Rental house electricity utilization abnormity early warning system and method based on data fusion

Abstract

The invention discloses a rental house electricity consumption abnormity early warning system and method based on data fusion, and relates to the technical field of electricity consumption abnormity early warning. Non-intrusive load decomposers, distributed temperature sensors and high-frequency current transformers are deployed at distribution boxes, sockets and cable nodes, and space-time alignment of current, temperature and load data is realized through a sliding time window algorithm; lSTM model parameters are improved by adopting an optimization algorithm, and the recognition accuracy of abnormal electric appliances is improved; constructing a digital twinborn model, mapping the power distribution network topology into a graph structure, dynamically calculating an abnormal propagation probability and visualizing a risk path; graded early warning is realized by fusing the dynamic weight model of the current deviation degree, the temperature rise gradient and the load sudden change. The recognition accuracy of the complex abnormal mode is improved; the visual pre-judgment of the risk propagation path is realized; and a complete closed loop from risk early warning to active protection is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of abnormal power consumption warning, and specifically relates to an abnormal power consumption warning system and method for rental housing based on data fusion. Background Art

[0002] With the acceleration of the urbanization process, the number of migrant workers going out has increased, and the number of rental houses has surged. Electrical fires have become the primary hidden danger threatening public safety.

[0003] Chinese Patent CN117037438B discloses a method and system for warning abnormal power consumption in buildings. The intelligent monitoring terminal performs multiple functions such as equipment monitoring, environmental monitoring, energy-saving monitoring, intelligent power distribution monitoring, leakage detection, energy consumption detection, and Internet of Things control; through these functions, comprehensive monitoring of building equipment, precise management of the power system, intelligent control of lighting equipment, real-time detection of residual current, accurate metering of energy consumption, continuous monitoring of the building environment, and efficient operation of energy efficiency management are realized; the ability to monitor the Wi-Fi mac information of each room in real time, and accurately identify different devices using the Wi-Fi mac recognition algorithm; perform synchronous operations on relevant data and warning prompts for abnormal power consumption; when abnormal power consumption is detected, the system will assist in abnormal handling operations based on the identified Wi-Fi mac information of different devices to ensure the safe and stable operation of the building electrical system.

[0004] In the prior art, only current and voltage thresholds are monitored, and complex abnormal patterns cannot be recognized; the data of electrical fire detectors, smart meters, and infrared thermal imagers are processed independently, and cross-device correlation analysis is not realized; damage has occurred when the alarm is triggered, and there is a lack of the ability to predict risks in advance. Summary of the Invention

[0005] (1) Technical Problems to be Solved In view of the problems in the related art, the present invention provides an abnormal power consumption warning method for rental housing based on data fusion. The present invention solves the problems of inability to recognize complex abnormal patterns, data islands, and lack of the ability to predict risks in advance through multi-source heterogeneous data fusion, machine learning technology, and dynamic hierarchical warning.

[0006] (2) Technical Solutions To solve the above technical problems, the present invention is realized through the following technical solutions: S1. Deploy sensors and collect data to obtain multi-source data of the rental housing; perform spatio-temporal alignment on the multi-source data of the rental housing to obtain fused data of the rental housing; S2. Extract features from the fused data of the rental housing and perform normalization processing to obtain a standard feature dataset of the rental housing; S3. Build an LSTM classification model. Use the current feature data, load feature data, and corresponding abnormal electrical appliance data in historical rental houses, and combine with an optimization algorithm to improve the LSTM classification model to obtain an optimized LSTM abnormal electrical appliance classification model; build a digital twin model of the rental house circuit; S4. According to the rental house standard feature dataset, combine the optimized LSTM abnormal electrical appliance classification model and the digital twin model of the rental house circuit to obtain real-time abnormal electrical appliance data and real-time abnormal propagation probability; and calculate the real-time risk value, and give an early warning according to the real-time risk value.

[0007] Preferably, the S1 includes the following steps: S11. Deploy non-intrusive load decomposers, distributed temperature sensors, and high-frequency current transformers in the distribution box, sockets, and cable channels of the rental house respectively to obtain multi-source sensors; S12. Collect data through the multi-source sensors to obtain multi-source data of the rental house; S13. Perform data spatio-temporal alignment on the multi-source data of the rental house to obtain fused data of the rental house; The above steps realize the three-dimensional acquisition and intelligent fusion of the electricity consumption data of the rental house through non-intrusive load decomposers, distributed temperature sensors, and high-frequency current transformers; can accurately identify hidden risks, and improve the abnormal positioning accuracy and detection response speed.

