Electric quantity accounting management system based on artificial intelligence

Abstract

The invention relates to the technical field of power management, in particular to an electric quantity accounting management system based on artificial intelligence, and the system comprises a dynamic electric quantity accounting module which constructs a dynamic electric quantity calculation model based on the historical operation data of user equipment, and outputs a theoretical electric quantity value; the multi-dimensional accounting difference analysis module is used for performing three-dimensional difference analysis on the theoretical electric quantity value and an actually measured electric quantity value of the metering terminal to generate an equipment-level difference coefficient matrix; the abnormal region association positioning module is used for generating an abnormal probability thermal distribution diagram of an abnormal current propagation path and positioning an abnormal electricity utilization physical region; and the self-adaptive checking strategy generation module is used for configuring checking parameters according to the abnormal probability thermal distribution diagram. According to the method, high-concealment abnormal sources such as multi-point cooperative electricity stealing and inverter faults can be effectively identified, the probability thermal distribution diagram and the main propagation path can be output, the spatial precision and the structural coherence of abnormal tracing are improved, and the bottleneck that noise disturbance and substantial abnormity cannot be distinguished through an existing method is broken through.

Description

Technical Field

[0001] The present invention relates to the technical field of power management, and in particular to an electric power accounting and management system based on artificial intelligence. Background Art

[0002] With the large-scale access of distributed energy, energy storage systems and power electronic equipment, the electricity consumption behavior on the user side has shown highly dynamic, nonlinear and strong disturbance characteristics. The traditional electricity accounting method based on fixed models has been difficult to adapt to the complex operating environment. In the existing electricity management system, the following technical bottlenecks are common: On the one hand, current electricity calculations mostly rely on active power accumulation under steady state or constant value models based on simplified assumptions, which cannot accurately reflect the changes in the operating status of the equipment under multiple working conditions. In particular, when faced with the charging and discharging pulses of energy storage equipment, the reverse current of photovoltaic inverters at night, or the harmonic distortion generated by power electronic devices, there are obvious calculation deviations.

[0003] On the other hand, the means of identifying abnormal power consumption mainly rely on static threshold alarms or historical comparison methods, which are difficult to capture hidden abnormal events with weak periodicity, high-frequency disturbances or cross-node collaborative behaviors. At the same time, the existing verification mechanism is usually based on manual verification, lacking intelligent scheduling and dynamic tracking capabilities for high-risk areas, resulting in low efficiency in abnormal tracing and poor positioning accuracy.

[0004] In addition, the current mainstream system lacks an effective connection between electricity accounting and grid topology, making it difficult to integrate abnormal characteristics with the physical connection relationship of equipment, resulting in a split between the data layer and the physical layer, seriously affecting the reliability of the overall decision-making closed loop. Summary of the invention

[0005] The present invention provides an artificial intelligence-based electricity accounting and management system, which integrates an intelligent electricity accounting and anomaly positioning method that integrates historical operation data modeling, real-time electrical characteristic perception and spatial topological structure analysis, so as to achieve detailed modeling of complex electricity consumption behaviors, accurate identification of abnormal propagation paths, and adaptive strategy verification of key areas, thereby improving the accuracy, responsiveness and intelligence of the electricity management system.

[0006] An electric power accounting and management system based on artificial intelligence, comprising: Dynamic power accounting module: builds a dynamic power calculation model based on the historical operation data of the user equipment, collects the actual voltage waveform, current harmonic components and power pulse sequence of the current power-consuming equipment in real time, and outputs the theoretical power value through the dynamic power calculation model; Multi-dimensional accounting difference analysis module: performs three-dimensional difference analysis on the theoretical power value and the actual power value measured by the metering terminal, wherein the three-dimensional difference analysis includes time domain cumulative deviation, frequency domain energy offset and pulse sequence similarity, and generates a device-level difference coefficient matrix; Abnormal area association positioning module: input the difference coefficient matrix into the spatial topology analysis network, combine the grid node impedance parameters and the user equipment location information, generate the abnormal probability thermal distribution map of the abnormal current propagation path, and locate the abnormal power consumption physical area; Adaptive verification strategy generation module: configures verification parameters according to the abnormal probability thermal distribution map, including: Enable high-frequency power quality monitoring mode for high-probability abnormal areas; Apply impedance disturbance test to adjacent node devices.

