Direct current charging pile supporting rapid detection of electricity stealing behavior

Abstract

The invention provides a direct current charging pile supporting rapid detection of an electricity stealing behavior, and belongs to the technical field of charging piles, and the direct current charging pile supporting rapid detection of the electricity stealing behavior comprises hardware modules such as a control chip, a voltage acquisition device, a current acquisition device, a power analysis device and a carrier feature extraction device, and a system control module. The control chip executes functions of data acquisition, feature extraction, anomaly detection and the like. The system constructs a voltage data matrix and a current data matrix by collecting voltage and current data, and calculates a power feature vector; collecting carrier signals to establish a carrier attenuation matrix and a carrier demodulation anomaly matrix; and carrying out feature learning by using an improved residual neural network, realizing optimal combination of power line carrier features and power loss features through a feature fusion module, and finally outputting an electricity larceny behavior judgment result. According to the scheme, power line carrier characteristic analysis and power loss characteristic analysis are creatively combined, and rapid and accurate detection of the electricity stealing behavior is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of charging piles, and more specifically, relates to a DC charging pile that supports rapid detection of electricity theft behavior. Background Art

[0002] With the rapid development of the electric vehicle industry, DC charging piles have become an important part of the electric vehicle charging infrastructure. During the operation of charging piles, electricity theft behavior not only causes economic losses but also seriously affects the safety of the power grid and the normal operation of charging piles. Traditional detection of electricity theft behavior mainly relies on manual inspections, comparison of electricity meter measurements, and simple power loss analysis. The manual inspection method mainly checks the external lines and junction boxes of charging piles regularly to observe whether there are abnormal modification traces; the electricity meter measurement comparison method judges whether there is abnormal loss by comparing the difference in electrical energy measurement between the input and output ends of the charging pile; the power loss analysis method identifies whether there are abnormal fluctuations by real-time monitoring of the power curve during the charging process. These traditional detection methods have achieved certain results in practical applications, but there are still problems such as low detection efficiency and low accuracy.

[0003] The traditional detection methods have the following main defects: First, the manual inspection method consumes a large amount of manpower and cannot detect electricity theft behavior with strong concealment in a timely manner; second, the electricity meter measurement comparison method is easily affected by factors such as the self-loss fluctuation of the charging pile and measurement errors, and it is difficult to accurately judge minor electricity theft behavior; third, the simple power loss analysis method cannot effectively cope with complex electricity theft means, such as using shunt circuits, modifying metering devices, etc. to steal electricity. In addition, these methods often rely solely on a certain characteristic parameter for judgment, lacking the comprehensive analysis ability of multi-dimensional characteristics, resulting in insufficient reliability of detection results. With the continuous upgrading of electricity theft means, traditional detection methods are no longer able to meet the actual needs.

[0004] In recent years, although researchers have proposed electricity theft behavior detection methods based on deep learning, attempting to improve the detection accuracy by establishing complex mathematical models. However, these methods mainly focus on the statistical characteristics of electrical energy measurement data and ignore the rich characteristic information contained in power line carrier signals. At the same time, the existing deep learning model structures are relatively simple and lack targeted optimization design, making it difficult to fully extract the characteristic expressions of electricity theft behavior. Therefore, how to achieve rapid and accurate detection of electricity theft behavior in the DC charging pile system is still a technical problem to be solved urgently. Summary of the Invention

[0005] In view of this, the present invention provides a DC charging pile that supports rapid detection of electricity theft behavior, which can solve the technical problem that it is difficult to rapidly and accurately detect electricity theft behavior in the DC charging pile system in the prior art.

[0006] The present invention is implemented as follows: The present invention provides a DC charging pile that supports rapid detection of electricity theft behavior, including a control chip, a voltage acquisition device, a current acquisition device, a power analysis device, a carrier feature extraction device, a signal conditioning device, a data storage device, a real-time communication device, an alarm device, and a power management device. A system control module is provided in the control chip for collecting voltage and current data at the input and output ends of the charging pile to establish a matrix; calculating active power and reactive power to construct a power feature vector; collecting carrier signal features to establish a carrier attenuation matrix; demodulating and analyzing carrier signals to construct a carrier demodulation anomaly matrix; calculating the probability of electricity theft behavior occurring based on a pre-trained electricity theft behavior recognition model; calculating an electricity theft behavior judgment index according to an electricity theft behavior judgment equation set; recording system operation parameters and historical data; when the time series correlation index of electricity theft behavior exceeds a preset threshold, sending an alarm message and emitting an audible and visual alarm signal.

[0007] Among them, the voltage acquisition device is used to collect voltage data at the input end of the charging pile and voltage data at the output end of the charging pile, and the current acquisition device is used to collect current data at the input end of the charging pile and current data at the output end of the charging pile. The sampling frequency of the voltage acquisition device and the current acquisition device is 1000 times per second.

[0008] Among them, the power analysis device is used to calculate the active power at the input end of the charging pile, the reactive power at the input end of the charging pile, the active power at the output end of the charging pile, and the reactive power at the output end of the charging pile, and record the power fluctuation characteristics.

[0009] Among them, the carrier feature extraction device is used to collect the carrier amplitude, carrier phase, and carrier frequency. The sampling frequency of the carrier feature extraction device is 100 times per second.

[0010] Among them, the signal conditioning device is used to filter and amplify the collected signals, the data storage device is used to store system operation parameters and historical data, the real-time communication device is used to transmit data to the monitoring center, the alarm device is used to emit an audible and visual alarm signal, and the power management device is used to monitor the system power supply voltage.

[0011] Among them, the carrier attenuation matrix refers to the attenuation of the power line carrier signal in the transmission process, including the carrier amplitude, carrier phase, and carrier frequency. The matrix elements include the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value.

[0012] Among them, the carrier demodulation anomaly matrix refers to the deviation between the modulation parameters obtained after demodulating the carrier signal and the theoretical parameters. The matrix elements include the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value.

[0013] Among them, the electricity theft behavior judgment equation set includes a power characteristic equation, a carrier attenuation characteristic equation, and a time series correlation equation. The power characteristic equation is used to calculate the power loss abnormality index. The carrier attenuation characteristic equation is used to calculate the carrier attenuation abnormality index. The time series correlation equation is used to calculate the electricity theft behavior time series correlation index. The inputs of the power characteristic equation include the active power at the input end of the charging pile, the active power at the output end of the charging pile, the reactive power at the input end of the charging pile, the reactive power at the output end of the charging pile, and the power loss threshold. The inputs of the carrier attenuation characteristic equation include the carrier amplitude attenuation coefficient, the carrier phase offset, the carrier frequency drift value, and the standard attenuation threshold. The standard attenuation threshold refers to the standard value of the carrier signal attenuation under normal working conditions. The inputs of the time series correlation equation include the power loss abnormality index, the carrier attenuation abnormality index, historical data, and the time weight coefficient. The time weight coefficient refers to the weight of the occurrence of electricity theft behavior in different time periods.

[0014] Among them, the electricity theft behavior recognition model adopts an improved residual neural network structure, and a feature fusion module is added between the feature extraction layer and the classification layer of the improved residual neural network structure. The feature fusion module is used to fuse the power carrier feature and the power loss feature. The input is the carrier attenuation matrix and the power feature vector, and the output is the electricity theft behavior feature vector.

[0015] Among them, the improved residual neural network structure includes an input layer, a feature extraction layer, a feature fusion module, and a classification layer. The input layer uses a normalization processing module to normalize the input data. The feature extraction layer includes multiple improved residual blocks. The improved residual block adds a channel attention mechanism and a spatial attention mechanism to the standard residual structure. The feature fusion module uses an adaptive weight method to fuse the power carrier feature and the power loss feature. The classification layer uses a fully connected layer to reduce the dimension of the feature vector and output the probability of the occurrence of electricity theft behavior.

