtain the power load identification results of the area to be monitored.

[0116] Optionally, by calling the operation instructions stored in memory A304, processor A303 is also used to execute any of the embodiments in the corresponding examples of the above-described power load monitoring method.

[0117] Based on the same technical concept described above, this application also proposes an electronic device, such as... Figure 4 As shown, the electronic device 400 includes: a memory B410, a processor B420, and a computer program A411 stored in the memory B410 and executable on the processor B420. When the processor B420 executes the computer program A411, it performs the following steps:

[0118] Acquire the current and voltage data of household loads in the area to be monitored and preprocess them to obtain characteristic data;

[0119] Multidimensional feature vectors are selected from the preprocessed feature data;

[0120] The optimal dataset is input into a pre-trained LightGBM model optimized by the ant colony algorithm to obtain the power load identification results of the area to be monitored.

[0121] Optionally, when processor B420 executes computer program A411, it can implement any of the embodiments in the corresponding examples of the above-described power load monitoring method.

[0122] It should be noted that the electronic device proposed in this application embodiment is a device used to implement the above-mentioned power load monitoring method. Therefore, based on the above-mentioned power load monitoring method proposed in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this application embodiment. Therefore, how the electronic device specifically implements the above-mentioned power load monitoring method will not be described in detail here. Any electronic device used by those skilled in the art to implement the above-mentioned power load monitoring method falls within the scope of protection of this application.

[0123] Based on the same technical concept described above, embodiments of this application also propose a computer-readable storage medium, such as... Figure 5 As shown, the computer-readable storage medium 500 stores a computer program B511, which, when executed by a processor, performs the following steps:

[0124] Acquire the current and voltage data of household loads in the area to be monitored and preprocess them to obtain characteristic data;

[0125] Multidimensional feature vectors are selected from the preprocessed feature data;

[0126] The optimal dataset is input into a pre-trained LightGBM model optimized by the ant colony algorithm to obtain the power load identification results of the area to be monitored.

[0127] Optionally, when the computer program B511 is executed by the processor, it can implement any of the embodiments corresponding to the above-described power load monitoring method.

[0128] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0129] To verify the power load identification performance of the LightGBM model (ACO-LightGBM algorithm) optimized by ant colony algorithm proposed in this application, typical classification and identification models were constructed using Python: K-Nearest Neighbor (KNN), Radial Basis Support Vector Machine (RBF SVM), and Decision Tree (DT) algorithms were used as comparison algorithms. The ACO-LightGBM algorithm and its comparison algorithms were applied to a unified benchmark test set (including single harmonic features, single VI trajectory image features, and combined indicators of harmonics and VI trajectory images, i.e., digital features and hybrid feature matrices based on VI trajectory and harmonics; digital features mainly refer to traditional features such as current span, number of self-intersections, number of peak values ​​in the middle section, instantaneous admittance change, and maximum and minimum distances; the hybrid matrix feature is a hybrid feature matrix formed by integrating harmonic features and the VI trajectory matrix, which can more effectively reflect the transient characteristics of load behavior) for lateral performance evaluation. Multidimensional classification performance analysis was performed, and the performance comparison results shown in Tables 3, 4, and 5 were obtained. Figure 8 The diagram shows the confusion matrix generated by using the ACO-LightGBM algorithm and the KNN algorithm to identify common loads and make statistical sense.

[0130] Table 3. Identification accuracy of various algorithms under different features

[0131] KNN RBF SVM DT ACO-LightGBM harmonic 60.99% 55.31% 60.61% 62.66% VI trajectory image 80.17% 44.51% 79.42% 82.51% Combined indicators (numerical characteristics) 85.28% 41.53% 74.77% 94.78% Combined Indicators (Hybrid Matrix) 84.17% 61.03% 71.66% 93.57%

[0132] Table 4 Running time of multiple algorithms under combined metrics

[0133] KNN RBF SVM DT ACO-LightGBM Combined indicators (numerical characteristics) 5.782s 20.667s 11.964s 16.336s Combined Indicators (Hybrid Matrix) 18.872s 35.789s 15.724s 70.258s

[0134] Table 5. Identification accuracy of ACO-LightGBM and KNN models

[0135] Electrical appliances Label Model Recognition rate Model Recognition rate air conditioner 1 KNN 59% ACO-LightGBM 84% fluorescent lamp 2 KNN 94% ACO-LightGBM 97% fan 3 KNN 62% ACO-LightGBM 82% refrigerator 4 KNN 57% ACO-LightGBM 73% hair dryer 5 KNN 85% ACO-LightGBM 93% heater 6 KNN 62% ACO-LightGBM 91% Incandescent lamp 7 KNN 85% ACO-LightGBM 94% computer 8 KNN 95% ACO-LightGBM 99% Micro-wave oven 9 KNN 100% ACO-LightGBM 100% Vacuum cleaner 10 KNN 92% ACO-LightGBM 92% washing machine 11 KNN 60% ACO-LightGBM 96%

[0136] Based on the results in Tables 3 to 5 and Figure 8It can be seen that the ACO-LightGBM algorithm proposed in this application embodiment is superior to existing load identification algorithms in terms of identification accuracy, and has outstanding performance in multidimensional classification.

