Smart meter fault warning method and system based on multi-parameter synchronous measurement

Abstract

The present invention discloses a fault warning method and system for smart meters based on multi-parameter synchronous measurement. The method includes synchronously collecting meter operation videos, reading data and environmental parameters. First, the video is optimized for illumination compensation, and image features are extracted through convolutional layers and fully connected layers. A comprehensive visual feature vector is constructed by combining the cross-attention mechanism with environmental parameters, and the overall shape of the meter is extracted to detailed features in layers; time series analysis is used to extract change trends and fluctuation characteristics from the reading data, taking into account the influence of environmental factors. Then, multi-source features are fused to generate a feature map reflecting the state of the meter; normal and fault state features are compared through similarity calculation to identify fault risks and issue warnings; ID matching and machine learning algorithms are used to evaluate the results, and historical data are combined to optimize fault warnings. By implementing the method of the present invention, multi-source information can be fused, which can significantly improve the speed and accuracy of fault diagnosis and ensure the continuous and stable operation of the power system.

Description

Technical Field

[0001] The present invention relates to a smart meter, and more particularly to a fault early warning method and system for a smart meter based on multi-parameter synchronous measurement. Background Art

[0002] With the rapid development of smart grids, smart meters are an indispensable component, and their stable operation is crucial to ensuring the overall efficiency and reliability of the power system. However, most current smart meter detection methods rely primarily on a single data dimension, such as relying solely on meter readings or simple video surveillance to assess the meter's operating status. While this approach can provide a certain degree of insight into the basic operating conditions of the meter, it falls short in the face of complex and changing real-world application environments. In particular, single-dimensional data collection struggles to fully capture the potential impact of these factors on meter performance, making it impossible to accurately determine whether the meter is operating optimally. This limitation means that in some cases, even if a meter fails or performance deteriorates, it may not be detected and addressed promptly due to a lack of sufficient information.

[0003] Traditional fault diagnosis methods are often limited to a single data source and lack the ability to synchronously collect and comprehensively analyze data from different sources. This significantly limits the efficiency and accuracy of fault detection. For example, methods based solely on meter readings may overlook meter errors caused by ambient temperature fluctuations, or simple video surveillance fails to account for the impact of varying lighting conditions on image recognition accuracy.

[0004] Therefore, it is necessary to design a new method to integrate multi-source information, which can significantly improve the speed and accuracy of fault diagnosis and ensure the continuous and stable operation of the power system. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a smart meter fault warning method and system based on multi-parameter synchronous measurement.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a smart meter fault early warning method based on multi-parameter synchronous measurement, comprising:

[0007] Synchronously collect meter operation videos, meter reading data, and environmental parameters;

[0008] Performing lighting compensation and optimization on the electricity meter operation video;

[0009] The image features of the lighting-optimized electricity meter operation video are extracted through convolutional and fully connected layers. A cross-attention mechanism is then combined with the environmental parameters to construct a comprehensive visual feature vector containing both image information and environmental factors. During the extraction process, a hierarchical approach is used to first extract the meter's overall shape, outline, and surrounding large features. The pattern and text on the meter's display, as well as detailed environmental features of the meter, are then extracted to produce the image feature vector.

[0010] Extracting change trends and fluctuation amplitude characteristics from the meter reading data using a time series analysis method to form a feature vector of the meter reading data, and analyzing the potential impact of environmental factors on the meter reading data;

[0011] Using multiple feature encoding layers to fuse the comprehensive visual feature vector, the meter reading data feature vector, and the feature vectors corresponding to the environmental parameters, to generate a prediction result that comprehensively reflects the meter status;

[0012] The prediction results that comprehensively reflect the status of the electricity meter and the feature description sets of normal and fault states are calculated for similarity to determine whether there is a fault risk and issue early warning information to obtain preliminary identification results;

[0013] Initial identification results are evaluated through ID matching and machine learning algorithms, and historical data is used to optimize the final fault warning results.

[0014] A further technical solution is: performing illumination compensation and optimization on the electric meter operation video includes:

[0015] Each frame of the meter operation video is input into a lighting adjustment model for lighting compensation and optimization, wherein the lighting adjustment model includes a local branch network for enhancing local feature details and a global branch network for generating a color correction matrix and gamma correction value and adjusting the entire meter image.

[0016] A further technical solution is as follows: the image features of the meter operation video after lighting optimization are extracted through the convolutional layer and the fully connected layer, and the cross-attention mechanism is combined with the environmental parameters to construct a comprehensive visual feature vector containing image information and environmental factors, including:

[0017] The video of the electricity meter running after lighting optimization is input into the convolutional neural network, which extracts image features through sliding convolution kernels and uses activation functions to introduce nonlinearity to process complex feature patterns.

[0018] The feature map extracted by the convolutional layer is flattened and input into the fully connected layer to learn the global feature representation of the image and output a feature vector containing the meter shape, display digits and symbols to obtain the image feature vector;

[0019] Encoding the environmental parameters into a vector form;

[0020] The encoded environmental parameter vector and the image feature vector are input into the cross-attention mechanism, the correlation is calculated to determine the influence weight of the environmental factors, and the image feature vector is weighted and adjusted according to the influence weight of the environmental factors to construct a comprehensive visual feature vector that integrates the image and environmental information.

[0021] A further technical solution is as follows: the overall shape, outline and surrounding features of the meter are first extracted in a layered manner, and then the pattern and words on the meter display screen and the detailed features of the meter's environment are extracted to obtain an image feature vector, including:

[0022] In the first several convolutional layers of the convolutional neural network, the overall shape and outline of the electric meter and the large features of the surrounding environment are extracted to obtain a feature map reflecting the overall appearance of the electric meter and the general environment, thereby obtaining a shallow feature map;

[0023] In the deep layer of the convolutional neural network, a smaller convolution kernel and step size are used to extract the detailed features of the pattern, words and surrounding environment of the electronic display screen, so as to obtain a feature map containing rich detailed information, thereby obtaining a deep feature map;

[0024] The shallow feature map and the deep feature map are spliced in the channel dimension, and the spliced feature map is input into the fully connected layer to output an image feature vector that integrates the meter shape, contour, surrounding features and display screen details.

[0025] A further technical solution is: using a time series analysis method to extract change trends and fluctuation amplitude characteristics from the meter reading data to form a feature vector of the meter reading data, and analyzing the potential impact of environmental factors on the meter reading data, including:

[0026] Normalizing the electric meter reading data;

[0027] Using sliding window technology and difference method, the trend slope and the change amount at adjacent time points are calculated based on the normalized meter reading data to reflect the long-term and short-term change trends of the meter readings. Within each sliding window, the standard deviation and range of the normalized meter reading data are calculated to obtain a meter reading data feature vector reflecting the operating status of the meter;

[0028] Extracting the change trend characteristics and fluctuation amplitude characteristics of the temperature and humidity data in the environmental parameters respectively to obtain an environmental extraction result;

[0029] Concatenating or weightedly fusing the feature vector reflecting the operating state of the electric meter with the environmental extraction result to form a fused comprehensive feature vector;

[0030] Through regression analysis and machine learning methods, a relationship model between meter readings and environmental parameters is established based on the fused comprehensive feature vector, and the influence of environmental factors on meter readings is analyzed.

[0031] A further technical solution is: using multiple feature coding layers to fuse the comprehensive visual feature vector, the meter reading data feature vector, and the feature vector corresponding to the environmental parameters to generate a prediction result that comprehensively reflects the meter status, including:

[0032] Normalizing the comprehensive visual feature vector, the electric meter reading data feature vector, and the feature vector corresponding to the environmental parameters respectively;

[0033] For the normalized vector, adjusting the number of channels of the feature vector corresponding to the environmental parameter, and performing spatial size adaptation on the comprehensive visual feature vector and the feature vector corresponding to the environmental parameter;

[0034] Concatenate the three adjusted feature vectors in the channel dimension to form a fused feature vector;

[0035] Performing a convolution operation on the fused feature vector to extract multi-scale features;

[0036] Using multiple feature coding layers to encode the multi-scale features, extract local features and reduce the spatial size, and perform weighted fusion of features at different levels during the encoding process to obtain an encoded feature map;

[0037] The encoded feature map is input into the decoupling head for prediction, and the output is a prediction result that comprehensively reflects the meter status.

