A load decomposition method, device, storage medium and computer device

Abstract

The application discloses a load decomposition method and device, a storage medium and a computer device, and relates to the field of intelligent power utilization. The method comprises the following steps: sampling the current and voltage of a to-be-tested load according to a preset sampling frequency to obtain an active power sequence and a reactive power sequence; extracting a variable-frequency air conditioner starting ramp feature from the active power sequence and the reactive power sequence to obtain an encoding sequence; splicing the active power sequence and the reactive power sequence in a window with the encoding sequence in the window in a feature dimension by a sliding window method to obtain a load feature; and inputting the load feature into a pre-trained variable-frequency air conditioner load decomposition model to obtain an active power value. The application inputs the active power sequence, the reactive power sequence and the encoding sequence into the model to directly obtain the active power of the variable-frequency air conditioner, enhances the generalization ability of the algorithm, and enables the algorithm to stably decompose the actual power consumption of the variable-frequency air conditioner in a scene under different environments.

Description

Technical Field

[0001] This application relates to the field of smart electricity use, and in particular to a load splitting method, apparatus, storage medium, and computer equipment. Background Technology

[0002] Non-Intrusive Load Monitoring (NILM) refers to installing monitoring equipment at the user bus inlet. Through intelligent algorithms, it decomposes the collected total load electrical quantities into independent operating data for each appliance and identifies the operating status of each appliance, thereby analyzing user electricity consumption behavior. Compared to invasive monitoring, NILM is more economical, reliable, and easier to implement. NILM can be divided into two types based on different technical approaches: event-based detection and non-event-based detection. The former aims to detect and classify the on / off events of appliances to achieve load monitoring, also known as load identification; the latter predicts the power sequence of target appliances or infers possible combinations of active appliances based on bus power sequences or other characteristics, also known as load decomposition.

[0003] Variable frequency air conditioners are smart home appliances that use variable frequency technology to control the compressor's speed, thereby achieving precise indoor temperature regulation. Compared to fixed frequency air conditioners, variable frequency air conditioners are more energy-efficient and can flexibly adjust their operating status according to actual needs, achieving more efficient cooling or heating effects. The operation of variable frequency air conditioners is complex; the continuous fluctuations in active and reactive power increase the design difficulty and complexity of load decomposition algorithms for them. When applying this technology to real-world scenarios, the algorithm's generalization ability is particularly important. However, related technologies cannot guarantee accurate and stable decomposition of the actual power consumption of variable frequency air conditioners in different environments, indicating poor generalization ability of the algorithms. Summary of the Invention

[0004] In view of this, this application provides a load decomposition method, apparatus, storage medium and computer equipment, the main purpose of which is to solve the technical problems that related technologies cannot ensure the accurate and stable decomposition of the actual power consumption of variable frequency air conditioners in different environments, and the algorithm has poor generalization ability.

[0005] According to one aspect of this application, a load decomposition method is provided, the method comprising:

[0006] The current and voltage of the load under test are sampled according to a preset sampling frequency to obtain several pairs of current and voltage data. Several pairs of active power and reactive power data are calculated based on each pair of current and voltage data. The active power and reactive power data are then aggregated to obtain an active power sequence and a reactive power sequence.

[0007] The variable frequency air conditioner start-up slope feature is extracted from the active power sequence and the reactive power sequence of the sample load. The variable frequency air conditioner start-up slope feature is encoded together with the other non-variable frequency air conditioner start-up slope features to obtain the encoded sequence of the sample load.

[0008] The active power sequence and reactive power sequence of the sample load are processed by the sliding window method. The active power sequence and reactive power sequence of the sample load within the window are concatenated with the encoded sequence of the sample load within the same window in the feature dimension to obtain the load features.

[0009] The load characteristics are input into a pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model.

[0010] Optionally, the logarithmic aggregation process is performed on the active power data and the reactive power data to obtain an active power sequence and a reactive power sequence, including:

[0011] All active power data are grouped according to a preset number of groups to obtain several active power data groups. The minimum value in each active power data group is extracted to obtain the active power sequence.

[0012] All reactive power data are grouped according to the preset number of groups to obtain several reactive power data groups. The minimum value in each reactive power data group is extracted to obtain the reactive power sequence.

[0013] Optionally, extracting the start-up slope characteristics of the variable frequency air conditioner from the active power sequence and the reactive power sequence includes:

[0014] The active power sequence and the reactive power sequence are processed by the sliding window method according to the first preset window length and the first preset step size to obtain the first active power sequence and the first reactive power sequence within the first window.

[0015] Calculate the correlation coefficient between the first active power sequence and the first reactive power sequence within the first window. If the correlation coefficient is greater than a preset correlation coefficient threshold, extract the first active power sequence and the first reactive power sequence within the first window as the start-up slope feature of the variable frequency air conditioner.

[0016] Optionally, the encoding of the variable frequency air conditioner's start-up slope feature and other non-variable frequency air conditioner start-up slope features to obtain an encoding sequence includes:

[0017] The data points corresponding to the first window that exhibit the variable frequency air conditioner start-up slope feature are encoded as 1, and the data points corresponding to the first window that do not exhibit the variable frequency air conditioner start-up slope feature are encoded as -1, thus obtaining the encoding sequence.