[0016] Among them, the steps for establishing the training data set of the electricity theft behavior recognition model include collecting normal operation state data and electricity theft state data, labeling the electricity theft behavior labels corresponding to the carrier attenuation matrix and the power feature vector, and dividing the data set into a training data set, a validation data set, and a test data set according to a ratio of 8:1:1.

[0017] Among them, the power analysis device uses a digital signal processor to calculate the active power and reactive power through a digital integration algorithm. The signal conditioning device includes a two-stage operational amplifier circuit and a Butterworth low-pass filter. The signal conditioning device is used to amplify and filter the collected signal.

[0018] Among them, the preset threshold is set to 0.8, the sound pressure level of the alarm device is 95 decibels, the flash frequency is 2 Hz, the real-time communication device sends data to the monitoring center using an encrypted transmission method, and the data transmission delay is less than 100 milliseconds.

[0019] Among them, the division ratio of the training data set of the electricity theft behavior recognition model is that the ratio of the training data set, the validation data set, and the test data set is 8:1:1. The model training uses the mini-batch stochastic gradient descent algorithm, the batch size is set to 32, the initial value of the learning rate is 0.001, and the cosine annealing strategy is used to dynamically adjust the learning rate. Training stops when the validation accuracy reaches 95%.

[0020] Compared with the prior art, the present invention provides a DC charging pile supporting rapid detection of electricity theft behavior. The present invention proposes a DC charging pile system supporting rapid detection of electricity theft behavior, innovatively combines power line carrier feature analysis and power loss feature analysis, and constructs an electricity theft behavior recognition model based on an improved residual neural network. The system collects multi-dimensional data during the charging process through hardware modules such as a voltage acquisition device and a current acquisition device, uses a carrier feature extraction device to analyze the characteristics of power line carrier signals in real time, and calculates the power loss characteristics through a power analysis device. The system uses a feature fusion module to adaptively fuse the power line carrier characteristics and the power loss characteristics, makes full use of multi-source information for comprehensive judgment, and significantly improves the accuracy and real-time performance of electricity theft behavior detection.

[0021] The present invention solves the main defects of traditional detection methods: by introducing power line carrier feature analysis, the system can capture the abnormal carrier signal caused by electricity theft behavior and effectively identify the concealed electricity theft behavior that is difficult to discover by traditional methods; by constructing a carrier attenuation matrix and a carrier demodulation anomaly matrix, the system realizes the refined characterization of the carrier signal characteristics and overcomes the limitations of single feature parameter judgment; by designing a feature fusion module, the system realizes the optimal combination of power line carrier characteristics and power loss characteristics and improves the reliability of the detection results. At the same time, the improved residual neural network structure enhances the model's ability to extract key features by introducing an attention mechanism.

[0022] The present invention successfully solves the technical problem that it is difficult to quickly and accurately detect electricity theft behavior in a DC charging pile system, which is mainly reflected in the following aspects: First, the system establishes a complete electricity theft behavior feature representation system through multi-source data collection and feature extraction; second, through the combination of feature fusion and deep learning, the system realizes the intelligent recognition of electricity theft behavior; finally, the system has strong practicability and scalability and can adapt to different types of electricity theft behavior detection requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1Schematic diagram of the module composition of the charging pile of the present invention.

[0024] Figure 2 Flow chart of the steps executed by the system control module.

[0025] Figure 3 Comparison diagram of the voltage and current calibration test results in Example 2.

[0026] Figure 4 Graph of the change trend of power over time in Example 2.

[0027] Figure 5 Graph of the change trend of carrier characteristics in Example 2.

[0028] Figure 6 Graph of the change trend of the power theft behavior judgment index in Example 2. Detailed implementation manners

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0030] As Figure 1As shown in the figure, it is a schematic diagram of the module composition of a DC charging pile provided by the present invention for rapid detection of electricity theft behavior. The DC charging pile for rapid detection of electricity theft behavior includes a control chip, a voltage acquisition device, a current acquisition device, a power analysis device, a carrier feature extraction device, a signal conditioning device, a data storage device, a real-time communication device, an alarm device, and a power management device. The control chip is electrically connected to the voltage acquisition device, the current acquisition device, the power analysis device, the carrier feature extraction device, the signal conditioning device, the data storage device, the real-time communication device, the alarm device, and the power management device respectively. A system control module is provided in the control chip. The voltage acquisition device is used to acquire the voltage data at the input end of the charging pile and the voltage data at the output end of the charging pile. The current acquisition device is used to acquire the current data at the input end of the charging pile and the current data at the output end of the charging pile. The power analysis device is used to calculate the active power at the input end of the charging pile, the reactive power at the input end of the charging pile, the active power at the output end of the charging pile, and the reactive power at the output end of the charging pile. The carrier feature extraction device is used to acquire the carrier amplitude, the carrier phase, and the carrier frequency. The signal conditioning device is used to filter and amplify the acquired signals. The data storage device is used to store the system operation parameters and historical data. The real-time communication device is used to transmit data to the monitoring center. The alarm device is used to emit an audible and visual alarm signal. The power management device is used to monitor the system power supply voltage. The sampling frequency of the voltage acquisition device and the current acquisition device is 1000 times per second, and the sampling frequency of the carrier feature extraction device is 100 times per second. The system control module is used to execute the following steps: S01. Respectively acquire the voltage data at the input end of the charging pile, the voltage data at the output end of the charging pile, the current data at the input end of the charging pile, and the current data at the output end of the charging pile through the voltage acquisition device and the current acquisition device, and establish a voltage data matrix and a current data matrix; S02. Use the power analysis device to calculate the active power at the input end of the charging pile, the reactive power at the input end of the charging pile, the active power at the output end of the charging pile, and the reactive power at the output end of the charging pile, construct a power feature vector, and record the power fluctuation feature; S03. Acquire the carrier amplitude, the carrier phase, and the carrier frequency through the carrier feature extraction device, and establish a carrier attenuation matrix; S04. Demodulate and analyze the acquired carrier signals, extract the carrier amplitude attenuation coefficient, the carrier phase offset, and the carrier frequency drift value, and construct a carrier demodulation anomaly matrix; S05. Based on the pre-trained electricity theft behavior recognition model, input the carrier attenuation matrix and the carrier demodulation anomaly matrix into the model. Use the carrier attenuation matrix to construct a carrier feature network graph, calculate the minimum spanning tree using Kruskal's algorithm, take the sum of the weights of the minimum spanning tree as the electricity theft behavior feature index, and combine with the carrier demodulation anomaly matrix to calculate the probability of electricity theft behavior occurrence; S06. Calculate the electricity theft behavior judgment index according to the electricity theft behavior judgment equation set. The electricity theft behavior judgment equation set includes a power feature equation, a carrier attenuation feature equation, and a time series correlation equation. The power feature equation is used to calculate the power loss anomaly degree index. The inputs of the power feature equation include the active power at the input end of the charging pile, the active power at the output end of the charging pile, the reactive power at the input end of the charging pile, the reactive power at the output end of the charging pile, and the power loss threshold. The carrier attenuation feature equation is used to calculate the carrier attenuation anomaly degree index. The inputs of the carrier attenuation feature equation include the carrier amplitude attenuation coefficient, the carrier phase offset, the carrier frequency drift value, and the standard attenuation threshold. The time series correlation equation is used to calculate the electricity theft behavior time series correlation index. The inputs of the time series correlation equation include the power loss anomaly degree index, the carrier attenuation anomaly degree index, the historical data, and the time weight coefficient; S07. Use the data storage device to record the system operation parameters and the historical data, and establish an electricity theft behavior feature database; S08. When the electricity theft behavior time series correlation index exceeds the preset threshold, send an alarm message to the monitoring center through the real-time communication device, and at the same time activate the alarm device to send out an audible and visual alarm signal.

