Multi-mode electricity utilization data fusion electricity meter metering abnormity identification method and system

Abstract

The invention provides an ammeter metering abnormity identification method and system based on multi-mode power utilization data fusion, and relates to the technical field of electric power metrology, and the method comprises the steps: collecting an actual power consumption set of intelligent power utilization equipment of a target user in a preset time zone, and carrying out the matching to obtain an equipment association degree set; according to the actual power consumption set and the equipment association degree set, predicting to obtain predicted total power consumption; performing analysis based on the intelligent electric equipment type set and the equipment association degree set to obtain prediction confidence; calculating a threshold adjustment coefficient according to the load fluctuation characteristics, the environment fluctuation characteristics and the prediction confidence in the preset time zone, correcting the initial error threshold, and outputting an adaptive error threshold; and calculating the electric quantity deviation amplitude according to the meter display consumed electric quantity of the electric energy meter of the target user in the preset time zone and the predicted total consumed electric quantity, and if the electric quantity deviation amplitude exceeds an adaptive error threshold, executing abnormal early warning. The technical problems of high false alarm rate and poor reliability of electricity meter metering abnormity identification in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electric power metering, in particular to a metering anomaly identification method and system for multi-modal power consumption data fusion. BACKGROUND

[0002] Metering anomaly identification of electric meters can timely correct metering deviations, avoid users mispaying fees due to electric meter faults or errors, reduce fee payment disputes, quickly locate metering anomaly points, reduce manual troubleshooting costs, improve operational efficiency, and also can early warn of electric meter aging, line hazards and other issues, and prevent power outages, fires and other safety accidents.

[0003] However, traditional electric metering anomaly identification usually relies on single displayed electric quantity data, is easily disturbed by factors such as power consumption equipment linkage and environmental changes, and has problems of high false positive rate and poor reliability.

[0004] There is an urgent need for a method that can fuse multi-modal power consumption data of users to achieve accurate electric metering anomaly identification. SUMMARY

[0005] The present application provides a metering anomaly identification method and system for multi-modal power consumption data fusion to solve the technical problems of high false positive rate and poor reliability in existing electric metering anomaly identification.

[0006] The technical solutions of the present application to solve the above technical problems are as follows: In a first aspect, the present application provides a metering anomaly identification method for multi-modal power consumption data fusion, comprising: collecting a set of actual consumption electric quantities of intelligent power consumption equipment of a target user in a preset time zone, and matching to obtain a set of device correlation degrees in the preset time zone; predicting and obtaining a predicted total consumption electric quantity of the target user in the preset time zone according to the set of actual consumption electric quantities and the set of device correlation degrees; analyzing to obtain a prediction confidence based on a set of intelligent power consumption equipment types and the set of device correlation degrees; calculating a threshold adjustment coefficient based on load fluctuation characteristics, environmental fluctuation characteristics in the preset time zone, and the prediction confidence, correcting an initial error threshold, and outputting an adaptive error threshold; calculating an electric quantity deviation amplitude based on a displayed consumption electric quantity of an electric energy meter of the target user in the preset time zone and the predicted total consumption electric quantity, and if the electric quantity deviation amplitude exceeds the adaptive error threshold, performing metering anomaly warning of the electric energy meter.

[0007] In a second aspect, the present application provides a metering anomaly identification system for multi-modal power consumption data fusion, comprising: The data acquisition module is configured to acquire an actual consumption power set of the smart power consumption equipment of the target user in a preset time zone and match to obtain a device correlation degree set in the preset time zone; The power prediction module is configured to predict and obtain a predicted total consumption power of the target user in the preset time zone according to the actual consumption power set and the device correlation degree set; The confidence degree acquisition module is configured to analyze and obtain a prediction confidence degree based on the smart power consumption equipment type set and the device correlation degree set; The threshold correction module is configured to calculate a threshold adjustment coefficient according to a load fluctuation feature, an environmental fluctuation feature in the preset time zone and the prediction confidence degree, correct an initial error threshold and output an adaptive error threshold; The early warning output module is configured to calculate a power deviation amplitude according to an indicated consumption power of the electric energy meter of the target user in the preset time zone and the predicted total consumption power, and perform metering abnormality early warning of the electric energy meter if the power deviation amplitude exceeds the adaptive error threshold.

[0008] The application has the following advantages: Compared with the prior art, the application first acquires an actual consumption power set of the smart power consumption equipment of the target user in a preset time zone and matches to obtain a device correlation degree set in the preset time zone, thereby providing a reliable data basis for subsequent model construction, abnormality identification and early warning. Secondly, the predicted total consumption power of the target user in the preset time zone is predicted and obtained according to the actual consumption power set and the device correlation degree set, and the actual consumption power set and the device correlation degree set are used to predict and output a reasonable total power consumption in the preset time zone, which is a reference value for subsequent abnormality identification. Thirdly, the prediction confidence degree is analyzed based on the smart power consumption equipment type set and the device correlation degree set, the prediction reliability of the current scene is evaluated based on historical data, the initial prediction confidence degree is obtained, and the prediction confidence degree reflecting the real prediction reliability of the current scene is obtained by combining the model prediction accuracy for correction, thereby improving the robustness of abnormality identification. Further, the threshold adjustment coefficient is calculated according to the load fluctuation feature, the environmental fluctuation feature in the preset time zone and the prediction confidence degree, the initial error threshold is corrected, and the adaptive error threshold is output, the threshold adjustment coefficient is obtained by quantifying the comprehensive fluctuation of the power consumption load, the environmental temperature and the prediction result reliability, the initial error threshold is amplified to reduce the misjudgment when the fluctuation is large, and the initial error threshold is reduced to improve the sensitivity when the fluctuation is small, thereby solving the problem that the traditional fixed threshold value is difficult to adapt to the complex power consumption scene. Finally, the power deviation amplitude is calculated according to the indicated consumption power of the electric energy meter of the target user in the preset time zone and the predicted total consumption power, and the metering abnormality early warning of the electric energy meter is performed if the power deviation amplitude exceeds the adaptive error threshold, thereby realizing the abnormality identification and early warning of the electric meter metering.

[0009] According to the technical scheme, the total consumed power in the current period is predicted according to the actual use of the smart power consumption equipment by the target user in the preset time zone, the confidence degree of the prediction result is analyzed, the initial error threshold is dynamically adjusted according to the fluctuation characteristics of the multi-modal data such as the environment and the prediction confidence degree, the adaptive error threshold is obtained, and the metering result deviation of the electric meter is identified and judged, so that the accuracy and reliability of the metering anomaly identification of the electric meter are improved. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A flowchart of a multi-modal power consumption data fusion electric metering anomaly identification method provided by the present application is shown. Figure 2 A structure diagram of a multi-modal power consumption data fusion electric metering anomaly identification system provided by the present application is shown.