[0018] Optionally, the step of processing the active power sequence and reactive power sequence using the sliding window method, and concatenating the active power sequence and reactive power sequence within the window with the coded sequence within the same window in the feature dimension to obtain the load features, includes:

[0019] The active power sequence and reactive power sequence are processed by the sliding window method according to the second preset window length and the second preset step size to obtain multiple second windows, as well as the second active power sequence, the second reactive power sequence and the coding sequence corresponding to each second window;

[0020] The second active power sequence, the second reactive power sequence, and the encoded sequence corresponding to the second window are concatenated to obtain multiple load characteristics corresponding to multiple second windows.

[0021] Optionally, the training method for the variable frequency air conditioning load decomposition model includes:

[0022] Obtain multiple sample loads;

[0023] For each sample load, the current and voltage of the sample load are sampled according to a preset sampling frequency to obtain several pairs of current data and voltage data. Based on each pair of current data and voltage data, several pairs of active power data and reactive power data of the sample load are calculated. The several pairs of active power data and reactive power data are aggregated to obtain the active power sequence and reactive power sequence of the sample load.

[0024] The variable frequency air conditioner start-up slope feature is extracted from the active power sequence and the reactive power sequence of the sample load. The variable frequency air conditioner start-up slope feature is encoded together with the other non-variable frequency air conditioner start-up slope features to obtain the encoded sequence of the sample load.

[0025] The active power sequence and reactive power sequence of the sample load are processed by the sliding window method. The active power sequence and reactive power sequence of the sample load within the window are concatenated with the encoded sequence of the sample load within the same window in the feature dimension to obtain the load characteristics of the sample load.

[0026] Based on the load characteristics of each sample load and the sample label corresponding to each load characteristic, a deep learning model is trained to obtain a variable frequency air conditioner load decomposition model, wherein the sample label is the active power value of the variable frequency air conditioner.

[0027] Optionally, the method for obtaining the sample labels includes:

[0028] While sampling the current and voltage of the sample load at a preset sampling frequency, the active power value of the variable frequency air conditioner is also collected.

[0029] Based on the correspondence between the active power sequence and reactive power sequence of the sample load and the active power value of the variable frequency air conditioner, the load characteristics of the sample load are assigned sample labels.

[0030] According to another aspect of this application, a load-dissipating apparatus is provided, the apparatus comprising:

[0031] The data acquisition module is used to sample the current and voltage of the load under test according to a preset sampling frequency, obtain several pairs of current data and voltage data, calculate several pairs of active power data and reactive power data based on each pair of current data and voltage data, and aggregate the several pairs of active power data and reactive power data to obtain active power sequence and reactive power sequence.

[0032] The encoding module is used to extract the variable frequency air conditioner start-up slope feature from the active power sequence and the reactive power sequence of the sample load, and encode the variable frequency air conditioner start-up slope feature and the other non-variable frequency air conditioner start-up slope features to obtain the encoded sequence of the sample load.

[0033] The load feature generation module is used to process the active power sequence and reactive power sequence of the sample load using a sliding window method, and to concatenate the active power sequence and reactive power sequence of the sample load within the window with the encoded sequence of the sample load within the same window in the feature dimension to obtain the load feature.

[0034] The load decomposition module is used to input the load characteristics into a pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model.

[0035] According to another aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described load decomposition method.

[0036] According to another aspect of this application, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described load decomposition method.

[0037] This application provides a load decomposition method, apparatus, storage medium, and computer equipment. The method includes sampling the current and voltage of the load to be tested at a preset sampling frequency to obtain several pairs of current and voltage data. Based on each pair of current and voltage data, several pairs of active power and reactive power data are calculated, and the active power and reactive power data are aggregated to obtain an active power sequence and a reactive power sequence. A variable frequency air conditioner start-up slope feature is extracted from the active power and reactive power sequences of the sample load. This variable frequency air conditioner start-up slope feature is encoded along with other non-variable frequency air conditioner start-up slope features to obtain an encoded sequence of the sample load. The active power and reactive power sequences of the sample load are processed using a sliding window method. The active power and reactive power sequences of the sample load within a window are concatenated with the encoded sequences of the sample load within the same window in a feature dimension to obtain load features. The load features are input into a pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model. This application removes the influence of environmental noise by aggregating active and reactive power data, reduces computational load and increases computational accuracy through the sliding window method, and accurately captures and analyzes the electrical characteristics of variable frequency air conditioners. It achieves accurate decomposition of the load power of variable frequency air conditioners, collects a large number of samples with labels of the actual active power values ​​of variable frequency air conditioners to train the load decomposition model, and inputs the active power sequence, reactive power sequence and coded sequence into the model to directly obtain the active power of variable frequency air conditioners, which enhances the generalization ability of the algorithm and enables it to stably decompose the actual power consumption of variable frequency air conditioners in different environments.

[0038] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0039] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0040] Figure 1 A schematic flowchart of a load decomposition method provided in an embodiment of this application is shown;

[0041] Figure 2 A flowchart illustrating another load decomposition method provided in an embodiment of this application is shown;

[0042] Figure 3A schematic diagram of the active and reactive power distribution of a variable frequency air conditioner provided in an embodiment of this application is shown.

[0043] Figure 4 This illustration shows a partially enlarged schematic diagram of the variable frequency air conditioner slope provided in an embodiment of this application;

[0044] Figure 5 A schematic diagram showing the decomposition results of the variable frequency air conditioning load decomposition model provided in an embodiment of this application is shown;

[0045] Figure 6 A schematic diagram of the architecture of the variable frequency air conditioning load decomposition model provided in an embodiment of this application is shown;

[0046] Figure 7 A schematic diagram of a load decomposition device provided in an embodiment of this application is shown;

[0047] Figure 8 This illustration shows a schematic diagram of a computer device for implementing a variable frequency air conditioning load decomposition method, according to an embodiment of this application. Detailed Implementation

[0048] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0049] The basic principle of non-intrusive load identification is to monitor signals such as voltage and current at the power inlet, and then use intelligent algorithms to process and analyze the collected data to identify the type, operating status, and energy consumption information of each electrical device. This technology can decompose the load within the entire system, breaking down the total electrical load into the electricity consumed by each electrical device, and providing information such as the operating time and start / stop status of each device.

