A direct-current charging pile supporting rapid detection of electricity stealing behavior

Abstract

The application provides a direct-current charging pile supporting rapid detection of electricity stealing behavior, and belongs to the technical field of charging piles. The direct-current charging pile supporting rapid detection of electricity stealing behavior comprises hardware modules such as a control chip, a voltage acquisition device, a current acquisition device, a power analysis device, a carrier feature extraction device and a system control module. The control chip performs functions such as data acquisition, feature extraction and abnormality detection. The system constructs a voltage data matrix and a current data matrix by acquiring voltage and current data, and calculates a power feature vector. The system acquires a carrier signal to establish a carrier attenuation matrix and a carrier demodulation abnormality matrix. Feature learning is performed by using an improved residual neural network, and the optimized combination of power carrier features and power loss features is realized through a feature fusion module, and finally, a judgment result of electricity stealing behavior is output. The scheme innovatively combines power carrier feature analysis and power loss feature analysis, and realizes rapid and accurate detection of electricity stealing behavior.

Description

Technical Field

[0001] This invention belongs to the field of charging pile technology, and more specifically, relates to a DC charging pile that supports rapid detection of electricity theft. Background Technology

[0002] With the rapid development of the electric vehicle industry, DC charging piles have become an important part of electric vehicle charging infrastructure. During the operation of charging piles, electricity theft not only causes economic losses but also seriously affects power grid security and the normal operation of the charging piles. Traditional methods for detecting electricity theft mainly rely on manual inspections, meter readings, and simple power loss analysis. Manual inspections primarily involve periodically checking the external wiring and junction boxes of the charging pile to observe for any signs of abnormal modification; meter readings compare the difference between the input and output power values ​​of the charging pile to determine if there is any abnormal loss; and power loss analysis monitors the power curve during charging in real time to identify any abnormal fluctuations. While these traditional detection methods have achieved some effectiveness in practical applications, they still suffer from low detection efficiency and accuracy.

[0003] Traditional detection methods suffer from the following major drawbacks: First, manual inspections are labor-intensive and unable to promptly detect highly concealed electricity theft. Second, meter comparison methods are easily affected by factors such as fluctuations in charging pile losses and measurement errors, making it difficult to accurately identify even minor instances of electricity theft. Third, simple power loss analysis methods are ineffective against complex electricity theft techniques, such as the use of shunt circuits or modified metering devices. Furthermore, these methods often rely on a single characteristic parameter for judgment, lacking the ability to comprehensively analyze multi-dimensional features, resulting in unreliable detection results. With the continuous upgrading of electricity theft methods, traditional detection methods are no longer sufficient to meet practical needs.

[0004] In recent years, although researchers have proposed deep learning-based methods for detecting electricity theft, attempting to improve detection accuracy by establishing complex mathematical models, these methods primarily focus on the statistical characteristics of electricity metering data, neglecting the rich feature information contained in power line carrier signals. Furthermore, existing deep learning models have relatively simple structures and lack targeted optimization designs, making it difficult to fully extract the feature representations of electricity theft behavior. Therefore, how to achieve rapid and accurate detection of electricity theft in DC charging pile systems remains a pressing technical challenge. Summary of the Invention

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

[0006] This invention is implemented as follows: This invention provides a DC charging pile supporting rapid detection of electricity theft, 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. The control chip includes a system control module, used to acquire voltage and current data from the charging pile's input and output terminals to establish a matrix; calculate active and reactive power to construct a power feature vector; acquire carrier signal features 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 based on a pre-trained electricity theft identification model; calculate electricity theft judgment indicators according to a set of electricity theft judgment equations; record system operating parameters and historical data; and when the time-series correlation index of electricity theft exceeds a preset threshold, send alarm information and issue an audible and visual alarm signal.

[0007] The voltage acquisition device is used to collect voltage data at the input end 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 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] 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 to record the power fluctuation characteristics.