[0011] In the drawings, the components represented by the numbers are as follows: The data acquisition module 11, the power prediction module 12, the confidence degree acquisition module 13, the threshold correction module 14, and the early warning output module 15. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0013] In the description of the present application, the terms "first" and "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0014] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0015] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for identifying meter metering anomalies based on multimodal electricity consumption data fusion, including: S10: Collect the actual power consumption set of the target user's smart electrical devices within the preset time zone, and match and obtain the device correlation set within the preset time zone; Traditional methods for identifying meter anomalies rely heavily on single meter readings, which are easily affected by factors such as user equipment linkage, load fluctuations, and environmental changes, resulting in high false alarm rates and poor reliability. To identify meter anomalies, it is necessary to first obtain the actual power consumption characteristics of smart devices.

[0016] To address the aforementioned issues, this application collects the actual power consumption set of smart electrical devices of target users within a preset time zone and matches it to obtain the device correlation set within the preset time zone.

[0017] Specifically, step S10 in the method includes: Collect the device types and actual power consumption of smart electrical devices used by the target user within a preset time zone to obtain a set of smart electrical device types and a set of actual power consumption. The smart electrical devices have energy consumption monitoring functions, and by default, the user is equipped with at least one smart electrical device. Using the time period-device correlation mapping table, the device correlation set is obtained by matching the preset time zone and the set of smart electrical equipment types.

[0018] In this embodiment, the device types and actual power consumption of smart devices used by the target user within a preset time zone are first collected to obtain a set of smart device types and a set of actual power consumption. The preset time zone is a relatively long time interval, such as 12 hours or a day. The smart devices have energy consumption monitoring functions, and by default, the user is equipped with at least one smart device. For example, if the preset time zone is a day, the device types of smart devices used by the target user within a day are collected, such as smart air conditioners, smart refrigerators, and smart water heaters, as well as the actual power consumption directly obtained from the smart devices, such as 10 kWh / day for smart air conditioners, 1 kWh / day for smart refrigerators, and 2 kWh / day for smart water heaters. This results in a set of smart device types: {smart air conditioners, smart refrigerators, smart water heaters, ...}, and a corresponding set of actual power consumption: {smart air conditioners 10 kWh / day, smart refrigerators 1 kWh / day, smart water heaters 2 kWh / day, ...}. These two sets constitute a basic profile of the user's electricity consumption behavior.

[0019] Secondly, using a time-cycle-device correlation mapping table, a device correlation set is obtained by matching the preset time zone and the set of smart electrical device types. The time-cycle-device correlation mapping table records the contribution weight of the power consumption of various smart electrical devices to the total power consumption over different time periods (e.g., monthly, quarterly). For example, if the preset time zone is a day in July and the set of smart electrical device types is {smart air conditioner, smart refrigerator, smart water heater}, then querying the time-cycle-device correlation mapping table yields a device correlation set of {0.5, 0.048, 0.1}. This device correlation set reflects the power consumption characteristics of various smart electrical devices within the preset time zone.

[0020] Specifically, the method for constructing the "time period-equipment correlation mapping table" includes: The year is divided according to a preset time step, multiple time periods are obtained, and the first time period is randomly selected. Based on the preset types of smart electrical devices, the average historical power consumption of several smart electrical devices in the target area within the first time period is calculated, and the average historical total power consumption is obtained. The ratio of the historical average power consumption to the historical average total power consumption is set as the device correlation degree, and a first device correlation degree set is calculated based on the several historical average power consumption values. By sequentially analyzing multiple device correlation sets over multiple time periods, a time period-device correlation mapping table is constructed.

[0021] In this embodiment, due to changes in environmental factors such as season and climate, the power consumption characteristics of the same smart electrical device will vary significantly across different time periods. For example, smart heating devices are used frequently in spring and winter, but very infrequently in summer and autumn. To improve accuracy when identifying meter reading anomalies, it is necessary to capture the temporal periodicity of power consumption by smart electrical devices. Furthermore, device correlation is the proportion of power consumption by smart electrical devices over a long period, reflecting the contribution weight of various smart electrical devices to the total power consumption. Moreover, device correlation exhibits a strong temporal periodicity; for example, smart heating devices are used frequently in spring and winter with a high device correlation, while their usage frequency is low in summer and autumn with a low device correlation. Therefore, a time period-device correlation mapping table can be constructed to capture the temporal periodicity of power consumption by smart electrical devices. Specifically: First, the year is divided according to a preset time step to obtain multiple time periods, and the first time period is randomly selected. The preset time step can be a relatively long time interval such as a month or a quarter. For example, the year is divided according to a preset time step (such as a month) to obtain 12 time periods, and July is randomly selected as the first time period.

[0022] Secondly, based on preset smart device types, the average historical power consumption of several smart devices in the target area during the first time period is statistically analyzed, and the average historical total power consumption is obtained. The preset smart device types can be determined based on the aforementioned set of smart device types. The target area includes the target user's residential community, building, and floor, etc., which can be dynamically selected by those skilled in the art based on actual circumstances. For example, based on the smart device type set, the preset smart device types are determined to be smart air conditioners, smart refrigerators, smart water heaters, etc. Then, the average historical power consumption of smart air conditioners, smart refrigerators, smart water heaters, etc., in the target area (e.g., the target user's residential community) during the first time period (e.g., July) is statistically analyzed. For example, the average power consumption of smart air conditioners is 250 kWh / month, smart refrigerators 24 kWh / month, and smart water heaters 50 kWh / month. The average historical total power consumption of the target area is then obtained as 500 kWh / month.

[0023] Next, the ratio of the historical average power consumption to the historical average total power consumption is set as the device correlation degree. A first device correlation degree set is calculated based on the several historical average power consumption values, where device correlation degree = historical average power consumption / historical average total power consumption. For example, if the average power consumption of a smart air conditioner in the target area is 250 kWh / month, a smart refrigerator is 24 kWh / month, and a smart water heater is 50 kWh / month, and the historical average total power consumption is 500 kWh / month, then the device correlation degree of the smart air conditioner = 250 / 500 = 0.5, the device correlation degree of the smart refrigerator = 24 / 500 = 0.048, and the device correlation degree of the smart air conditioner = 50 / 500 = 0.1, resulting in the first device correlation degree set {0.5, 0.048, 0.1}. The device correlation degree reflects the contribution weight of the power consumption of various smart electrical devices to the total power consumption. The larger the device correlation degree, the greater the contribution weight of the power consumption of that smart electrical device to the total power consumption.