[0050] Key technologies for non-intrusive load decomposition include the following aspects: Data measurement: Acquiring signals such as voltage and current at the power inlet through monitoring equipment as input data. Preprocessing: Performing preprocessing operations such as cleaning and filtering on the input data to eliminate noise and interference and improve data accuracy. Event detection: Utilizing algorithms to detect switching events of electrical equipment, such as power on / off actions. Load feature extraction: Extracting parameters that characterize the features of different electrical equipment from the preprocessed data, such as active power and reactive power. Load feature matching and identification: Matching the extracted load features with a known electrical equipment feature library to identify the type and operating status of each electrical equipment. Decomposition output: Integrating and outputting the energy consumption information of each identified electrical equipment to obtain the decomposition results of each load within the entire system.

[0051] Currently, edge intelligent devices are constantly collecting electricity consumption data from tens of thousands of user nodes, generating billions of data points every day. The complexity of the electricity consumption behavior patterns contained within this data far exceeds what human observation can capture. Therefore, it is necessary to take a data-driven approach to automatically extract high-quality features from massive amounts of electricity consumption data, and ultimately accurately classify and identify various electrical appliances and decompose their power consumption in various real-world scenarios.

[0052] In one embodiment, such as Figure 1 As shown, a load decomposition method is provided, including the following steps:

[0053] 101. Sample the current and voltage of the load to be tested according to the preset sampling frequency to obtain several pairs of current data and voltage data. Calculate several pairs of active power data and reactive power data based on each pair of current data and voltage data. Then, aggregate the several pairs of active power data and reactive power data to obtain an active power sequence and a reactive power sequence.

[0054] 102. Extract the variable frequency air conditioner start-up slope feature from the active power sequence and the reactive power sequence of the sample load, and encode the variable frequency air conditioner start-up slope feature and the other non-variable frequency air conditioner start-up slope features to obtain the encoded sequence of the sample load;

[0055] 103. The active power sequence and reactive power sequence of the sample load are processed by the sliding window method, and the active power sequence and reactive power sequence of the sample load in the window are concatenated with the encoded sequence of the sample load in the same window in the feature dimension to obtain the load features.

[0056] 104. Input the load characteristics into the pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model.

[0057] In the above embodiments, the load to be tested refers to the load of the total power supply to which the variable frequency air conditioner belongs, which is to be detected in this application. The current and voltage of the total power supply are obtained through sampling equipment. The sampling equipment can be a data acquisition device, a programmable AC variable frequency power supply, a dedicated power monitoring instrument, an oscilloscope or signal analyzer with sampling function, an embedded system, or a customized data acquisition module. It is only necessary to ensure that the sampling rate of the equipment meets the preset sampling frequency. Equipment with appropriate accuracy and resolution can be selected first to ensure the accuracy of the data. Consider the interface and communication protocol provided by the equipment to facilitate integration and data exchange with other systems or equipment. Select equipment that is easy to use and maintain, and ensure that it has good reliability and stability. The active power and reactive power of the total power supply are calculated based on the current and voltage, which are the original data containing the variable frequency air conditioner load required for subsequent load decomposition.

[0058] By simultaneously sampling the current and voltage of the load under test at a preset sampling frequency, and calculating the active and reactive power at the same time based on this data, the integrity and consistency of the data are ensured. This comprehensive data acquisition method provides a comprehensive and reliable data foundation for subsequent load decomposition. Simultaneously acquiring current and voltage data and calculating power based on this data avoids measurement errors caused by time differences. This precise data acquisition and calculation method improves the accuracy of power data, making load decomposition based on this data more accurate and reliable.

[0059] In the above embodiment, the variable frequency air conditioner start-up slope feature is the prior knowledge for recognizing the variable frequency air conditioner start-up event. That is, if the variable frequency air conditioner start-up slope feature exists, it can be considered that the variable frequency air conditioner has been started and started running. In order to perform subsequent load decomposition, this moment is distinguished from all other moments and encoded into different feature vectors for comparative analysis.

[0060] By using the identified start-up slope characteristics of the inverter air conditioner as prior knowledge to determine the start-up event of the inverter air conditioner, this prior knowledge can be used in subsequent data analysis to quickly and accurately identify the start-up event of the inverter air conditioner, so that the power data can be clearly divided into the start-up process and the non-start-up process of the inverter air conditioner.

[0061] In the above embodiments, the sliding window method is an algorithm widely used in many fields such as image processing, computer vision, and natural language processing. It involves traversing a given dataset using a fixed-size or variable-size window (also called a sliding window). At each window location, specific operations or calculations are performed, such as classification, detection, and feature extraction. In some algorithms, the sliding window method can reduce time complexity and improve algorithm efficiency. The active power sequence, reactive power sequence, and feature vectors within the same window obtained through the sliding window method constitute the direct data required for decomposition. Feature concatenation refers to directly connecting two or more feature vectors along their feature axes to form a new, longer feature vector. This operation is commonly used in machine learning or deep learning to increase the input feature dimension of the model, potentially improving model performance. In practical applications, this operation can be used to integrate multiple different information sources or feature types. For example, in power load forecasting, in addition to active and reactive power data, there may be other features such as temperature, humidity, and time type. By concatenating these features along the feature dimension, a more comprehensive input can be provided to the model, with the aim of improving the model's predictive performance.