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

[0010] The signal conditioning device is used to filter and amplify the acquired signal; the data storage device is used to store system operating parameters and historical data; the real-time communication device is used to transmit data to the monitoring center; the alarm device is used to issue audible and visual alarm signals; and the power management device is used to monitor the system power supply voltage.

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

[0012] 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.

[0013] The set of equations for judging electricity theft 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 anomaly index, the carrier attenuation characteristic equation is used to calculate the carrier attenuation anomaly index, and the time-series correlation equation is used to calculate the time-series correlation index of electricity theft. The inputs to the power characteristic equation include the active power at the charging pile input end, the active power at the charging pile output end, the reactive power at the charging pile input end, the reactive power at the charging pile output end, and the power loss threshold. The inputs to 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, where the standard attenuation threshold refers to the standard value of carrier signal attenuation under normal operating conditions. The inputs to the time-series correlation equation include the power loss anomaly index, the carrier attenuation anomaly index, historical data, and the time weight coefficient, where the time weight coefficient refers to the weight of electricity theft occurring in different time periods.

[0014] The electricity theft identification model employs an improved residual neural network structure, with a feature fusion module 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 power line carrier features and power loss features; the input is the power line carrier attenuation matrix and the power feature vector, and the output is the electricity theft behavior feature vector.

[0015] 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 standardization processing module to normalize the input data. The feature extraction layer includes multiple improved residual blocks, which add channel attention and spatial attention mechanisms to the standard residual structure. The feature fusion module uses an adaptive weighting method to fuse power line carrier features and power loss features. The classification layer uses a fully connected layer to reduce the dimensionality of the feature vector and outputs the probability of electricity theft.

[0016] The steps for establishing the training dataset for the electricity theft identification model include collecting normal operation status data and electricity theft status data, labeling the electricity theft behavior tags corresponding to the carrier attenuation matrix and power feature vector, and dividing the dataset into training dataset, validation dataset and test dataset in an 8:1:1 ratio.

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

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

[0019] The training dataset of the electricity theft identification model is divided into training dataset, validation dataset, and test dataset in a ratio of 8:1:1. The model training adopts the mini-batch stochastic gradient descent algorithm with a batch size of 32 and an initial learning rate of 0.001. The learning rate is dynamically adjusted using a cosine annealing strategy, and training stops when the validation accuracy reaches 95%.

[0020] Compared with existing technologies, this invention provides a DC charging pile that supports rapid detection of electricity theft. This invention proposes a DC charging pile system that innovatively combines power line carrier feature analysis with power loss feature analysis to construct an electricity theft identification model based on an improved residual neural network. The system collects multi-dimensional data during the charging process through hardware modules such as voltage and current acquisition devices. It uses a carrier feature extraction device to analyze power line carrier signal characteristics in real time and calculates power loss characteristics through a power analysis device. The system employs a feature fusion module to adaptively fuse power line carrier features and power loss characteristics, fully utilizing multi-source information for comprehensive judgment, significantly improving the accuracy and real-time performance of electricity theft detection.

[0021] This invention addresses the main shortcomings of traditional detection methods: by introducing power line carrier feature analysis, the system can capture carrier signal anomalies caused by electricity theft, effectively identifying covert electricity theft that is difficult to detect using traditional methods; by constructing a power line carrier attenuation matrix and a power line demodulation anomaly matrix, the system achieves refined characterization of carrier signal features, overcoming the limitations of judging by a single feature parameter; by designing a feature fusion module, the system achieves an optimized combination of power line carrier features and power loss features, improving the reliability of the detection results. Simultaneously, the improved residual neural network structure, by introducing an attention mechanism, enhances the model's ability to extract key features.

[0022] This invention successfully solves the technical problem of the difficulty in quickly and accurately detecting electricity theft in DC charging pile systems, mainly in the following aspects: First, the system establishes a complete feature representation system for electricity theft through multi-source data acquisition and feature extraction; second, it achieves intelligent identification of electricity theft by combining feature fusion and deep learning; finally, the system has strong practicality and scalability, and can adapt to different types of electricity theft detection needs. Attached Figure Description

[0023] Figure 1This is a schematic diagram of the module composition of the charging pile of the present invention.