[0007] Optionally, the dynamic electricity accounting module specifically includes: Historical data feature mining: Multi-dimensional decomposition and processing of historical operation data of user equipment, including: using adaptive wavelet packet decomposition to extract the energy proportion of steady-state components and transient components of the voltage waveform, establishing the correlation matrix between current harmonic components and equipment load rate through harmonic coupling analysis, and using dynamic time warping algorithm to align power pulse sequence patterns under different working conditions; Dynamic power calculation model construction: The processed historical operation data is input into the hybrid prediction network, which includes: The LSTM timing prediction unit based on the equipment aging curve is used to generate the basic power prediction value. The fully connected network embedded with the frequency domain attention mechanism corrects the prediction deviation caused by harmonic distortion. Based on the pulse feature compensator, the prediction weight is dynamically adjusted according to the power pulse sequence collected in real time. The heterogeneous data acquisition unit deployed on the edge synchronously captures the zero-crossing distortion rate and phase jitter parameters of the voltage waveform, the 3rd to 39th order content and phase angle distribution of the current harmonics, and the amplitude mutation gradient and time interval entropy of the power pulse. Theoretical value calculation: input the real-time collected data into the dynamic power calculation model to obtain the theoretical power value.

[0008] Optionally, the theoretical power value calculation includes executing: Adaptive filtering based on the equipment operation phase can eliminate the measurement noise caused by ambient temperature and humidity; Integrate the correlation features of voltage, current and pulse through spatiotemporal feature fusion algorithm; The output includes the theoretical power value within the normal operating fluctuation range, and the theoretical power value is dynamically updated as the equipment ages.

[0009] Optionally, the multi-dimensional accounting difference analysis module specifically includes: Time domain cumulative deviation calculation: The theoretical power value and the measured power value are compared in a sliding manner in a preset time window, and the dynamic time warping algorithm is used to align the asynchronously sampled power sequence, calculate the cumulative deviation in each window, and generate a time domain deviation vector; Frequency domain energy offset detection: Decompose the current harmonic components of the theoretical value and the measured value using the frequency domain decomposition algorithm, calculate the energy spectrum density ratio, extract the energy offset index of each harmonic, and construct the frequency domain offset vector; Pulse sequence similarity evaluation: Based on the structure matching algorithm, the distance distribution between the theoretical value and the measured pulse sequence is matched, the phase synchronization error of the pulse amplitude mutation point is calculated, the KL divergence of the pulse interval distribution is quantified, and the pulse similarity vector is generated; Generation of difference coefficient matrix: The time domain deviation vector, frequency domain offset vector, and pulse similarity vector are tensor-concatenated, and dimension differences are eliminated through normalization processing weighted by feature importance, and a third-order difference coefficient matrix with the output dimension of [device number × timestamp × difference type] is output.

[0010] Optionally, the frequency domain energy offset detection specifically includes: In the frequency domain analysis stage, the corresponding current harmonic components in the theoretical and measured electrical quantity sequences are first extracted, and wavelet packet decomposition is used to perform multi-scale frequency band analysis on each group of harmonic signals to extract the energy distribution characteristics within the preset sensitive frequency band (3kHz-15kHz). The sensitive frequency band is selected to cover the operating frequency range of typical power electronic equipment. After the extraction is completed, the energy density of the theoretical and measured data within the sensitive frequency band is quantitatively calculated respectively, and the offset index of each order of harmonics is extracted based on the relative offset degree of the two. The energy offset results of all orders are summarized to construct a frequency domain offset vector.

[0011] Optionally, in the pulse sequence similarity evaluation, the structure matching algorithm adopts the Hausdorff distance algorithm to perform position matching on the mutation point sets in the two sequences and identify the phase synchronization error.

[0012] Optionally, the abnormal area association positioning module specifically includes: Grid topology modeling: Build a grid node connection relationship topology map based on user equipment location information, annotate the line impedance parameters between nodes, superimpose the reverse current constraint conditions of the distributed power access point on the topology map, and generate an electrification topology model including the impedance matrix and the node admittance matrix; Abnormal propagation simulation: Map the difference coefficient matrix to the corresponding nodes of the power grid topology model; perform abnormal propagation deduction based on the graph neural network (GNN), and the calculation of abnormal propagation deduction includes: i. Calculate the attenuation factor of abnormal current according to the node impedance parameters; ii. Capture cross-region abnormal correlation features through multi-head attention mechanism; iii. Monte Carlo method is used to simulate the diffusion path of abnormal current in the topological network; Probability distribution generation: Count the frequency of abnormal currents in each line in the simulated propagation, calculate the abnormal current residence probability value based on the line impedance parameters, generate an abnormal probability thermal distribution map covering the entire network, and mark the suspicious line set whose probability value exceeds the preset abnormal residence probability threshold; Physical area positioning: perform spatial clustering analysis on the abnormal probability thermal distribution map to identify abnormal probability clustering areas; define the physical boundaries of abnormal power consumption based on the user equipment location and the grid topology connection relationship.

[0013] Optionally, the abnormal area association positioning module further includes outputs including suspicious device identification and abnormal propagation main path, wherein: The suspicious device identification is based on the grid node connected to the suspicious line set, and the grid node is bound to the actual user device to form a suspicious device identification set, indicating a potential abnormal power source head or affected terminal; The main path of abnormal propagation is achieved by recording the node paths and their order experienced in each round of propagation during the abnormal diffusion process of Monte Carlo simulation. Among all simulated paths, the frequency of occurrence of each path is counted, and the path sequence with the highest cumulative occurrence frequency is selected as the main path of abnormal propagation. The output main path sequence of abnormal propagation is a structured and ordered node list, which reflects the main propagation trajectory of abnormal information in the power grid.

