Equipment energy consumption state monitoring method based on electric signals of multi-energy complementary energy supply system

Abstract

The invention provides an equipment energy consumption state monitoring method based on electric signals of a multi-energy complementary energy supply system, and the method comprises the steps: obtaining time sequence electric signals of all energy consumption equipment, carrying out the preprocessing, extracting multi-dimensional features, constructing feature vectors, and carrying out the collection to obtain an original sample set; disturbance enhancement is applied to the original sample in the known state, and an abnormal behavior agent sample set is generated; the method comprises the following steps: constructing an Encoder-Classifier architecture model on the basis of a two-stage model training mechanism; acquiring electric signals in real time, outputting the electric signals according to a fixed window, preprocessing and extracting features, and inputting the features into the model to obtain monitoring results. According to the method, the electric signal time sequence characteristic and the equipment operation rule are combined, the abnormal behavior agent sample can be automatically generated without an abnormal label, and a complex equipment scene is adapted; through double-stage training of data enhancement and similarity learning, the model recognizes the state and abnormity of equipment, and intelligent support is provided for safe operation and energy efficiency improvement of a multi-energy system.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology, and specifically to a method for monitoring the energy consumption status of equipment based on electrical signals from a multi-energy complementary power supply system. Background Technology

[0002] With the continuous advancement of carbon reduction goals and the accelerated pace of energy transition, multi-energy complementary energy supply systems are gradually being widely used in industrial parks, public buildings, and large commercial complexes. In these systems, key energy-consuming equipment, such as electric boilers, heat pumps, refrigeration units, and energy storage devices, exhibit complex and variable operating characteristics under different conditions. Their energy efficiency, operating status, and safety have a significant impact on the stable operation and energy utilization efficiency of the entire system. Therefore, achieving status identification and abnormal energy consumption behavior monitoring of key energy-consuming equipment has become a crucial supporting technology for improving the intelligent operation level of multi-energy systems.

[0003] However, existing technologies for monitoring energy-consuming equipment have significant drawbacks. Due to the diverse types of equipment in multi-energy systems, the operating mechanisms and changing patterns of different devices vary considerably, making it difficult for traditional identification methods that rely on specific sensors to collect and analyze data to be fully adaptable. Furthermore, the increasing number of monitoring devices also leads to higher costs in building the operation and control system. Moreover, based on practical needs, different analysis models must be built for different devices, and the significant increase in monitoring data will seriously affect the system's operational response speed. Furthermore, the need to add a large number of monitoring devices for each energy-consuming device will significantly increase the probability of false alarms, thereby affecting the overall stable operation efficiency of the system.

[0004] In summary, a new method for monitoring energy-consuming equipment is needed. This method should be based on existing hardware infrastructure, using data monitoring combined with intelligent analysis to identify the energy consumption status of equipment and separate abnormal energy consumption situations. It should be applicable to complex and diverse equipment scenarios, thereby achieving cost-effective and efficiency-enhancing intelligent monitoring. Summary of the Invention

[0005] This invention addresses the problems existing in the prior art by providing a method for key equipment status identification and abnormal behavior detection that combines data augmentation and similarity learning mechanisms, providing intelligent support for the safe operation and energy efficiency improvement of multi-energy systems.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a method for monitoring the energy consumption status of equipment based on electrical signals of a multi-energy complementary power supply system, comprising the following steps: Acquire the timing electrical signals of each energy-consuming device in the energy supply system; The timing electrical signal is preprocessed to obtain a preprocessed electrical signal; Multidimensional features are extracted from the preprocessed electrical signal, and feature vectors of the energy-consuming devices are constructed based on the multidimensional features; the feature vectors of each energy-consuming device are combined to obtain the original sample set. Perturbation enhancement operations are applied to multiple original samples with known energy consumption states to generate a proxy sample set of abnormal behavior; A device energy consumption status monitoring model is constructed based on a two-stage model training mechanism; the device energy consumption status monitoring model is an Encoder-Classifier architecture model, which includes a shared feature extraction network and a linear classifier. The system collects electrical signals from various energy-consuming devices in the power supply system in real time and outputs real-time timing electrical signals based on a fixed window. The real-time time-series electrical signal is preprocessed and its features are extracted before being input into the equipment energy status monitoring model to obtain monitoring results.