[0038] A further technical solution is to calculate the similarity of the prediction result reflecting the meter status and the normal and fault status feature description set, determine whether there is a fault risk and issue a warning message to obtain a preliminary identification result, including:

[0039] Establish a description set of the normal operating state characteristics and fault state characteristics of the electricity meter;

[0040] For the description set, the text thinking chain method is used to simplify the complex description into expressions composed of simple words and relationships;

[0041] The simplified description set is segmented and converted into word vectors, and then a multi-head self-attention mechanism is used to generate text feature vectors;

[0042] The prediction result that comprehensively reflects the meter status is calculated by cosine similarity calculation with the text feature vectors of normal and fault status to obtain the calculation result;

[0043] Based on the calculation results, it is determined whether there is a failure risk and an early warning message is issued to obtain a preliminary identification result.

[0044] A further technical solution is to evaluate the preliminary identification results through ID matching and machine learning algorithms, and optimize the final fault warning results using historical data, including:

[0045] Assigning a unique ID to the preliminary recognition result in each frame of the image;

[0046] Calculating the correlation between the preliminary identification results of the same ID;

[0047] updating or confirming the preliminary recognition result of the current frame according to the preliminary recognition result of the subsequent frame;

[0048] If the recognition result matches the fault characteristics and the correlation is high, the credibility of the warning information is enhanced; if the correlation is low, it is determined to be a misidentification and the warning information is downgraded or discarded;

[0049] The suspected fault results from similarity calculation and the results after ID tracking optimization are comprehensively analyzed and combined with the historical operating data and fault records of the meter. A machine learning algorithm is used for re-evaluation, and the importance of features is analyzed to accurately determine the fault type and risk level, thereby optimizing the final fault warning results.

[0050] The further technical solution is: after evaluating the preliminary identification results through ID matching and machine learning algorithms and optimizing the final fault warning results using historical data, it also includes:

[0051] Specific maintenance recommendations are given based on the final fault warning results, including on-site inspection, component replacement or complete replacement.

[0052] The present invention also provides a smart meter fault warning system based on multi-parameter synchronous measurement, comprising:

[0053] A collection unit, used to synchronously collect meter operation videos, meter reading data, and environmental parameters;

[0054] an optimization unit, configured to perform illumination compensation and optimization on the electricity meter operation video;

[0055] A construction unit is configured to extract image features from the illumination-optimized electricity meter operation video using convolutional layers and fully connected layers, and to construct a comprehensive visual feature vector comprising image information and environmental factors by combining the environmental parameters with a cross-attention mechanism. During the extraction process, a hierarchical approach is adopted to first extract the overall shape, outline, and surrounding large features of the electricity meter, and then extract the pattern and text on the electricity meter display screen and detailed environmental features of the electricity meter to obtain an image feature vector.

[0056] an analysis unit, configured to extract change trends and fluctuation amplitude characteristics from the meter reading data using a time series analysis method, form a feature vector of the meter reading data, and analyze the potential impact of environmental factors on the meter reading data;

[0057] a fusion unit, configured to fuse the comprehensive visual feature vector, the meter reading data feature vector, and the feature vector corresponding to the environmental parameters using multiple feature coding layers to generate a prediction result that comprehensively reflects the state of the meter;

[0058] A calculation and judgment unit is used to perform similarity calculation on the prediction result that comprehensively reflects the state of the electric meter and the normal and fault state feature description set, determine whether there is a fault risk and issue an early warning information to obtain a preliminary identification result;

[0059] The evaluation unit is used to evaluate the preliminary identification results through ID matching and machine learning algorithms, and use historical data to optimize the final fault warning results.

[0060] The beneficial effects of the present invention compared with the prior art are as follows: the present invention synchronously collects meter operation videos, reading data and environmental parameters, first performs illumination compensation and optimization processing on the video, and adopts a layered approach to extract image features ranging from the overall shape of the meter to the details of the display screen through convolutional layers and fully connected layers, while combining the cross-attention mechanism to fuse environmental factors to construct a comprehensive visual feature vector; on the other hand, time series analysis is used to extract change trends and fluctuation amplitude features from the meter reading data, and the influence of environmental factors is evaluated to form a reading data feature vector; subsequently, multiple feature encoding layers are used to fuse the above-mentioned visual features, reading data features and environmental parameter features to generate a prediction result that comprehensively reflects the meter status; by calculating the similarity between this prediction result and the normal and fault state feature description set, the fault risk is identified and warned, and the preliminary results are further evaluated through ID matching and machine learning algorithms, and the final fault warning result is optimized using historical data; this method integrates multi-source information, which not only significantly improves the speed and accuracy of fault diagnosis, but also ensures the continuous and stable operation of the power system.

[0061] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1A flow chart of a smart meter fault warning method based on multi-parameter synchronous measurement provided by an embodiment of the present invention;

[0064] Figure 2 A schematic block diagram of a smart meter fault warning system based on multi-parameter synchronous measurement provided by an embodiment of the present invention;

[0065] Figure 3 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0067] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0068] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0069] It should be further understood that the term "and / or" used in the present description and appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0070] See also Figure 1 , Figure 1This is a schematic flow chart of a smart meter fault warning method based on multi-parameter synchronous measurement, provided by an embodiment of the present invention. This method is applied to a server that interacts with smart meters. The server synchronously collects meter operation videos, reading data, and environmental parameters. It then applies illumination compensation and optimization, convolutional layers, and fully connected layers combined with a cross-attention mechanism to extract image features. This method captures hierarchical features from the overall to the detailed features of the meter. Time series analysis is used to extract the changing trends and fluctuations in the reading data, taking into account the influence of environmental factors. Furthermore, multiple feature encoding layers are used to fuse comprehensive visual feature vectors, meter reading data feature vectors, and environmental parameter feature vectors to generate a prediction result that comprehensively reflects the meter status. Preliminary recognition results are evaluated through similarity calculation and ID matching techniques, and historical data is used to optimize the final fault warning result. This entire process effectively integrates multi-source information, significantly improving the speed and accuracy of fault diagnosis while also providing timely warning information, ensuring the continued stable operation of the power system and providing specific maintenance recommendations for subsequent actions.

[0071] Figure 1 FIG is a flow chart of a smart meter fault warning method based on multi-parameter synchronous measurement provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S110 to S170.

[0072] S110 , synchronously collecting meter operation video, meter reading data, and environmental parameters.

[0073] In this embodiment, based on the application scenario and monitoring requirements, a camera is installed in a suitable location near the meter box or in the meter installation area, such as on the front or side of the meter box, to clearly capture key features such as the meter display and indicator lights. A high-resolution camera with night vision capabilities and waterproof and dustproof design is selected to ensure stable operation and capture clear meter video in various lighting and weather conditions. The camera captures real-time meter operation video and transmits the video data to a data acquisition server or local storage device via wired (e.g., network cable, coaxial cable) or wireless (e.g., Wi-Fi, 4G / 5G) methods, while ensuring the stability and consistency of the video capture frame rate.

[0074] Connect the data acquisition device to the meter's communication interface, such as RS-485, RS-232, or Modbus, to ensure the device can stably read meter output data, including voltage, current, power, and energy. Based on actual needs and meter operating characteristics, set a reasonable data collection frequency, such as once every minute or every hour, to capture real-time changes in meter readings. Convert and organize the collected meter readings into a standardized data format (such as JSON or CSV) for subsequent data processing and analysis.

[0075] Define the environmental scope to be monitored, including the microenvironment surrounding the meter installation location (such as the interior of the meter box and the local area around the meter) and the broader external environment (such as the overall climate conditions in the installation area). Select appropriate environmental sensors, such as temperature, humidity, and pressure sensors, requiring them to possess high precision, high sensitivity, excellent stability, and interference resistance to accurately measure environmental parameters. Install the environmental sensors at the selected monitoring locations, ensuring they are properly positioned and installed. Debug them to ensure they function properly and accurately collect environmental parameter data. Use a data acquisition system to collect environmental parameter data output by the environmental sensors in real time and synchronize it with the meter's operating video and meter readings to ensure all data is on the same time scale, providing a foundation for subsequent comprehensive analysis. Preprocess the collected environmental parameter data, including noise removal, filtering, and smoothing, to improve data accuracy and reliability.