[0062] The sliding window method segments data by continuously moving the window, capturing local changes and trends. Concatenating data from multiple windows can smooth out noise and outliers to some extent, enhancing the robustness of the feature data. This allows the model to maintain stable performance when faced with complex and variable load data. By concatenating active power sequences, reactive power sequences, and time feature vectors within the window, the dimensionality and complexity of load feature data can be greatly enriched. This multi-dimensional feature data better reflects the operating status and changing trends of the load, providing a more comprehensive and accurate information foundation for subsequent load identification, classification, and prediction. Using this concatenated load feature data as input to a deep learning model can significantly improve model performance. This is because the model can learn more useful feature information, thereby more accurately identifying load types, predicting load demand, or optimizing load control strategies.

[0063] In the above embodiments, the stitched data can be input into the variable frequency air conditioner load decomposition model to directly obtain the predicted variable frequency air conditioner active power curve. The output method can be selected from data format, visualization output method, model file output method, specific software format output method, or API interface output method, including but not limited to CSV / Excel, JSON, XML, PMML / ONNX / TensorFlow, charts, images / PDF, BIM model output, other professional software formats, and outputting the model prediction results as a service to external systems through interfaces such as RESTfulAPI or GraphQL.

[0064] The variable frequency air conditioner load decomposition model can output the active power value when the variable frequency air conditioner is running, which makes it easy for users to operate and manage the output results, thereby achieving accurate prediction of the actual active power value when the variable frequency air conditioner is running.

[0065] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the implementation process of this embodiment, another load decomposition method is provided, such as... Figure 2 As shown, the method includes the following steps:

[0066] 201. Sample the current and voltage of the load under test according to the preset sampling frequency to obtain several pairs of current data and voltage data. Calculate several pairs of active power data and reactive power data based on each pair of current data and voltage data.

[0067] In the above embodiments, the preset sampling frequency can be determined by the sampling frequency required in actual applications. For example, the load current and voltage of the main power supply to the variable frequency air conditioner under test can be sampled at a frequency of 10Hz. Figure 3The diagram shows the active and reactive power distribution of a variable frequency air conditioner. Data was collected and preprocessed. After obtaining several current and voltage data points through sampling equipment, the active and reactive power of the corresponding original load were calculated based on each set of current and voltage data. The calculation formula is as follows:

[0068]

[0069]

[0070] Where U represents voltage, I represents current, φ represents the angle between voltage and current, P represents active power, and Q represents reactive power.

[0071] 202. Group all active power data according to the preset grouping number to obtain several active power data groups, extract the minimum value in each active power data group to obtain an active power sequence; group all reactive power data according to the preset grouping number to obtain several reactive power data groups, extract the minimum value in each reactive power data group to obtain a reactive power sequence.

[0072] In the above embodiments, real-world power consumption scenarios are often accompanied by a large amount of data noise, so it is necessary to perform noise reduction preprocessing on the data. In this embodiment, active power data and reactive power data are aggregated to obtain active power sequence and reactive power sequence, thereby reducing the impact of environmental noise. Specifically, this can be achieved by grouping all active power data and all reactive power data according to a preset number of groups. The preset number of groups can be determined according to the needs of actual application. For example, 10 data points per second can be used as the number of groups to form several active power and reactive power data groups. The minimum value in each data group is extracted and aggregated to form active power sequence and reactive power sequence.

[0073] This application uses a data preprocessing method of taking the minimum value and aggregating to remove noise, which can filter out the main noise interference while retaining the main waveform characteristics of the variable frequency air conditioner, ensuring that the subsequent variable frequency air conditioner identification process is more accurate and effective.

[0074] 203. Process the active power sequence and the reactive power sequence using the sliding window method according to the first preset window length and the first preset step size to obtain the first active power sequence and the first reactive power sequence within the first window; calculate the correlation coefficient between the first active power sequence and the first reactive power sequence within the first window; if the correlation coefficient is greater than the preset correlation coefficient threshold, extract the first active power sequence and the first reactive power sequence within the first window as the start-up slope feature of the variable frequency air conditioner.

[0075] In the above embodiments, the start-up slope characteristics of the variable frequency air conditioner are obtained by the sliding window method and correlation coefficient calculation. The active power sequence and reactive power sequence are divided according to the first preset window length and the first preset step size. The preset window length and step size can be determined according to the needs of actual application, for example, a window length of 30s and a step size of 5s are selected. The correlation coefficient between the first active power sequence and the first reactive power sequence within the first window is calculated. The choice of correlation coefficient can be determined according to the needs of actual application, for example, the Pearson correlation coefficient. The value of the Pearson correlation coefficient is between -1 and 1. According to the value range of the Pearson correlation coefficient, the correlation strength between two variables can be judged: 0.8-1.0 indicates extremely strong correlation, 0.6-0.8 indicates strong correlation, 0.4-0.6 indicates moderate correlation, 0.2-0.4 indicates weak correlation, and 0.0-0.2 indicates extremely weak correlation or no correlation.

[0076] The formula for calculating the Pearson correlation coefficient is as follows:

[0077]

[0078] Where Corr is the Pearson correlation coefficient, p is the active power sequence, q is the reactive power sequence, t is time, and T is the window length. The mean of the active power sequence within the window. This represents the mean of the reactive power sequence within the window.