[0014] Optionally, the adaptive verification strategy generation module specifically includes: When the abnormal probability value of a certain area in the abnormal probability thermal distribution map exceeds the preset abnormal residence probability threshold, a monitoring mode switching instruction is sent to the metering terminal to which the area belongs to execute: i. Increase the voltage / current sampling frequency to 5-10 times the original frequency; ii. Synchronously enable the real-time tracking mode of harmonic components to capture the 3rd to 39th order harmonic phase mutations; iii. Deploy transient event detectors on the edge side to record voltage sag / swell waveform segments; Impedance perturbation test execution: Apply multi-band impedance perturbation to the adjacent node devices with the largest abnormal probability gradient change in the abnormal probability thermal distribution diagram.

[0015] Optionally, the multi-band impedance disturbance includes: i. Inject 0.1-2kHz swept frequency test signal through controllable current source; ii. Theoretical impedance response calculation: based on topological model and impedance matrix , calculate the theoretical response impedance spectrum; Acquisition of measured impedance response: Record the voltage of each node after the test signal is injected to obtain the measured response impedance spectrum; iii. Abnormal offset determination: Calculate the Euclidean distance between the theoretical response impedance spectrum and the measured response impedance spectrum; iv. Compare the Euclidean distance between the measured impedance spectrum and the theoretical impedance spectrum, and calculate the abnormal node offset percentage. When the abnormal node offset percentage exceeds the second threshold (12%-18%), it is marked as "impedance abnormality associated device". The impedance offset is the relative deviation between the measured impedance value Zmeas and the theoretical expected impedance Zref.

[0016] Beneficial effects of the present invention: The present invention introduces a historical feature mining mechanism based on wavelet decomposition, harmonic coupling analysis and dynamic time warping, constructs a hybrid power model that integrates LSTM time series prediction, frequency domain attention correction and pulse compensation, and improves the dynamic response modeling capability of complex power consumption behaviors (such as energy storage equipment charging and discharging, variable load fluctuations). Compared with the traditional static power model, the proposed method can accurately reflect the power fluctuation law under equipment aging, harmonic pollution and non-steady-state operation conditions, and realize high-precision theoretical power output with adaptive correction capability, providing a traceable dynamic benchmark for subsequent difference analysis.

[0017] The present invention proposes a graph neural network anomaly propagation modeling method that integrates the difference coefficient matrix and the power grid topology structure. By introducing the impedance attenuation factor, the multi-head attention mechanism and the Monte Carlo diffusion simulation, an anomaly recognition system that cooperates with the electrical properties and the data-driven algorithm is constructed. This mechanism can not only effectively identify highly concealed anomaly sources such as multi-point collaborative electricity theft and inverter failure, but also output a probabilistic thermal distribution map and a main propagation path, thereby improving the spatial accuracy and structural coherence of anomaly tracing, and breaking through the bottleneck that the existing methods cannot distinguish between noise disturbances and actual anomalies.

[0018] Based on the abnormal area positioning results, the present invention further constructs a gradient intelligent verification strategy system. Through the conditional activation of the high-frequency power quality monitoring mode and the multi-band impedance disturbance injection mechanism, it realizes the refined multi-dimensional review of key areas, and can dynamically adjust the sampling frequency, monitoring dimension and detection intensity to ensure that verification resources are allocated according to risk priority, thereby improving the overall verification efficiency and positioning accuracy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 The management system of the embodiment of the present invention provides you with a module diagram; Figure 2 Schematic diagram of an adaptive verification strategy generation module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0022] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0023] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0024] like Figure 1-Figure 2 As shown, an electric power accounting and management system based on artificial intelligence includes: Dynamic power accounting module: builds a dynamic power calculation model based on the historical operation data of user equipment, collects the actual voltage waveform, current harmonic components and power pulse sequence of the current power-consuming equipment in real time, and outputs the theoretical power value through the dynamic power calculation model; Multi-dimensional accounting difference analysis module: performs three-dimensional difference analysis on the theoretical power value and the actual power value measured by the metering terminal. The three-dimensional difference analysis includes time domain cumulative deviation, frequency domain energy offset and pulse sequence similarity, and generates a device-level difference coefficient matrix; Abnormal area association positioning module: The difference coefficient matrix is ​​input into the spatial topology analysis network, and combined with the grid node impedance parameters and user equipment location information, an abnormal probability thermal distribution map of the abnormal current propagation path is generated to locate the abnormal power consumption physical area; Adaptive verification strategy generation module: configures verification parameters according to the abnormal probability thermal distribution map, including: Enable high-frequency power quality monitoring mode for high-probability abnormal areas; Apply impedance disturbance test to adjacent node devices.