[0024] Finally, multiple device correlation sets for various time periods are analyzed sequentially to construct a time period-device correlation mapping table. For example, following the same method used to obtain the first device correlation set in the preceding steps, all time periods are traversed, and the statistics and calculations are repeated. Finally, the device correlation sets for all periods are integrated to form a complete time period-device correlation mapping table. For instance, the correlation of smart air conditioners in July is 0.5, reflecting increased air conditioner usage frequency in summer, while the correlation of smart air conditioners in October is 0.05, reflecting decreased air conditioner usage frequency in autumn. Thus, the time period-device correlation mapping table can capture the seasonal and periodic patterns of power consumption of smart electrical devices, enabling subsequent anomaly identification models to dynamically adapt to the power consumption characteristics of different time periods and improve prediction accuracy.

[0025] In summary, compared to existing technologies, this application collects the actual power consumption set of smart devices of target users within a preset time zone and matches it to obtain the device correlation set within the preset time zone. This provides a reliable data foundation for subsequent model building and anomaly identification and early warning.

[0026] S20: Based on the actual power consumption set and the device correlation set, predict the total power consumption of the target user in the preset time zone; When identifying meter reading anomalies, it is necessary to obtain the predicted total electricity consumption of the target user within a preset time zone. This prediction is used as a benchmark for anomaly identification and compared with the electricity consumption displayed on the meter to determine whether there is an anomaly in the meter reading.

[0027] To address the aforementioned issues, this application predicts the power consumption of a target user within a preset time zone based on the actual power consumption set and the device correlation set.

[0028] Specifically, step S20 in the method includes: Based on the target user's historical electricity consumption logs, multiple sample power consumption sets and multiple sample device correlation sets are collected. The historical total power consumption of different sample power consumption sets and sample device correlation sets in historical time zones is collected and set as the sample total power consumption set. The time interval between the historical time zone and the preset time zone is the same. Using the multiple sets of power consumption from samples and the multiple sets of device correlation from samples as input, and using the total power consumption set of samples as supervision, the machine learning model is trained until convergence to obtain a power consumption predictor, and the prediction accuracy of the model is verified. Using the power consumption predictor, the predicted total power consumption of the target user within a preset time zone is obtained based on the actual power consumption set and the device correlation set.

[0029] In this embodiment of the application, firstly, based on the target user's historical electricity consumption logs, multiple sample electricity consumption sets and multiple sample device correlation sets are collected, and the historical total electricity consumption of different sample electricity consumption sets and sample device correlation sets in historical time zones is collected and set as the sample total electricity consumption set. The time interval between the historical time zone and the preset time zone is the same. The historical total electricity consumption can be monitored by equipment such as transformers, feeders, and distribution boxes in the power grid to obtain the total electricity consumption of a certain area or community, and the total electricity consumption of each household can be obtained. For example, if the preset time zone interval is one day in July, the actual electricity consumption of multiple smart devices in July of the past three years is collected from the target user's historical electricity consumption logs at daily intervals. For example, the actual electricity consumption on a certain day in July in history is used to obtain a sample electricity consumption set {smart air conditioner 12 kWh, smart refrigerator 0.9 kWh, smart water heater 3 kWh}, and the device correlation degree on that day is used to obtain a sample device correlation degree set {smart air conditioner correlation degree 0.51, smart refrigerator correlation degree 0.05, smart air conditioner correlation degree 0.08}. Then, the historical total electricity consumption corresponding to the sample electricity consumption set and the sample device correlation degree set is collected through power grid monitoring data as 20 kWh to obtain the sample total electricity consumption. This step is repeated to obtain multiple sample electricity consumption sets, multiple sample device correlation degree sets, and the sample total electricity consumption set.

[0030] Secondly, using the multiple sample power consumption sets and multiple sample device correlation sets as input, and the total sample power consumption set as supervision, the machine learning model is trained until convergence to obtain the power consumption predictor, and the model prediction accuracy is verified. For example, the training process of the power consumption predictor can be implemented through the following technical path: 1. Training data preparation: Divide the multiple sample power consumption sets, multiple sample device correlation sets, and the total sample power consumption set into training set, validation set, and test set in a ratio of 7:1.5:1.5. 2. Model construction: A prediction model can be constructed using a Long Short-Term Memory (LSTM) network, mainly composed of an input layer, hidden layer, and output layer. The input layer receives the actual power consumption set and device correlation set. The hidden layer uses bidirectional LSTM units (with 128 hidden neurons) to process temporal information and dynamically weights the influence of different devices through an attention mechanism. Then, a Dropout layer (ratio 0.3) is connected to prevent overfitting and enhance generalization ability. The output layer outputs the predicted total power consumption of the user within a preset time zone through fully connected neurons. 3. Model Training: Multiple sample power consumption sets and multiple sample device correlation sets from the training set are input into the model. The total power consumption set is used as supervision. The Adam optimizer (initial learning rate 0.001) and mean squared error (MSE) loss function are used for training. LSTM weights, attention weights, and fully connected layer parameters are updated through backpropagation. Early stopping strategies (stopping if there is no improvement in loss after 5 consecutive validation set iterations) and a learning rate decay mechanism (halving the learning rate during loss plateaus) can be introduced to improve model stability. After multiple iterations of training and testing, the model is considered converged when its prediction accuracy on the validation set reaches over 95%, resulting in a trained power consumption predictor. Further, the model's prediction accuracy is verified using an independent test set. For example, the power consumption predictor is invoked, inputting multiple sample power consumption sets and multiple sample device correlation sets from the test set. Multiple total power consumption predictions are generated, and by comparing these with the total power consumption sets in the test set, the model's prediction accuracy is calculated to be 96%.

[0031] For example, a pre-trained power consumption predictor is invoked, and the actual power consumption set (e.g., {smart air conditioner 10 kWh / day, smart refrigerator 1 kWh / day, smart water heater 2 kWh / day}) and the device correlation set (e.g., {0.5, 0.048, 0.1}) are input to predict the total power consumption of the target user in the preset time zone as 20 kWh.

[0032] In summary, compared to existing technologies, this application predicts the target user's power consumption within a preset time zone based on the actual power consumption set and the device correlation set. Thus, the predicted total power consumption within the preset time zone, based on the actual power consumption set and the device correlation set, serves as the benchmark for subsequent anomaly identification.

[0033] S30: The prediction confidence level is obtained based on the analysis of the set of smart electrical equipment types and the set of equipment correlation. Traditional prediction confidence assessments often rely on the overall prediction accuracy of the model, but ignore the differences in the model in specific scenarios. The same model may have significant differences in prediction performance under different device types and scenarios with different degrees of correlation. For example, a model may have a prediction accuracy of 95% for low-power devices, but its prediction accuracy for high-power, highly correlated devices may only be 70%, which makes the confidence level unable to reflect the true prediction reliability of the current scenario.

[0034] To address the aforementioned issues, this application analyzes the set of smart electrical device types and the set of device correlations to obtain the prediction confidence level.