[0079] The presence of a variable frequency air conditioner startup slope within the first window is determined by the Pearson correlation coefficient. The method involves calculating the Pearson correlation coefficient between the active and reactive power sequences within each slid-through first window, and comparing this correlation coefficient with a preset correlation coefficient threshold. This threshold can be determined based on practical application needs; for example, it could be set to 0.9. If the Pearson correlation coefficient between the active and reactive power sequences within a certain first window is greater than 0.9, then a variable frequency air conditioner startup slope is considered to exist within that first window. Figure 4 As shown, a magnified schematic diagram of the slope of the variable frequency air conditioner is provided.

[0080] By combining the sliding window method with Pearson correlation coefficient calculation, a specific pattern of change between active and reactive power generated during the startup of a variable frequency air conditioner can be accurately identified, namely the startup slope characteristic. This method ensures the accuracy and reliability of the detection. The preset first window length, first step length, and Pearson correlation coefficient threshold can all be adjusted according to the needs of actual applications, making this method highly flexible and adaptable to meet the detection requirements of different environments and equipment. Furthermore, this method only requires sequential data of active and reactive power, which is easy to acquire and process.

[0081] 204. Encode the data points corresponding to the first window that have the variable frequency air conditioner start-up slope feature as 1, and encode the data points corresponding to the first window that do not have the variable frequency air conditioner start-up slope feature as -1 to obtain the encoding sequence.

[0082] In the above embodiment, the data points corresponding to the first window exhibiting the variable frequency air conditioner start-up slope characteristic are encoded as 1, and the data points corresponding to the first window without the variable frequency air conditioner start-up slope characteristic are encoded as -1. This is based on the results previously obtained using the sliding window method and Pearson correlation coefficient. Based on these encodings, an encoded sequence of the same length as the power sequence can be obtained.

[0083] 205. The active power sequence and reactive power sequence are processed by the sliding window method according to the second preset window length and the second preset step size to obtain multiple second windows, and a second active power sequence, a second reactive power sequence and a coding sequence corresponding to each second window; the second active power sequence, the second reactive power sequence and the coding sequence corresponding to the second window are spliced ​​together to obtain multiple load characteristics corresponding to multiple second windows.

[0084] In the above embodiments, the active power sequence and reactive power sequence are processed using a sliding window method according to a second preset window length and a second preset step size. The preset second window length and second step size can be determined according to the needs of actual applications, such as a second window length of 2 minutes and a second step size of 30 seconds, thus obtaining a coding sequence length of 1200 corresponding to the second window. When obtaining the second active power sequence and the second reactive power sequence within the second window, the coding sequence within the second window can be obtained according to the correspondence between the active power sequence, the reactive power sequence and the coding sequence. For each second window, the second active power sequence, the second reactive power sequence and the coding sequence within the second window are concatenated along the feature dimension to obtain the load characteristics of each second window. Multiple load characteristics are obtained based on multiple second window load characteristics. These multiple load characteristics are input into the variable frequency air conditioner load decomposition model to obtain the load decomposition result of the variable frequency air conditioner, i.e., the active power value of the variable frequency air conditioner. The active power value of the variable frequency air conditioner can be output in the form of a chart, for example, the active power curve corresponding to the active power value of the variable frequency air conditioner can be output.

[0085] By setting a second preset window length and a second preset step size, the sliding window method can flexibly capture the dynamic changes in load data. Especially for frequently adjusted load equipment such as variable frequency air conditioners, the sliding window method can more accurately reflect power fluctuations. The application of the sliding window method makes data processing more efficient. By using a sliding window, the computational load can be reduced and processing speed increased without losing important information. The second active power sequence, the second reactive power sequence, and the coded sequence within the second window are concatenated to form the load characteristics of each second window. This feature extraction method not only preserves the integrity of the original data but also provides more contextual information to the load decomposition model through the introduction of the coded sequence, helping to improve the accuracy of load decomposition.

[0086] 206. Input the load characteristics into the pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model.

[0087] like Figure 5 As shown in the diagram, a schematic diagram of the load decomposition results of the variable frequency air conditioner load decomposition model is provided. The information in the diagram shows that the predicted data output by the variable frequency air conditioner load decomposition model of this application is almost identical to the actual active power value during the actual operation of the variable frequency air conditioner. This high degree of consistency is not only reflected in the power fluctuations during the stable operation of the variable frequency air conditioner, but also in the critical stages of load changes during the start-up and shutdown of the variable frequency air conditioner. The model can accurately capture the sharp rise and fall of power, which is completely consistent with the actual situation. Therefore, the prediction results of the model can be considered good.

[0088] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the implementation process of this embodiment, a training method for a variable frequency air conditioning load decomposition model is provided. This method includes the following steps:

[0089] Multiple sample loads are acquired. For each sample load, its current and voltage are sampled at a preset sampling frequency to obtain several pairs of current and voltage data. Based on each pair of current and voltage data, several pairs of active and reactive power data are calculated for the sample load. These pairs of active and reactive power data are then aggregated to obtain the active and reactive power sequences of the sample load. The variable frequency air conditioner start-up slope feature is extracted from the active and reactive power sequences of the sample load. This feature is then encoded along with the start-up slope features of other non-variable frequency air conditioners to obtain the encoded sequence of the sample load. The active and reactive power sequences of the sample load are then processed using a sliding window method. The sequence involves concatenating the active power sequence and reactive power sequence of the sample load within a window with the encoded sequence of the sample load within the same window along the feature dimension to obtain the load features of the sample load. Based on the load features of each sample load and the sample label corresponding to each load feature, a deep learning model is trained to obtain a variable frequency air conditioner load decomposition model, wherein the sample label is the active power value of the variable frequency air conditioner. The method for obtaining the sample label includes: simultaneously sampling the current and voltage of the sample load at a preset sampling frequency, and collecting the active power value of the variable frequency air conditioner; assigning sample labels to the load features of the sample load according to the correspondence between the active power sequence and reactive power sequence of the sample load and the active power value of the variable frequency air conditioner.