[0076] Specifically, the meter operation video refers to the meter operation video collected by a device installed near the smart meter over a period of time. The video content includes image information of key parts such as the meter appearance, display screen, indicator lights, etc., to monitor the surface condition of the meter, whether the display is normal, the flashing frequency of the indicator lights, and other intuitive states.

[0077] Meter reading data refers to the reading data of the smart meter itself collected synchronously during the same period, including electrical parameters such as voltage, current, power, and electric energy, which is used to analyze electrical performance indicators such as the meter's measurement accuracy and power change trends.

[0078] Environmental parameters refer to the synchronous collection of meter operating environment parameters, including meteorological data such as temperature, humidity, and air pressure, as well as geographic information of the meter's location and the distribution of surrounding facilities. These parameters can be used to assess the potential impact of the environment on meter operation. For example, high temperature may cause accelerated aging of meter components, and high humidity may cause degradation of insulation performance.

[0079] To improve the accuracy of data synchronization, a high-precision timestamp mechanism can be introduced, and all acquisition devices can be calibrated using a unified time reference (such as GPS time). In addition, a dedicated time synchronization protocol (such as PTP - Precision Time Protocol) can be used to ensure that data from different sources can be accurately aligned.

[0080] S120: Perform illumination compensation and optimization on the electricity meter operation video.

[0081] In this embodiment, each frame of the meter operation video is input into the illumination adjustment model for illumination compensation and optimization, wherein the illumination adjustment model includes a local branch network for enhancing local feature details and a global branch network for generating a color correction matrix and gamma correction value and adjusting the entire meter image.

[0082] Each frame of the meter operation video is fed into a lighting adjustment model for lighting compensation and optimization. This model includes a local branch network for enhancing local feature details, such as the digits and scale markings on the meter display, and a global branch network responsible for generating a color correction matrix and gamma correction values. These values are then used to adjust the brightness, contrast, and color balance of the entire meter image to accommodate image acquisition in different lighting environments, ensuring a clearer and more accurate image reflecting the meter's actual status.

[0083] The local branch network focuses on the local features of the image and performs a point-by-point linear transformation on the feature map of the input image through a series of 1×1 convolutional layers. This transformation can enhance the characteristics of specific pixels, thereby better preserving and highlighting important details in the image.

[0084] The global branch network is responsible for capturing and generating global illumination adjustment parameters, including color correction matrices and gamma correction values. These parameters are used to adjust the overall lighting conditions of the image, making the image consistent and high-quality across different environments.

[0085] Each frame of the meter operation video is input into the lighting adjustment model.

[0086] The local branch network generates a predicted multiplication map and addition map by performing an independent linear transformation on each pixel of the feature map. These two maps serve as the query (Q) for subsequent feature fusion.

[0087] After processing by the local branch network, the local details in the image are significantly enhanced.

[0088] The global branch network starts with each frame and first generates a value vector (V) representing the global information through a fully connected layer. Then, another fully connected layer generates a key vector (K).

[0089] K and V are combined with the query (Q) generated by PEM through a cross-attention mechanism to form the final color correction matrix and gamma correction value.

[0090] The color correction matrix and gamma correction values generated by the local branch network and the global branch network are applied to the original input image. Specifically, the original image is reshaped into a form that fits these parameters, and then the color and brightness of the image are finely adjusted through matrix operations and gamma correction.

[0091] The final output is a lighting-adjusted image that maintains the original details while having better lighting uniformity and visual effects.

[0092] By combining local and global branch networks, the team addresses the issue of poor image quality under varying lighting conditions, particularly for image acquisition in dynamic environments. The enhanced image quality helps improve the performance of subsequent meter image recognition and analysis tasks, such as meter reading recognition and fault detection.

[0093] In another embodiment, an adaptive adjustment mechanism can be added to enable the illumination adjustment model to dynamically adjust its parameter settings based on the specific conditions of the current image. Furthermore, the robustness of the model can be enhanced by adding more training samples, especially those covering a variety of complex lighting conditions.

[0094] S130. Extract image features of the lighting-optimized electricity meter operation video through convolutional layers and fully connected layers, and use a cross-attention mechanism in combination with the environmental parameters to construct a comprehensive visual feature vector containing image information and environmental factors. In the extraction process, a hierarchical approach is adopted to first extract the overall shape, outline, and surrounding large features of the electricity meter, and then extract the pattern and words on the electricity meter display screen and the environmental detail features of the electricity meter to obtain an image feature vector.

[0095] In this embodiment, the comprehensive visual feature vector refers to a feature vector that integrates image features and environmental parameter information, and can comprehensively represent the visual features and operating environment of the electric meter.

[0096] Image feature vector refers to the feature vector extracted by convolutional neural network, which contains image information such as meter shape and display content.

[0097] In one embodiment, the aforementioned step S130 may include steps S131 to S134 .

[0098] S131. Input the electricity meter operation video after lighting optimization into the convolutional neural network, extract image features by sliding the convolution kernel, and use the activation function to introduce nonlinearity to process complex feature patterns.

[0099] The lighting-optimized electricity meter operation video is fed frame by frame into a convolutional neural network, ensuring uniform image size for each frame. The convolutional layer extracts image features by sliding the convolution kernel, while also introducing nonlinearity using activation functions to process complex feature patterns. The first few layers of the convolutional neural network extract the meter's overall shape, outline, and large-scale features of its surroundings. This creates a feature map reflecting the meter's overall appearance and general surroundings. This helps the model understand the meter's position and general state within the environment.

[0100] S132. Flatten the feature map extracted by the convolutional layer and input it into the fully connected layer to learn the global feature representation of the image and output a feature vector containing the shape of the meter, the display screen numbers and the symbol information to obtain the image feature vector.

[0101] In this embodiment, the feature map extracted by the convolutional layer is flattened and fed into the fully connected layer to learn the global feature representation of the image. The fully connected layer outputs a feature vector containing the meter's shape, display digits, and symbols, thereby generating an image feature vector. In the deep layers of the convolutional neural network, a smaller convolution kernel and step size are used to extract detailed features of the meter's display pattern, text, and surrounding environment, generating a feature map rich in detailed information. These detailed features help further analyze the meter's operating environment, such as the texture of the meter box and the layout of surrounding pipelines.

[0102] S133. Encode the environmental parameters into a vector form.

[0103] In this embodiment, environmental parameters (such as temperature, humidity, etc.) are quantized and normalized, and then encoded into a vector form of a fixed length.

[0104] S134. Input the encoded environmental parameter vector and the image feature vector into the cross-attention mechanism, calculate the correlation to determine the environmental factor influence weight, and adjust the image feature vector according to the environmental factor influence weight to construct a comprehensive visual feature vector that integrates the image and environmental information.

[0105] In this embodiment, the encoded environmental parameter vector is fed into a cross-attention mechanism along with the image feature vector. This step aims to dynamically assess the weight of each environmental factor's impact on the image features, ensuring that the resulting feature vector fully reflects the specific impact of environmental conditions on the meter's operating status.

[0106] The image feature vector is weighted based on the influence of environmental factors determined by the cross-attention mechanism. This process helps highlight the features that are more critical in specific environments, providing more accurate data support for subsequent analysis.

[0107] Through the above steps, a comprehensive visual feature vector is obtained that combines image features with environmental parameter information. This vector not only includes the physical characteristics of the meter itself (such as shape and the numbers on the display), but also integrates factors affecting its operating status (such as temperature and humidity). This enables it to provide a comprehensive and accurate description of the meter's operating status based on multiple factors.

[0108] In another embodiment, after obtaining a preliminary comprehensive visual feature vector, it is further processed using multiple feature encoding layers. Upsampling and downsampling techniques are used to adjust the spatial dimensions of feature maps at different scales to ensure their consistency. These feature maps are then fused together through concatenation or weighted addition to form a final fused feature map. This approach effectively integrates information from different layers, improving the model's ability to understand and process multi-scale data.

[0109] In addition, the above-mentioned layered approach is used to first extract the overall shape, outline, and surrounding features of the meter, and then extract the pattern, words, and environmental details of the meter display screen to obtain the image feature vector, including:

[0110] In the first several convolutional layers of the convolutional neural network, the overall shape and outline of the electric meter and the large features of the surrounding environment are extracted to obtain a feature map reflecting the overall appearance of the electric meter and the general environment, thereby obtaining a shallow feature map;

[0111] In the deep layer of the convolutional neural network, a smaller convolution kernel and step size are used to extract the detailed features of the pattern, words and surrounding environment of the electronic display screen, so as to obtain a feature map containing rich detailed information, thereby obtaining a deep feature map;

[0112] The shallow feature map and the deep feature map are spliced in the channel dimension, and the spliced feature map is input into the fully connected layer to output an image feature vector that integrates the meter shape, contour, surrounding features and display screen details.