[0008] Preferably, the S13 includes the following steps: S131. Stamp the multi-source data of the rental house with timestamps based on the clock of the edge computing node; S132. Set the time window to a , and slice the multi-source data of the rental house according to the time window a to compensate for the transmission delay of the multi-source sensors and obtain compensated multi-source data of the rental house; S133. Establish a spatio-temporal mapping relationship, and process the compensated multi-source data of the rental house through the spatio-temporal mapping relationship to obtain fused data of the rental house; The above steps realize the deep fusion of multi-source data by establishing a precise spatio-temporal reference system, effectively solve the spatio-temporal misalignment problem of current, temperature, and load data in complex scenarios of rental houses, and improve the alignment accuracy of multi-source heterogeneous data.

[0009] Preferably, the S2 includes the following steps: S21. Extract the features in the fused data of the rental house to obtain a fused feature dataset of the rental house ; among them, b 1. b 2 and b3 respectively represent real-time current characteristic data, real-time temperature characteristic data, and real-time load characteristic data; the current characteristic data includes fundamental wave effective value, harmonic content, and waveform distortion rate, the temperature characteristic data includes regional temperature difference gradient ΔT and temperature rise rate dT / dt, and the load characteristic data includes sudden change in active power ΔP and equipment fingerprint matching degree; S22. Normalize the characteristic data in the rental housing fusion characteristic data set to obtain a rental housing standard characteristic data set; The above steps extract three major categories of multi-dimensional characteristics of current, temperature, and load from spatio-temporal fusion data, and use normalization processing to eliminate the dimension difference, forming a standard characteristic data set; effectively solving the problems of single characteristic dimension and inconsistent data scale in traditional monitoring systems, and laying a data foundation for accurate risk quantification.

[0010] Preferably, the said S3 includes the following steps: S31. Construct an LSTM classification model, and set the initial weight value of the said LSTM classification model to c 1, and the initial bias value to c 2; S32. Collect the current characteristic data and load characteristic data in historical rental housing and the corresponding abnormal electrical appliance data to obtain historical rental housing electrical appliance data; divide the historical rental housing electrical appliance data into historical rental housing electrical appliance training data and historical rental housing electrical appliance test data; S33. Set the training error threshold d , and the maximum number of training iterations to e ; use the historical rental housing electrical appliance training data to repeatedly train the LSTM classification model; after each round of training, calculate the training error f of the LSTM classification model, and adjust the initial weight value and initial bias value according to the training error; When f ≤ d or the maximum number of training iterations e is reached, obtain the trained LSTM abnormal electrical appliance classification model; S34. Use the historical rental housing electrical appliance test data to test the trained LSTM abnormal electrical appliance classification model, and after the test is completed, obtain an optimized LSTM abnormal electrical appliance classification model; S35. Construct a digital twin model of the rental housing circuit; The above steps include constructing an abnormal electrical appliance classification model based on LSTM and a digital twin model of the rental housing circuit; training the LSTM network with historical current and load data, and achieving high-precision anomaly recognition through dynamically adjusting weight biases and iterative optimization; constructing a digital twin system for the rental housing circuit; leveraging the time-series processing ability of LSTM to accurately capture the abnormal characteristics of electrical appliances and improve the classification accuracy; realizing the real-time mapping and simulation of the circuit state through digital twin technology, enhancing the immediacy and visualization of anomaly diagnosis; and the automatic detection mechanism can significantly reduce the cost of manual inspection, effectively prevent the risk of electrical fires in rental housing, and improve the safety management efficiency.

[0011] Preferably, the S34 includes the following steps: S341. Set the test accuracy threshold as g , input the historical rental housing electrical appliance test data into the trained LSTM abnormal electrical appliance classification model to obtain the test results, and calculate the test accuracy according to the test results h ; S342. If h ≥ g , regard the trained LSTM abnormal electrical appliance classification model as the optimized LSTM abnormal electrical appliance classification model; otherwise, use the optimization algorithm to find the optimal weight value and optimal bias value of the trained LSTM abnormal electrical appliance classification model, and regard the optimal weight value and optimal bias value as the weight value and bias value of the trained LSTM abnormal electrical appliance classification model to obtain the optimized LSTM abnormal electrical appliance classification model; The above steps verify the model performance by adopting historical electricity consumption data. When the test accuracy is lower than the threshold, the network weights and bias values are automatically adjusted through the optimization algorithm strategy to make the recognition accuracy of the model for illegal electrical appliances in rental housing pass.