[0007] Optionally, the two-stage model training includes the following steps: The first stage involves closed-set supervised classification training based on the original sample set. In the second stage, a training pair is constructed by combining the original sample set and the abnormal behavior proxy sample set, and a similarity supervision loss is introduced based on the training pair to perform joint modeling to obtain the equipment energy status monitoring model.

[0008] Optionally, in the first stage, the optimization objective is a classification loss function.

[0009] Optionally, in the first stage, the loss weight coefficient is set to 0; After the first stage of training is completed, the loss weight coefficient is adjusted to 1 for use in the second stage of training.

[0010] Optionally, the multidimensional features include time-domain features, frequency-domain features, and statistical features.

[0011] Optionally, the perturbation enhancement operation includes at least one of amplitude perturbation, time-scale stretching and compression, Gaussian noise perturbation, frequency domain coefficient perturbation, or random masking of characteristic channels.

[0012] Optionally, the following steps may also be included: The monitoring results include normal behavior and abnormal behavior; When the monitoring result indicates normal behavior, output the device status category; When the monitoring result indicates abnormal behavior, an alarm mechanism is triggered, and the probability of abnormal behavior and the corresponding abnormal behavior label are output.

[0013] Optionally, obtaining the monitoring results further includes the following steps: Calculate the sum of class confidence scores for the monitored samples in the known categories, and calculate the anomaly confidence score based on the sum of class confidence scores; then compare the anomaly confidence score with a preset threshold. When the anomaly confidence score is higher than the preset threshold, the category of the monitored sample as abnormal behavior is output. If the anomaly confidence score is not higher than the preset threshold, the monitored sample is determined to be a normal behavior category.

[0014] Optionally, the timing electrical signal includes at least one of current, voltage, active power, or reactive power.

[0015] Optionally, the preprocessing of the time-series electrical signal includes at least one of denoising, outlier removal, or normalization.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention enables the monitoring of the status of various energy-consuming devices through electrical signal monitoring and analysis, providing a novel status monitoring method. Furthermore, by combining the timing characteristics of energy-consuming electrical signals with the operating rules of the equipment, it achieves the ability to automatically generate abnormal behavior proxy samples after disturbance enhancement, without relying on abnormal labels, and is applicable to status identification and abnormal monitoring methods in complex and diverse equipment scenarios. Moreover, by combining data augmentation and similarity learning mechanisms for two-stage model training, the model can identify equipment status and abnormal behavior, providing intelligent support for the safe operation and energy efficiency improvement of multi-energy systems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the original data of voltage, current and active power of the load event segment in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the voltage, current, and reactive current curves of the energy-saving lamp device in Embodiment 1 of the present invention; Figure 4 This is a statistical chart of recognition scores for each device type in Embodiment 1 of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0021] It is worth noting that, unless otherwise specified, the methods used in this invention are all conventional methods; and the raw materials and equipment used are all conventional commercially available products, and their sources are not specifically limited.

[0022] It should also be noted that, for ease of understanding, the method steps in the specific embodiments of the present invention are described in a certain order, but those skilled in the art can change the order of the steps according to actual needs, so this should not be used as a limiting condition; further, in the description of the following specific embodiments, the superscripts and subscripts of each parameter should be understood as distinguishing marks of similar identifiers in accordance with common interpretations, representing the parameters of the related or corresponding devices, and should not be understood as specific models or special marks.

[0023] Please see Figure 1 As shown, this application provides a method for monitoring the energy consumption status of equipment based on electrical signals of a multi-energy complementary power supply system, which includes the following steps: First, the timing electrical signals of each energy-consuming device in the energy supply system are acquired. Optionally, this application collects electrical signals such as current, voltage, active power, and reactive power of each energy-consuming device during operation through smart meters and / or sensors, and combines them with sampling time-marked electrical signal data to obtain timing electrical signals.

[0024] Timing signals can fully capture the operating characteristics of equipment under different conditions such as startup, stable operation, and load switching, and are the core data foundation for subsequent status analysis and anomaly identification. To ensure the signal coverage of the details of equipment operation, the sampling frequency is set to 1Hz or higher to ensure that no characteristic information of key operating states is missed.

[0025] The acquired time-series electrical signals are preprocessed. Optionally, signal preprocessing includes at least one of denoising, outlier removal, or normalization. This application specifically uses a combination of the above three methods for preprocessing. Denoising is used to eliminate interference from environmental electromagnetic noise and sensor thermal noise during signal transmission, specifically including wavelet transform denoising or moving average denoising. Outlier removal is used to remove isolated outlier data points caused by instantaneous sensor failures or sudden line interference, achieved through the 3σ criterion or box plot method. Normalization maps electrical signal indicators of different magnitudes, such as current in amperes and power in kilowatts, to the same data range of [0,1] or [-1,1], avoiding excessive bias of the model towards large-scale features due to numerical differences, which would affect the training effect.