[0024] Figure 2 This is a flowchart of the steps executed by the system control module.

[0025] Figure 3 This is a comparison chart of the voltage and current calibration test results in Example 2.

[0026] Figure 4 This is a graph showing the power change over time in Example 2.

[0027] Figure 5 This is a graph showing the changing trend of carrier characteristics in Example 2.

[0028] Figure 6 This is a trend chart of the indicators for judging electricity theft in Example 2. Detailed Implementation

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

[0030] like Figure 1The diagram shows the module composition of a DC charging pile supporting rapid detection of electricity theft, as provided by this invention. This DC charging pile 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. The control chip contains a system control module. The voltage acquisition device is used to acquire voltage data at the charging pile's input and output terminals. The current acquisition device... The system includes a data acquisition device for collecting current data at the charging pile's input and output terminals; a power analysis device for calculating the active power, reactive power, and output power of the charging pile; a carrier feature extraction device for acquiring carrier amplitude, phase, and frequency; a signal conditioning device for filtering and amplifying the acquired signals; a data storage device for storing system operating parameters and historical data; a real-time communication device for transmitting data to the monitoring center; an alarm device for issuing audible and visual alarm signals; and a power management device for monitoring the system's power supply voltage. The voltage and current acquisition devices have a sampling frequency of 1000 times per second, and the carrier feature extraction device has a sampling frequency of 100 times per second. The system control module executes the following steps:

[0031] S01. The voltage acquisition device and the current acquisition device 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, and establish a voltage data matrix and a current data matrix.

[0032] S02. 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 using the power analysis device, construct a power feature vector, and record the power fluctuation characteristics.

[0033] S03. Collect the carrier amplitude, carrier phase and carrier frequency through the carrier feature extraction device, and establish a carrier attenuation matrix;

[0034] S04. Demodulate and analyze the acquired carrier signal, extract the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value, and construct a carrier demodulation anomaly matrix.

[0035] S05. Based on the pre-trained electricity theft behavior identification model, the carrier attenuation matrix and the carrier demodulation anomaly matrix are input into the model. The carrier feature network graph is constructed using the carrier attenuation matrix. The minimum spanning tree is calculated using the Kruskal algorithm. The sum of the weights of the minimum spanning tree is used as the electricity theft behavior feature index. The probability of electricity theft behavior is calculated by combining the carrier demodulation anomaly matrix.

[0036] S06. Calculate the electricity theft behavior judgment index based on the set of electricity theft behavior judgment equations. The set of electricity theft behavior judgment equations 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 anomaly index. The input of the power characteristic equation includes the active power at the charging pile input end, the active power at the charging pile output end, the reactive power at the charging pile input end, the reactive power at the charging pile output end, and the power loss threshold. The carrier attenuation characteristic equation is used to calculate the carrier attenuation anomaly index. The input of the carrier attenuation characteristic equation includes 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 time-series correlation index of electricity theft behavior. The input of the time-series correlation equation includes the power loss anomaly index, the carrier attenuation anomaly index, the historical data, and the time weighting coefficient.

[0037] S07. Using the data storage device, record the system operating parameters and the historical data to establish a database of electricity theft behavior characteristics;

[0038] S08. When the time-series correlation index of the electricity theft exceeds a preset threshold, an alarm message is sent to the monitoring center through the real-time communication device, and the alarm device is activated to issue an audible and visual alarm signal.

[0039] The carrier attenuation matrix refers to the attenuation of the carrier amplitude, carrier phase, and carrier frequency of the power line carrier signal during transmission. 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. Matrix elements include the carrier amplitude attenuation coefficient, the carrier phase offset, and the carrier frequency drift value. The power loss anomaly index refers to the degree of power loss anomaly at the input and output ends of the charging pile. The carrier attenuation anomaly index refers to the degree of power line carrier signal attenuation anomaly. The standard attenuation threshold refers to the standard value of carrier signal attenuation under normal operating conditions. The time weighting coefficient refers to the weight of electricity theft behavior occurring in different time periods.