[0090] In the above embodiments, the modeling method of the provided load decomposition method uses the actual active power value generated by the variable frequency air conditioner during actual operation as the basic data of the model. First, the acquired model input data is merged with the corresponding label data to form a complete training dataset. The label data contains active power values, which are obtained manually. While collecting the original load data for training, the operating data of the variable frequency air conditioner is also collected. Within a specified time period, the actual active power value of the variable frequency air conditioner during operation is recorded, and this value is used as a label, which is associated with the load characteristics used for training in the corresponding time period. Subsequently, the model is trained using the training dataset to obtain a comprehensive and accurate variable frequency air conditioner load decomposition model.

[0091] The core feature of this model is that it combines frequency domain analysis with adaptive graph convolution technology. By combining these two technologies, the model can accurately capture the complex dependencies between multivariate time series. In particular, considering the characteristics of variable frequency air conditioners running for a long time and with complex power changes, this design enables the model to accurately predict the actual active power value of variable frequency air conditioners during load operation.

[0092] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, and in order to fully illustrate the implementation process of this embodiment, as follows: Figure 6 As shown, a schematic diagram of the variable frequency air conditioning load decomposition model is provided. This application selects MSGNet as the basic architecture of the variable frequency air conditioning load decomposition model. MSGNet is an advanced deep learning model designed to handle multivariate time series forecasting problems. MSGNet effectively extracts significant periodic patterns using frequency domain analysis and decomposes the time series into different time scales, enabling the model to capture the inter-series correlations more meticulously at different time scales. The model introduces adaptive graph convolutional layers to autonomously learn diverse inter-series correlations within each time scale. This adaptability allows MSGNet to adapt to different data distributions and patterns. MSGNet employs a self-attention mechanism to capture intra-series dependencies, further enhancing the model's ability to understand time series data. MSGNet has the ability to automatically learn interpretable multi-scale inter-series correlations and exhibits strong generalization ability even when applied to out-of-distribution samples.

[0093] Furthermore, as Figure 1 , Figure 2 The specific implementation of the method shown in this embodiment provides a load decomposition device, such as... Figure 7 As shown, the device includes: a data acquisition module 71, an encoding module 72, a load characteristic generation module 73, and a load decomposition module 74.

[0094] The data acquisition module 71 is used to sample the current and voltage of the load under test according to a preset sampling frequency, obtain several pairs of current data and voltage data, calculate several pairs of active power data and reactive power data based on each pair of current data and voltage data, and perform aggregation processing on the several pairs of active power data and reactive power data to obtain active power sequence and reactive power sequence.

[0095] The encoding module 72 is used to extract the variable frequency air conditioner start-up slope feature from the active power sequence and the reactive power sequence of the sample load, and encode the variable frequency air conditioner start-up slope feature and the other non-variable frequency air conditioner start-up slope features to obtain the encoded sequence of the sample load.

[0096] The load feature generation module 73 is used to process the active power sequence and the reactive power sequence of the sample load using the sliding window method, and to concatenate the active power sequence and reactive power sequence of the sample load in the window with the encoded sequence of the sample load in the same window in the feature dimension to obtain the load feature.

[0097] The load decomposition module 74 is used to input the load characteristics into the pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model.

[0098] In specific application scenarios, in the step of aggregating the active power data and reactive power data using logarithmic pairs to obtain active power sequences and reactive power sequences, the data acquisition module 71 can be specifically used to: group all active power data according to a preset number of groups to obtain several active power data groups, extract the minimum value in each active power data group, and obtain an active power sequence; group all reactive power data according to the preset number of groups to obtain several reactive power data groups, extract the minimum value in each reactive power data group, and obtain a reactive power sequence.

[0099] In a specific application scenario, in the step of extracting the start-up slope feature of the variable frequency air conditioner from the active power sequence and the reactive power sequence, the encoding module 72 can be specifically used to: process the active power sequence and the reactive power sequence according to a first preset window length and a first preset step size using a sliding window method to obtain a first active power sequence and a first reactive power sequence within a first window; calculate the correlation coefficient between the first active power sequence and the first reactive power sequence within the first window; if the correlation coefficient is greater than a preset correlation coefficient threshold, extract the first active power sequence and the first reactive power sequence within the first window as the start-up slope feature of the variable frequency air conditioner.

[0100] In a specific application scenario, in the step of encoding the start-up slope feature of the variable frequency air conditioner and the start-up slope feature of other non-variable frequency air conditioners, the encoding module 72 can be specifically used to: encode the data point corresponding to the first window that has the start-up slope feature of the variable frequency air conditioner as 1, and encode the data point corresponding to the first window that does not have the start-up slope feature of the variable frequency air conditioner as -1, to obtain an encoding sequence.