[0025] The dynamic electricity accounting module specifically includes: 1. Historical data feature mining: Multi-dimensional decomposition of historical operation data of user devices, including: 1.1 Steady-state and transient energy ratio extraction: Adopt adaptive wavelet packet decomposition algorithm to extract voltage waveform The steady-state component With transient component , and calculate its energy share: , It is the energy proportion of the steady-state component in the voltage waveform, reflecting the energy proportion of the periodic and continuous stable part of the voltage signal. It is the energy proportion of the transient component in the voltage waveform, reflecting the energy proportion of the non-stationary components caused by switching, disturbance, etc. 1.2. Harmonic-load factor correlation matrix construction: Use the harmonic coupling analysis algorithm to establish the current harmonic components Equipment load factor The correlation matrix : , Indicates Subharmonic current components, is the equipment load factor; 1.3. Dynamic alignment of pulse sequences: Use the dynamic time warping (DTW) algorithm to align power pulse sequences under different working conditions : ,in, They represent the power pulse sequences under the 𝑖th and 𝑗th historical conditions respectively.

[0026] 2. Construction of power calculation model: Input the above processed historical data into the hybrid prediction network, including: 2.1. Time series prediction unit: LSTM network based on equipment aging curve, input historical feature vector , output basic power forecast value : ; 2.2. Frequency Domain Attention Correction Module: A fully connected network with frequency domain attention mechanism embedded in it. After correction, the output is: , FC-Attn represents a fully connected network with frequency domain attention mechanism (Fully Connected + Attention); 2.3. Pulse characteristic compensator: based on the currently acquired power pulse sequence Calculate the compensation factor , and adjust the output: .

[0027] 3. Real-time data fusion: The following features are captured synchronously through the edge-side heterogeneous data acquisition module: Zero-crossing distortion rate of voltage waveform Phase jitter parameters ; The 3rd to 39th order content of current harmonics And phase angle distribution , For the The phase angle of the subharmonics; The amplitude gradient of the power pulse is abrupt and time interval entropy .

[0028] 4. Theoretical value calculation: Input the above real-time collection characteristics into the dynamic power calculation model and execute: 4.1. Adaptive filtering: Apply adaptive filters based on the current operating stage of the device and environmental conditions (temperature and humidity) , filter out noise interference: , represents the environmental noise adaptive filter; 4.2. Spatial-temporal feature fusion calculation: using fusion function ,Integrate the voltage, current, power pulse and other features to generate a fusion feature vector: ; 4.3. Output theoretical power value: Based on the above fusion feature vector, output the theoretical power prediction value including the dynamic fluctuation range : , Represents a hybrid prediction network.

[0029] Predictor receives the fusion function The obtained comprehensive feature vector is used to call the constructed dynamic power model structure, and the output result combines the dynamic power theoretical value of timing, frequency domain and pulse characteristics, that is, The detailed process is as follows: 4.3.1. Feature access phase: receiving The integrated fusion feature vector , which integrates: Voltage, current and power pulse signals after adaptive filtering; Auxiliary features such as spatial phase jitter, harmonic phase distribution, pulse interval entropy, etc. Contextual information that characterizes the equipment's aging and operating stage.

[0030] 4.3.2. Main process of model prediction: calling the constructed hybrid prediction network model, mainly including: a) LSTM time series prediction submodule: Utilize historical feature data and current fusion features ; Output the basic power forecast value of the device at the current stage , reflecting the aging trend and periodic load characteristics.

[0031] b) Frequency domain attention correction submodule: enter and harmonic load matrix ; Weighted correction for the error caused by harmonic distortion, output .

[0032] c) Pulse characteristic compensator: Model the current power pulse characteristics (such as mutation gradient, interval entropy) and output the pulse compensation term ,correct the fluctuations caused by sudden loads.

[0033] 4.3.3. Final theoretical power output: Integrate the above three results to generate the final theoretical power prediction value : .

[0034] The multi-dimensional accounting difference analysis module specifically includes: 1. Time domain cumulative deviation calculation: 1.1. Theoretical power value The measured power value Using sliding window Make a comparison, is the length of the sliding time window; 1.2. Use dynamic time warping (DTW) to align the asynchronously sampled power sequences and obtain the aligned sequences: ; 1.3. Calculate the cumulative deviation within each window : , represents the time index within the sliding window, is the cumulative deviation corresponding to time 𝑡; 1.4. The deviations of all windows are combined into a time domain deviation vector : ;in, is the theoretical power value, is the measured power value, is the power sequence aligned by DTW.

[0035] 2. Frequency domain energy offset detection: 2.1. Yes and The current harmonic components are decomposed by wavelet packets to extract the frequency interval kHz,15kHz]; 2.2. Calculate the energy spectral density ratio within the frequency band : , is the energy spectrum density of the measured harmonic current signal in the frequency band 𝑓, represents the energy spectrum density of the theoretical harmonic current signal in the frequency band 𝑓; Extracting harmonic orders Energy shift index , forming a frequency domain offset vector : ,in, For the Subharmonics in the frequency band The energy spectral density within are the theoretical and measured harmonic current signals, Indicates Frequency domain energy shift index of subharmonics.