[0113] In this embodiment, a video of an electricity meter operating after lighting optimization is input frame by frame into a convolutional neural network. In the first several convolutional layers of the convolutional neural network, a convolution kernel is used to convolve the image to extract the overall shape, outline, and large features of the surrounding environment of the electricity meter. The convolution operation extracts spatial features by applying a filter (convolution kernel) to the input data. The convolution layer performs downsampling, that is, reducing the size of the feature map through the stride parameter and increasing the number of channels. This step helps reduce the spatial dimension of the data while enhancing the expressive power of the features. After the convolution operation, the feature map enters the batch normalization layer, which helps accelerate the network convergence process and enhance the network stability. Next, the ReLU activation function is applied to each pixel in each feature map to increase the nonlinearity of the network, enabling it to learn more complex data patterns.

[0114] After the above processing, a feature map reflecting the overall appearance and general environment of the electric meter is obtained, thereby obtaining a shallow feature map.

[0115] In the deep part of the convolutional neural network, a smaller convolution kernel and step size are used to perform further convolution operations on the shallow feature map. This step can extract the detailed features of the electronic display screen pattern, text and surrounding environment, and obtain a feature map containing rich detailed information. The C2f layer is a network structure specially designed to extract higher-level features, which is used to further extract feature information, including specific feature combinations or other advanced feature engineering methods. Using residual links, the original input information is passed directly to the subsequent layers for concat splicing. This technology allows more original input information to be retained, helps avoid information loss and alleviate the gradient vanishing problem, while enhancing the network's deep feature expression capabilities.

[0116] After the above processing, deep feature maps containing rich detail information are obtained. These feature maps can capture the subtle features of the electricity meter and its surrounding environment.

[0117] The shallow feature map and the deep feature map are concatenated along the channel dimension. This step integrates feature information from different levels to form a comprehensive feature representation. The concatenated feature map contains both large features such as the meter's overall shape and outline, as well as detailed features such as the display pattern and text, providing a more comprehensive description of the meter and its surroundings. The concatenated feature map is flattened and input into the fully connected layer. The fully connected layer learns the global feature representation of the image through a large number of neuronal connections, outputting an image feature vector that integrates the meter's shape, outline, surrounding features, and display details. This step further integrates all important information in the image to form a comprehensive description of the image content.

[0118] S140 , extracting change trends and fluctuation amplitude characteristics from the meter reading data using a time series analysis method to form a feature vector of the meter reading data, and analyzing the potential impact of environmental factors on the meter reading data.

[0119] In this embodiment, the meter reading data feature vector refers to a feature vector formed by extracting the change trend and fluctuation amplitude characteristics from the meter reading data through a time series analysis method, and is used to reflect the long-term and short-term change trends and fluctuations of the meter reading.

[0120] In one embodiment, the aforementioned step S140 may include steps S141 to S145.

[0121] S141. Normalize the meter reading data.

[0122] In this embodiment, the meter reading data is normalized to standardize its numerical range, usually to the interval 0 and 1. The purpose of this step is to eliminate the dimensional differences between data of different dimensions and make the data have a unified scale to prepare for subsequent feature extraction.

[0123] S142. Using sliding window technology and differential method, the trend slope and the change amount at adjacent time points are calculated based on the normalized meter reading data to reflect the long-term and short-term change trends of the meter readings. Within each sliding window, the standard deviation and range of the normalized meter reading data are calculated to obtain a characteristic vector of the meter reading data reflecting the operating status of the meter.

[0124] In this embodiment, a sliding window technique is used to process the normalized meter reading data. The size of the sliding window can be set according to actual needs, for example, to 10 time points. The sliding window moves across the data sequence at each time point, covering a certain number of consecutive data points each time.

[0125] Difference method: Use the difference method to calculate the change between adjacent time points. For each time point t, calculate the difference between it and the previous time point t-1, that is, This step can reflect the short-term trend of meter readings.

[0126] Within each sliding window, a linear regression method is used to calculate the trend slope. The data points within the sliding window are fitted to a straight line, and the slope of this line is the trend slope. The trend slope reflects the long-term trend of the meter reading within that window.

[0127] Within each sliding window, the standard deviation and range of the normalized meter readings are calculated. The standard deviation measures the dispersion of the data points, while the range is the difference between the maximum and minimum values. These two indicators can reflect the fluctuation of the meter readings.

[0128] S143 , respectively extracting the change trend characteristics and fluctuation amplitude characteristics of the temperature and humidity data in the environmental parameters to obtain an environmental extraction result.

[0129] In this embodiment, the environmental extraction result refers to the change trend characteristics and fluctuation amplitude characteristics of the temperature and humidity data in the environmental parameters.

[0130] The trend characteristics and fluctuation amplitude characteristics of temperature and humidity data in environmental parameters are extracted respectively. For temperature and humidity data, the sliding window technique and difference method can also be used to calculate their trend slope and the amount of change between adjacent time points, as well as the standard deviation and range.

[0131] S144: Concatenate or weightedly fuse the feature vector reflecting the operating status of the electric meter with the environment extraction result to form a fused comprehensive feature vector.

[0132] In this embodiment, the feature vector reflecting the meter's operating status is concatenated or weighted fused with the environmental extraction result to form a fused comprehensive feature vector. Concatenation directly connects the two feature vectors, while weighted fusion performs a weighted sum of the two feature vectors based on a predetermined weight.

[0133] S145. Through regression analysis and machine learning methods, a relationship model between meter readings and environmental parameters is established based on the fused comprehensive feature vector, and the influence of environmental factors on meter readings is analyzed.

[0134] Using regression analysis and machine learning methods, a relationship model between meter readings and environmental parameters is established based on the fused comprehensive feature vector. For example, a multivariate linear regression model can be used to analyze the linear relationship between meter readings and temperature and humidity, or nonlinear models such as decision trees and random forests can be used to capture more complex relationships. Using this established relationship model, the impact of environmental factors on meter readings can be analyzed. For example, the impact of temperature changes on meter accuracy or the impact of humidity changes on meter reading fluctuations can be analyzed.

[0135] S150 , using multiple feature coding layers to fuse the comprehensive visual feature vector, the meter reading data feature vector, and the feature vector corresponding to the environmental parameters, to generate a prediction result that comprehensively reflects the meter status.

[0136] In this embodiment, the prediction result that comprehensively reflects the state of the electricity meter refers to providing the best estimation result of the current state of the electricity meter based on the model's learning and understanding of the characteristics of the electricity meter and its environment.

[0137] In one embodiment, the aforementioned step S150 may include steps S151 to S156.

[0138] S151 . Normalize the comprehensive visual feature vector, the electric meter reading data feature vector, and the feature vector corresponding to the environmental parameters respectively.

[0139] In this embodiment, the comprehensive visual feature vector, the meter reading data feature vector, and the environmental parameter feature vector are normalized. The purpose of normalization is to unify the scale of data from different sources, eliminate the impact of different dimensions, and make subsequent processing more accurate and effective. Common normalization methods include Min-Max normalization and Z-score normalization.

[0140] S152 . For the normalized vector, adjust the number of channels of the feature vector corresponding to the environmental parameter, and perform spatial size adaptation on the comprehensive visual feature vector and the feature vector corresponding to the environmental parameter.

[0141] In this embodiment, the number of channels of the environmental parameter feature vector is adjusted to be consistent with the number of channels of the comprehensive visual feature vector and the meter reading data feature vector. The comprehensive visual feature vector and the environmental parameter feature vector are spatially adapted to ensure that they have the same spatial size as the meter reading data feature vector.

[0142] Specifically, the number of channels of the environmental parameter feature vector is adjusted to 1 to simplify the feature vector and highlight its importance. A convolution operation (such as 1×1 convolution) can be used to adjust the number of channels.

[0143] Adapt the spatial size of the comprehensive visual feature vector and the environmental parameter feature vector to ensure that they have the same spatial size as other feature vectors (such as the meter reading data feature vector). This can be achieved through upsampling or downsampling techniques, such as using nearest neighbor interpolation for upsampling or using maximum pooling for downsampling.