[0012] Preferably, the S35 includes the following steps: S351. Collect the physical parameters and connection relationship data of the distribution network to obtain circuit data; S352. Represent the distribution network as a graph structure according to the circuit data, where the nodes in the graph represent devices and the edges in the graph represent the connections between devices; S353. Extract features for each node and edge to obtain circuit feature data; the circuit feature data includes the rated current and voltage of the device; S354. Use GraphCNN to learn the topological features of the network and generate the distribution network topology graph; The above steps build a digital twin model of the power distribution system. Through circuit topology digitization, graph structure transformation, feature engineering extraction, and GraphCNN topology learning, the power distribution network is mapped into a topology graph containing device nodes and connection edges. The rated current and impedance characteristic feature vectors are extracted, and the GraphCNN is used to mine hidden association rules, improving the prediction accuracy of the abnormal propagation path.

[0013] Preferably, the steps of using an optimization algorithm to find the optimal weight value and optimal bias value of the trained LSTM abnormal electrical appliance classification model in S342 include the following steps: S3421. Construct a whale population k , and set the scale of the whale population to j , then the whale population , where k i represents the i th whale in the whale population; S3422. Randomly generate an initial position set of the whale population according to the weight value and bias value, where p i1 represents the initial one-dimensional position of the i th whale, and p i2 represents the initial two-dimensional position of the i th whale; S3423. Start the iterative operation, and set the maximum number of optimization iterations to q ; in each round of iteration, update the positions of each whale in the initial position set of the whale population according to the test accuracy; and obtain the global best whale individual and global best target in the whale population in each round of iteration; S3424. When h ≥ g or the maximum number of optimization iterations q is reached, the whale population stops the iterative operation to obtain the optimal weight value and optimal bias value; The above steps finely tune the weights and biases of the LSTM model by introducing the whale optimization algorithm; the optimization of the LSTM model is realized by initializing the whale population, generating the initial positions, and iterative updating; the classification accuracy of the model is significantly improved; the parameter convergence is accelerated through the population collaborative iteration mechanism, reducing the model training time and calculation cost.

[0014] Preferably, S4 includes the following steps: S41. Extract the current feature data and load feature data from the rental housing standard feature dataset, and input them into the optimized LSTM classification model to obtain real-time abnormal electrical appliance data; S42. Extract the current feature data and temperature feature data from the rental housing standard feature dataset, combine with the digital twin model to obtain the real-time distribution network topology diagram, and calculate the real-time abnormal propagation probability. R The calculation formula is as follows: ; Where R represents the probability that the abnormality propagates from node i to node j . w ij represents the connection weight from node i to j . represents the current change amount on node i ; I rated represents the rated current of node i ; S43. Extract the real-time current feature data, real-time temperature feature data, and real-time load feature data from the rental housing standard feature dataset to obtain the real-time current deviation data T 1. Real-time temperature gradient data T 2 and real-time load mutation frequency data T 3; Set the risk weights of the feature data in the rental housing standard feature dataset under different electrical appliances, abnormal propagation probabilities, seasons, and times to obtain the risk weight set; according to the real-time abnormal electrical appliance data, real-time abnormal propagation probability, current season, and time, obtain the real-time current deviation data weight t 1. Real-time temperature gradient data weight t 2 and real-time load mutation frequency data weight t 3; S44. Calculate the real-time risk value according to the feature data in the rental housing standard feature dataset and the weights corresponding to each feature data W The calculation formula is as follows: ; S45. Set the first-level warning threshold as Z 1, the second-level warning threshold as Z 2, the third-level warning threshold as Z 3, and the fourth-level warning threshold as Z 4; when W < Z 1, no warning is required; when Z 1 ≤ W < Z 2, execute the first-level warning, and the mobile APP pushes a prompt of abnormality; when Z 2 ≤ W < Z3. Execute secondary warning and automatically cut off non-essential loads; when Z 3 ≤ W < Z 4. Execute tertiary warning and link with fire-fighting equipment for standby start-up; when Z 4 ≤ W , execute quaternary warning, cut off power supply for the whole circuit and trigger audible and visual alarms; The above steps realize intelligent prevention and control of electricity safety in rental houses by constructing a dynamic risk assessment and hierarchical warning system. Through integrating and optimizing the LSTM abnormal electrical appliance classification model and GraphCNN topology prediction, combining current deviation degree, temperature rise gradient, and load mutation, dynamically adjust the risk weight, and introduce the topological propagation probability value to correct the warning threshold, so as to improve the accuracy of risk assessment.