[0036] 3. Pulse sequence similarity evaluation: 3.1. Definition of theoretical pulse sequence The measured pulse sequence ; 3.2. Matching Theoretical Pulse Sequences Using Hausdorff Distance The set of pulse amplitude mutation points The measured pulse sequence The set of pulse amplitude mutation points : , Represents the Hausdorff distance, which is used to measure the maximum and minimum distance difference between two point sets (theoretical and measured pulse mutation point sets). is the set of mutation points extracted from the theoretical pulse sequence, is the set of mutation points extracted from the measured pulse sequence, is any mutation point in the theoretical sequence, is any mutation point in the measured sequence, Indicate point With point The Euclidean distance between them, sup is the supremum (i.e. "maximum value"), taking the maximum value in the set, and inf is the infimum (i.e. "minimum value"), taking the minimum value in the set. It means taking the larger of the maximum and minimum distances in the two directions as the final symmetry distance measure.

[0037] 3.3. Calculation of phase synchronization error at the mutation point First, extract the position points where the amplitude changes suddenly from the theoretical pulse sequence and the measured pulse sequence. These position points represent typical electrical events such as equipment start-up and stop, current impact or state switching. The extraction method can be based on the power change rate or complete, and the output is two time point sets, corresponding to the mutation moments in the theoretical sequence and the measured sequence respectively. Then, establish a one-to-one correspondence between the two mutation point sets. A matching strategy based on the minimum time distance is adopted, that is, each mutation point in the theoretical sequence is paired with the mutation point closest in time in the measured sequence. After the mutation point pairing is completed, the time difference between each pair of mutation points is calculated. In order to convert the time difference into a phase quantity with periodicity and comparability, the average period of the pulse sequence is introduced as a standardized scale. That is, the time difference of each pair of mutation points is divided by the period length and multiplied by the angle range corresponding to the complete period, and it is mapped to an angle quantity, so as to obtain the relative phase offset value of each pair of mutation points. Finally, the phase offsets of all mutation point pairs are statistically summarized, and the average value is taken as the overall phase synchronization error, which is used to reflect the degree of synchronization between the theoretically predicted mutation behavior and the actual observation in the time dimension.

[0038] 3.4. Statistical theory and measured pulse interval probability distribution , calculate KL divergence: , is the Kullback-Leibler divergence between the two; 3.5. Constructing the pulse similarity vector : ,in, represents the set of theoretical and measured pulse mutation points, represents the phase synchronization error, is the KL divergence of the pulse interval distribution.

[0039] 4. Generation of coefficient of difference matrix: 4.1. Concatenate the above three types of features into a joint feature tensor: ; 4.2. Normalize and weight the tensor (with feature importance as weight) : , represents the 𝑖th type of difference features (such as time domain, frequency domain, pulse), is the weighting factor of the 𝑖th class of difference features, is the normalized weighted difference feature tensor; 4.3. Output the final third-order matrix of difference coefficients : ; in, is the device number index, 𝑡 is the timestamp index, 𝑐 is the difference type (time domain, frequency domain, pulse), is the difference tensor after concatenation, is the feature weight factor, is the normalized difference tensor.

[0040] The abnormal area association positioning module specifically includes: 1. Grid topology modeling: 1.1. Build an electrical topology diagram: According to the geographical location information and power distribution structure of the user equipment, establish a node connection relationship topology diagram: ,in, Represents a collection of nodes (devices, transformers, access points), Indicates the electrical connection edge; 1.2. Line parameter annotation: each edge Associated impedance value , construct the impedance matrix : , Representation Node arrive The line impedance; 1.3. Reverse current constraint modeling: access point nodes with distributed power generation , define the reverse current boundary condition: , represents the set of nodes containing distributed generation, Representation Node The reverse current, Representation Node The reverse current threshold; 1.4. Admittance matrix generation: deriving the node admittance matrix from the topological structure and impedance relationship , forming an electrification topology model: .