[0035] Specifically, step S30 in the method includes: Using the set of smart electrical equipment types and the set of equipment correlation as comparison constraints, historical power consumption monitoring logs of the target area are retrieved, and several historical predicted total power consumption and several historical actual power consumption are collected. Several prediction deviation ranges are calculated based on the total historical predicted power consumption and the actual historical power consumption. The percentage of prediction deviation ranges that are less than a preset deviation index is counted and set as the initial prediction confidence level. The initial prediction confidence is corrected based on the model's prediction accuracy, and the prediction confidence is output.

[0036] In this embodiment, the historical power consumption monitoring logs of the target area are first retrieved using the set of smart electrical device types and the set of device correlation as comparison constraints. Several historical predicted total power consumption and several historical actual power consumption are collected. The historical power consumption monitoring logs contain several sets of smart electrical device types and device correlation sets of different users in the target area during historical time, as well as the historical predicted total power consumption and historical actual power consumption output by the corresponding power consumption predictor. The historical actual power consumption can be monitored by devices such as transformers, feeders, and distribution boxes in the power grid to obtain the total power consumption of a certain area or community, and the total power consumption of each household can be obtained. For example, using a set of smart electrical device types (such as {smart air conditioner, smart refrigerator, smart water heater}) and a set of device correlation degrees (such as {0.5, 0.048, 0.1}) as comparison constraints, historical power consumption monitoring logs of the target area (such as the community where the target user is located) are retrieved, and several historical predicted total power consumption (such as 21 kWh, 20.5 kWh, 22 kWh, 19.8 kWh, 21.3 kWh, ...) and several corresponding historical actual power consumption (such as 20 kWh, 22 kWh, 23 kWh, 21 kWh, 20 kWh, ...) are collected.

[0037] Secondly, based on the aforementioned historical predicted total power consumption and the aforementioned historical actual power consumption, several prediction deviation ranges are calculated. The percentage of prediction deviation ranges less than a preset deviation index is counted and set as the initial prediction confidence level. The prediction deviation range is calculated as follows: Prediction deviation range = (Historical predicted total power consumption - Historical actual power consumption) / Historical actual power consumption. The preset deviation index indicates that prediction results deviating from this range are acceptable. This application recommends setting it to 0.06, but those skilled in the art can adjust it dynamically according to the actual situation. For example, there are 100 historical predicted total power consumption data points: 21 kWh, 20.5 kWh, 22 kWh, 19.8 kWh, 21.3 kWh, etc., and corresponding historical actual power consumption data points: 20 kWh, 22 kWh, 23 kWh, 21 kWh, 20 kWh, etc., also totaling 100 data points. Then, the prediction deviation range = 21 kWh - 20 kWh / 20 kWh = 0.05, the prediction deviation range = 20.5 kWh - 22 kWh / 22 kWh = 0.068, etc., totaling 100 data points. If the preset deviation index is 0.06, and there are 82 prediction deviation ranges less than the preset deviation index, then the initial prediction confidence level = 82 / 100 = 0.82. The initial prediction confidence level can reflect the prediction reliability of the power consumption predictor in similar historical scenarios. The higher the initial prediction confidence level, the more reliable the prediction result.

[0038] Finally, the initial prediction confidence is corrected based on the model prediction accuracy, and the prediction confidence is output. The prediction confidence = model prediction accuracy * initial prediction confidence. The model prediction accuracy reflects the prediction accuracy and generalization ability of the power consumption predictor itself. Correcting the initial prediction confidence based on the model prediction accuracy combines the prediction reliability of the power consumption predictor in similar historical scenarios with the model's generalization ability. For example, if the model prediction accuracy is 96% and the initial prediction confidence is 0.82, then the prediction confidence = 96% * 0.82 = 0.787. The prediction confidence reflects the true prediction reliability of the current scenario. A higher prediction confidence indicates stronger reliability of the current predicted total power consumption, and vice versa.

[0039] In summary, compared to existing technologies, this application obtains the prediction confidence level based on the analysis of a set of intelligent electrical device types and the set of device correlation. Thus, based on historical data, the prediction reliability of the current scenario is specifically evaluated to obtain an initial prediction confidence level. This is then corrected by combining the model's prediction accuracy to obtain a prediction confidence level that reflects the true prediction reliability of the current scenario, thereby improving the robustness of anomaly identification.

[0040] S40: Calculate the threshold adjustment coefficient based on the load fluctuation characteristics, environmental fluctuation characteristics and the prediction confidence level in the preset time zone, correct the initial error threshold, and output the adaptation error threshold; The deviation between the displayed electricity consumption and the predicted total electricity consumption is calculated. If the deviation exceeds the allowable error threshold, the electricity meter is considered to be malfunctioning. This error threshold should not be fixed because actual load fluctuations, environmental fluctuations, and prediction confidence all affect the predicted total electricity consumption. Specifically: Load fluctuation characteristics reflect the inherent stability of electricity consumption behavior. When a user's electricity load fluctuates drastically, it means that the electricity consumption pattern is highly random. For example, household users frequently start and stop appliances such as water heaters and induction cookers. This unstable electricity consumption behavior will reduce the reference value of historical electricity consumption patterns for the present, and the prediction model will find it difficult to accurately capture instantaneous changes, resulting in increased prediction deviation. Conversely, if the load fluctuation is gentle, the electricity consumption pattern is more in line with historical patterns, and the prediction accuracy will naturally be higher.

[0041] Environmental (especially temperature) fluctuations reflect the intensity of external environmental disturbances. Sudden temperature rises and falls can directly change the operating time of equipment such as air conditioners and heaters. For example, air conditioners run continuously during hot summer days, while heating loads surge during cold waves. If environmental fluctuations are severe, the power consumption patterns of equipment will differ significantly from those of the same period in history. If the prediction model does not adequately adapt to such sudden environmental changes, the prediction bias will increase.

[0042] Prediction confidence reflects the reliability of model predictions; a low confidence level means that the model's prediction results in the current scenario are more uncertain.

[0043] To address the aforementioned issues, this application calculates a threshold adjustment coefficient based on the load fluctuation characteristics, environmental fluctuation characteristics, and prediction confidence level within a preset time zone, corrects the initial error threshold, and outputs an adaptation error threshold.

[0044] Specifically, step S40 in the method includes: The active power sequence of target users within a preset time zone is collected for load fluctuation analysis, and the load fluctuation coefficient is output. The load fluctuation coefficient is the ratio of the standard deviation of active power to the mean of active power. An environmental temperature sequence of a target area within a preset time zone is collected for environmental fluctuation analysis, and an environmental fluctuation coefficient is output. The environmental fluctuation coefficient is the ratio of the standard deviation of the environmental temperature to the mean of the environmental temperature. The threshold adjustment coefficient is calculated based on the load fluctuation coefficient, environmental fluctuation coefficient, and prediction confidence level.