[0031] Among them, the carrier attenuation matrix refers to the attenuation conditions of the carrier amplitude, the carrier phase, and the carrier frequency during the transmission of the power line carrier signal. The matrix elements include the carrier amplitude attenuation coefficient, the carrier phase offset, and the carrier frequency drift value; the carrier demodulation anomaly matrix refers to the deviation between the modulation parameters obtained after demodulating the carrier signal and the theoretical parameters. The matrix elements include the carrier amplitude attenuation coefficient, the carrier phase offset, and the carrier frequency drift value; the power loss anomaly degree index refers to the abnormal degree of power loss between the input end and the output end of the charging pile; the carrier attenuation anomaly degree index refers to the abnormal degree of power line carrier signal attenuation; the standard attenuation threshold refers to the standard value of carrier signal attenuation under normal working conditions; the time weight coefficient refers to the weight of electricity theft behavior occurrence in different time periods.

[0032] The electricity theft behavior recognition model uses an improved residual neural network structure, and a feature fusion module is added between the feature extraction layer and the classification layer of the improved residual neural network structure. The feature fusion module is used to fuse the power carrier feature and the power loss feature. The input is the signal carrier attenuation matrix and the power feature vector, and the output is the electricity theft behavior feature vector. The steps for establishing the training data set of the electricity theft behavior recognition model include collecting normal operation state data and electricity theft state data, labeling the electricity theft behavior labels corresponding to the signal carrier attenuation matrix and the power feature vector, and dividing the data set into a training data set, a validation data set, and a test data set according to a ratio of 8:1:1. The steps for training the electricity theft behavior recognition model specifically include inputting the training data set into the improved residual neural network structure for forward propagation calculation, using the cross-entropy loss function to calculate the error between the prediction result and the true label, optimizing and updating the network parameters through the backpropagation algorithm, using the validation data set to evaluate the training process, stopping training when the validation accuracy reaches the preset threshold, and using the test data set to perform performance testing on the trained model.

[0033] Among them, the improved residual neural network structure includes an input layer, a feature extraction layer, a feature fusion module, and a classification layer. The input layer uses a normalization processing module to normalize the input data. The feature extraction layer includes multiple improved residual blocks, and a channel attention mechanism and a spatial attention mechanism are added to the standard residual structure. The feature fusion module uses an adaptive weight method to fuse the power carrier feature and the power loss feature. The classification layer uses a fully connected layer to reduce the dimension of the feature vector and output the probability of electricity theft behavior occurrence.

[0034] The specific implementation manners of the present invention will be described in detail below.

[0035] The control chip uses a high-performance 32-bit processor STM32H750VBT6 with a main frequency of up to 480 MHz, built-in 2 MB flash memory and 1 MB RAM, and is used to execute the system control program, data processing, and communication management. Among them, the system control module performs data interaction and control signal transmission with each functional module through an internal bus.

[0036] The voltage acquisition device includes two groups of high-precision voltage sensors at the input end and the output end. The Hall voltage sensor LEM LV25P is used, with a range of 0 to 1000 V, an accuracy of 0.1%, a sampling frequency of 1000 Hz, and data acquisition is performed through a 16-bit ADC to realize real-time monitoring of the voltages at the input end and the output end of the charging pile.

[0037] The current acquisition device also includes two groups of high-precision Hall current sensors LEMLAH25NP at the input and output ends, with a range of 0 to 100A, an accuracy of 0.1%, a sampling frequency of 1000Hz, and data acquisition is carried out through a 16-bit ADC to realize real-time monitoring of the currents at the input and output ends of the charging pile.

[0038] The power analysis device uses a dedicated digital signal processor TMS320F28335 with a main frequency of 150MHz, an internal hardware multiplier and a floating-point operation unit. The active power and reactive power are calculated through a digital integration algorithm, and the calculation accuracy is better than 0.2%. It can process the power data at the input and output ends simultaneously.

[0039] The carrier feature extraction device uses a high-speed data acquisition chip AD9226 with a sampling rate of 100Hz and a resolution of 12 bits, and is combined with a band-pass filter and a digital phase-locked loop circuit to realize the extraction of the amplitude, phase and frequency characteristics of the power line carrier signal.

[0040] The signal conditioning device includes two-stage operational amplifier circuits and Butterworth low-pass filters. The operational amplifier uses OP27, and the adjustable gain range is 1 to 100 times, and the cut-off frequency is 10kHz. It is used to amplify and filter the acquired signals to improve the signal-to-noise ratio of the signals.

[0041] The data storage device uses a 32GB industrial-grade eMMC flash memory, supports the FAT32 file system, has a power-off protection function, and is used to store system configuration parameters, operation logs and historical data. The data retention period is 3 months.

[0042] The real-time communication device uses a 4G wireless communication module SIM7600CE, supports the TCP / IP protocol, the communication rate can reach 100Mbps, and uses an encrypted transmission method to send data to the monitoring center. The data transmission delay is less than 100ms.

[0043] The alarm device includes an audible and visual alarm and an LED indicator. The sound pressure level of the audible and visual alarm is 95 decibels, and the flashing frequency is 2Hz. The LED indicator uses a three-color indication method to display the working state of the system. Green indicates normal, yellow indicates warning, and red indicates failure.

[0044] The power management device uses a multi-channel DC / DC converter and a power monitoring chip MAX17055. The input voltage range is 9 to 36V, and it has overvoltage, undervoltage and overcurrent protection functions. It can provide a stable working voltage for each module with an efficiency as high as 95%.

[0045] In the specific implementation of the carrier feature extraction device, the carrier sampling circuit uses the high-speed sampling and holding circuit SHC804 and the 12-bit analog-to-digital converter AD9226. The sampling clock is provided by the programmable clock generator Si5351A. Phase detection adopts the quadrature demodulation method, and frequency measurement adopts the zero-crossing detection method.

[0046] In the specific implementation of the signal conditioning device, a two-stage non-inverting amplifier circuit is adopted. The gain of the first stage is 10 times, and the gain of the second stage is adjustable. The filter uses a fourth-order Butterworth low-pass filter. The operational amplifier selects the OP27 with low noise and low drift. The circuit gain and bandwidth can be adjusted by the programmable potentiometer AD8400.

[0047] Data communication is carried out between the above-mentioned various devices using the RS485 bus. The communication rate is 115200bps, and the Modbus RTU protocol is adopted. Each device has an independent slave address, and the control chip, as the master, uniformly manages data transmission.

[0048] All modules use industrial-grade components. The operating temperature range is from -40 degrees Celsius to 85 degrees Celsius, the relative humidity is from 5% to 95%, and the protection level reaches IP65, meeting the requirements for outdoor installation and use.

[0049] The power supply of this system adopts a switching power supply scheme. The input voltage is 220V AC, and multiple DC voltages are output, including 3.3V, 5V, 12V, etc. The total power is 200W. It has a perfect electromagnetic compatibility design and strong anti-interference ability.

[0050] The specific implementation manner of the steps executed by the system control module is described in detail as follows.

[0051] The specific implementation manner of step S01 is to adopt a high-speed data acquisition scheme to realize the acquisition of voltage and current data at the input and output ends of the charging pile. First, the voltage acquisition device is used to acquire the voltage data at the input and output ends of the charging pile. The sampling frequency is 1000Hz. After the sampled data is converted by a 16-bit ADC, it is stored in the voltage data buffer. At the same time, the current acquisition device is used to acquire the current data at the input and output ends of the charging pile. The sampling frequency is also 1000Hz. After the sampled data is converted by a 16-bit ADC, it is stored in the current data buffer. Then, digital filtering processing is performed on the acquired voltage data and current data. The moving average filtering algorithm is used to eliminate sampling noise, and the length of the filtering window is 16 points. Finally, the processed data is arranged in chronological order to construct a voltage data matrix and a current data matrix. The number of rows of the matrix represents the sampling time points, and the number of columns represents the measurement parameters, including the input voltage, output voltage, input current, and output current. The purpose of this step is to obtain the basic voltage and current data during the operation of the charging pile, providing data support for subsequent power analysis and anomaly detection.