[0015] An electricity anomaly warning system for rental houses based on data fusion, which is used to realize the above-mentioned electricity anomaly warning method based on data fusion, includes a multi-source data acquisition and spatio-temporal fusion module, a multi-dimensional feature extraction and standardization module, an intelligent classification and digital twin modeling module, and a dynamic risk assessment and hierarchical warning module; The multi-source data acquisition and spatio-temporal fusion module is used to deploy a multi-source sensor network at the distribution box, socket, and cable channel through a non-intrusive load decomposer, a distributed temperature sensor, and a high-frequency current transformer to collect multi-source data of rental houses in real time; use the edge computing node clock synchronization technology to timestamp the multi-source data, and compensate for the transmission delay through time window slicing, establish a spatio-temporal mapping relationship, and fuse the scattered multi-source data to obtain the fused data of rental houses; The multi-dimensional feature extraction and standardization module is used to extract current features, temperature features, and load feature data from the fused data of rental houses; eliminate the influence of different dimensions through normalization processing to obtain the standard feature dataset of rental houses; The intelligent classification and digital twin modeling module is used to construct an LSTM classification model improved by the whale optimization algorithm, use historical current and load feature data, improve the generalization ability of the model through adaptive weight optimization, and realize the accurate identification of abnormal electrical appliances; based on GraphCNN, model the distribution network topology, transform the device connection relationship into a graph structure, extract parameters such as rated current and voltage as node features, and generate a dynamically updated digital twin model; realize the digital mapping of circuit topology features; The dynamic risk assessment and hierarchical warning module is used to dynamically quantify the risk level through real-time abnormal electrical appliance identification, abnormal propagation probability calculation, and multi-dimensional risk value assessment, and execute a quaternary warning strategy.

[0016] (III) Beneficial effects: The present invention has the following beneficial effects: Through the multi-source sensor collaborative deployment and spatio-temporal alignment technology, the present invention solves the problem of data islands in existing monitoring systems; the present invention deploys non-intrusive load decomposers, distributed temperature sensors, and high-frequency current transformers in distribution boxes, sockets, and cable channels, performs spatio-temporal slicing and mapping compensation on multi-dimensional data such as current, temperature, and load, constructs a fused data cube, and realizes precise spatio-temporal correlation of cross-device data; improves the recognition accuracy of complex abnormal patterns.

[0017] The present invention combines the whale optimization algorithm with the LSTM abnormal electrical appliance classification model to solve the problem of abnormal electrical appliance classification; by constructing a training set containing dynamic features such as fundamental wave effective value and harmonic content, and using the whale algorithm to adaptively optimize the weights and biases of LSTM, the classification accuracy of the LSTM abnormal electrical appliance classification model is improved; effectively solves the pain point that the existing technology cannot identify hidden electrical fire hazards.

[0018] The present invention constructs a digital twin early warning system based on GraphCNN, which solves the lag defect in the existing technology; by converting the distribution network topology into a graph structure, extracting the rated parameters of the equipment as node features, and dynamically calculating the abnormal propagation probability through the real-time current change amount and the connection weight between nodes; realizes the visual prediction of the risk propagation path.

[0019] The present invention establishes a multi-dimensional risk quantification model and a hierarchical response mechanism to form a closed-loop prevention and control system; by designing a dynamic weight function including current deviation degree, temperature gradient, and load mutation frequency, and combining environmental factors such as season and time period to construct a calculation model of the risk value; through hierarchical disposal strategies such as mobile APP push, non-important load removal, and full-loop power-off, realizes a complete closed-loop from risk early warning to active protection.

[0020] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic flow chart of the method for early warning of abnormal electricity consumption in rental houses based on data fusion according to the present invention; Figure 2 It is a schematic flow chart of obtaining an optimized LSTM abnormal electrical appliance classification model in the method for early warning of abnormal electricity consumption in rental houses based on data fusion according to the present invention; Figure 3Schematic flow chart of obtaining the rental housing circuit twin model in the rental housing power consumption anomaly warning method based on data fusion of the present invention; Figure 4 Schematic flow chart of realizing the hierarchical warning of rental housing power consumption anomalies in the rental housing power consumption anomaly warning method based on data fusion of the present invention; Figure 5 Schematic diagram of the modules of the rental housing power consumption anomaly warning system based on data fusion of the present invention. Detailed implementation manners

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0024] In the description of the present invention, it should be understood that the terms "open hole", "upper", "lower", "top", "middle", "inner", etc. indicating the orientation or position relationship are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the components or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0025] Embodiment 1:

[0026] Please refer to Figure 1 , Figure 2 , Figure 3 and Figure 4 , the present invention discloses a rental housing power consumption anomaly warning method based on data fusion, including the following steps: S1. Deploy sensors and collect data to obtain multi-source data of the rental housing; perform spatio-temporal alignment on the multi-source data of the rental housing to obtain the fusion data of the rental housing; The S1 includes the following steps: S11. Deploy non-intrusive load decomposers, distributed temperature sensors, and high-frequency current transformers at the distribution box, sockets, and cable channels of the rental housing respectively to obtain multi-source sensors; S12. Collect data through the multi-source sensors to obtain multi-source data of the rental housing; S13. Perform data spatio-temporal alignment on the multi-source data of the rental housing to obtain the fusion data of the rental housing; The S13 includes the following steps: S131. Stamp the multi-source data of the rental housing with timestamps based on the clock of the edge computing node; S132. Set the time window to a , and according to the time window aSlice the multi-source data of the rental house, compensate for the transmission delay of the multi-source sensors, and obtain the compensated multi-source data of the rental house; S133. Establish a spatio-temporal mapping relationship, and process the compensated multi-source data of the rental house through the spatio-temporal mapping relationship to obtain the fused data of the rental house; Take a rental house illegally accessing a high-power electric welding machine (declared power 3000W, actual peak power 8200W) as an example; The traditional method did not perform data fusion and finally did not detect any abnormalities; The present invention performs more multi-source data association. The current transformer detects a sudden increase in harmonic content (THD from 5% → 23%), the temperature sensor finds that the ΔT in the cable channel is 15 °C / m (the baseline is 3 °C / m), and the load decomposer identifies that the fingerprint matching degree of the "unknown load" is only 41%; The system calculates the risk value W = 76 (Z2 = 70) 7 minutes after the abnormal start, triggers a secondary warning and cuts off the non-essential loads to avoid fire.