[0137] Furthermore, Table 6 shows the accuracy of the ACO-LightGBM algorithm used in the embodiments of this application for identification using hybrid matrix features and digital features, as well as... Figure 9 The statistical results of load identification using the ACO-LightGBM algorithm based on mathematical features are shown.

[0138] Table 6. Identification accuracy of the ACO-LightGBM algorithm using hybrid matrix features and numerical features.

[0139] Electrical appliances Based on the recognition rate of hybrid matrix features / % Based on mathematical feature recognition rate / % air conditioner 84% 86% fluorescent lamp 97% 97% fan 82% 87% refrigerator 73% 84% hair dryer 93% 97% heater 91% 91% Incandescent lamp 94% 94% computer 99% 100% Micro-wave oven 100% 100% Vacuum cleaner 92% 97% washing machine 96% 97%

[0140] From Table 6 and Figure 9 It can be seen that the ACO-LightGBM algorithm proposed in this application improves the accuracy of load identification by 11% compared with that of mixed matrix features using digital features. Furthermore, it has a high model generalization ability in the identification process based on both mixed matrix features and digital features, indicating that the ACO-LightGBM algorithm proposed in this application has better environmental adaptability in practical applications.

[0141] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A non-intrusive adjustable power load monitoring method, characterized in that, include: Acquire the current and voltage data of household loads in the area to be monitored and preprocess them to obtain characteristic data; Multidimensional feature vectors are selected from the feature data; The multidimensional feature vector is input into a pre-trained LightGBM model optimized based on the ant colony algorithm to obtain the power load identification results of the area to be monitored.

2. The non-intrusive adjustable power load monitoring method according to claim 1, characterized in that, The preprocessing includes: Resampling, noise filtering, steady-state period segmentation, amplitude normalization, and feature extraction.

3. The non-intrusive adjustable power load monitoring method according to claim 1, characterized in that, The step of selecting multi-dimensional feature vectors from the feature data to form the optimal dataset includes: Calculate the segmentation equity for each feature, sort the features by importance based on the segmentation equity, and remove low-importance features whose cumulative contribution is below a threshold; Principal component analysis is used to compress the feature dimensions, forming a multidimensional feature vector.

4. A non-intrusive adjustable power load monitoring method according to any one of claims 1-3, characterized in that, The training process of the LightGBM model optimized based on the ant colony algorithm includes: Extract typical residential load data from the benchmark dataset and preprocess it; Select the optimal dataset based on the preprocessed data; The optimal parameter combination of the LightGBM model is determined by iterative optimization using the ant colony algorithm. The LightGBM model takes a multidimensional feature vector as input and outputs a device identification label.

5. The non-intrusive adjustable power load monitoring method according to claim 4, characterized in that, The method of using ant colony optimization to iteratively optimize and determine the optimal parameter combination for the LightGBM model includes: Initialize the ant colony algorithm parameters and the model parameter search space; In each iteration, the ant selects a set of parameter combinations, trains the LightGBM model using the optimal dataset, and calculates the accuracy as the fitness. Each ant searches the model parameter search space and selects parameter values ​​based on pheromones and heuristic information. Update the pheromone based on the quality of the parameter combination chosen by the ant; Determine whether the number of iterations has reached the upper limit or whether the accuracy has converged. If so, the currently selected parameter combination is taken as the optimal parameter combination of the LightGBM model; otherwise, continue to the next round of iteration.

6. The non-intrusive adjustable power load monitoring method according to claim 4, characterized in that, The selection of parameter values ​​based on pheromones and heuristic information includes: A dynamic adaptive heuristic factor is adopted, that is, the weights are dynamically adjusted according to the parameter contributions during the iteration process; A hybrid global and local update algorithm is used to update pheromones during the iteration process.

7. The non-intrusive adjustable power load monitoring method according to claim 4, characterized in that, The process of extracting typical residential load data from the benchmark dataset and preprocessing it includes: Extract current and voltage data, as well as device tags, of typical residential loads from the benchmark dataset; Preprocessing of current and voltage data of typical household loads yields multidimensional feature data; The preprocessing mentioned above includes: Resampling, noise filtering, steady-state period segmentation, amplitude normalization, and feature extraction.

8. The non-intrusive adjustable power load monitoring method according to claim 7, characterized in that, The process of selecting the optimal dataset based on the preprocessed data includes: Calculate the segmentation equity for each feature, sort the features by importance based on the segmentation equity, and propose low-importance features with a cumulative contribution below a threshold; Principal component analysis is used to compress feature dimensions and generate an optimal dataset, which includes multidimensional feature vectors and their labels.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the non-intrusive adjustable power load monitoring method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the non-intrusive adjustable power load monitoring method according to any one of claims 1-7.

Images