[0041] 2. Abnormal propagation simulation: 2.1. Difference data mapping: Difference coefficient matrix Mapping to grid nodes: , which means extracting the difference tensor slices of the 𝑣th node at all times and all difference types, Represents the input feature vector of node 𝑣 in the power grid topology graph, which is used for the subsequent anomaly propagation simulation in the graph neural network. It integrates all the difference features related to the node. Agg() represents the "aggregation operation" and is used to compress the three-dimensional tensor into a one-dimensional vector; 2.2. Graph Neural Network Propagation Modeling: Based on the graph neural network propagation mechanism, calculate the abnormal state vector of each node ,The propagation update rules are as follows: i. Attenuation factor calculation: , the attenuation factor reflects the attenuation degree of abnormal current propagation between nodes, Representation Node The propagation attenuation factor is represents the impedance attenuation coefficient; ii. Multi-head attention fusion: ;in, For Node In the The updated feature representation in the layer graph neural network (i.e., the state vector after this round of propagation), For adjacent nodes In the The feature vector of the layer represents the state information of the previous round. For Node The adjacent node set of represents the grid nodes directly connected to it (determined according to the topological structure), For the Attention Head-to-Head The attention weight of the node For Node The degree of information influence under this channel is obtained through softmax normalization. For the The linear transformation matrix corresponding to the attention head is used to map the input features to the attention space of the channel. The number of attention heads in the multi-head attention mechanism. Each attention head learns a different weighting method for neighbor features, which improves the model's ability to model multiple relationships in complex graph structures. Indicates that all The features output by the attention heads are concatenated to form a node The final feature representation of .

[0042] iii. Anomaly diffusion simulation: Repeat the above propagation process Next, the Monte Carlo sampling strategy is introduced to simulate each initial disturbance and record the possible propagation paths of abnormal current, where Is a node The input feature vector of (derived from the coefficient of difference matrix), Indicates Layer node status, Representation Node The set of adjacent nodes of For simulation rounds.

[0043] 3. Probability distribution generation: 3.1. Propagation frequency statistics: for each edge exist The frequency of abnormal propagation in the simulation Perform normalized statistics: ; 3.2. Calculation of residence probability: Introduce line impedance weight and calculate abnormal residence probability: ; 3.3. Thermal distribution generation: Summarize the residence probabilities of all lines to form an abnormal probability thermal distribution map , and filter out those that exceed the abnormal residence probability threshold Suspicious line collection: ,in, For Line The frequency of abnormal transmission, is the propagation frequency, To propagate the residence probability, is the impedance correction factor, Indicates the abnormal residence probability threshold, ranging from 65% to 70%. is the abnormal probability thermal distribution diagram, A collection of suspicious lines.

[0044] 4. Physical area positioning: 4.1. Spatial cluster analysis: Based on the distribution of suspicious lines in the abnormal heat map, clustering is performed on its topological structure to identify regional sub-maps with abnormal concentration trends: , Indicates A topological subgraph of clustering areas; 4.2. Boundary demarcation and device mapping: Combine the grid connection relationship with the geographical location of the user's equipment, map the abnormal area to a specific physical coordinate space, and perform polygonal approximation on the regional boundary; 4.3. Positioning report output: Output a comprehensive report including the following contents: Suspicious device identification collection ; Abnormal propagation main path sequence .

[0045] The abnormal area association positioning module also includes outputs including suspicious device identification and the main path of abnormal propagation, where: The suspicious device identification is based on the grid nodes connected to the suspicious line set, binding the grid nodes with the actual user devices to form a suspicious device identification set, indicating potential abnormal power heads or affected terminals; The main path of anomaly propagation is determined by recording the node paths and their order experienced in each round of propagation during the anomaly diffusion process of Monte Carlo simulation. Among all simulated paths, the frequency of occurrence of each path is counted, and the path sequence with the highest cumulative occurrence frequency is selected as the main path of anomaly propagation. The output main path sequence of anomaly propagation is a structured and ordered node list, which reflects the main propagation trajectory of abnormal information in the power grid.

[0046] The adaptive verification strategy generation module specifically includes: When the abnormal probability value of a certain area in the abnormal probability thermal distribution map exceeds the preset abnormal residence probability threshold, a monitoring mode switching instruction is sent to the metering terminal to which the area belongs to execute: i. Increase the voltage / current sampling frequency to 5-10 times the original frequency; ii. Synchronously enable the real-time tracking mode of harmonic components to capture the 3rd to 39th order harmonic phase mutations; iii. Deploy transient event detectors on the edge side to record voltage sag / swell waveform segments; Impedance perturbation test execution: Apply multi-band impedance perturbation to the adjacent node devices with the largest abnormal probability gradient change in the abnormal probability thermal distribution diagram.

[0047] Multi-band impedance perturbations include: i. Inject 0.1-2kHz swept frequency test signal through controllable current source; ii. Theoretical impedance response calculation: based on topological model and impedance matrix , calculate the theoretical response impedance spectrum: ; Obtaining the measured impedance response: Record the voltage of each node after the test signal is injected to obtain the measured response impedance spectrum: ; iii. Abnormal offset determination: Calculate the Euclidean distance between the theoretical response impedance spectrum and the measured response impedance spectrum; iv. Compare the Euclidean distance between the measured impedance spectrum and the theoretical impedance spectrum, and calculate the abnormal node offset percentage. When the abnormal node offset percentage exceeds the second threshold (12%-18%), it is marked as "impedance abnormality associated device". The impedance offset is the relative deviation between the measured impedance value Zmeas and the theoretical expected impedance Zref.