[0027] S2. Extract features from the fused data of the rental house and perform normalization processing to obtain the standard feature dataset of the rental house; The S2 includes the following steps: S21. Extract features from the fused data of the rental house to obtain the fused feature dataset of the rental house ; where b 1, b 2 and b 3 respectively represent real-time current feature data, real-time temperature feature data and real-time load feature data; the current feature data includes fundamental wave effective value, harmonic content, waveform distortion rate, the temperature feature data includes regional temperature difference gradient ΔT, temperature rise rate dT / dt, and the load feature data includes sudden change in active power ΔP, equipment fingerprint matching degree; S22. Perform normalization processing on the feature data in the fused feature dataset of the rental house to obtain the standard feature dataset of the rental house; S3. Construct an LSTM classification model, use the current feature data and load feature data in the historical rental house and the corresponding abnormal electrical appliance data, and combine with an optimization algorithm to improve the LSTM classification model to obtain an optimized LSTM abnormal electrical appliance classification model; Construct a digital twin model of the rental house circuit; The S3 includes the following steps: S31. Construct an LSTM classification model, and set the initial weight value of the LSTM classification model to c 1, the initial bias value is c 2; S32. Collect the current feature data, load feature data, and corresponding abnormal electrical appliance data in historical rental houses to obtain historical rental house electrical appliance data; divide the historical rental house electrical appliance data into historical rental house electrical appliance training data and historical rental house electrical appliance test data; S33. Set the training error threshold d . The maximum number of training iterations is e ; use the historical rental house electrical appliance training data to repeatedly train the LSTM classification model; after each round of training, calculate the training error f of the LSTM classification model, and adjust the initial weight value and initial bias value according to the training error; When f ≤ d or the maximum number of training iterations e is reached, obtain the trained LSTM abnormal electrical appliance classification model; S34. Use the historical rental house electrical appliance test data to test the trained LSTM abnormal electrical appliance classification model. After the test is completed, obtain the optimized LSTM abnormal electrical appliance classification model; Adopt 15,000 groups of electrical appliance operation data (including abnormal electrical appliances such as electric blankets and modified air conditioners) collected from 2021 to 2024, set the initial weight c1 = 0.01, bias c2 = 0.1, training error threshold d = 0.01, and maximum number of iterations e = 500; after being optimized by the whale optimization algorithm (population size j = 50, number of iterations q = 200), the model classification accuracy rate is increased from 85.3% to 96.7% (test set n = 3000); The harmonic recognition rate of the traditional SVM model is 72.1%, the power mutation detection rate is 68.5%, and the fingerprint matching degree is 65.3%; The harmonic recognition rate of the LSTM model before optimization is 83.6%, the power mutation detection rate is 79.2%, and the fingerprint matching degree is 81.4%; The harmonic recognition rate of the optimized LSTM abnormal electrical appliance classification model of the present invention is 95.8%, the power mutation detection rate is 93.7%, and the fingerprint matching degree is 97.1%; The said S34 includes the following steps: S341. Set the test accuracy threshold as g , input the historical rental house electrical appliance test data into the trained LSTM abnormal electrical appliance classification model to obtain the test result, and calculate the test accuracy h according to the test result; S342. If h ≥ g, use the trained LSTM abnormal electrical appliance classification model as the optimized LSTM abnormal electrical appliance classification model; otherwise, use an optimization algorithm to find the optimal weight value and optimal bias value of the trained LSTM abnormal electrical appliance classification model, and use the optimal weight value and optimal bias value as the weight value and bias value of the trained LSTM abnormal electrical appliance classification model to obtain the optimized LSTM abnormal electrical appliance classification model; The use of an optimization algorithm to find the optimal weight value and optimal bias value of the trained LSTM abnormal electrical appliance classification model in S342 includes the following steps: S3421. Construct a whale population k , set the size of the whale population to j , then the whale population , where k i represents the i th whale in the whale population; S3422. Randomly generate an initial position set of the whale population according to the weight value and bias value , where p i1 represents the initial one-dimensional position of the i th whale, p i2 represents the initial two-dimensional position of the i th whale; S3423. Start the iterative operation, and set the maximum number of optimization iterations to q ; In each round of iteration, update the positions of each whale in the initial position set of the whale population according to the test accuracy; and obtain the global best whale individual and global best target in the whale population in each round of iteration. S3424. When h ≥ g or reach the maximum number of optimization iterations q , the whale population stops the iterative operation to obtain the optimal weight value and optimal bias value; S35. Construct a digital twin model of the rental housing circuit; The S35 includes the following steps: S351. Collect the physical parameters and connection relationship data of the distribution network to obtain circuit data; S352. Represent the distribution network as a graph structure according to the circuit data, where the nodes in the graph represent devices, and the edges in the graph represent the connections between devices; S353. Extract features for each node and edge to obtain circuit feature data; the circuit feature data includes the rated current and voltage of the device; S354. Use GraphCNN to learn the topological features of the network and generate a topological graph of the distribution network; S4. Based on the rental housing standard feature dataset, combined with the optimized LSTM abnormal electrical appliance classification model and the rental housing circuit digital twin model, obtain real-time abnormal electrical appliance data and real-time abnormal propagation probability, calculate the real-time risk value, and issue a warning according to the real-time risk value; The said S4 includes the following steps: S41. Extract the current feature data and load feature data from the rental housing standard feature dataset, and input them into the optimized LSTM classification model to obtain real-time abnormal electrical appliance data; S42. Extract the current feature data and temperature feature data from the rental housing standard feature dataset, combine with the digital twin model to obtain the real-time distribution network topology diagram, and calculate the real-time abnormal propagation probability R The calculation formula is as follows, ; Where R represents the probability that the abnormality propagates from node i to node j ; w ij represents the connection weight from node i to j ; represents the current change amount on node i ; I rated represents the rated current of node i ; S43. Extract the real-time current feature data, real-time temperature feature data and real-time load feature data from the rental housing standard feature dataset to obtain real-time current deviation data T 1. Real-time temperature gradient data T 2 and real-time load mutation frequency data T 3; Set the risk weights of the feature data in the rental housing standard feature dataset under different electrical appliances, abnormal propagation probabilities, seasons and times to obtain a risk weight set; according to the real-time abnormal electrical appliance data, real-time abnormal propagation probability, current season and time, obtain the real-time current deviation data weight t 1. Real-time temperature gradient weight data t 2 and real-time load mutation frequency data weight t 3; S44. Calculate the real-time risk value according to the feature data in the rental housing standard feature dataset and the weights corresponding to each feature data W The calculation formula is as follows, ; S45. Set the first-level warning threshold as Z 1, and the second-level warning threshold as Z2. The third-level warning threshold is Z 3. The fourth-level warning threshold is Z 4; When W < Z 1, no warning is required; When Z 1 ≤ W < Z 2, execute the first-level warning, and the mobile APP pushes a prompt of abnormality; When Z 2 ≤ W < Z 3, execute the second-level warning, and automatically cut off non-essential loads; When Z 3 ≤ W < Z 4, execute the third-level warning, and link the fire-fighting equipment to prepare for startup; When Z 4 ≤ W , execute the fourth-level warning, cut off the power supply of the entire circuit and trigger an audible and visual alarm.

[0028] Taking a rental house as an example: The detected real-time current deviation degree T1 of the rental house is 1.8 (rated value 1.0), the cable channel ΔT is 12 °C / m (normal <5 °C / m), the load mutation frequency T3 is 8 times / hour (baseline 2 times), the system calculates the risk value W = 83.6 (Z3 = 80), and triggers the third-level warning. After on-site verification, it is confirmed that it is an illegally modified instant water heater, and its actual power (8500W) exceeds the declared power (3000W) by 283%. The warning response time is 17 minutes earlier than the traditional current threshold method; Comparison with the traditional method: In the same test environment (n = 30 rooms, monitoring period 1 month), the method of this patent is significantly superior to the existing technology in key indicators: The abnormal electrical appliance recognition rate of the traditional threshold method is 61.2%, the false alarm rate is 23.4%, the warning response time is 28.5 min, and the overload fault location accuracy is 3.2 m; The abnormal electrical appliance recognition rate of the method of the present invention is 95.3%, the false alarm rate is 4.1%, the warning response time is 9.2 min, and the overload fault location accuracy is 0.4 m.