[0045] In this embodiment, the active power sequence of the target user within a preset time zone is first collected for load fluctuation analysis, and a load fluctuation coefficient is output. The load fluctuation coefficient is the ratio of the standard deviation of active power to the mean of active power, i.e., load fluctuation coefficient = standard deviation of active power / mean of active power. The active power sequence is a set of active power continuously recorded in chronological order, which can be collected by a smart energy meter or a power meter. Exemplarily, the active power sequence of the target user within the preset time zone is collected at a certain sampling frequency. The sampling frequency can be determined according to the preset time zone. For example, if the preset time zone is one day, the sampling frequency can be set to 1 hour; if the preset time zone is 12 hours, the sampling frequency can be set to 0.5 hours. Those skilled in the art can dynamically set the frequency according to actual conditions. For example, if the preset time zone is one day, the active power (in kW) sequence of the target user within one day is collected at a sampling frequency of 1 hour: [1.2, 1.5, 1.8, 2.0, 1.9, 1.7, ...], which contains a total of 24 data points. Then, the data are summed and the mean is calculated to obtain an average active power of 1.5kW. The average active power reflects the average electricity consumption intensity of the user within the preset time zone. The standard deviation of the active power sequence is then calculated to be 0.75kW. The standard deviation of active power reflects the severity of power fluctuations. Finally, the load fluctuation coefficient is calculated as 0.75 / 1.5 = 0.5. The load fluctuation coefficient can reflect the overall dispersion of power deviation from the mean within the preset time zone.

[0046] Secondly, the ambient temperature sequence of the target area within a preset time zone is collected for environmental fluctuation analysis, and the ambient fluctuation coefficient is output. The ambient fluctuation coefficient is the ratio of the standard deviation of the ambient temperature to the mean of the ambient temperature, i.e., ambient fluctuation coefficient = standard deviation of ambient temperature / mean of ambient temperature. For example, the ambient temperature sequence of the target area within the preset time zone is collected at a certain sampling frequency. The sampling frequency can be determined according to the preset time zone; for example, if the preset time zone is one day, the sampling frequency can be set to 1 hour; if the preset time zone is 12 hours, the sampling frequency can be set to 0.5 hours. Those skilled in the art can dynamically set the frequency according to actual conditions. The ambient temperature can be collected by temperature sensors deployed in the target area. For example, if the preset time zone is one day, the ambient temperature (in °C) sequence of the target user within one day is collected at a sampling frequency of 1 hour: [23, 24, 27, 29, 31, 32, 33, 34, ...], which contains a total of 24 data points. Then, the average value is calculated by summing the data points, and the average ambient temperature is 29 °C. The standard deviation of the ambient temperature is then calculated to be 3.48 °C. The standard deviation of the ambient temperature reflects the degree of fluctuation of the ambient temperature. Finally, the ambient fluctuation coefficient is calculated as 3.48 / 29 = 0.12. The ambient fluctuation coefficient can reflect the degree of interference of the ambient temperature on the power supply within the preset time zone.

[0047] Finally, the threshold adjustment coefficient is calculated based on the load fluctuation coefficient, environmental fluctuation coefficient, and prediction confidence level. The threshold adjustment coefficient can comprehensively reflect the interference of the load adjustment coefficient, environmental adjustment coefficient, and confidence level adjustment coefficient on the power consumption. The larger the threshold adjustment coefficient, the more serious the interference on the power consumption. The error threshold should be appropriately relaxed to reduce misjudgment.

[0048] Specifically, the phrase "calculating the threshold adjustment coefficient based on the load fluctuation coefficient, environmental fluctuation coefficient, and prediction confidence level" includes: The ratio of the load fluctuation coefficient to the historical average load fluctuation coefficient of the target user is set as the load adjustment coefficient; The ratio of the environmental fluctuation coefficient to the average historical environmental fluctuation coefficient of the target area is set as the environmental adjustment coefficient. The ratio of the historical predicted confidence level of the target user to the predicted confidence level is set as the confidence level adjustment coefficient; The threshold adjustment coefficient is calculated by weighting the load adjustment coefficient, environmental adjustment coefficient, and confidence adjustment coefficient.

[0049] In this embodiment, the ratio of the load fluctuation coefficient to the historical average load fluctuation coefficient of the target user is first set as the load adjustment coefficient, where the load adjustment coefficient = load fluctuation coefficient / historical average load fluctuation coefficient. If the load adjustment coefficient > 1, it indicates that the current load fluctuation is more severe than the historical average, and the threshold needs to be relaxed to reduce misjudgment; conversely, the threshold should be tightened. The historical average load fluctuation coefficient can be obtained by collecting several load fluctuation coefficients over a historical period and calculating the average value using the same method as the aforementioned steps. For example, if the historical average load fluctuation coefficient of the target user over the past month is 0.4, reflecting the long-term stability of the target user's electricity consumption, and the load fluctuation coefficient of the current preset time zone is 0.5, then the load adjustment coefficient = 0.5 / 0.4 = 1.25, indicating that the current user's electricity load fluctuation is more severe than the historical average level, and the interference of load fluctuation on electricity consumption is enhanced, so the error threshold can be appropriately relaxed.

[0050] Secondly, the ratio of the environmental fluctuation coefficient to the historical average environmental fluctuation coefficient of the target area is set as the environmental adjustment coefficient, where the environmental adjustment coefficient = environmental fluctuation coefficient / historical average environmental fluctuation coefficient. If the environmental adjustment coefficient > 1, it indicates that the current environmental temperature fluctuation is more severe than in the past, and the threshold needs to be relaxed to reduce misjudgment. The historical average environmental fluctuation coefficient can be obtained by collecting several environmental fluctuation coefficients during the same historical period and calculating the average value using the same method as the previous steps. For example, if the historical average environmental fluctuation coefficient of the target area in the same period of the past 3 years (such as July) is 0.1, reflecting the normal stability of the temperature in July in the target area, and the environmental fluctuation coefficient of the target area in the current preset time zone is 0.12, then the environmental adjustment coefficient = 0.12 / 0.1 = 1.2, indicating that the current environmental temperature fluctuation is more severe than the historical average, and the interference of temperature on electricity consumption is enhanced, so the error threshold can be appropriately relaxed.

[0051] Next, the ratio of the historical prediction confidence average to the predicted confidence level is set as the confidence adjustment coefficient, where the confidence adjustment coefficient = historical prediction confidence average / predicted confidence level. If the confidence adjustment coefficient > 1, it indicates that the current prediction confidence level is lower than the historical average level, and the threshold needs to be relaxed to reduce misjudgments. The historical prediction confidence average can be obtained by collecting several prediction confidence levels over a historical period and calculating the average, using the same method as the previous steps. For example, if the historical prediction confidence average of the target user over the past 3 months is 0.75, reflecting the average reliability of the model's historical predictions for that user, and the current prediction confidence level for the preset time zone is 0.787, then the confidence adjustment coefficient = 0.75 / 0.787 = 0.95. The confidence adjustment coefficient < 1 indicates that the current prediction reliability is better than the historical average, and the predicted value output by the model is more reliable, so the error threshold can be appropriately tightened.