[0026] After the above preprocessing, a stable preprocessed electrical signal that can be directly used for feature extraction can be obtained.

[0027] Multidimensional features are extracted based on preprocessed electrical signals. These multidimensional features include time-domain features, frequency-domain features, and statistical features. Time-domain features reflect the signal's variation over time, such as peak value, valley value, mean, variance, and waveform factor. Frequency-domain features are extracted after converting the time-domain signal to the frequency domain using Fourier transform or wavelet transform, reflecting the signal's energy distribution characteristics in different frequency bands, such as dominant frequency amplitude and band energy proportion. Statistical features are a quantitative description of the overall statistical regularity of the signal, such as skewness (reflecting the degree of asymmetry in signal distribution), kurtosis (reflecting the steepness of signal distribution), and kurtosis. The extracted multidimensional features are combined according to device type and signal category to construct feature vectors for each energy-consuming device. The construction process can be represented as follows: ; In the formula, For the first i Feature vectors of energy-consuming equipment in Taiwan For the equipment k Statistical representations of time-series characteristics, such as the mean of current signals and the amplitude of the dominant frequency of voltage signals; n This represents the total number of feature dimensions.

[0028] The feature vectors of all power-consuming devices are collected to form the original sample set for model training.

[0029] To address the problem that abnormal energy consumption behaviors in real-world scenarios are sudden and non-repetitive, leading to a scarcity of abnormal samples and difficulty in learning abnormal features by the model, a perturbation enhancement operation is applied to the original samples of multiple known energy consumption states to generate a proxy sample set of abnormal behaviors, which is a pseudo-unseen class sample used to represent "unknown" behaviors.

[0030] Furthermore, the perturbation enhancement operation of this application includes at least one of the following: amplitude perturbation, time-scale stretching and compression, Gaussian noise perturbation, frequency domain coefficient perturbation, and random occlusion of feature channels; the specific implementation is as follows: Amplitude perturbation randomly multiplies the signal amplitude of the original sample by an amplitude factor. This is to simulate instantaneous fluctuations in device voltage or current; Time-scale stretching and compression stretches or compresses the signal's time axis by a ratio of 0.9 to 1.1 to simulate instantaneous changes in the equipment's operating speed, such as slight fluctuations in motor speed. Gaussian noise perturbation adds a noise term to the original signal. If it follows a Gaussian distribution with a mean of 0 and a variance of 0.01 to 0.05, it can simulate electromagnetic interference in actual operation. Frequency domain coefficient perturbation randomly fine-tunes the frequency domain coefficients of the signal after Fourier transform (the adjustment range does not exceed 10% of the original coefficients) to simulate the frequency characteristic changes caused by unstable operating conditions of the simulated equipment. Feature channel random occlusion randomly sets 10% to 20% of the feature dimensions in the feature vector to zero to simulate a scenario where local sensor data is temporarily missing.

[0031] All perturbation operations ensure that the generated samples have significant differences from the original class characteristics, but conform to the physical operating laws of the equipment, such as the current amplitude not being negative and the power not exceeding the rated range of the equipment.

[0032] Furthermore, this application uses amplitude perturbation and Gaussian perturbation as examples to generate abnormal behavior proxy samples. It can be represented as: ; In the formula, This is an operation to enhance the disturbance.

[0033] The set of abnormal behavior proxy samples constitutes the abnormal behavior proxy sample set required for training. Abnormal behavior is usually characterized by suddenness, non-repetition, and unpredictability. The lack of sufficient labeled abnormal data leads to the insufficient generalization ability of conventional supervised learning methods when identifying unknown behaviors. The above operation in this application can solve the problem of scarce abnormal samples, that is, the difficulty in building an effective supervised model.

[0034] A device energy consumption status monitoring model is constructed based on a two-stage model training mechanism. This model employs an Encoder-Classifier architecture, comprising a shared feature extraction network (Encoder()) and a linear classifier (Classifier()). The model output dimension is... ,in, The number of categories with known device states. The number of pre-assigned abnormal behavior proxy sample categories, i.e. the number of pseudo-unseen categories, is usually set to 3 to 5 categories based on the device type and the number of historical abnormal types.