[0029] Embodiment 2: Please refer to Figure 5 , a rental house electricity consumption abnormality warning system based on data fusion, used to implement the above-mentioned rental house electricity consumption abnormality warning method based on data fusion, including a multi-source data acquisition and spatio-temporal fusion module, a multi-dimensional feature extraction and standardization module, an intelligent classification and digital twin modeling module, and a dynamic risk assessment and hierarchical warning module; The multi-source data acquisition and spatio-temporal fusion module is used to deploy a multi-source sensor network in distribution boxes, sockets and cable channels through non-intrusive load decomposers, distributed temperature sensors and high-frequency current transformers to collect multi-source data of rental houses in real time; use the edge computing node clock synchronization technology to timestamp the multi-source data, and compensate for the transmission delay through time window slicing, establish a spatio-temporal mapping relationship, and fuse the scattered multi-source data to obtain the fused data of rental houses; The multi-dimensional feature extraction and standardization module is used to extract current features, temperature features and load feature data from the fused data of rental houses; eliminate the influence of different dimensions through normalization processing to obtain the standard feature dataset of rental houses; The intelligent classification and digital twin modeling module is used to construct an LSTM classification model improved by the whale optimization algorithm, use historical current and load feature data, improve the generalization ability of the model through adaptive weight optimization, and achieve accurate identification of abnormal electrical appliances; model the distribution network topology based on GraphCNN, transform the device connection relationship into a graph structure, extract parameters such as rated current and voltage as node features, and generate a dynamically updated digital twin model; realize the digital mapping of circuit topology features; The dynamic risk assessment and classification warning module is used to dynamically quantify the risk level through real-time abnormal electrical appliance identification, abnormal propagation probability calculation and multi-dimensional risk value assessment, and execute a four-level warning strategy.

[0030] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0031] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principle and practical application of the invention, so that those skilled in the art can understand and utilize the invention well.

Claims

1. A method for early warning of abnormal electricity consumption in rental housing based on data fusion, characterized in that, It includes the following steps: S1. Deploy sensors and collect data to obtain multi-source data of rental houses; perform spatio-temporal alignment on the multi-source data of rental houses to obtain fused data of rental houses; S2. Extract features from the fused data of rental houses and perform normalization processing to obtain a standard feature dataset of rental houses; S3. Build an LSTM classification model, use the current feature data, load feature data and corresponding abnormal electrical appliance data in historical rental houses, and combine with an optimization algorithm to improve the LSTM classification model to obtain an optimized LSTM abnormal electrical appliance classification model; build a digital twin model of the rental house circuit; S4. According to the standard feature dataset of rental houses, combine with the optimized LSTM abnormal electrical appliance classification model and the digital twin model of the rental house circuit to obtain real-time abnormal electrical appliance data and real-time abnormal propagation probability; and calculate the real-time risk value and give an early warning according to the real-time risk value.

2. The method for abnormal electricity consumption warning of rental housing based on data fusion according to claim 1, characterized in that, The S1 includes the following steps: S11. Deploy non-intrusive load decomposers, distributed temperature sensors and high-frequency current transformers at the distribution box, sockets and cable channels of rental houses respectively to obtain multi-source sensors; S12. Collect data through the multi-source sensors to obtain multi-source data of rental houses; S13. Perform data spatio-temporal alignment on the multi-source data of rental houses to obtain fused data of rental houses.

3. The method for abnormal electricity consumption warning of rental housing based on data fusion according to claim 2, wherein, The S13 includes the following steps: S131. Stamp timestamps on the multi-source data of rental houses based on the clock of the edge computing node; S132. Set the time window as a , and slice the multi-source data of the rental housing according to the time window a to compensate for the transmission delay of the multi-source sensors and obtain the compensated multi-source data of the rental housing; S133. Establish a spatio-temporal mapping relationship, and process the compensated multi-source data of rental houses through the spatio-temporal mapping relationship to obtain fused data of rental houses.

4. The method for abnormal electricity consumption warning of rental housing based on data fusion according to claim 1, wherein The S2 includes the following steps: S21. Extract features from the rental housing integration data to obtain a rental housing integration feature dataset ; among them, b 1. b 2 and b 3 respectively represent real-time current feature data, real-time temperature feature data, and real-time load feature data; S22. Perform normalization processing on the feature data in the fused feature dataset of rental houses to obtain a standard feature dataset of rental houses.