[0052] The specific implementation of step S02 is to calculate the active power and reactive power at the input and output ends of the charging pile based on the collected voltage data matrix and current data matrix. First, a power analysis device is used to synchronously sample the voltage and current data. The sampling period is 20 milliseconds, and 50 points are collected in each period. Then, the discrete Fourier transform algorithm is used to calculate the fundamental wave components and harmonic components of the voltage and current signals. The active power is calculated based on the fundamental wave components, and the reactive power is calculated based on the harmonic components. Next, a power feature vector is constructed, and the vector elements include the input active power, input reactive power, output active power, and output reactive power. Finally, by calculating the power difference between adjacent time points, the power fluctuation characteristics are recorded. The power fluctuation threshold is set to 5% of the rated power. The purpose of this step is to analyze the power transmission characteristics of the charging pile and provide power feature data for detecting electricity theft behavior.

[0053] The specific implementation of step S03 is to collect the amplitude, phase, and frequency characteristics of the power line carrier signal through a carrier feature extraction device. First, a band-pass filter is used to preprocess the carrier signal. The center frequency of the filter is 100 kHz, and the bandwidth is 20 kHz. Then, the quadrature demodulation method is used to extract the amplitude and phase information of the carrier signal. The sampling frequency is 100 Hz. Next, the zero-crossing detection method is used to measure the frequency of the carrier signal, and the measurement accuracy is better than 0.1 Hz. Finally, the collected carrier feature data is arranged in chronological order to construct a carrier attenuation matrix. The matrix elements include the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value. The purpose of this step is to obtain the transmission characteristics of the power line carrier signal and provide carrier feature data for detecting electricity theft behavior.

[0054] The specific implementation of step S04 is to demodulate and analyze the collected carrier signal to extract the abnormal characteristics of the carrier modulation parameters. First, a digital phase-locked loop is used to synchronously demodulate the carrier signal. The bandwidth of the phase-locked loop is 1 kHz, and the locking time is less than 1 millisecond. Then, the deviation between the actual demodulation parameters and the theoretical parameters is calculated, including the amplitude attenuation coefficient, phase offset, and frequency drift value. Among them, the logarithmic ratio method is used to calculate the amplitude attenuation coefficient, the phase difference method is used to calculate the phase offset, and the frequency difference method is used to calculate the frequency drift value. Next, the calculated parameter deviation values are arranged in chronological order to construct a carrier demodulation abnormal matrix. Finally, feature statistical analysis is performed on the abnormal matrix to calculate the mean, variance, and change rate of each parameter. The purpose of this step is to analyze the abnormal situation of the modulation characteristics of the carrier signal and provide demodulation feature data for detecting electricity theft behavior.

[0055] The specific implementation of step S05 is to conduct the probability analysis of power theft behavior based on a pre-trained power theft behavior recognition model. First, the carrier attenuation matrix and the carrier demodulation anomaly matrix are used as the model inputs. Then, the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value in the carrier attenuation matrix are used to construct a carrier feature network graph. The nodes of the network graph represent the carrier features at different time points, and the weights of the edges represent the difference degrees of the carrier features at adjacent time points. Next, the Kruskal algorithm is used to calculate the minimum spanning tree of the network graph. The time complexity of the algorithm is O(ElogV), where E is the number of edges and V is the number of vertices. The sum of the weights of the minimum spanning tree is used as the power theft behavior feature index, which reflects the degree of temporal variation of the carrier features. Finally, combined with the anomaly parameters in the carrier demodulation anomaly matrix, the weighted summation method is used to calculate the probability of power theft behavior. The weight coefficients are determined through model training. The purpose of this step is to make a preliminary judgment on power theft behavior using machine learning methods.

[0056] The specific implementation of step S06 is to calculate the power theft behavior judgment index according to the power theft behavior judgment equations. First, the power loss anomaly index is calculated based on the power feature equation. This equation uses the active power and reactive power at the input end and the output end as input variables. The power loss threshold is set to 5%. When the power loss exceeds the threshold, the anomaly index increases. Then, the carrier attenuation anomaly index is calculated based on the carrier attenuation feature equation. This equation uses the carrier amplitude attenuation coefficient, phase offset, and frequency drift value as input variables. The standard attenuation threshold is determined according to the actual line parameters. Next, the power theft behavior temporal correlation index is calculated based on the temporal correlation equation. This equation uses the power loss anomaly index, the carrier attenuation anomaly index, and historical data as input variables. The time weight coefficient uses an exponential decay function. The weight of the most recent time point is 1, and the weight of the historical time points decreases as the time interval increases. Finally, the calculation results of the three equations are comprehensively evaluated to obtain the comprehensive judgment index of power theft behavior. The purpose of this step is to quantitatively evaluate power theft behavior from multiple perspectives.

[0057] The specific implementation of step S07 is to use the data storage device to record the system operation parameters and historical data, and establish a power theft behavior feature database. First, the system operation parameters are stored in chronological order, including basic data such as voltage, current, and power, as well as derivative data such as carrier features and anomaly indicators. The data storage period is 1 minute. Then, the historical data is compressed and stored. A lossless compression algorithm is used to reduce the storage space occupancy, and the compression ratio is not less than 5 to 1. Next, a data index structure is established, and the B-tree index method is used to improve the data query efficiency. Finally, the data is backed up regularly to ensure data security. The purpose of this step is to provide historical data support for power theft behavior analysis.

[0058] The specific implementation of step S08 is to implement the alarm function for electricity theft behavior. First, it is determined whether the time-series correlation index of the electricity theft behavior exceeds a preset threshold, which is set to 0.8. When the index exceeds the threshold, the alarm program is started, and then an alarm message is sent to the monitoring center through a real-time communication device. The alarm message includes a timestamp, location information, abnormal index value, etc. An encrypted transmission method is used to ensure data security. Then, the alarm device is started to emit an audible and visual alarm signal, with a sound pressure level of 95 decibels and a flash frequency of 2 Hz. Finally, the alarm event is recorded, including the alarm time, alarm reason, and processing result. The purpose of this step is to detect and handle electricity theft behavior in a timely manner.

[0059] The specific implementation of the electricity theft behavior recognition model uses an improved residual neural network structure. First, at the input layer, a normalization processing module is used to normalize the input data. The batch normalization method is adopted to calculate the mean and variance of the data, and then a normalization transformation is performed. Then, multiple improved residual blocks are set in the feature extraction layer. Each residual block contains two 3×3 convolutional layers and a skip connection. A channel attention mechanism and a spatial attention mechanism are added to the standard residual structure. The channel attention uses a squeeze-and-excitation network structure, and the spatial attention uses an adaptive pooling method. Then, a feature fusion module is added between the feature extraction layer and the classification layer, and the power carrier feature and power loss feature are fused using an adaptive weight method. The weight coefficient is automatically learned through the backpropagation algorithm. Finally, at the classification layer, a fully connected layer is used to reduce the dimensionality of the feature vector and output the probability of the occurrence of electricity theft behavior.

[0060] The training dataset of the electricity theft behavior recognition model is established by combining laboratory simulation and on-site collection. First, different types of electricity theft behaviors are simulated in the laboratory environment, and the corresponding carrier features and power features data are collected. Then, the data in the normal operating state is collected at the actual charging pile site. Next, the collected data is labeled to clearly identify the types of electricity theft behaviors. Finally, the dataset is divided into a training set, a validation set, and a test set according to the ratio of 8:1:1.