[0035] (1) First stage; The first stage of training is closed-set supervised classification training based on the original sample set. The core objective is to enable the model to learn the feature distribution of known device states, and the optimization objective is the cross-entropy loss function, calculated as follows: ; In the formula, For cross-entropy loss, N This represents the number of original samples for the first stage of training. For the first i The true labels of each original sample are uniquely coded, such as "refrigeration unit operating normally" corresponding to... =[1,0,...,0], Dimensions and Number of Known Classes Consistent For the model to the first i The predicted probability distribution of each original sample is output by the linear classifier Classifier(), where each dimension corresponds to the predicted probability of a known class. .

[0036] To ensure that the first-stage model focuses solely on classifying known classes, this application sets the weight coefficient λ of the subsequent similarity supervision loss to 0, optimizing only the cross-entropy loss. Through this training stage, the model can initially acquire the ability to accurately classify known device states. Training typically ends when the cross-entropy loss tends to stabilize on the validation set, such as when the loss decreases by less than 0.001 over five consecutive epochs.

[0037] After the first phase of training is completed, the loss weight coefficient λ is adjusted to 1, and the second phase of training begins.

[0038] (2) Second stage; The core of the second stage is to enable the model to learn the feature differences between known class samples and abnormal behavior proxy samples, and to construct training pairs with similarity by combining the original sample set and the abnormal behavior proxy sample set. Furthermore, a similarity-supervised loss is introduced for joint modeling.

[0039] Among them, constructing the joint training set is ; Furthermore, define any two samples , ( The similarity of labels to () is calculated using the following formula: ; In the formula, For the samplei With sample j Label similarity, for example: when two samples both belong to "normal cooling of heat pump" (known class) or both belong to "proxies generated by voltage amplitude perturbation" (abnormal behavior proxy class), Conversely, it is 0.

[0040] Define the model's feature encoding and similarity supervision loss for samples: Samples , The feature codes output by the shared feature extraction network Encoder() are as follows: , The cosine similarity between the two is This reflects the degree of similarity between two samples in the feature space. The similarity supervision loss uses a binary similarity supervision loss, the formula of which is: ; In the formula, For similarity supervision loss, M Let P be the total number of training pairs constructed in the second stage, and P be the set of training pairs. The final joint loss function is: ; In the formula, L For the total loss, Cross-entropy loss; during training, set , After initial convergence, switch to .

[0041] Through phased training, the model retains its ability to classify known categories while learning the feature boundaries between known classes and abnormal behavior proxy samples, laying the foundation for subsequent abnormal behavior detection.

[0042] Because real-world scenarios are open-ended, monitoring methods often suffer from the "open-category problem." During industrial operations, equipment states or abnormal behaviors unseen during model training, such as soft faults, misoperations, and electrical interference, frequently occur. These fall under the category of Open Set Recognition (OSR), and traditional classification models tend to misclassify such unknown categories as known ones. Furthermore, a range of monitoring technologies, including traditional classifiers, anomaly detection models, and clustering methods, such as equipment anomaly detection methods based on Support Vector Machines (SVM), Isolation Forests, and Autoencoders, generally suffer from weak ability to identify unknown categories, poor model transferability, and strong dependence on specific equipment, making it difficult to meet the intelligent monitoring needs of multi-functional systems across equipment and operating conditions. This application, however, utilizes the aforementioned two-stage training method and open-set recognition technology, employing mechanisms such as discriminative boundary expansion, contrastive learning, and adversarial example generation to detect unknown categories, demonstrating robustness in open environments.

[0043] In practical monitoring scenarios, electrical signals from various energy-consuming devices in the energy supply system are acquired in real time, and real-time time-series electrical signals are output based on a fixed time window. The window length is set to 30-60 seconds to ensure complete coverage of the characteristics of one minimum operating cycle of the device; the sliding step size is set to 10 seconds to balance real-time performance and data redundancy, avoiding monitoring delays. The real-time time-series electrical signals undergo the above preprocessing steps (denoising, outlier removal, and normalization), and the corresponding time-domain features, frequency-domain features, and statistical features are extracted. The extracted features are then organized according to the feature dimensions and order of the original sample set and input into the trained device energy consumption status monitoring model to obtain the predicted probability distribution output by the model.