[0052] Finally, the threshold adjustment coefficient is calculated by weighting the load adjustment coefficient, environmental adjustment coefficient, and confidence adjustment coefficient. The threshold adjustment coefficient is calculated as follows: Threshold adjustment coefficient = w1 * load adjustment coefficient + w2 * environmental adjustment coefficient + w3 * confidence adjustment coefficient, where w1, w2, and w3 are the weights of the load adjustment coefficient, environmental adjustment coefficient, and confidence adjustment coefficient, respectively, and w1 + w2 + w3 = 1. w1, w2, and w3 can be dynamically set according to the actual characteristics of the target user and the target area. For example, if the target area has a large temperature difference between day and night, and the electricity consumption of the target user is easily affected by the ambient temperature, the weight of the environmental adjustment coefficient w2 can be appropriately increased. Those skilled in the art can set these values ​​dynamically according to the actual situation. For example, w1, w2, and w3 can be set to 0.3, 0.4, and 0.3 respectively. For example, if w1, w2, and w3 are set to 0.3, 0.4, and 0.3 respectively, the calculated load adjustment coefficient is 1.25, the environmental adjustment coefficient is 1.2, and the confidence adjustment coefficient is 0.95. Then, the threshold adjustment coefficient = 0.3*1.25 + 0.4*1.2 + 0.3*0.95 = 1.14. The threshold adjustment coefficient can comprehensively reflect the interference of the load adjustment coefficient, environmental adjustment coefficient, and confidence adjustment coefficient on the power consumption. The larger the threshold adjustment coefficient, the more serious the interference on the power consumption. The error threshold should be appropriately relaxed to reduce misjudgment.

[0053] Furthermore, the initial error threshold is corrected by a threshold adjustment coefficient to output an adaptation error threshold, where the adaptation error threshold = threshold adjustment coefficient * initial error threshold. The initial error threshold can be dynamically set according to the actual sensitivity of the target user's electricity meter. For example, the initial error threshold can be set to ±2%, which can be dynamically set by those skilled in the art according to actual conditions. For example, if the threshold adjustment coefficient is 1.14 and the initial error threshold is ±2%, then the adaptation error threshold = 1.14 * (±2%) = ±2.28%. Thus, the initial error threshold is appropriately relaxed because the current fluctuations in electricity load and ambient temperature are higher than the historical average, and the threshold needs to be relaxed to avoid misjudging deviations caused by normal fluctuations as metering anomalies. If the target user's load fluctuation coefficient drops to 0.3 (below the historical average) and the ambient temperature fluctuation coefficient drops to 0.08 (below the historical average), the threshold adjustment coefficient may be less than 1. In this case, the error threshold can be tightened to improve the sensitivity of anomaly identification.

[0054] In summary, compared to existing technologies, this application calculates a threshold adjustment coefficient based on load fluctuation characteristics, environmental fluctuation characteristics, and the prediction confidence level within a preset time zone. This coefficient corrects the initial error threshold and outputs an adaptation error threshold. Thus, by quantifying the comprehensive fluctuations of electricity load, ambient temperature, and prediction result reliability, a threshold adjustment coefficient is obtained. When fluctuations are large, the initial error threshold is amplified to reduce misjudgments; when fluctuations are small, the initial error threshold is reduced to improve sensitivity. This solves the problem that traditional fixed thresholds are difficult to adapt to complex electricity usage scenarios.

[0055] S50: Calculate the power consumption deviation based on the power consumption displayed on the target user's electricity meter within the preset time zone and the predicted total power consumption. If the power consumption deviation exceeds the adaptation error threshold, execute the metering anomaly warning for the electricity meter.

[0056] In the aforementioned steps, the predicted total power consumption is obtained by the power consumption predictor based on the actual power consumption set of the target user in the preset time zone and the device correlation set. The initial error threshold is corrected by the threshold adjustment coefficient to obtain the adaptation error threshold. Therefore, the power deviation range can be calculated based on the power consumption displayed by the electricity meter of the target user in the preset time zone and the predicted total power consumption. If the power deviation range exceeds the adaptation error threshold, the metering anomaly warning is executed. The power deviation range is calculated as follows: power deviation range = (power consumption displayed by the electricity meter - predicted total power consumption) / power consumption displayed by the electricity meter.

[0057] For example, if the predicted total electricity consumption is 20 kWh and the electricity consumption displayed on the meter is 20.8 kWh, then the electricity deviation is (20.8-20) / 20.8=4%. If the adaptation error threshold is ±2.28%, and the electricity deviation exceeds the adaptation error threshold, the metering anomaly warning is executed. In this way, the abnormality identification and anomaly warning of the metering are realized.

[0058] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application first collects the actual power consumption set of smart devices of target users within a preset time zone, and then matches and obtains the device correlation set within the preset time zone. This provides a reliable data foundation for subsequent model building and anomaly identification and early warning.

[0059] Secondly, this application predicts the target user's power consumption within a preset time zone based on the actual power consumption set and the device correlation set. Thus, the predicted total power consumption within the preset time zone, based on the actual power consumption set and the device correlation set, serves as the benchmark for subsequent anomaly identification.

[0060] Furthermore, this application obtains the prediction confidence level based on the set of smart electrical equipment types and the set of equipment correlation. Thus, the prediction reliability of the current scenario is specifically evaluated based on historical data to obtain an initial prediction confidence level. This is then corrected by combining the model's prediction accuracy to obtain a prediction confidence level that reflects the true prediction reliability of the current scenario, thereby improving the robustness of anomaly identification.

[0061] Furthermore, this application calculates a threshold adjustment coefficient based on the load fluctuation characteristics, environmental fluctuation characteristics, and prediction confidence level within a preset time zone, corrects the initial error threshold, and outputs an adaptation error threshold. Thus, by quantifying the comprehensive fluctuations of electricity load, ambient temperature, and prediction result reliability, a threshold adjustment coefficient is obtained. When fluctuations are large, the initial error threshold is amplified to reduce misjudgments; when fluctuations are small, the initial error threshold is reduced to improve sensitivity. This solves the problem that traditional fixed thresholds are difficult to adapt to complex electricity consumption scenarios.

[0062] Finally, this application calculates the power consumption deviation based on the displayed power consumption and predicted total power consumption of the target user's electricity meter within the preset time zone. If the power consumption deviation exceeds the adaptation error threshold, an abnormal metering warning is issued, thus realizing the identification and warning of abnormal metering.

[0063] Through the above technical solution, this application predicts the total electricity consumption in the current period based on the actual usage of smart electrical equipment by target users in a preset time zone, analyzes the confidence level of the prediction results, and then dynamically adjusts the initial error threshold according to the fluctuation characteristics of multimodal data such as the environment and prediction confidence level to obtain the adaptation error threshold. Based on this, the deviation of the electricity meter measurement result is identified and judged, thereby improving the accuracy and reliability of electricity meter measurement anomaly identification.