[0144] S153. Concatenate the three adjusted feature vectors in the channel dimension to form a fused feature vector.

[0145] In this embodiment, a fused feature vector is formed by concatenating or weightedly fusing feature vectors from multiple different sources (such as image features, meter reading data features, and environmental parameter features) along the channel dimension after preprocessing, including normalization, channel number adjustment, and spatial scale adaptation. This fused feature vector integrates multiple pieces of information to more comprehensively and accurately describe the meter's operating status.

[0146] The normalized and resized comprehensive visual feature vector, the meter reading data feature vector, and the environmental parameter feature vector are concatenated along the channel dimension to form a fused feature vector containing multi-source information. This process allows the model to simultaneously consider information from images, meter readings, and environmental conditions, improving prediction accuracy.

[0147] A hybrid structure of maximum and average pooling is used for downsampling to produce an adjusted environmental feature map. Convolution is then used to adjust the number of channels for the shallow meter shape features and the deep meter screen pattern and text features. Upsampling is then performed using nearest neighbor interpolation to produce an adjusted meter feature map. Finally, a convolution operation is performed on the adjusted environmental feature map and the adjusted meter feature map, and the concatenation is performed along the channel dimension to produce a fused feature vector.

[0148] S154: Perform a convolution operation on the fused feature vector to extract multi-scale features.

[0149] In this example, a convolutional layer is used to process the fused feature vector, capturing features at different scales by designing different filter sizes. This multi-scale feature extraction helps the model better understand complex scenes, especially when the target object (such as an electricity meter) is surrounded by numerous interference factors.

[0150] S155. Use multiple feature coding layers to encode the multi-scale features, extract local features and reduce the spatial size, and perform weighted fusion on features at different levels during the encoding process to obtain an encoded feature map.

[0151] In this embodiment, the extracted multi-scale features are further processed using multiple feature encoding layers, aiming to enhance important features while reducing redundant information. These encoding layers can adopt different structures (such as residual networks and densely connected networks) to effectively reduce the spatial size of the data while preserving key information.

[0152] During the encoding process, by weighted fusion of features at different levels, the effective information at each level can be more flexibly integrated to form a more comprehensive and representative feature representation.

[0153] Multiple feature encoding layers are used to encode multi-scale features. Convolutional layers extract local features, and pooling layers reduce the spatial size of feature maps to extract important features. During the encoding process, features at different levels are weighted and fused to produce the encoded feature map.

[0154] S156: Input the encoded feature map into the decoupling head for prediction, and output a prediction result that comprehensively reflects the state of the electricity meter.

[0155] In this embodiment, the encoded feature map is finally fed into the decoupling head, where a series of convolutional and pooling layers perform the final classification or regression task, outputting a comprehensive prediction of the meter's status. This feature map not only contains basic information about the meter's status but also incorporates the impact of its operating environment, providing strong support for subsequent maintenance decisions.

[0156] Specifically, the encoded feature map is fed into the decoupling head for prediction. Convolution is performed within the decoupling head, and the sigmoid function is used for normalization. The model is optimized using the BCDELoss loss function to ensure that the model can effectively learn from complex features and produce predictions suitable for classification tasks.

[0157] The fused feature map is fed into the decoupling head for prediction. Through a series of convolutional, pooling, and fully connected layer operations, the decoupling head learns and extracts the category probability distribution corresponding to each position, ultimately outputting the category with the highest probability as the prediction result to determine the meter's operating status.

[0158] Through these steps, the entire process not only fully mines and utilizes data from different modalities, but also significantly improves the model's understanding and prediction accuracy for complex scenarios through multi-level and multi-angle feature extraction and fusion. This is of great significance in fields such as smart grid management and urban infrastructure monitoring.

[0159] S160: Calculate similarity between the prediction result that comprehensively reflects the state of the electric meter and the normal and fault state feature description sets to determine whether there is a fault risk and issue a warning message to obtain a preliminary identification result.

[0160] In this embodiment, the preliminary recognition result refers to the meter status judgment result obtained by calculating the similarity between the prediction result and the feature description set, including the fault risk prompt and status description.

[0161] In one embodiment, the aforementioned step S160 may include steps S161 to S165 .

[0162] S161. Establish a description set of the normal operating state characteristics and fault state characteristics of the electricity meter.

[0163] In this embodiment, the normal operating state description is to collect images, readings, and environmental parameter features of the normal operation of the meter to construct a normal state feature description set. For example, "the display screen is intact, the reading is stable, and the ambient temperature and humidity are normal."

[0164] Specifically, we collect image features (e.g., intact display, normal flashing indicator lights), reading data features (e.g., stable readings, no abnormal fluctuations), and environmental parameter features (e.g., normal temperature and humidity ranges) when the meter is operating normally. This constructs a feature description of the meter's normal operating status. For example, a normal operating status might be described as "display pixels intact, readings stable within the normal power range, and ambient temperature between 20°C and 30°C."

[0165] Fault Status Description: For each fault type (such as measurement error, display failure, and communication failure), a detailed description is created from three dimensions: image, reading, and environmental parameter characteristics. This creates a fault status feature description set. For example, "Measurement error: Reading deviation exceeds 5%, and error fluctuates significantly in high temperature environments."

[0166] Specifically, for each fault type (such as measurement error fault, display fault, communication fault, etc.), a detailed description is made from multiple dimensions such as image features, reading data features, and environmental parameter features to form a fault feature description template containing multi-parameter features.

[0167] Metering error faults: Characteristic descriptions may include: the deviation between the meter reading and the actual power calculation value exceeds a set threshold, or the error changes abnormally at a specific ambient temperature. For example, "The deviation between the meter reading and the actual power calculation value exceeds 5%, and the error increases at an ambient temperature of 35°C."

[0168] Display failure: Feature descriptions may include image features such as missing display pixels, blurred content, and abnormal indicator light flashing frequency, while the corresponding meter reading may not be affected. For example, "There are missing pixels on the display, the indicator light flashes abnormally, but the meter reading does not change significantly."

[0169] Communication failure: Characteristic descriptions may include communication interruption between the meter and the system, data transmission delay or error, etc. For example, "Communication between the meter and the system is interrupted, and data transmission delay exceeds 10 seconds."

[0170] S162. For the description set, use the text thinking chain method to simplify the complex description into an expression composed of simple words and relationships.

[0171] In this embodiment, complex fault feature descriptions are converted into long sentences composed of simple words and relationships to more efficiently match and compare with the actual extracted feature vectors. For example, "The meter reading deviates from the actual power calculation value by more than 5%, and the error increases at an ambient temperature of 35°C" can be simplified to "The reading deviates by more than 5%, and the error increases at 35°C."

[0172] S163. Segment the simplified description set and convert it into word vectors, and then use the multi-head self-attention mechanism to generate text feature vectors.

[0173] In this example, a pre-trained text encoder (such as BERT or Word2Vec) is used to convert the segmented words into fixed-dimensional word vectors. For example, the BERT model is used to convert "reading deviation exceeds 5%" into a 512-dimensional word vector.

[0174] The word vector obtained above is used as input and passed to the multi-head self-attention mechanism.

[0175] The multi-head self-attention mechanism calculates attention scores between different word vectors to measure their relevance. For example, in the description "reading deviation exceeds 5%, high temperature error increases," "reading deviation" and "high temperature error" may have high attention scores, indicating that they are significantly related in fault diagnosis.

[0176] Based on the attention scores, a weighted summation of word vectors is performed to generate a higher-level text feature vector. This feature vector not only retains the original text information but also incorporates contextual semantic associations, enabling it to more accurately represent fault characteristics.

[0177] S164 , performing cosine similarity calculation on the prediction result that comprehensively reflects the state of the electric meter and the text feature vectors of the normal and fault states to obtain a calculation result.

[0178] In this embodiment, if the prediction result is the output of a classification task, such as the probability of the fault type being "metering error fault" being 0.8, this probability value is directly used as the feature vector. If it is a multi-classification task, one-hot encoding can be used to convert the class label into a feature vector. For example, a normal state is represented as [1,0], and a metering error fault is represented as [0,1]. If the prediction result is the output of a regression task, such as a meter reading of 120.5kWh, it can be mapped to a discrete category. For example, set 100kWh to 150kWh as the normal range, map it to the category "normal", and convert it into the corresponding feature vector.