5. The method for abnormal electricity consumption warning of rental housing based on data fusion according to claim 1, wherein The S3 includes the following steps: S31. Construct an LSTM classification model and set the initial weight value of the LSTM classification model to c 1. The initial bias value is c 2; S32. Collect the current feature data, load feature data and corresponding abnormal electrical appliance data in historical rental houses to obtain historical rental house electrical appliance data; divide the historical rental house electrical appliance data into historical rental house electrical appliance training data and historical rental house electrical appliance test data; S33. Set the training error threshold d , set the maximum number of training iterations to e ; use the historical rental housing electrical appliance training data to repeatedly train the LSTM classification model; after each round of training, calculate the training error of the LSTM classification model f , and adjust the initial weight value and initial bias value according to the training error; When f ≤ d or the maximum number of training iterations is reached e a trained LSTM abnormal electrical appliance classification model is obtained; S34. Use the historical rental house electrical appliance test data to test the trained LSTM abnormal electrical appliance classification model. After the test is completed, obtain an optimized LSTM abnormal electrical appliance classification model; S35. Build a digital twin model of the rental house circuit.

6. The method for warning of abnormal electricity consumption in rental housing based on data fusion according to claim 5, characterized in that, The S34 includes the following steps: S341. Set the test accuracy threshold to g , input the historical rental housing electrical appliance test data into the trained LSTM abnormal electrical appliance classification model to obtain the test results, and calculate the test accuracy according to the test results h ; S342. If h ≥ g , use the trained LSTM abnormal electrical appliance classification model as the optimized LSTM abnormal electrical appliance classification model; otherwise, use an optimization algorithm to find the optimal weight value and optimal bias value of the trained LSTM abnormal electrical appliance classification model, and use the optimal weight value and optimal bias value as the weight value and bias value of the trained LSTM abnormal electrical appliance classification model to obtain the optimized LSTM abnormal electrical appliance classification model.

7. The method for early warning of abnormal electricity consumption in rental housing based on data fusion according to claim 5, characterized in that, The S35 includes the following steps: S351. Collect the physical parameters and connection relationship data of the distribution network to obtain circuit data; S352. Represent the distribution network as a graph structure according to the circuit data, where the nodes in the graph represent devices and the edges in the graph represent the connections between devices; S353. Extract features for each node and edge to obtain circuit feature data; S354. Use GraphCNN to learn the topological features of the network and generate a topological graph of the distribution network.

8. The method for abnormal electricity consumption warning of rental housing based on data fusion according to claim 6, wherein, The steps of using the optimization algorithm to find the optimal weight value and optimal bias value of the trained LSTM abnormal electrical appliance classification model in S342 include the following steps: S3421. Construct a whale population k , set the size of the whale population to j , then the whale population , where k i represents the i -th whale in the whale population; S3422. Randomly generate the initial position set of the whale population according to the weight value and the bias value , where p i1 represents the initial one-dimensional position of the i th whale, p i2 represents the initial two-dimensional position of the i th whale; S3423. Start the iterative operation and set the maximum number of optimization iterations to q ; During each iteration, update the positions of each whale in the initial position set of the whale population according to the test accuracy; and obtain the globally best whale individual and the globally best objective in the whale population during each iteration. S3424. When h ≥ g or the optimized maximum number of iterations is reached q the whale population stops the iterative operation to obtain the optimal weight value and the optimal bias value.

9. The method for abnormal electricity consumption warning of rental housing based on data fusion according to claim 1, wherein The S4 includes the following steps: S41. Extract the current feature data and load feature data from the rental housing standard feature dataset, and input them into the optimized LSTM classification model to obtain real-time abnormal electrical appliance data; S42. Extract the current feature data and temperature feature data from the rental housing standard feature dataset, combine with the digital twin model to obtain the real-time distribution network topology diagram, and calculate the real-time abnormal propagation probability; S43. Extract the real-time current feature data, real-time temperature feature data and real-time load feature data from the rental housing standard feature dataset to obtain the real-time current deviation data, real-time temperature gradient data and real-time load mutation frequency data; Set the risk weight set; S44. Calculate the real-time risk value according to the feature data in the rental housing standard feature dataset and the risk weight set; S45. Set up a hierarchical early warning mechanism, and conduct hierarchical early warning according to the real-time risk value combined with the hierarchical early warning mechanism.

10. The rental housing electricity consumption abnormal warning system based on data fusion is characterized in that, A system for implementing the rental housing electricity consumption abnormal early warning method based on data fusion according to any one of claims 1-9, the system includes a rental housing electricity consumption abnormal early warning system based on data fusion for implementing the above-mentioned rental housing electricity consumption abnormal early warning method based on data fusion, including a multi-source data acquisition and spatio-temporal fusion module, a multi-dimensional feature extraction and standardization module, an intelligent classification and digital twin modeling module, and a dynamic risk assessment and hierarchical early warning module.

Images