[0064] Example 2, as Figure 2 As shown, based on the same inventive concept as the multimodal electricity consumption data fusion method for identifying meter metering anomalies provided in Embodiment 1, this embodiment of the invention also provides a multimodal electricity consumption data fusion system for identifying meter metering anomalies, including: The data acquisition module 11 is used to collect the actual power consumption set of the smart electrical devices of the target user within the preset time zone, and to match and obtain the device correlation set within the preset time zone. The power prediction module 12 is used to predict the total power consumption of the target user in a preset time zone based on the actual power consumption set and the device correlation set. Confidence acquisition module 13 is used to obtain the predicted confidence based on the set of smart electrical equipment types and the set of equipment correlation. Threshold correction module 14 is used to calculate the threshold adjustment coefficient based on the load fluctuation characteristics, environmental fluctuation characteristics and the prediction confidence in the preset time zone, correct the initial error threshold, and output the adaptation error threshold. The early warning output module 15 is used to calculate the power consumption deviation based on the power consumption displayed on the electricity meter of the target user in the preset time zone and the predicted total power consumption. If the power consumption deviation exceeds the adaptation error threshold, the module executes an early warning for metering abnormality of the electricity meter.

[0065] Specifically, the data acquisition module 11 is used for: Collect the device types and actual power consumption of smart electrical devices used by the target user within a preset time zone to obtain a set of smart electrical device types and a set of actual power consumption. The smart electrical devices have energy consumption monitoring functions, and by default, the user is equipped with at least one smart electrical device. Using the time period-device correlation mapping table, the device correlation set is obtained by matching the preset time zone and the set of smart electrical equipment types.

[0066] Specifically, the method for constructing the "time period-equipment correlation mapping table" includes: The year is divided according to a preset time step, multiple time periods are obtained, and the first time period is randomly selected. Based on the preset types of smart electrical devices, the average historical power consumption of several smart electrical devices in the target area within the first time period is calculated, and the average historical total power consumption is obtained. The ratio of the historical average power consumption to the historical average total power consumption is set as the device correlation degree, and a first device correlation degree set is calculated based on the several historical average power consumption values. By sequentially analyzing multiple device correlation sets over multiple time periods, a time period-device correlation mapping table is constructed.

[0067] Specifically, the power prediction module 12 is used for: Based on the target user's historical electricity consumption logs, multiple sample power consumption sets and multiple sample device correlation sets are collected. The historical total power consumption of different sample power consumption sets and sample device correlation sets in historical time zones is collected and set as the sample total power consumption set. The time interval between the historical time zone and the preset time zone is the same. Using the multiple sets of power consumption from samples and the multiple sets of device correlation from samples as input, and using the total power consumption set of samples as supervision, the machine learning model is trained until convergence to obtain a power consumption predictor, and the prediction accuracy of the model is verified. Using the power consumption predictor, the predicted total power consumption of the target user within a preset time zone is obtained based on the actual power consumption set and the device correlation set.

[0068] The confidence acquisition module 13 is specifically used for: Using the set of smart electrical equipment types and the set of equipment correlation as comparison constraints, historical power consumption monitoring logs of the target area are retrieved, and several historical predicted total power consumption and several historical actual power consumption are collected. Several prediction deviation ranges are calculated based on the total historical predicted power consumption and the actual historical power consumption. The percentage of prediction deviation ranges that are less than a preset deviation index is counted and set as the initial prediction confidence level. The initial prediction confidence is corrected based on the model's prediction accuracy, and the prediction confidence is output.

[0069] Specifically, the threshold correction module 14 is used for: The active power sequence of target users within a preset time zone is collected for load fluctuation analysis, and the load fluctuation coefficient is output. The load fluctuation coefficient is the ratio of the standard deviation of active power to the mean of active power. An environmental temperature sequence of a target area within a preset time zone is collected for environmental fluctuation analysis, and an environmental fluctuation coefficient is output. The environmental fluctuation coefficient is the ratio of the standard deviation of the environmental temperature to the mean of the environmental temperature. The threshold adjustment coefficient is calculated based on the load fluctuation coefficient, environmental fluctuation coefficient, and prediction confidence level.

[0070] Further, the phrase "calculating the threshold adjustment coefficient based on the load fluctuation coefficient, environmental fluctuation coefficient, and prediction confidence level" includes: The ratio of the load fluctuation coefficient to the historical average load fluctuation coefficient of the target user is set as the load adjustment coefficient; The ratio of the environmental fluctuation coefficient to the average historical environmental fluctuation coefficient of the target area is set as the environmental adjustment coefficient. The ratio of the historical predicted confidence level of the target user to the predicted confidence level is set as the confidence level adjustment coefficient; The threshold adjustment coefficient is calculated by weighting the load adjustment coefficient, environmental adjustment coefficient, and confidence adjustment coefficient.

[0071] The warning output module 15 is specifically used for: The power consumption deviation is calculated based on the power consumption displayed on the electricity meter of the target user in the preset time zone and the predicted total power consumption. If the power consumption deviation exceeds the adaptation error threshold, the metering anomaly warning is issued.

[0072] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application first uses a data acquisition module to collect the actual power consumption set of the target user's smart devices within a preset time zone, and matches it to obtain the device correlation set within the preset time zone, providing a reliable data foundation for subsequent model building and anomaly identification and early warning. Secondly, through a power consumption prediction module, the predicted total power consumption of the target user within the preset time zone is predicted based on the actual power consumption set and the device correlation set. A reasonable total power consumption for the preset time zone is predicted and output based on the actual power consumption set and the device correlation set, serving as the benchmark value for subsequent anomaly identification. Thirdly, through a confidence level acquisition module, the prediction confidence level is obtained based on the smart device type set and the device correlation set. The prediction reliability of the current scenario is specifically evaluated based on historical data to obtain an initial prediction confidence level. This initial confidence level is then corrected by combining it with the model's prediction accuracy to obtain a prediction confidence level that reflects the true prediction reliability of the current scenario, thus improving the robustness of anomaly identification. Furthermore, through the threshold correction module, a threshold adjustment coefficient is calculated based on the load fluctuation characteristics, environmental fluctuation characteristics, and prediction confidence within a preset time zone. This coefficient corrects the initial error threshold and outputs an adaptation error threshold. By quantifying the comprehensive fluctuations of electricity load, ambient temperature, and prediction reliability, the threshold adjustment coefficient is obtained. When fluctuations are large, the initial error threshold is amplified to reduce misjudgments; when fluctuations are small, the initial error threshold is lowered to improve sensitivity. This solves the problem that traditional fixed thresholds are difficult to adapt to complex electricity consumption scenarios. Finally, through the early warning output module, the electricity consumption deviation is calculated based on the displayed electricity consumption of the target user's electricity meter and the predicted total electricity consumption within a preset time zone. If the electricity consumption deviation exceeds the adaptation error threshold, an electricity meter metering anomaly warning is issued, realizing the identification and early warning of meter metering anomalies. Thus, by integrating multimodal data such as actual power consumption of smart devices, device correlation, load fluctuations, and environmental characteristics, the accuracy and reliability of electricity metering anomaly identification are improved.