[0179] The cosine similarity formula is used to calculate the similarity between the comprehensive prediction result feature vector and the normal and fault state text feature vectors. If the comprehensive prediction result feature vector is [0.8, 0.2] and the normal state text feature vector is [1, 0], the similarity is 0.8.

[0180] S165. Determine whether there is a failure risk based on the calculation result and issue a warning message to obtain a preliminary identification result.

[0181] In this embodiment, the similarity threshold is set according to actual needs. For example, the similarity threshold for a normal state is set to 0.8, and the similarity threshold for a fault state is set to 0.7.

[0182] The calculated similarity is compared with the set threshold. For example, if the similarity between the comprehensive prediction result and the normal state is 0.85, which is higher than the threshold of 0.8, the meter is judged to be in a normal state. If the similarity with the metering error fault state is 0.75, which is higher than the threshold of 0.7, the meter is judged to have a metering error fault risk.

[0183] If the similarity with the fault signature exceeds a set threshold, the system outputs the corresponding fault warning information, including the fault type, location, and risk level. For example, the output might be "Fault Type: Meter Error Fault, Location: Electricity Metering Chip, Risk Level: High."

[0184] If the similarity with the normal features is higher than the set threshold, it is judged to be in a normal state and a normal operation prompt message can be output.

[0185] If the similarity falls below a set threshold, the system continues to monitor and update the data to further analyze the meter status. For example, if the similarity between the comprehensive prediction result and the normal state and all known fault states falls below the threshold, the system continues to collect data, updates the feature vector, and recalculates and judges the similarity.

[0186] S170. Evaluate the preliminary identification results through ID matching and machine learning algorithms, and use historical data to optimize the final fault warning results.

[0187] In this embodiment, the final fault warning result is the final judgment on the fault status of the meter after comprehensively considering the preliminary recognition results, similarity calculation, ID tracking and historical data, which includes key information such as fault type, location, risk level, etc.

[0188] In one embodiment, the aforementioned step S170 may include steps S171 to S175.

[0189] S171, assigning a unique ID to the preliminary recognition result in each frame of the image;

[0190] S172, calculating the correlation between the preliminary recognition results of the same ID;

[0191] S173. Update or confirm the preliminary recognition result of the current frame according to the preliminary recognition result of the subsequent frame;

[0192] S174. If the recognition result matches the fault characteristics and the correlation is high, the credibility of the warning information is enhanced. If the correlation is low, it is determined to be a misidentification and the warning information is downgraded or discarded.

[0193] S175. Comprehensively analyze the suspected fault results of similarity calculation and the results after ID tracking optimization, and combine them with the historical operation data and fault records of the meter; use machine learning algorithms to re-evaluate and analyze the importance of features to accurately determine the fault type and risk level, so as to optimize the final fault warning results.

[0194] In this embodiment, a unique ID is assigned to each preliminary fault recognition result in each frame of the image to ensure that each detection frame can be uniquely identified, which facilitates subsequent tracking and matching.

[0195] Perform ID matching on the detection frames in all frames and find the detection frames with the same ID, which are considered to be the representation of the same target at different time points.

[0196] By comparing the IDs of detection boxes in different frames, we find detection boxes with the same ID. These detection boxes are considered to be representations of the same object at different time points. We calculate the Intersection over Union (IoU) value between the detection boxes with the same ID in the current frame and the next frame to measure the degree of change in the object's position between the two frames.

[0197] If the correlation (IoU value) is greater than the set threshold, the recognition result of the next frame is used as the intermediate recognition result of the current frame. It is assumed that the detection boxes in the two frames represent the same target, and the recognition result of the next frame is used to update the recognition result of the current frame.

[0198] If the correlation is lower than the set threshold, the detection frame and the corresponding recognition result are discarded, and it is considered that the detection frames in the two frames may not represent the same target, or the target has changed significantly.

[0199] If the frame after the current frame is not the last frame, the next frame will be used as the current frame, and the ID matching and association calculation will be repeated. If the frame after the current frame is the last frame, all intermediate recognition results will be determined as the final recognition result.

[0200] A comprehensive analysis is performed by combining the suspected fault results from similarity calculation with the optimized ID tracking results. Furthermore, the meter's historical operating data and fault records provide richer context for the evaluation.

[0201] Use machine learning algorithms (such as Bayesian networks and decision trees) to re-evaluate the comprehensive features. Train the model based on historical data to learn the mapping between different features and fault types. Use the machine learning model to analyze the importance of features and identify the features that have the greatest impact on fault warnings.

[0202] For suspected fault characteristics that are highly similar to historical fault patterns, the warning level is increased and maintenance personnel are promptly notified for inspection. For suspected fault characteristics that occur occasionally and have little correlation with historical faults, they are determined to be temporary operational fluctuations and no formal warning is issued or only a low-level warning is issued.

[0203] Through this comprehensive analysis and machine learning evaluation approach, faults can be identified more accurately, false alarms can be reduced, and the reliability and practicality of the fault warning system can be improved.

[0204] In one embodiment, the above method further includes:

[0205] After evaluating the preliminary identification results through ID matching and machine learning algorithms and optimizing the final fault warning results using historical data, the following also applies:

[0206] Specific maintenance recommendations are given based on the final fault warning results, including on-site inspection, component replacement or complete replacement.

[0207] Based on the final fault warning results, specific maintenance suggestions are given. These suggestions can be divided into several categories according to different fault levels:

[0208] On-site inspection: For cases where anomalies are detected but the specific cause is unclear, it is recommended to dispatch a technician to conduct a detailed on-site inspection. This can avoid unnecessary component replacement or complete replacement, and can also help identify and resolve problems in a timely manner.

[0209] Part replacement: If the fault warning clearly points to a problem with a specific component, such as a malfunctioning sensor or a damaged display, a recommendation should be made to replace the corresponding component. This recommendation is usually accompanied by a detailed diagnostic report explaining why the component needs to be replaced and how to select a suitable replacement.

[0210] Complete replacement: In some extreme cases, when the overall performance of the meter has significantly degraded and the cost of repair is too high, it is recommended to replace the entire device with a new one. This recommendation is often based on long-term historical data analysis showing that the device is nearing or exceeding its expected service life.

[0211] The above-mentioned smart meter fault warning method based on multi-parameter synchronous measurement synchronously collects meter operation video, reading data and environmental parameters, first performs illumination compensation and optimization processing on the video, and uses a layered approach to extract image features from the overall shape of the meter to the details of the display screen through convolutional layers and fully connected layers. At the same time, a cross-attention mechanism is combined to integrate environmental factors to construct a comprehensive visual feature vector; on the other hand, time series analysis is used to extract change trends and fluctuation amplitude characteristics from the meter reading data, and the impact of environmental factors is evaluated to form a reading data feature vector; then, multiple feature encoding layers are used to fuse the above visual features, reading data features and environmental parameter features to generate a prediction result that comprehensively reflects the meter status; by calculating the similarity between this prediction result and the set of normal and fault state feature descriptions, fault risks are identified and warnings are issued, and then the preliminary results are further evaluated through ID matching and machine learning algorithms, and the final fault warning results are optimized using historical data; this method integrates multi-source information, not only significantly improving the speed and accuracy of fault diagnosis, but also ensuring the continuous and stable operation of the power system.

[0212] Figure 2 FIG is a schematic block diagram of a smart meter fault warning system 300 based on multi-parameter synchronous measurement provided by an embodiment of the present invention. Figure 2 As shown, corresponding to the above-mentioned smart meter fault warning method based on multi-parameter synchronous measurement, the present invention also provides a smart meter fault warning system 300 based on multi-parameter synchronous measurement. The smart meter fault warning system 300 based on multi-parameter synchronous measurement includes a unit for executing the above-mentioned smart meter fault warning method based on multi-parameter synchronous measurement, and the system can be configured in a server. Specifically, please refer to Figure 2 The smart meter fault warning system 300 based on multi-parameter synchronous measurement includes a collection unit 301, an optimization unit 302, a construction unit 303, an analysis unit 304, a fusion unit 305, a calculation and judgment unit 306 and an evaluation unit 307.