[0061] In the model training process, the mini-batch stochastic gradient descent algorithm is first adopted. The batch size is set to 32, and the initial value of the learning rate is 0.001. The cosine annealing strategy is used to dynamically adjust the learning rate. Then, the cross-entropy loss function is used to calculate the error between the prediction result and the true label. Next, the gradient is calculated through the backpropagation algorithm and the network parameters are updated. The gradient clipping method is used to prevent gradient explosion. Finally, the validation set is used to evaluate the training effect, and the training is stopped when the validation accuracy reaches 95%. The test set is used to test the performance of the trained model, and the test accuracy requirement is not less than 93%.

[0062] In the present invention, multiple calculation processes and matrix constructions are involved. The following details each calculation process:

[0063] The construction of the voltage data matrix and the current data matrix is represented as follows: ; ; In the formula, and respectively represent the input - end voltage and the output - end voltage at the th sampling moment, with the unit of volt; and respectively represent the input - end current and the output - end current at the th sampling moment, with the unit of ampere; is the number of sampling points, and its value is 1000.

[0064] The construction process of the voltage data matrix and the current data matrix: First, consider the sampling - frequency requirement. According to the Nyquist sampling theorem, the sampling frequency is set to 1000 Hz, which can effectively capture signal changes below 500 Hz; the number of rows of the matrix corresponds to the sampling time points, and the number of columns corresponds to the measured parameters. Two columns are used to store the data of the input end and the output end respectively, which is convenient for subsequent calculation and processing; each matrix element is obtained through 16 - bit ADC sampling, and the sampling accuracy is 0.1%, meeting the measurement requirements.

[0065] The calculation of active power and reactive power is represented as follows: ; ; ; ; In the formula, and are the active powers of the input end and the output end respectively, with the unit of watt; and are the reactive powers of the input end and the output end respectively, with the unit of var; and are the phase angles of the input end and the output end respectively, with the unit of radian; is the number of sampling points within the calculation period, and its value is 50; , , , are the measurement - error correction coefficients, and the range is 0.01 - 0.05.

[0066] The detailed derivation process of the calculation of active power and reactive power is described as follows: Based on the definition of power, the active power is the average value of the product of the instantaneous values of voltage and current; considering the phase difference in actual measurement, the phase - angle terms and ; To improve the calculation accuracy, a 50-point digital integration algorithm is adopted; a measurement error correction coefficient is added to , which is obtained through experimental calibration; this calculation method can obtain active power and reactive power simultaneously, with high calculation efficiency.

[0067] The calculation of carrier feature extraction is expressed as follows: ; ; ; In the formula, is the carrier amplitude, with the unit of volt; is the carrier phase, with the unit of radian; is the carrier frequency, with the unit of hertz; and are the in-phase component and the quadrature component respectively; , , are the measurement noise correction terms, with the range of 0.001 to 0.01.

[0068] The calculation and derivation process of carrier feature extraction is described in detail as follows: Using the principle of quadrature demodulation, the carrier signal is decomposed into the in-phase component and the quadrature component ; The amplitude calculation adopts the vector modulus method, the phase calculation adopts the arctangent function; the frequency calculation adopts the phase derivative method, which can track the frequency change in real time; the noise correction terms , , are introduced and obtained through experimental statistics; this method has high anti-noise ability.

[0069] The construction of the carrier attenuation matrix is expressed as follows: ; In the formula, is the amplitude attenuation coefficient at the th sampling moment, dimensionless, with the range of 0 to 1; is the phase offset at the th sampling moment, with the unit of radian; is the frequency drift value at the th sampling moment, with the unit of hertz; is the number of sampling points, with the value of 100.

[0070] The detailed process of constructing the carrier attenuation matrix is described as follows: The three column vectors of the matrix respectively correspond to amplitude attenuation, phase shift, and frequency drift; each parameter is represented by a relative value for easy comparison and analysis; the number of sampling points is set to 100, which can meet the requirements of feature extraction; this matrix comprehensively reflects the transmission characteristics of the carrier signal.

[0071] The construction of the carrier demodulation anomaly matrix is expressed as follows: ; In the formula, represents the amplitude attenuation anomaly value; represents the phase shift anomaly value; represents the frequency drift anomaly value; , , are respectively the amplitude attenuation coefficient, phase shift amount, and frequency drift value under the standard working state.

[0072] The weight calculation of the electricity theft behavior characteristic network graph is expressed as follows: ; In the formula, is the edge weight between node and node in the network graph; and respectively represent the th feature component of node and node ; is the smoothing factor, with a value of 0.01.

[0073] The idea of calculating the weight of the electricity theft behavior characteristic network graph: The Euclidean distance is used to measure the similarity between nodes in the feature space; the smoothing factor is introduced to avoid the weight being zero; this calculation method can effectively describe the spatio-temporal correlation of features.

[0074] The expression of the power characteristic equation is as follows: ; In the formula, is the power loss anomaly index, dimensionless; is the standard active power loss rate, with a value of 0.05; is the standard reactive power loss rate, with a value of 0.08; and are the weight coefficients and satisfy ; is the correction term, with a range of 0 to 0.1.

[0075] The idea of establishing the power characteristic equation is as follows: The relative deviation of input and output power is used to represent the degree of abnormality; the influences of active power and reactive power are considered separately; the weight coefficients are determined by experimental optimization; the correction term is used to compensate for measurement errors; this equation can accurately reflect the abnormality of power transmission.

[0076] The expression of the carrier attenuation characteristic equation is as follows: ; In the formula, is the carrier attenuation abnormality index, dimensionless; , , are the characteristic weight coefficients and satisfy ; is the correction term, with a range of 0 to 0.1.

[0077] The idea of establishing the carrier attenuation characteristic equation is as follows: The method of multi-characteristic weighted average is adopted; the characteristic weight coefficients are obtained by optimizing a large amount of experimental data; the correction term is used to improve the robustness of the equation; this equation comprehensively evaluates the degree of abnormality of the carrier signal.

[0078] The expression of the time series correlation equation is as follows: ; In the formula, is the time series correlation index of electricity theft behavior, dimensionless; and are respectively the power loss abnormality index and the carrier attenuation abnormality index at the th historical moment; is the time weight coefficient, is the attenuation coefficient, with a value of 0.1; is the length of historical data, with a value of 10; is the correction term, with a range of 0 to 0.1.

[0079] The idea of establishing the time series correlation equation is as follows: The exponential decay weight is used to reflect the time correlation; the attenuation coefficient is determined by experiments; the length of historical data is set to 10 time points; the correction term is used to compensate for the cumulative error; this equation effectively describes the time series characteristics of electricity theft behavior.

[0080] The construction principles and significance of the above equations are as follows: The power characteristic equation takes into account the abnormal losses of active power and reactive power, describes the degree of abnormality in the form of relative deviation, introduces a weight coefficient to achieve feature fusion, and adds a correction term to improve robustness; the carrier attenuation characteristic equation comprehensively evaluates the degree of abnormality of the carrier signal by using the method of multi-feature weighted average, and the feature weight coefficient is determined by experimental optimization; the time-series correlation equation uses an exponentially decaying weight to reflect the influence degree of historical data on the current judgment, and the weight is smaller for a longer time span, which is more in line with the actual situation.

[0081] Optionally, the calculation in the improved residual neural network is as follows: ; In the formula, is the output of the improved residual block; is the input feature; is the mapping function of two layers of 3×3 convolution; is the channel attention weight; is the spatial attention weight.

[0082] Optionally, the calculation of the feature fusion module is as follows: ; In the formula, is the fused feature vector; is the power carrier feature vector; is the power loss feature vector; and are adaptive weight coefficients and satisfy .

[0083] Optionally, the calculation of the probability of electricity theft behavior is as follows: ; In the formula, is the probability of electricity theft behavior; is the time-series correlation index; is the total weight of the minimum spanning tree; and are model parameters; is the bias term.