[0073] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0074] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0077] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0078] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0079] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying meter metering anomalies based on multimodal electricity consumption data fusion, characterized in that, The methods include: Collect the actual power consumption set of smart electrical devices of target users within a preset time zone, and match and obtain the device correlation set within the preset time zone; Based on the actual power consumption set and the device correlation set, the predicted total power consumption of the target user in the preset time zone is obtained. The prediction confidence level is obtained based on the analysis of the smart electrical equipment type set and the equipment correlation set; The threshold adjustment coefficient is calculated based on the load fluctuation characteristics, environmental fluctuation characteristics and prediction confidence in the preset time zone, and the initial error threshold is corrected to output the adaptation error threshold. The power consumption deviation is calculated based on the power consumption displayed on the electricity meter of the target user in the preset time zone and the predicted total power consumption. If the power consumption deviation exceeds the adaptation error threshold, the metering anomaly warning is issued.

2. The method for identifying meter metering anomalies based on multimodal electricity consumption data fusion according to claim 1, characterized in that, Collect the actual power consumption set of smart devices of target users within a preset time zone, and match and obtain the device correlation set within the preset time zone, including: Collect the device types and actual power consumption of smart electrical devices used by the target user within a preset time zone to obtain a set of smart electrical device types and a set of actual power consumption. The smart electrical devices have energy consumption monitoring functions, and by default, the user is equipped with at least one smart electrical device. Using the time period-device correlation mapping table, the device correlation set is obtained by matching the preset time zone and the set of smart electrical equipment types.

3. The method for identifying meter metering anomalies based on multimodal electricity consumption data fusion according to claim 2, characterized in that, The method for constructing the time period-device correlation mapping table includes: The year is divided according to a preset time step, multiple time periods are obtained, and the first time period is randomly selected. Based on the preset types of smart electrical devices, the average historical power consumption of several smart electrical devices in the target area within the first time period is calculated, and the average historical total power consumption is obtained. The ratio of the historical average power consumption to the historical average total power consumption is set as the device correlation degree, and a first device correlation degree set is calculated based on the several historical average power consumption values. By sequentially analyzing multiple device correlation sets over multiple time periods, a time period-device correlation mapping table is constructed.

4. The method for identifying meter metering anomalies based on multimodal electricity consumption data fusion according to claim 1, characterized in that, Based on the actual power consumption set and the device correlation set, the predicted total power consumption of the target user within a preset time zone is obtained, including: Based on the target user's historical electricity consumption logs, multiple sample power consumption sets and multiple sample device correlation sets are collected. The historical total power consumption of different sample power consumption sets and sample device correlation sets in historical time zones is collected and set as the sample total power consumption set. The time interval between the historical time zone and the preset time zone is the same. Using the multiple sets of power consumption from samples and the multiple sets of device correlation from samples as input, and using the total power consumption set of samples as supervision, the machine learning model is trained until convergence to obtain a power consumption predictor, and the prediction accuracy of the model is verified. Using the power consumption predictor, the predicted total power consumption of the target user within a preset time zone is obtained based on the actual power consumption set and the device correlation set.

5. The method for identifying meter metering anomalies based on multimodal electricity consumption data fusion according to claim 4, characterized in that, The prediction confidence level is obtained based on the analysis of the smart electrical equipment type set and the equipment correlation set, including: Using the set of smart electrical equipment types and the set of equipment correlation as comparison constraints, historical power consumption monitoring logs of the target area are retrieved, and several historical predicted total power consumption and several historical actual power consumption are collected. Several prediction deviation ranges are calculated based on the total historical predicted power consumption and the actual historical power consumption. The percentage of prediction deviation ranges that are less than a preset deviation index is counted and set as the initial prediction confidence level. The initial prediction confidence is corrected based on the model's prediction accuracy, and the prediction confidence is output.

6. The method for identifying meter metering anomalies based on multimodal electricity consumption data fusion according to claim 1, characterized in that, The threshold adjustment coefficient is calculated based on the load fluctuation characteristics, environmental fluctuation characteristics, and prediction confidence level within the preset time zone, including: The active power sequence of target users within a preset time zone is collected for load fluctuation analysis, and the load fluctuation coefficient is output. The load fluctuation coefficient is the ratio of the standard deviation of active power to the mean of active power. An environmental temperature sequence of a target area within a preset time zone is collected for environmental fluctuation analysis, and an environmental fluctuation coefficient is output. The environmental fluctuation coefficient is the ratio of the standard deviation of the environmental temperature to the mean of the environmental temperature. The threshold adjustment coefficient is calculated based on the load fluctuation coefficient, environmental fluctuation coefficient, and prediction confidence level.

7. The method for identifying meter metering anomalies based on multimodal electricity consumption data fusion according to claim 6, characterized in that, The threshold adjustment coefficient is calculated based on the load fluctuation coefficient, environmental fluctuation coefficient, and prediction confidence level, including: The ratio of the load fluctuation coefficient to the historical average load fluctuation coefficient of the target user is set as the load adjustment coefficient; The ratio of the environmental fluctuation coefficient to the average historical environmental fluctuation coefficient of the target area is set as the environmental adjustment coefficient. The ratio of the historical predicted confidence level of the target user to the predicted confidence level is set as the confidence level adjustment coefficient; The threshold adjustment coefficient is calculated by weighting the load adjustment coefficient, environmental adjustment coefficient, and confidence adjustment coefficient.

8. A multimodal electricity consumption data fusion-based electricity meter anomaly identification system, characterized in that, For performing the method according to any one of claims 1-7, comprising: The data acquisition module is used to collect the actual power consumption of smart electrical devices of target users within a preset time zone, and to match and obtain the device correlation set within the preset time zone. The power prediction module is used to predict the total power consumption of the target user in a preset time zone based on the actual power consumption set and the device correlation set. The confidence level acquisition module is used to obtain the predicted confidence level based on the set of smart electrical equipment types and the set of equipment correlation. The threshold correction module is used to calculate the threshold adjustment coefficient based on the load fluctuation characteristics, environmental fluctuation characteristics and the prediction confidence in the preset time zone, correct the initial error threshold, and output the adaptation error threshold. The early warning output module is used to calculate the power consumption deviation based on the power consumption displayed on the electricity meter of the target user in the preset time zone and the predicted total power consumption. If the power consumption deviation exceeds the adaptation error threshold, the module will execute an early warning for metering anomalies of the electricity meter.

Images