[0101] In a specific application scenario, in the step of processing active power sequence and reactive power sequence using the sliding window method, the load feature generation module 73 can be specifically used to: process the active power sequence and reactive power sequence according to the second preset window length and the second preset step size using the sliding window method to obtain multiple second windows, and a second active power sequence, a second reactive power sequence and a coding sequence corresponding to each second window; and concatenate the second active power sequence, the second reactive power sequence and the coding sequence corresponding to the second window to obtain multiple load features corresponding to multiple second windows.

[0102] In specific application scenarios, the load decomposition device provided in this application further includes a model training module. In training the variable frequency air conditioner load decomposition model, the model training module can specifically be used for:

[0103] Multiple sample loads are acquired. For each sample load, the current and voltage are sampled at a preset sampling frequency to obtain several pairs of current and voltage data. Based on each pair of current and voltage data, several pairs of active and reactive power data are calculated for the sample load. The active and reactive power data are then aggregated to obtain the active power sequence and reactive power sequence of the sample load. The variable frequency air conditioner start-up slope characteristic is extracted from the active power sequence and reactive power sequence of the sample load. This variable frequency air conditioner start-up slope characteristic is compared with other non-variable frequency air conditioner start-up slope characteristics. The variable frequency air conditioner starts with a gradual slope feature and is encoded to obtain the encoded sequence of the sample load. The active power sequence and reactive power sequence of the sample load are processed by the sliding window method. The active power sequence and reactive power sequence of the sample load within the window are concatenated with the encoded sequence of the sample load within the same window in the feature dimension to obtain the load feature of the sample load. Based on the load feature of each sample load and the sample label corresponding to each load feature, a deep learning model is trained to obtain the variable frequency air conditioner load decomposition model, wherein the sample label is the active power value of the variable frequency air conditioner.

[0104] The method for obtaining the sample tags is to collect the active power value of the variable frequency air conditioner while sampling the current and voltage of the sample load at a preset sampling frequency; and to assign sample tags to the load characteristics of the sample load based on the correspondence between the active power sequence and reactive power sequence of the sample load and the active power value of the variable frequency air conditioner.

[0105] Based on the above, Figure 1 , Figure 2 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 1 , Figure 2 The load decomposition method is shown.

[0106] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), including several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the methods described in the various implementation scenarios of this application.

[0107] Based on the above, Figure 1 , Figure 2The method shown, and Figure 3 , Figure 4 and Figure 5 The variable frequency air conditioner load splitting device embodiment shown above, in order to achieve the above objective, is as follows: Figure 8 As shown, this embodiment also provides a computer device for decomposing variable frequency air conditioning loads. Specifically, it can be a personal computer, server, smartphone, tablet, smartwatch, or other network device. This computer device includes a storage medium and a processor; the storage medium stores computer programs and an operating system; the processor executes the computer programs to achieve the above-described... Figures 1 to 5 The method shown.

[0108] Optionally, the computer device may also include internal memory, a communication interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, a display screen, and input devices such as a keyboard. The communication interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0109] Those skilled in the art will understand that the computer equipment structure for variable frequency air conditioning load decomposition provided in this embodiment does not constitute a limitation on the computer equipment, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0110] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the aforementioned computer hardware and the software resources to be identified, supporting the operation of information processing programs and other software and / or programs to be identified. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing computer device.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0112] This application provides a load decomposition method, apparatus, storage medium, and computer equipment. The method includes sampling the current and voltage of the load to be tested at a preset sampling frequency to obtain several pairs of current and voltage data. Based on each pair of current and voltage data, several pairs of active power and reactive power data are calculated, and the active power and reactive power data are aggregated to obtain an active power sequence and a reactive power sequence. A variable frequency air conditioner start-up slope feature is extracted from the active power and reactive power sequences of the sample load. This variable frequency air conditioner start-up slope feature is encoded along with other non-variable frequency air conditioner start-up slope features to obtain an encoded sequence of the sample load. The active power and reactive power sequences of the sample load are processed using a sliding window method. The active power and reactive power sequences of the sample load within a window are concatenated with the encoded sequences of the sample load within the same window in a feature dimension to obtain load features. The load features are input into a pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model. This application achieves accurate decomposition of the load power of variable frequency air conditioners by precisely capturing and analyzing their electrical characteristics. It collects a large number of samples with labels of the actual active power of variable frequency air conditioners to train the load decomposition model. By inputting the active power sequence, reactive power sequence and coded sequence into the model, the active power of variable frequency air conditioners can be directly obtained, which enhances the generalization ability of the algorithm and enables it to stably decompose the actual power consumption of variable frequency air conditioners in different environments.

[0113] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0114] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A load decomposition method, characterized in that, The method includes: The current and voltage of the load under test are sampled according to a preset sampling frequency to obtain several pairs of current and voltage data. Several pairs of active power and reactive power data are calculated based on each pair of current and voltage data. The active power and reactive power data are then aggregated to obtain an active power sequence and a reactive power sequence. The variable frequency air conditioner start-up slope feature is extracted from the active power sequence and the reactive power sequence. The variable frequency air conditioner start-up slope feature and the other non-variable frequency air conditioner start-up slope features are encoded to obtain the encoded sequence of the sample load. The active power sequence and reactive power sequence of the sample load are processed by the sliding window method. The active power sequence and reactive power sequence of the sample load within the window are concatenated with the encoded sequence of the sample load within the same window in the feature dimension to obtain the load features. The load characteristics are input into a pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model. The training method of the variable frequency air conditioner load decomposition model includes: Obtain multiple sample loads; For each sample load, the current and voltage of the sample load are sampled according to a preset sampling frequency to obtain several pairs of current data and voltage data. Based on each pair of current data and voltage data, several pairs of active power data and reactive power data of the sample load are calculated. The several pairs of active power data and reactive power data are aggregated to obtain the active power sequence and reactive power sequence of the sample load. The variable frequency air conditioner start-up slope feature is extracted from the active power sequence and the reactive power sequence of the sample load. The variable frequency air conditioner start-up slope feature is encoded together with the other non-variable frequency air conditioner start-up slope features to obtain the encoded sequence of the sample load. The active power sequence and reactive power sequence of the sample load are processed by the sliding window method. The active power sequence and reactive power sequence of the sample load within the window are concatenated with the encoded sequence of the sample load within the same window in the feature dimension to obtain the load characteristics of the sample load. Based on the load characteristics of each sample load and the sample label corresponding to each load characteristic, a deep learning model is trained to obtain a variable frequency air conditioner load decomposition model, wherein the sample label is the active power value of the variable frequency air conditioner.