[0044] The determination of monitoring results is achieved by calculating the sum of category confidence scores and monitoring the anomaly confidence scores of the samples. Firstly, in calculating the sum of category confidence scores, for any sample... Calculate its classification probability: ; ; In the formula, c The category for known device status.

[0045] Next, calculate the anomaly confidence score. The formula is: ; The anomaly confidence score reflects the probability that a sample exhibits abnormal behavior. The anomaly confidence score is then compared with a preset threshold. τ Comparison; where the threshold τBy selecting the validation set and AUROC curve, the threshold can be determined. τ Optimization is performed, specifically by comparing different τ The abnormal sample identification rate and normal sample accuracy on the validation set are selected when the AUROC value is maximized. τ The value is generally between 0.6 and 0.8.

[0046] when (Abnormal energy consumption behavior) indicates that the sample The corresponding energy-consuming equipment exhibits abnormal behavior, and the category of the abnormal behavior is output; when When the monitored sample is determined to be in normal behavior, the category of normal behavior is output, and the known category with the highest predicted probability is selected as the equipment status category.

[0047] For monitoring results that are determined to be normal behavior, the status category of the equipment is directly output to provide data support for energy efficiency management of the power supply system, such as load scheduling and energy consumption optimization. For monitoring results identified as abnormal behavior, an alarm mechanism is immediately triggered, such as system pop-up notifications, audible and visual alarms, and remote push notifications to maintenance personnel. At the same time, the probability of abnormal behavior, i.e., the abnormal confidence score, and the corresponding abnormal behavior label are output, i.e., the abnormal behavior proxy sample category most similar to the monitored sample features. Additionally, the feature space similarity between the abnormal behavior and the most recently known class is also output, i.e., the cosine similarity between the feature code of the abnormal sample and the feature code of the most recently known class sample, to provide a reference for subsequent manual interpretation or expert analysis of the cause of the anomaly.

[0048] During model deployment, to adapt to the edge computing environment of industrial sites, the shared feature extraction network Encoder() can be quantized and deployed to embedded edge devices (such as edge gateways and industrial controllers). The quantization accuracy is usually set to 8 bits or 16 bits. While ensuring that the model accuracy loss is less than 3%, the computing power and storage usage of the devices are reduced, and the real-time response speed is improved. To avoid false alarms caused by single data fluctuations, the alarm logic is set to trigger an alarm when the monitoring results of three consecutive sliding windows are all judged as abnormal behavior. At the same time, an adaptive model update mechanism is set. When the cumulative number of abnormal behavior proxy samples, i.e., Open samples, reaches a preset threshold, such as 50 samples for each abnormal proxy category, these samples can be used as new abnormal behavior proxy samples and added to the training set for semi-supervised backfill training to update the model parameters and further improve the model's ability to identify new types of abnormal behavior.

[0049] Example 1; To verify the effectiveness of the method in this embodiment, the publicly available load dataset PLAID is used as the verification basis. This dataset contains voltage and current waveform data of many common household and commercial appliances under various conditions, and is suitable for constructing experimental scenarios of typical load behavior of multi-energy devices.

[0050] The validation task selected eight common devices from the dataset, including air conditioners (AC), energy-saving lamps (CFL), refrigerators, hair dryers, heaters, incandescent bulbs (ILB), laptops, and vacuum cleaners, constructing them as known class samples. In addition, three unseen device classes (Unknown class), such as fans, microwave ovens, and washing machines, were randomly selected as representatives of simulated "abnormal energy consumption behavior." These unseen categories were simulated as unknown devices or abnormal power consumption behaviors that might occur during the testing phase.

[0051] During the experiment, the original sampled signals were first preprocessed, including normalization, filtering, and sliding window segmentation, and time-domain and frequency-domain features were extracted to form a unified sample vector. Then, according to the method proposed in this application, a training set of known categories and pseudo-unseen category augmented samples were constructed for two-stage training: the first stage identified the state of the standard known devices, and the second stage introduced pseudo-unseen samples to enhance the open set recognition capability.

[0052] In extracting time-domain and frequency-domain features to form a unified sample vector, taking the "off-on" load event of an energy-saving lamp sample as an example: when extracting the corresponding features, first collect the total raw voltage, current, and active power signals of the user side in the load event segment shown in Figure 2, and then preprocess them; furthermore, the voltage signal interruption is present in the electrical signal change corresponding to the load event in the figure. Voltage signal recovery Current signal conduction Current signal interruption Extract the voltage of the energy-saving lamp based on this load change event. Current reactive current Curves, with units of (V), (A), and (A), respectively, such as Figure 3 As shown in the image.