[0084] Specifically, the principle of the present invention is as follows: The core technical principle of the present invention is based on the following scientific understanding: Electricity theft behavior will simultaneously affect the transmission characteristics of power carrier signals and the power loss characteristics of the system. In terms of power carrier signals, electricity theft behavior changes the circuit topology, resulting in abnormal attenuation, phase shift, and frequency drift of the carrier signal during transmission. By establishing a signal carrier attenuation matrix, the system can quantitatively describe these abnormal characteristics and provide reliable characteristic indicators for the detection of electricity theft behavior. In terms of power loss characteristics, electricity theft behavior will cause power imbalance between the input and output ends of the charging pile, and this imbalance is manifested as abnormal changes in active power and reactive power. By constructing a power characteristic vector, the system can capture these abnormal change characteristics.

[0085] The present invention uses an improved residual neural network structure for feature learning and pattern recognition, and this structural design is based on the basic principles of deep learning. First, residual connections can effectively alleviate the problem of gradient disappearance in deep networks and ensure the training effect of the model; second, the introduction of channel attention mechanism and spatial attention mechanism enables the model to adaptively focus on important features and improve the efficiency of feature extraction; finally, the feature fusion module realizes the intelligent fusion of power carrier features and power loss features by learning the optimal feature combination weights.

[0086] The technical solution of the present invention follows the logical chain of "multi-source perception - feature extraction - feature fusion - intelligent recognition", and each link is closely connected and mutually supported. The design of the hardware module ensures the comprehensiveness and accuracy of data collection, and the design of the software module ensures the intelligence and reliability of data processing and decision-making. Through systematic design and optimization, the present invention realizes high efficiency and high accuracy in the detection of electricity theft behavior.

[0087] The following provides a specific Embodiment 1 of the present invention, and the specific implementation of each step in this Embodiment 1 is described in detail as follows.

[0088] In Embodiment 1 of the present invention, the control chip uses a high-performance 32-bit processor STM32H750VBT6 with a main frequency of 480 MHz, built-in 2 MB of flash memory and 1 MB of RAM, which is used to execute the system control program, data processing, and communication management. Among them, the system control module conducts data interaction and control signal transmission with each functional module through the internal bus, and the bus clock frequency is 240 MHz.

[0089] The voltage acquisition device includes two groups of high-precision Hall voltage sensors LEM LV25P at the input end and the output end, with a measurement range of 0 to 1000 V, an accuracy of 0.1%, a sampling frequency of 1000 Hz, and data acquisition is performed through a 16-bit ADC to achieve real-time monitoring of the voltages at the input and output ends of the charging pile. The sampled data satisfies the following voltage data matrix format: ; In the formula, and respectively represent the input - end voltage and the output - end voltage at the th sampling moment, with the unit of volt.

[0090] The current acquisition device uses a high - precision Hall current sensor LEM LAH25NP, with a range of 0 to 100 A, an accuracy of 0.1%, a sampling frequency of 1000 Hz, and data acquisition is carried out through a 16 - bit ADC to realize real - time monitoring of the input - end and output - end currents of the charging pile. The sampling data satisfies the following current data matrix format: ; In the formula, and respectively represent the input - end current and the output - end current at the th sampling moment, with the unit of ampere.

[0091] The power analysis device uses a dedicated digital signal processor TMS320F28335, with a main frequency of 150 MHz, an internal hardware multiplier and a floating - point operation unit. The active power and reactive power are calculated through a digital integration algorithm, and the calculation accuracy is better than 0.2%. It can process the power data of the input - end and output - end simultaneously. The power calculation satisfies the following equations: ; ; ; ; In the formula, and are the active powers of the input - end and output - end respectively, with the unit of watt; and are the reactive powers of the input - end and output - end respectively, with the unit of var; and are the phase angles of the input - end and output - end respectively, with the unit of radian; is the number of sampling points within the calculation period, with a value of 50; , , , are measurement error correction coefficients, with a range of 0.01 to 0.05.

[0092] The carrier feature extraction device uses a high-speed data acquisition chip AD9226 with a sampling rate of 100 Hz and a resolution of 12 bits. It is combined with a band-pass filter and a digital phase-locked loop circuit to achieve the extraction of the amplitude, phase, and frequency features of the power line carrier signal. The center frequency of the filter is 100 kHz and the bandwidth is 20 kHz. The carrier feature extraction satisfies the following equations: ; ; ; In the formula, is the carrier amplitude, with the unit of volt; is the carrier phase, with the unit of radian; is the carrier frequency, with the unit of hertz; and are the in-phase component and the quadrature component respectively; , , are the measurement noise correction terms, with the range from 0.001 to 0.01.

[0093] The signal conditioning device includes two-stage operational amplifier circuits and a Butterworth low-pass filter. The operational amplifier uses OP27 with an adjustable gain range from 1 to 100 times and a cut-off frequency of 10 kHz. It is used to amplify and filter the acquired signal to improve the signal-to-noise ratio. The gain of the first stage is 10 times, and the gain of the second stage is adjustable. The filter uses a fourth-order Butterworth low-pass filter. The operational amplifier selects OP27 with low noise and low drift. The circuit gain and bandwidth can be adjusted by a programmable potentiometer AD8400.

[0094] The data storage device uses a 32GB industrial-grade eMMC flash memory, supports the FAT32 file system, stores data in a cyclic write mode, automatically deletes the earliest data when the storage space is insufficient, and has a power-off protection function. It is used to store system configuration parameters, operation logs, and historical data, and the data retention period is 3 months.

[0095] The real-time communication device uses a 4G wireless communication module SIM7600CE, supports the TCP / IP protocol, the communication rate can reach 100 Mbps, uses the RSA encryption algorithm to encrypt the transmitted data, the key length is 2048 bits, and the data transmission delay is less than 100 ms.

[0096] The alarm device includes an audible and visual alarm and an LED indicator. The sound pressure level of the audible and visual alarm is 95 decibels, and the flash frequency is 2 Hz. The LED indicator uses a three-color indication method to display the system working status. Green indicates normal, yellow indicates warning, and red indicates failure.

[0097] The power management device uses multiple DC / DC converters and the power monitoring chip MAX17055. The input voltage range is 9 to 36V, with an overvoltage protection threshold of 36V, an undervoltage protection threshold of 9V, and an overcurrent protection threshold of 10A. It can provide a stable working voltage for each module, and the conversion efficiency is as high as 95%.

[0098] Data communication is carried out between the above-mentioned devices using the RS485 bus. The communication rate is 115200bps, and the Modbus RTU protocol is adopted. Each device has an independent slave address, and the control chip, as the master, uniformly manages data transmission. The communication frame format includes device address, function code, data area, and check bit.

[0099] All modules use industrial-grade components. The operating temperature range is from -40 degrees Celsius to 85 degrees Celsius, and the relative humidity is 5% to 95%. The protection level reaches IP65, meeting the requirements for outdoor installation and use. The shell is made of aluminum alloy material and is treated with surface plastic spraying, with good corrosion resistance.

[0100] The power supply of this system adopts a switching power supply scheme. The input voltage is 220V AC, and multiple DC voltages are output, including 3.3V, 5V, 12V, etc. The total power is 200W, and it has a perfect electromagnetic compatibility design, meeting the national standard requirements in terms of radiation emission, conducted emission, and immunity.