[0040] The electricity theft detection model employs an improved residual neural network structure, adding a feature fusion module between the feature extraction layer and the classification layer of the improved residual neural network structure. This feature fusion module fuses power line carrier characteristics and power loss characteristics, taking the power line carrier attenuation matrix and the power feature vector as inputs and outputting an electricity theft behavior feature vector. The steps for establishing the training dataset for the electricity theft detection model include collecting normal operation data and electricity theft status data, labeling the electricity theft behavior with the corresponding power line carrier attenuation matrix and power feature vector, and dividing the dataset into training, validation, and test datasets in an 8:1:1 ratio. The training steps for the electricity theft detection model specifically include inputting the training dataset into the improved residual neural network structure for forward propagation calculation, using the cross-entropy loss function to calculate the error between the predicted result and the true label, optimizing and updating the network parameters using the backpropagation algorithm, evaluating the training process using the validation dataset, stopping training when the validation accuracy reaches a preset threshold, and performing performance testing on the trained model using the test dataset.

[0041] 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, which add channel attention and spatial attention mechanisms to the standard residual structure. The feature fusion module uses an adaptive weighting method to fuse power line carrier features and power loss features. The classification layer uses a fully connected layer to reduce the dimensionality of the feature vectors and outputs the probability of electricity theft.

[0042] The specific embodiments of the present invention will be described in detail below.

[0043] The control chip uses a high-performance 32-bit processor STM32H750VBT6 with a main frequency of up to 480MHz. It has 2MB of flash memory and 1MB of RAM built in, which are used to execute system control programs, data processing and communication management. The system control module interacts with each functional module and transmits control signals through the internal bus.

[0044] The voltage acquisition device includes two sets of high-precision voltage sensors at the input and output ends. It adopts the Hall voltage sensor LEM LV25P, with a range of 0 to 1000V, an accuracy of 0.1%, and a sampling frequency of 1000Hz. Data acquisition is performed through a 16-bit ADC to realize real-time monitoring of the voltage at the input and output ends of the charging pile.

[0045] The current acquisition device also includes two sets of high-precision Hall current sensors, LEMLAH25NP, at the input and output ends. The range is 0 to 100A, the accuracy is 0.1%, and the sampling frequency is 1000Hz. Data acquisition is performed through a 16-bit ADC to realize real-time monitoring of the current at the input and output ends of the charging pile.

[0046] The power analysis device uses a dedicated digital signal processor TMS320F28335 with a main frequency of 150MHz. It has a built-in hardware multiplier and floating-point unit, and calculates active and reactive power through a digital integration algorithm with a calculation accuracy better than 0.2%. It can process power data from both the input and output terminals simultaneously.

[0047] The carrier feature extraction device uses the high-speed data acquisition chip AD9226 with a sampling rate of 100Hz and a resolution of 12 bits. Combined with a bandpass filter and a digital phase-locked loop circuit, it can extract the amplitude, phase and frequency features of the power line carrier signal.

[0048] The signal conditioning device includes a two-stage operational amplifier circuit and a Butterworth low-pass filter. The operational amplifier is an OP27 with an adjustable gain range of 1 to 100 times and a cutoff frequency of 10kHz. It is used to amplify and filter the acquired signal to improve the signal-to-noise ratio.

[0049] The data storage device uses 32GB industrial-grade eMMC flash memory, supports the FAT32 file system, and has power-loss protection. It is used to store system configuration parameters, operation logs, and historical data, with a data retention period of 3 months.

[0050] The real-time communication device uses the SIM7600CE 4G wireless communication module, which supports the TCP / IP protocol and has a communication rate of up to 100Mbps. It sends data to the monitoring center using encrypted transmission, and the data transmission delay is less than 100ms.

[0051] 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 show the system's working status: green indicates normal, yellow indicates warning, and red indicates malfunction.