[0053] During the validation phase, the PLAID dataset was divided into a training set and a test set. The training set contained only the eight known device categories, while the test set contained a mixture of known and unseen categories. When identifying test set samples using the method described in this application, most unknown categories were correctly identified as "open categories," i.e., abnormal energy usage behaviors, thereby significantly reducing the risk of misclassifying unknown behaviors as known categories.

[0054] The results of the examples show that the proposed method achieves an accuracy of over 96% in known device status identification tasks, and also demonstrates good recognition capabilities in detecting open-class samples, with a recognition rate exceeding 94%. Figure 4 As shown, compared with traditional Softmax classifiers, One-ClassSVM, and isolated forest methods, this application significantly improves the detection capability of unknown behaviors and has a lower false positive rate, indicating that it is more robust and generalizable in real-world multi-energy supply systems.

[0055] Further analysis of the prediction results revealed that the "abnormal confidence score" for abnormal energy consumption behavior was significantly higher than that for normal samples, indicating that the model can effectively construct the discrimination boundary and perform abnormal alarm judgment. Overall, the monitoring method proposed in this application demonstrates excellent performance on the standard load dataset PLAID, exhibiting good transferability and engineering applicability.

[0056] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A method for monitoring the energy consumption status of equipment based on electrical signals of a multi-energy complementary power supply system, characterized in that: Includes the following steps: Acquire the timing electrical signals of each energy-consuming device in the energy supply system; The timing electrical signal is preprocessed to obtain a preprocessed electrical signal; Multidimensional features are extracted from the preprocessed electrical signal, and feature vectors of the energy-consuming devices are constructed based on the multidimensional features; the feature vectors of each energy-consuming device are combined to obtain the original sample set. Perturbation enhancement operations are applied to multiple original samples with known energy consumption states to generate a proxy sample set of abnormal behavior; A device energy consumption status monitoring model is constructed based on a two-stage model training mechanism; the device energy consumption status monitoring model is an Encoder-Classifier architecture model, which includes a shared feature extraction network and a linear classifier. The system collects electrical signals from various energy-consuming devices in the power supply system in real time and outputs real-time timing electrical signals based on a fixed window. The real-time time-series electrical signal is preprocessed and its features are extracted before being input into the equipment energy status monitoring model to obtain monitoring results.

2. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 1, characterized in that: The two-stage model training includes the following steps: The first stage involves closed-set supervised classification training based on the original sample set. In the second stage, a training pair is constructed by combining the original sample set and the abnormal behavior proxy sample set, and a similarity supervision loss is introduced based on the training pair to perform joint modeling to obtain the equipment energy status monitoring model.

3. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 2, characterized in that: In the first stage, the optimization objective is the classification loss function.

4. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 3, characterized in that: In the first stage, the loss weight coefficient is set to 0; After the first stage of training is completed, the loss weight coefficient is adjusted to 1 for use in the second stage of training.

5. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 1, characterized in that: The multidimensional features include time-domain features, frequency-domain features, and statistical features.

6. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 1, characterized in that: The perturbation enhancement operation includes at least one of the following: amplitude perturbation, time-scale stretching and compression, Gaussian noise perturbation, frequency domain coefficient perturbation, or random masking of feature channels.

7. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 1, characterized in that: It also includes the following steps: The monitoring results include normal behavior and abnormal behavior; When the monitoring result indicates normal behavior, output the device status category; When the monitoring result indicates abnormal behavior, an alarm mechanism is triggered, and the probability of abnormal behavior and the corresponding abnormal behavior label are output.

8. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 7, characterized in that: Obtaining the monitoring results also includes the following steps: Calculate the sum of class confidence scores for the monitored samples in the known categories, and calculate the anomaly confidence score based on the sum of class confidence scores; then compare the anomaly confidence score with a preset threshold. When the anomaly confidence score is higher than the preset threshold, the category of the monitored sample as abnormal behavior is output. If the anomaly confidence score is not higher than the preset threshold, the monitored sample is determined to be a normal behavior category.

9. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 1, characterized in that: The timing electrical signal includes at least one of current, voltage, active power, or reactive power.

10. The equipment energy consumption status monitoring method based on the electrical signal of a multi-energy complementary power supply system according to claim 1 or 9, characterized in that: The preprocessing of the time-series electrical signal includes at least one of denoising, outlier removal, or normalization.

Images