[0101] In this Embodiment 1, the specific implementation manners of steps S01 to S08 are as follows:

[0102] The specific implementation manner of step S01 is to adopt a high-speed data acquisition scheme to collect the voltage and current data at the input and output ends of the charging pile. The voltage data at the input and output ends of the charging pile is collected through a voltage acquisition device, with a sampling frequency of 1000Hz. The sampled data is converted by a 16-bit ADC and then stored in the voltage data buffer. At the same time, the current data at the input and output ends of the charging pile is collected through a current acquisition device, with a sampling frequency of 1000Hz. The sampled data is converted by a 16-bit ADC and then stored in the current data buffer. The collected voltage data and current data are respectively constructed into a voltage data matrix and a current data matrix, and the matrix format satisfies: ; ; In the formula, and respectively represent the input-end voltage and output-end voltage at the th sampling moment, with the unit of volt; and respectively represent the input-end current and output-end current at the th sampling moment, with the unit of ampere; is the number of sampling points, with a value of 1000; the sampled data is processed by a moving average filtering algorithm, and the filtering window length is 16 points to eliminate sampling noise and improve data quality.

[0103] The specific implementation of step S02 is based on the collected voltage data matrix and current data matrix to calculate the active power and reactive power at the input and output ends of the charging pile. The calculation equations are as follows: ; ; ; ; In the formula, and are the active powers at the input and output ends respectively, with the unit of watt; and are the reactive powers at the input and output ends respectively, with the unit of var; and are the phase angles at the input and output ends respectively, with the unit of radian; is the number of sampling points within the calculation period, with a value of 50; , , , are the measurement error correction coefficients, with a range of 0.01 to 0.05; a power feature vector is constructed based on the calculation results, and the power fluctuation characteristics are recorded by calculating the power difference between adjacent time points. The power fluctuation threshold is set to 5% of the rated power.

[0104] The specific implementation of step S03 is to collect the amplitude, phase, and frequency characteristics of the power line carrier signal through a carrier feature extraction device. First, a band-pass filter is used to preprocess the carrier signal. The center frequency of the filter is 100 kHz, and the bandwidth is 20 kHz. Then, an orthogonal demodulation method is used to extract the carrier signal characteristics: ; ; ; In the formula, is the carrier amplitude, with the unit of volt; is the carrier phase, with the unit of radian; is the carrier frequency, with the unit of hertz; and are the in-phase component and the quadrature component respectively; , , To measure the noise correction term, with a range from 0.001 to 0.01; the sampling frequency is 100 Hz, and AD9226 is used for data acquisition. The collected characteristic data is used to construct a signal carrier attenuation matrix: ; In the formula, is the amplitude attenuation coefficient at the th sampling moment, dimensionless, with a range from 0 to 1; is the phase offset at the th sampling moment, in radians; is the frequency drift value at the th sampling moment, in hertz; is the number of sampling points, with a value of 100.

[0105] The specific implementation of step S04 is to perform demodulation analysis on the collected carrier signal, extract the abnormal characteristics of the carrier modulation parameters, use a digital phase-locked loop to synchronously demodulate the carrier signal, the bandwidth of the phase-locked loop is 1 kHz, and the locking time is less than 1 millisecond. Calculate the deviation between the actual demodulation parameters and the theoretical parameters, and construct a carrier demodulation abnormal matrix: ; In the formula, represents the amplitude attenuation abnormal value; represents the phase offset abnormal value; represents the frequency drift abnormal value; , , are the amplitude attenuation coefficient, phase offset, and frequency drift value under the standard working state respectively; perform characteristic statistical analysis on the abnormal matrix, and calculate the mean, variance, and change rate of each parameter.

[0106] The specific implementation of step S05 is to perform power theft behavior probability analysis based on a pre-trained power theft behavior recognition model. Use the signal carrier attenuation matrix and the carrier demodulation abnormal matrix as the model input, and use the characteristics in the signal carrier attenuation matrix to construct a carrier characteristic network graph. The calculation of the edge weights of the network graph satisfies: ; In the formula, is the edge weight between node and node in the network graph; and respectively represent the th and th characteristic components of node ; Let \(\alpha\) be the smoothing factor, with a value of 0.01. The Kruskal algorithm is used to calculate the minimum spanning tree of the network graph. The time complexity of the algorithm is \(O(E\log V)\), where \(E\) is the number of edges and \(V\) is the number of vertices. The total weight of the minimum spanning tree is used as the power theft behavior characteristic index. Combining the abnormal parameters in the carrier demodulation abnormal matrix, the following equation is used to calculate the probability of power theft behavior: ; In the formula, is the probability of power theft behavior; is the time series correlation index; is the total weight of the minimum spanning tree; and are model parameters; is the bias term.

[0107] The specific implementation of step S06 is to calculate the power theft behavior judgment index according to the power theft behavior judgment equation set. The power characteristic equation is: ; In the formula, is the power loss abnormality index, dimensionless; is the standard active power loss rate, with a value of 0.05; is the standard reactive power loss rate, with a value of 0.08; and are weight coefficients, and satisfy ; is the correction term, with a range of 0 to 0.1. The carrier attenuation characteristic equation is: ; In the formula, is the carrier attenuation abnormality index, dimensionless; , , are characteristic weight coefficients, and satisfy ; is the correction term, with a range of 0 to 0.1. The time series correlation equation is: ; In the formula, is the power theft behavior time series correlation index, dimensionless; and are respectively the power loss abnormality index and the carrier attenuation abnormality index at the th historical moment; is the time weight coefficient, is the attenuation coefficient, with a value of 0.1; is the length of historical data, with a value of 10; is the correction term, with a range of 0 to 0.1.

[0108] The specific implementation of step S07 is to use a data storage device to record system operation parameters and historical data, establish a database of electricity theft behavior characteristics, store the system operation parameters in chronological order, including basic data such as voltage, current, and power, as well as derivative data such as carrier characteristics and abnormal indicators. The data storage period is 1 minute. Compress and store the historical data, and use a lossless compression algorithm to reduce the storage space occupancy. The compression ratio is not less than 5:1. Establish a data index structure and use the B-tree index method to improve the data query efficiency.

[0109] The specific implementation of step S08 is to implement the alarm function for electricity theft behavior. Judge whether the time series correlation index of electricity theft behavior exceeds the preset threshold of 0.8. When the index exceeds the threshold, send an alarm message to the monitoring center through a real-time communication device. The alarm message includes a timestamp, location information, abnormal index values, etc. Use the RSA encryption algorithm for data encryption, and the key length is 2048 bits. At the same time, start the alarm device to send out an audible and visual alarm signal, with a sound pressure level of 95 decibels and a flash frequency of 2 Hz.

[0110] The structure of the electricity theft behavior recognition model is based on an improved residual neural network, and its calculation satisfies: ; In the formula, is the output of the improved residual block; is the input feature; is the mapping function of two layers of 3×3 convolution; is the channel attention weight; is the spatial attention weight; The calculation of the feature fusion module satisfies: ; In the formula, is the fused feature vector; is the power carrier feature vector; is the power loss feature vector; and are adaptive weight coefficients, and satisfy .

[0111] The training process of the model uses the mini-batch stochastic gradient descent algorithm, with a batch size of 32. The initial value of the learning rate is 0.001. Use the cosine annealing strategy to dynamically adjust the learning rate. The loss function uses the cross-entropy function. Calculate the gradient through the backpropagation algorithm and update the network parameters. Use the gradient clipping method to prevent gradient explosion. Stop training when the validation accuracy reaches 95%, and the test accuracy requirement is not less than 93%.

[0112] The training dataset is established by combining laboratory simulation and on-site collection. Different types of electricity theft behaviors are simulated in the laboratory environment to collect corresponding carrier characteristics and power characteristic data, and data under normal operating conditions are collected at the actual charging pile site. The collected data is labeled to clearly identify the types of electricity theft behaviors, and the dataset is divided into a training set, a validation set, and a test set according to a ratio of 8:1:1.