[0213] The collection unit 301 is used to synchronously collect the meter operation video, meter reading data and environmental parameters; the optimization unit 302 is used to perform illumination compensation and optimization on the meter operation video; the construction unit 303 is used to extract the image features of the meter operation video after illumination optimization through the convolution layer and the fully connected layer, and use the cross attention mechanism in combination with the environmental parameters to construct a comprehensive visual feature vector containing image information and environmental factors, wherein, in the extraction process, a hierarchical method is adopted to first extract the overall shape, outline and surrounding large features of the meter, and then extract the pattern, wording and environmental detail features of the meter display screen to obtain the image feature vector; the analysis unit 304 is used to use the time series analysis method to extract the image features from the meter reading data Extract the change trend and fluctuation amplitude characteristics to form the meter reading data feature vector, and analyze the potential impact of environmental factors on the meter reading data; the fusion unit 305 is used to use multiple feature coding layers to fuse the comprehensive visual feature vector, the meter reading data feature vector and the feature vector corresponding to the environmental parameters to generate a prediction result that comprehensively reflects the meter status; the calculation and judgment unit 306 is used to calculate the similarity of the prediction result that comprehensively reflects the meter status and the normal and fault state feature description set, determine whether there is a fault risk and issue an early warning information to obtain a preliminary recognition result; the evaluation unit 307 is used to evaluate the preliminary recognition result through ID matching and machine learning algorithm, and use historical data to optimize the final fault warning result.

[0214] In one embodiment, the optimization unit 302 is used to input each frame of the electric meter operation video into the illumination adjustment model for illumination compensation and optimization, wherein the illumination adjustment model includes a local branch network for enhancing local feature details and a global branch network for generating a color correction matrix and gamma correction value and adjusting the entire electric meter image.

[0215] In one embodiment, the construction unit 303 is configured to:

[0216] The lighting-optimized electricity meter operation video is input into a convolutional neural network, image features are extracted by sliding the convolution kernel, and nonlinearity is introduced using an activation function to process complex feature patterns; the feature map extracted by the convolution layer is flattened and input into the fully connected layer, which learns the global feature representation of the image and outputs a feature vector containing the meter shape, display screen numbers and symbol information to obtain an image feature vector; the environmental parameters are encoded into a vector form; the encoded environmental parameter vector and the image feature vector are input into a cross-attention mechanism, the correlation is calculated to determine the influence weight of the environmental factors, and the image feature vector is weighted and adjusted according to the influence weight of the environmental factors to construct a comprehensive visual feature vector that integrates the image and environmental information.

[0217] In one embodiment, the construction unit 303 is further configured to:

[0218] In the first several convolutional layers of the convolutional neural network, the overall shape, outline and large features of the surrounding environment of the electricity meter are extracted to obtain a feature map reflecting the overall appearance of the electricity meter and the general environment, thereby obtaining a shallow feature map; in the deep layer of the convolutional neural network, a smaller convolution kernel and step size are used to extract the detailed features of the electricity meter screen pattern, wording and surrounding environment, thereby obtaining a feature map containing rich detailed information, thereby obtaining a deep feature map; the shallow feature map and the deep feature map are spliced in the channel dimension, and the spliced feature map is input into the fully connected layer to output an image feature vector that integrates the electricity meter shape, outline, surrounding features and display screen details.

[0219] In one embodiment, the analysis unit 304 is configured to:

[0220] Normalize the meter reading data; adopt sliding window technology and difference method to calculate the trend slope and the change amount of adjacent time points based on the normalized meter reading data, reflecting the long-term and short-term change trends of the meter readings; calculate the standard deviation and range of the normalized meter reading data in each sliding window to obtain a meter reading data feature vector reflecting the operation status of the meter; extract the change trend characteristics and fluctuation amplitude characteristics of the temperature and humidity data in the environmental parameters respectively to obtain an environmental extraction result; splice or weightedly fuse the feature vector reflecting the operation status of the meter with the environmental extraction result to form a fused comprehensive feature vector; establish a relationship model between the meter readings and the environmental parameters based on the fused comprehensive feature vector through regression analysis and machine learning methods, and analyze the influence of environmental factors on the meter readings.

[0221] In one embodiment, the fusion unit 305 is further configured to:

[0222] The comprehensive visual feature vector, the electric meter reading data feature vector, and the feature vectors corresponding to the environmental parameters are respectively normalized; for the normalized vectors, the number of channels of the feature vector corresponding to the environmental parameters is adjusted, and the spatial dimensions of the comprehensive visual feature vector and the feature vectors corresponding to the environmental parameters are adapted; the three adjusted feature vectors are spliced in the channel dimension to form a fused feature vector; a convolution operation is performed on the fused feature vector to extract multi-scale features; the multi-scale features are encoded using multiple feature encoding layers to extract local features and reduce the spatial dimension, and features at different levels are weightedly fused during the encoding process to obtain an encoded feature map; the encoded feature map is input into a decoupling head for prediction, and a prediction result that comprehensively reflects the state of the electric meter is output.

[0223] In one embodiment, the calculation and judgment unit 306 is configured to:

[0224] Establish a description set of the normal operating state characteristics and fault state characteristics of the meter; for the description set, use the text thinking chain method to simplify the complex description into an expression composed of simple words and relationships; perform word segmentation on the simplified description set and convert it into a word vector, and then use the multi-head self-attention mechanism to generate a text feature vector; calculate the cosine similarity between the prediction result that comprehensively reflects the meter status and the text feature vectors of the normal and fault states to obtain a calculation result; determine whether there is a fault risk based on the calculation result and issue a warning information to obtain a preliminary recognition result.

[0225] In one embodiment, the evaluation unit 307 is configured to:

[0226] Assign a unique ID to the preliminary recognition result in each frame of the image; calculate the correlation between the preliminary recognition results of the same ID; update or confirm the preliminary recognition result of the current frame based on the preliminary recognition results of the subsequent frame; when the recognition result meets the fault characteristics and the correlation is high, the credibility of the warning information is enhanced; if the correlation is low, it is determined to be a misidentification, and the warning information is downgraded or discarded; comprehensively analyze the suspected fault results of the similarity calculation and the results after ID tracking optimization, and combine them with the historical operation data and fault records of the meter; use machine learning algorithms for re-evaluation, analyze the importance of features to accurately determine the fault type and risk level, and optimize the final fault warning result.

[0227] In one embodiment, the system further comprises:

[0228] The suggestion unit is used to give specific maintenance suggestions based on the final fault warning results, including on-site inspection, component replacement or complete replacement.

[0229] It should be noted that technical personnel in the relevant field can clearly understand that the specific implementation process of the above-mentioned smart meter fault warning system 300 based on multi-parameter synchronous measurement and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and conciseness of the description, it will not be repeated here.

[0230] The smart meter fault warning system 300 based on multi-parameter synchronous measurement can be implemented in the form of a computer program. Figure 3 Runs on the computer equipment shown.

[0231] See also Figure 3 , Figure 3 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0232] See Figure 3The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0233] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, enable the processor 502 to execute a smart meter fault early warning method based on multi-parameter synchronous measurement.

[0234] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0235] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a smart meter fault early warning method based on multi-parameter synchronous measurement.

[0236] The network interface 505 is used to communicate with other devices through the network. Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0237] The processor 502 is configured to run a computer program 5032 stored in a memory to implement all steps of the smart meter fault early warning method based on multi-parameter synchronous measurement.

[0238] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0239] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0240] Therefore, the present invention further provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes all steps of the smart meter fault early warning method based on multi-parameter synchronous measurement.

[0241] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0242] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0243] In the several embodiments provided herein, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementations may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0244] The steps in the method of the embodiment of the present invention may be adjusted in order, combined, or deleted as needed. The units in the system of the embodiment of the present invention may be combined, divided, or deleted as needed. In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0245] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (such as a personal computer, terminal, or network device) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0246] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A smart meter fault warning method based on multi-parameter synchronous measurement, characterized in that: include: Synchronously collect meter operation videos, meter reading data, and environmental parameters; Performing lighting compensation and optimization on the electricity meter operation video; The image features of the lighting-optimized electricity meter operation video are extracted through convolutional and fully connected layers. A cross-attention mechanism is then combined with the environmental parameters to construct a comprehensive visual feature vector containing both image information and environmental factors. During the extraction process, a hierarchical approach is used to first extract the meter's overall shape, outline, and surrounding large features. The pattern and text on the meter's display, as well as detailed environmental features of the meter, are then extracted to produce the image feature vector. Extracting change trends and fluctuation amplitude characteristics from the meter reading data using a time series analysis method to form a feature vector of the meter reading data, and analyzing the potential impact of environmental factors on the meter reading data; Using multiple feature encoding layers to fuse the comprehensive visual feature vector, the meter reading data feature vector, and the feature vectors corresponding to the environmental parameters, to generate a prediction result that comprehensively reflects the meter status; The prediction results that comprehensively reflect the status of the electricity meter and the feature description sets of normal and fault states are calculated for similarity to determine whether there is a fault risk and issue early warning information to obtain preliminary identification results; Initial identification results are evaluated through ID matching and machine learning algorithms, and historical data is used to optimize the final fault warning results.

2. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 1 is characterized in that: The performing illumination compensation and optimization on the electric meter operation video includes: Each frame of the meter operation video is input into a lighting adjustment model for lighting compensation and optimization, wherein the lighting adjustment model includes a local branch network for enhancing local feature details and a global branch network for generating a color correction matrix and gamma correction value and adjusting the entire meter image.

3. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 1 is characterized in that: The image features of the meter operation video after lighting optimization are extracted through the convolutional layer and the fully connected layer, and the cross-attention mechanism is combined with the environmental parameters to construct a comprehensive visual feature vector containing image information and environmental factors, including: The video of the electricity meter running after lighting optimization is input into the convolutional neural network, which extracts image features through sliding convolution kernels and uses activation functions to introduce nonlinearity to process complex feature patterns. The feature map extracted by the convolutional layer is flattened and input into the fully connected layer to learn the global feature representation of the image and output a feature vector containing the meter shape, display digits and symbols to obtain the image feature vector; Encoding the environmental parameters into a vector form; The encoded environmental parameter vector and the image feature vector are input into the cross-attention mechanism, the correlation is calculated to determine the influence weight of the environmental factors, and the image feature vector is weighted and adjusted according to the influence weight of the environmental factors to construct a comprehensive visual feature vector that integrates the image and environmental information.

4. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 3 is characterized in that: The hierarchical method first extracts the overall shape, outline and surrounding features of the meter, and then extracts the pattern and words on the meter display screen and the detailed features of the meter's environment to obtain an image feature vector, including: In the first several convolutional layers of the convolutional neural network, the overall shape and outline of the electric meter and the large features of the surrounding environment are extracted to obtain a feature map reflecting the overall appearance of the electric meter and the general environment, thereby obtaining a shallow feature map; In the deep layer of the convolutional neural network, a smaller convolution kernel and step size are used to extract the detailed features of the pattern, words and surrounding environment of the electronic display screen, so as to obtain a feature map containing rich detailed information, thereby obtaining a deep feature map; The shallow feature map and the deep feature map are spliced in the channel dimension, and the spliced feature map is input into the fully connected layer to output an image feature vector that integrates the meter shape, contour, surrounding features and display screen details.

5. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 1 is characterized in that: The method of extracting the change trend and fluctuation amplitude characteristics from the meter reading data using a time series analysis method to form a feature vector of the meter reading data, and analyzing the potential impact of environmental factors on the meter reading data, includes: Normalizing the electric meter reading data; Using sliding window technology and difference method, the trend slope and the change amount at adjacent time points are calculated based on the normalized meter reading data to reflect the long-term and short-term change trends of the meter readings. Within each sliding window, the standard deviation and range of the normalized meter reading data are calculated to obtain a meter reading data feature vector reflecting the operating status of the meter; Extracting the change trend characteristics and fluctuation amplitude characteristics of the temperature and humidity data in the environmental parameters respectively to obtain an environmental extraction result; Concatenating or weightedly fusing the feature vector reflecting the operating state of the electric meter with the environmental extraction result to form a fused comprehensive feature vector; Through regression analysis and machine learning methods, a relationship model between meter readings and environmental parameters is established based on the fused comprehensive feature vector, and the influence of environmental factors on meter readings is analyzed.

6. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 1 is characterized in that: The method of using multiple feature coding layers to fuse the comprehensive visual feature vector, the feature vector of the meter reading data, and the feature vector corresponding to the environmental parameters to generate a prediction result that comprehensively reflects the state of the meter includes: Normalizing the comprehensive visual feature vector, the electric meter reading data feature vector, and the feature vector corresponding to the environmental parameters respectively; For the normalized vector, adjusting the number of channels of the feature vector corresponding to the environmental parameter, and performing spatial size adaptation on the comprehensive visual feature vector and the feature vector corresponding to the environmental parameter; Concatenate the three adjusted feature vectors in the channel dimension to form a fused feature vector; Performing a convolution operation on the fused feature vector to extract multi-scale features; Using multiple feature coding layers to encode the multi-scale features, extract local features and reduce the spatial size, and perform weighted fusion of features at different levels during the encoding process to obtain an encoded feature map; The encoded feature map is input into the decoupling head for prediction, and the output is a prediction result that comprehensively reflects the meter status.

7. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 1 is characterized in that: The prediction result reflecting the status of the electric meter and the set of normal and fault state feature descriptions are comprehensively calculated for similarity, and whether there is a fault risk is determined and an early warning message is issued to obtain a preliminary identification result, including: Establish a description set of the normal operating state characteristics and fault state characteristics of the electricity meter; For the description set, the text thinking chain method is used to simplify the complex description into expressions composed of simple words and relationships; The simplified description set is segmented and converted into word vectors, and then a multi-head self-attention mechanism is used to generate text feature vectors; The prediction result that comprehensively reflects the meter status is calculated by cosine similarity calculation with the text feature vectors of normal and fault status to obtain the calculation result; Based on the calculation results, it is determined whether there is a failure risk and an early warning message is issued to obtain a preliminary identification result.

8. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 1 is characterized in that: The ID matching and machine learning algorithm are used to evaluate the preliminary identification results, and the historical data is used to optimize the final fault warning results, including: Assigning a unique ID to the preliminary recognition result in each frame of the image; Calculating the correlation between the preliminary identification results of the same ID; updating or confirming the preliminary recognition result of the current frame according to the preliminary recognition result of the subsequent frame; If the recognition result matches the fault characteristics and the correlation is high, the credibility of the warning information is enhanced; if the correlation is low, it is determined to be a misidentification and the warning information is downgraded or discarded; The suspected fault results from similarity calculation and the results after ID tracking optimization are comprehensively analyzed and combined with the historical operating data and fault records of the meter. A machine learning algorithm is used for re-evaluation, and the importance of features is analyzed to accurately determine the fault type and risk level, thereby optimizing the final fault warning results.

9. The smart meter fault warning method based on multi-parameter synchronous measurement according to claim 1 is characterized in that: After evaluating the preliminary identification results through ID matching and machine learning algorithms and optimizing the final fault warning results using historical data, the following also applies: Specific maintenance recommendations are given based on the final fault warning results, including on-site inspection, component replacement or complete replacement.

10. The intelligent meter fault warning system based on multi-parameter synchronous measurement is characterized by: include: A collection unit, used to synchronously collect meter operation videos, meter reading data, and environmental parameters; an optimization unit, configured to perform illumination compensation and optimization on the electricity meter operation video; A construction unit is configured to extract image features from the illumination-optimized electricity meter operation video using convolutional layers and fully connected layers, and utilize a cross-attention mechanism in combination with the environmental parameters to construct a comprehensive visual feature vector comprising image information and environmental factors. During the extraction process, a hierarchical approach is employed to first extract the overall shape, outline, and surrounding macro features of the electricity meter, and then extract the pattern and text on the meter display screen and detailed environmental features of the meter to obtain an image feature vector. an analysis unit, configured to extract change trends and fluctuation amplitude characteristics from the meter reading data using a time series analysis method, form a feature vector of the meter reading data, and analyze the potential impact of environmental factors on the meter reading data; a fusion unit, configured to fuse the comprehensive visual feature vector, the meter reading data feature vector, and the feature vector corresponding to the environmental parameters using multiple feature coding layers to generate a prediction result that comprehensively reflects the state of the meter; A calculation and judgment unit is used to perform similarity calculation on the prediction result that comprehensively reflects the state of the electric meter and the normal and fault state feature description set, determine whether there is a fault risk and issue an early warning information to obtain a preliminary identification result; The evaluation unit is used to evaluate the preliminary identification results through ID matching and machine learning algorithms, and use historical data to optimize the final fault warning results.

Images