[0048] The above measures achieve accurate identification and technical closed-loop verification of suspected abnormal electricity usage behaviors by implementing refined and enhanced online verification operations on metering terminals and key node equipment in high-risk areas.

[0049] The high-frequency monitoring part is used to improve the time and frequency resolution of anomaly capture, thereby identifying micro-disturbances and harmonic phase mutations that cannot be detected under traditional sampling; The impedance perturbation test part achieves accurate identification of the deviation of the physical layer electrical properties of suspicious nodes by comparing the injection-response characteristics with the offset of the theoretical model.

[0050] By increasing the sampling frequency and tracking harmonics, the detection capability of abnormal features such as transient disturbances, voltage distortion and harmonic phase mutations is enhanced, enabling the system to quickly perceive and diagnose sudden and high-frequency anomalies. By injecting multi-band disturbances and comparing impedance responses, it is possible to accurately determine whether the device has abnormal electrical characteristics or physical access deviations, avoiding misjudgment based on statistical indicators alone and enhancing the physical credibility of positioning.

[0051] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0052] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An electric power accounting and management system based on artificial intelligence, characterized in that: include: Dynamic power accounting module: builds a dynamic power calculation model based on the historical operation data of the user equipment, collects the actual voltage waveform, current harmonic components and power pulse sequence of the current power-consuming equipment in real time, and outputs the theoretical power value through the dynamic power calculation model; Multi-dimensional accounting difference analysis module: performs three-dimensional difference analysis on the theoretical power value and the actual power value measured by the metering terminal, wherein the three-dimensional difference analysis includes time domain cumulative deviation, frequency domain energy offset and pulse sequence similarity, and generates a device-level difference coefficient matrix; Abnormal area association positioning module: input the difference coefficient matrix into the spatial topology analysis network, combine the grid node impedance parameters and the user equipment location information, generate the abnormal probability thermal distribution map of the abnormal current propagation path, and locate the abnormal power consumption physical area; Adaptive verification strategy generation module: configures verification parameters according to the abnormal probability thermal distribution map, including: Enable high-frequency power quality monitoring mode for high-probability abnormal areas; Apply impedance disturbance test to adjacent node devices.

2. The power accounting and management system based on artificial intelligence according to claim 1 is characterized in that: The dynamic electricity accounting module specifically includes: Historical data feature mining: Multi-dimensional decomposition and processing of historical operation data of user equipment, including: using adaptive wavelet packet decomposition to extract the energy proportion of steady-state components and transient components of the voltage waveform, establishing the correlation matrix between current harmonic components and equipment load rate through harmonic coupling analysis, and using dynamic time warping algorithm to align power pulse sequence patterns under different working conditions; Dynamic power calculation model construction: The processed historical operation data is input into the hybrid prediction network, which includes: The LSTM timing prediction unit based on the equipment aging curve is used to generate the basic power prediction value. The fully connected network embedded with the frequency domain attention mechanism corrects the prediction deviation caused by harmonic distortion. Based on the pulse feature compensator, the prediction weight is dynamically adjusted according to the power pulse sequence collected in real time. The heterogeneous data acquisition units deployed on the edge synchronously capture the zero-crossing distortion rate and phase jitter parameters of the voltage waveform, the 3rd to 39th order content and phase angle distribution of the current harmonics, the amplitude mutation gradient and time interval entropy of the power pulse sequence; Theoretical value calculation: input the real-time collected data into the dynamic power calculation model to obtain the theoretical power value.

3. The power accounting and management system based on artificial intelligence according to claim 2 is characterized in that: The theoretical power value calculation includes executing: Adaptive filtering based on the equipment operation phase can eliminate the measurement noise caused by ambient temperature and humidity; Integrate the correlation features of voltage, current and pulse through spatiotemporal feature fusion algorithm; The output includes the theoretical power value within the normal operating fluctuation range, and the theoretical power value is dynamically updated as the equipment ages.

4. The power accounting and management system based on artificial intelligence according to claim 1 is characterized in that: The multi-dimensional accounting difference analysis module specifically includes: Time domain cumulative deviation calculation: The theoretical power value and the measured power value are compared in a sliding manner in a preset time window, and the dynamic time warping algorithm is used to align the asynchronously sampled power sequence, calculate the cumulative deviation in each window, and generate a time domain deviation vector; Frequency domain energy offset detection: Decompose the current harmonic components of the theoretical value and the measured value using the frequency domain decomposition algorithm, calculate the energy spectrum density ratio, extract the energy offset index of each harmonic, and construct the frequency domain offset vector; Pulse sequence similarity evaluation: Based on the structure matching algorithm, the distance distribution between the theoretical value and the measured pulse sequence is matched, the phase synchronization error of the pulse amplitude mutation point is calculated, the KL divergence of the pulse interval distribution is quantified, and the pulse similarity vector is generated; Generation of difference coefficient matrix: The time domain deviation vector, frequency domain offset vector, and pulse similarity vector are tensor-concatenated, and dimension differences are eliminated through normalization processing weighted by feature importance, and a third-order difference coefficient matrix with the output dimension of [device number × timestamp × difference type] is output.