2. The method according to claim 1, characterized in that, The logarithmic algorithm aggregates the active power data and the reactive power data to obtain an active power sequence and a reactive power sequence, including: All active power data are grouped according to a preset number of groups to obtain several active power data groups. The minimum value in each active power data group is extracted to obtain the active power sequence. All reactive power data are grouped according to the preset number of groups to obtain several reactive power data groups. The minimum value in each reactive power data group is extracted to obtain the reactive power sequence.

3. The method according to claim 2, characterized in that, The extraction of the variable frequency air conditioner start-up slope characteristics from the active power sequence and the reactive power sequence includes: The active power sequence and the reactive power sequence are processed by the sliding window method according to the first preset window length and the first preset step size to obtain the first active power sequence and the first reactive power sequence within the first window. Calculate the correlation coefficient between the first active power sequence and the first reactive power sequence within the first window. If the correlation coefficient is greater than a preset correlation coefficient threshold, extract the first active power sequence and the first reactive power sequence within the first window as the start-up slope feature of the variable frequency air conditioner.

4. The method according to claim 3, characterized in that, The encoding of the variable frequency air conditioner's start-up slope characteristic and the other non-variable frequency air conditioner's start-up slope characteristics yields an encoding sequence, including: The data points corresponding to the first window that exhibit the variable frequency air conditioner start-up slope feature are encoded as 1, and the data points corresponding to the first window that do not exhibit the variable frequency air conditioner start-up slope feature are encoded as -1, thus obtaining the encoding sequence.

5. The method according to claim 4, characterized in that, The process of processing active and reactive power sequences using the sliding window method, concatenating the active and reactive power sequences within a window with the coded sequences within the same window along the feature dimension to obtain load features, includes: The active power sequence and reactive power sequence are processed by the sliding window method according to the second preset window length and the second preset step size to obtain multiple second windows, as well as the second active power sequence, the second reactive power sequence and the coding sequence corresponding to each second window; The second active power sequence, the second reactive power sequence, and the encoded sequence corresponding to the second window are concatenated to obtain multiple load characteristics corresponding to multiple second windows.

6. The method according to claim 1, characterized in that, The method for obtaining the sample labels includes: While sampling the current and voltage of the sample load at a preset sampling frequency, the active power value of the variable frequency air conditioner is also collected. Based on the correspondence between the active power sequence and reactive power sequence of the sample load and the active power value of the variable frequency air conditioner, the load characteristics of the sample load are assigned sample labels.

7. A load-distribution device, characterized in that, The device includes: The data acquisition module is used to sample the current and voltage of the load under test according to a preset sampling frequency, obtain several pairs of current data and voltage data, calculate several pairs of active power data and reactive power data based on each pair of current data and voltage data, and aggregate the several pairs of active power data and reactive power data to obtain active power sequence and reactive power sequence. The encoding module is used to extract the variable frequency air conditioner start-up slope feature from the active power sequence and the reactive power sequence, and encode the variable frequency air conditioner start-up slope feature and the other non-variable frequency air conditioner start-up slope features to obtain the encoded sequence of the sample load. The load feature generation module is used to process the active power sequence and reactive power sequence of the sample load using a sliding window method, and to concatenate the active power sequence and reactive power sequence of the sample load within the window with the encoded sequence of the sample load within the same window in the feature dimension to obtain the load feature. The load decomposition module is used to input the load characteristics into the pre-trained variable frequency air conditioner load decomposition model to obtain the active power value of the variable frequency air conditioner output by the variable frequency air conditioner load decomposition model. The model training module is used to acquire multiple sample loads. For each sample load, the current and voltage of the sample load are sampled at a preset sampling frequency to obtain several pairs of current and voltage data. Based on each pair of current and voltage data, several pairs of active power and reactive power data of the sample load are calculated. The active power and reactive power data pairs are then aggregated to obtain the active power sequence and reactive power sequence of the sample load. The variable frequency air conditioner start-up slope characteristics are extracted from the active power sequence and reactive power sequence of the sample load. The sample load is encoded by combining the start-up slope features of other non-inverter air conditioners. The active power sequence and reactive power sequence of the sample load are then processed using a sliding window method. The active power sequence and reactive power sequence of the sample load within the window are concatenated with the encoded sequence of the sample load within the same window along the feature dimension to obtain the load features of the sample load. Based on the load features of each sample load and the sample label corresponding to each load feature, a deep learning model is trained to obtain the inverter air conditioner load decomposition model, where the sample label is the active power value of the inverter air conditioner.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images