[0113] To better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: The R & D team first conducted a calibration test on the voltage and current collection devices of the charging pile. The test results are shown in Table 1: Table 1 Calibration Test Results of Voltage and Current Collection Devices

[0114] Figure 3 The comparison of the voltage and current calibration test results is shown. After the calibration test is completed, the R & D team conducted a calibration experiment on the power analysis device and used a standard power source for testing. The experimental results are shown in Table 2: Table 2 Calibration Experiment Results of Power Analysis Device

[0115] During the actual operation process, an abnormal situation occurred in a charging pile on January 15, 2025. The voltage and current data recorded by the system are shown in Table 3: Table 3 Voltage and Current Data under Abnormal Conditions

[0116] Based on the collected data, the system calculates the power feature vector, and the calculation results are shown in Table 4: Table 4 Power Feature Calculation Results

[0117] Figure 4 The change trend of power over time is shown. Multiple line charts are used to show the changes in input and output active power and reactive power, clearly showing the dynamic change process of power parameters. At the same time, the carrier characteristic data collected by the system is shown in Table 5: Table 5 Carrier Characteristic Data

[0118] Figure 5 The change trend of carrier characteristics is shown. Multiple coordinate axes are used to show the changes in amplitude attenuation coefficient, phase offset, and frequency drift value over time, highlighting the abnormal changes in carrier characteristics. The results calculated by the system according to the electricity theft behavior judgment equations are shown in Table 6: Table 6 Calculation Results of Electricity Theft Behavior Judgment Indicators

[0119] Figure 6 It shows the changing trend of the electricity theft behavior judgment indicators. A line chart is used to show the changes in the power loss abnormality degree, carrier attenuation abnormality degree, and time series correlation indicators, and the warning threshold line is marked, visually showing the judgment process of electricity theft behavior. When the time series correlation indicator exceeds the preset threshold of 0.8, the system determines that an electricity theft behavior has occurred, sends an alarm message to the monitoring center through a real-time communication device, and at the same time activates an audible and visual alarm.

[0120] Through the inspection of this charging pile, it is found that someone has privately connected a shunt device between the output end of the charging pile and the electric vehicle, resulting in part of the electric energy being stolen. This incident shows that the system of the present invention can effectively detect electricity theft behavior and send an alarm in a timely manner.

[0121] In this Embodiment 2, the traditional electricity theft detection method mainly relies on the difference between input and output powers for judgment, is easily affected by measurement errors and load fluctuations, and has a high false alarm rate. While the present invention adopts a method combining power carrier characteristics and power characteristics, realizes feature fusion by improving the residual neural network, and introduces time series correlation analysis, significantly improving the detection accuracy. Specifically, it is manifested in the following aspects: 1. The detection accuracy rate is increased from 85% of the traditional method to 93%. 2. The false alarm rate is reduced from 15% to 5%. 3. The detection response time is shortened from 10 seconds to 1 second. 4. It can identify various types of electricity theft behaviors and has stronger adaptability. 5. It has the ability of adaptive learning and can continuously optimize the detection model.

[0122] In this Embodiment 2, 15000 groups of normal data and 5000 groups of electricity theft data are used in the training process of the electricity theft behavior recognition model, and are divided into a training set, a validation set, and a test set according to the ratio of 8:1:1. During the training process, the learning rate starts from 0.001 and is adjusted using the cosine annealing strategy. The validation accuracy reaches 95.2% and the test accuracy is 93.8% at the 1000th round of iteration, meeting the system requirements.

[0123] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Table 7 below.

[0124] Table 7 Variable Explanation Table

[0125] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A DC charging pile system for rapid detection of electricity theft behavior, comprising a control chip, a voltage acquisition device, a current acquisition device, a power analysis device, a carrier feature extraction device, a signal conditioning device, a data storage device, a real-time communication device, an alarm device and a power management device, characterized in that, The control chip is provided with a system control module, which is used to collect the voltage and current data at the input and output ends of the charging pile to establish a matrix; calculate the active power and reactive power to construct a power feature vector; collect the carrier signal characteristics to establish a carrier attenuation matrix; demodulate and analyze the carrier signal to construct a carrier demodulation anomaly matrix; calculate the probability of electricity theft behavior based on a pre-trained electricity theft behavior recognition model; calculate the electricity theft behavior judgment index according to the electricity theft behavior judgment equation set; record the system operation parameters and historical data; when the time sequence correlation index of the electricity theft behavior exceeds the preset threshold, send an alarm message and emit an audible and visual alarm signal.

2. The DC charging pile system for rapid detection of electricity theft behavior according to claim 1, characterized in that, The voltage acquisition device is used to collect the voltage data at the input end of the charging pile and the voltage data at the output end of the charging pile, and the current acquisition device is used to collect the current data at the input end of the charging pile and the current data at the output end of the charging pile.

3. The DC charging pile system for rapid detection of electricity theft behavior according to claim 2, characterized in that, The power analysis device is used to calculate the active power at the input end of the charging pile, the reactive power at the input end of the charging pile, the active power at the output end of the charging pile, and the reactive power at the output end of the charging pile, and record the power fluctuation characteristics.

4. The DC charging pile system for rapid detection of electricity theft behavior according to claim 3, characterized in that, The carrier feature extraction device is used to collect the carrier amplitude, carrier phase, and carrier frequency.

5. The DC charging pile system for rapid detection of electricity theft behavior according to claim 4, characterized in that The signal conditioning device is used to filter and amplify the collected signal, the data storage device is used to store the system operation parameters and historical data, the real-time communication device is used to transmit data to the monitoring center, the alarm device is used to emit an audible and visual alarm signal, and the power management device is used to monitor the system power supply voltage.

6. The DC charging pile system for rapid detection of electricity theft behavior according to claim 5, characterized in that, The carrier attenuation matrix refers to the attenuation of the power line carrier signal in the transmission process in terms of carrier amplitude, carrier phase, and carrier frequency. The matrix elements include the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value.

7. The DC charging pile system for rapid detection of electricity theft behavior according to claim 6, characterized in that, The carrier demodulation anomaly matrix refers to the deviation between the modulation parameters obtained after demodulating the carrier signal and the theoretical parameters. The matrix elements include the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value.

8. The DC charging pile system for rapid detection of electricity theft behavior according to claim 7, characterized in that, The electricity theft behavior judgment equation set includes a power feature equation, a carrier attenuation feature equation, and a time sequence correlation equation. The power feature equation is used to calculate the power loss anomaly index, the carrier attenuation feature equation is used to calculate the carrier attenuation anomaly index, and the time sequence correlation equation is used to calculate the time sequence correlation index of the electricity theft behavior; the inputs of the power feature equation include the active power at the input end of the charging pile, the active power at the output end of the charging pile, the reactive power at the input end of the charging pile, the reactive power at the output end of the charging pile, and the power loss threshold; the inputs of the carrier attenuation feature equation include the carrier amplitude attenuation coefficient, carrier phase offset, carrier frequency drift value, and standard attenuation threshold, and the standard attenuation threshold refers to the standard value of the carrier signal attenuation in the normal working state; the inputs of the time sequence correlation equation include the power loss anomaly index, the carrier attenuation anomaly index, historical data, and time weight coefficient, and the time weight coefficient refers to the weight of the occurrence of electricity theft behavior in different time periods.

9. The DC charging pile system for rapid detection of electricity theft behavior according to claim 8, characterized in that, The electricity theft behavior recognition model adopts an improved residual neural network structure, and a feature fusion module is added between the feature extraction layer and the classification layer of the improved residual neural network structure. The feature fusion module is used to fuse the power carrier feature and the power loss feature.

10. A DC charging pile system for rapid detection of electricity theft behavior according to claim 9, characterized in that, The steps for establishing the training data set of the electricity theft behavior recognition model include collecting normal operation state data and electricity theft state data, annotating the electricity theft behavior labels corresponding to the carrier attenuation matrix and the power feature vector, and dividing the data set into a training data set, a validation data set, and a test data set according to a ratio of 8:1:1.

Images