5. The power accounting and management system based on artificial intelligence according to claim 4 is characterized in that: The frequency domain energy offset detection specifically includes: In the frequency domain analysis stage, the corresponding current harmonic components in the theoretical and measured electrical quantity sequences are first extracted, and multi-scale frequency band analysis is performed on each group of harmonic signals using wavelet packet decomposition to extract the energy distribution characteristics within a preset sensitive frequency band. The sensitive frequency band is selected to cover the operating frequency range of typical power electronic equipment. After the extraction is completed, the energy density of the theoretical and measured data within the sensitive frequency band is quantitatively calculated, and the offset index of each order of harmonics is extracted based on the relative offset degree of the two. The energy offset results of all orders are summarized to construct a frequency domain offset vector.

6. The power accounting and management system based on artificial intelligence according to claim 4 is characterized in that: In the pulse sequence similarity evaluation, the structure matching algorithm adopts the Hausdorff distance algorithm to perform position matching on the mutation point sets in the two sequences and identify the phase synchronization error.

7. The power accounting and management system based on artificial intelligence according to claim 1 is characterized in that: The abnormal area association positioning module specifically includes: Grid topology modeling: Build a grid node connection relationship topology map based on user equipment location information, annotate the line impedance parameters between nodes, superimpose the reverse current constraint conditions of the distributed power access point on the topology map, and generate an electrification topology model including the impedance matrix and the node admittance matrix; Abnormal propagation simulation: Map the difference coefficient matrix to the corresponding nodes of the power grid topology model; perform abnormal propagation deduction based on the graph neural network, and the calculation of abnormal propagation deduction includes: i. Calculate the attenuation factor of abnormal current according to the node impedance parameters; ii. Capture cross-region abnormal correlation features through multi-head attention mechanism; iii. Monte Carlo method is used to simulate the diffusion path of abnormal current in the topological network; Probability distribution generation: Count the frequency of abnormal currents in each line in the simulated propagation, calculate the abnormal current residence probability value based on the line impedance parameters, generate an abnormal probability thermal distribution map covering the entire network, and mark the suspicious line set whose probability value exceeds the preset abnormal residence probability threshold; Physical area positioning: perform spatial clustering analysis on the abnormal probability thermal distribution map to identify abnormal probability clustering areas; define the physical boundaries of abnormal power consumption based on the user equipment location and the grid topology connection relationship.

8. The power accounting and management system based on artificial intelligence according to claim 7 is characterized in that: The abnormal area association positioning module also includes outputs including suspicious device identification and abnormal propagation main path, wherein: The suspicious device identification is based on the grid node connected to the suspicious line set, and the grid node is bound to the actual user device to form a suspicious device identification set, indicating a potential abnormal power source head or affected terminal; The main path of abnormal propagation is achieved by recording the node paths and their order experienced in each round of propagation during the abnormal diffusion process of Monte Carlo simulation. Among all simulated paths, the frequency of occurrence of each path is counted, and the path sequence with the highest cumulative occurrence frequency is selected as the main path of abnormal propagation. The output main path sequence of abnormal propagation is a structured and ordered node list, which reflects the main propagation trajectory of abnormal information in the power grid.

9. The power accounting and management system based on artificial intelligence according to claim 1 is characterized in that: The adaptive verification strategy generation module specifically includes: When the abnormal probability value of a certain area in the abnormal probability thermal distribution map exceeds the preset abnormal residence probability threshold, a monitoring mode switching instruction is sent to the metering terminal to which the area belongs to execute: i. Increase the voltage / current sampling frequency to 5-10 times the original frequency; ii. Synchronously enable the real-time tracking mode of harmonic components to capture the 3rd to 39th order harmonic phase mutations; iii. Deploy transient event detectors on the edge side to record voltage sag / swell waveform segments; Impedance perturbation test execution: Apply multi-band impedance perturbation to the adjacent node devices with the largest abnormal probability gradient change in the abnormal probability thermal distribution diagram.

10. The power accounting and management system based on artificial intelligence according to claim 9 is characterized in that: The multi-band impedance disturbance includes: i. Inject 0.1-2kHz swept frequency test signal through controllable current source; ii. Theoretical impedance response calculation: based on topological model and impedance matrix , calculate the theoretical response impedance spectrum; Acquisition of measured impedance response: Record the voltage of each node after the test signal is injected to obtain the measured response impedance spectrum; iii. Abnormal offset determination: Calculate the Euclidean distance between the theoretical response impedance spectrum and the measured response impedance spectrum; iv. Compare the Euclidean distance between the measured impedance spectrum and the theoretical impedance spectrum, calculate the abnormal node offset percentage, and when the abnormal node offset percentage exceeds the second threshold, mark it as "abnormal impedance associated device".

Images