[0052] 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. It has overvoltage, undervoltage and overcurrent protection functions and can provide stable operating voltage for each module with an efficiency of up to 95%.

[0053] In the specific implementation of the carrier feature extraction device, the carrier sampling circuit adopts the high-speed sample-and-hold circuit SHC804 and the 12-bit analog-to-digital converter AD9226. The sampling clock is provided by the programmable clock generator Si5351A. The phase detection adopts the quadrature demodulation method, and the frequency measurement adopts the zero-crossing detection method.

[0054] In the specific implementation of the signal conditioning device, a two-stage in-phase amplifier circuit is adopted. The first stage has a gain of 10 times, and the second stage has an adjustable gain. The filter adopts a fourth-order Butterworth low-pass filter, and the operational amplifier is selected as the low-noise, low-drift OP27. The circuit gain and bandwidth can be adjusted by the programmable potentiometer AD8400.

[0055] The above devices communicate with each other via an RS485 bus at a rate of 115200bps using the Modbus RTU protocol. Each device has an independent slave address, and the control chip acts as the master to manage data transmission.

[0056] All modules use industrial-grade components, operate in a temperature range of -40 degrees Celsius to 85 degrees Celsius, have a relative humidity range of 5% to 95%, and have an IP65 protection rating, meeting the requirements for outdoor installation and use.

[0057] The system uses a switching power supply with an input voltage of 220V AC and multiple output DC voltages, including 3.3V, 5V, and 12V, with a total power of 200W. It has a complete electromagnetic compatibility design and strong anti-interference capability.

[0058] The specific implementation method of the system control module execution steps is described in detail below.

[0059] The specific implementation of step S01 involves using a high-speed data acquisition scheme to collect voltage and current data at the input and output terminals of the charging pile. First, voltage data at the input and output terminals of the charging pile is collected using a voltage acquisition device at a sampling frequency of 1000Hz. The sampled data is then converted by a 16-bit ADC and stored in a voltage data buffer. Simultaneously, current data at the input and output terminals of the charging pile is collected using a current acquisition device, also at a sampling frequency of 1000Hz. The sampled data is also converted by a 16-bit ADC and stored in a current data buffer. Then, the collected voltage and current data undergo digital filtering, employing a moving average filtering algorithm to eliminate sampling noise. The filtering window length is 16 points. Finally, the processed data is arranged in chronological order to construct voltage and current data matrices. The number of rows in the matrix represents the sampling time point, and the number of columns represents the measurement parameters, including input voltage, output voltage, input current, and output current. The purpose of this step is to obtain basic voltage and current data during the operation of the charging pile, providing data support for subsequent power analysis and anomaly detection.

[0060] The specific implementation of step S02 is based on the collected voltage and current data matrices to calculate the active and reactive power at the input and output ends of the charging pile. First, the voltage and current data are synchronously sampled using a power analysis device with a sampling period of 20 milliseconds and 50 points are collected in each period. Then, the fundamental and harmonic components of the voltage and current signals are calculated using the discrete Fourier transform algorithm. The active power is calculated based on the fundamental component, and the reactive power is calculated based on the harmonic components. Next, a power feature vector is constructed, with vector elements including active power at the input end, reactive power at the input end, active power at the output end, and reactive power at the output end. Finally, 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. 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.

[0061] The specific implementation of step S03 involves acquiring the amplitude, phase, and frequency characteristics of the power line carrier signal using a carrier feature extraction device. First, a bandpass filter is used to preprocess the carrier signal. The center frequency of the filter is 100kHz, and the bandwidth is 20kHz. Then, orthogonal demodulation is used to extract the amplitude and phase information of the carrier signal. The sampling frequency is 100Hz. Next, the zero-crossing detection method is used to measure the frequency of the carrier signal, with a measurement accuracy better than 0.1Hz. Finally, the acquired carrier feature data is arranged in chronological order to construct a carrier attenuation matrix. The matrix elements include the carrier amplitude attenuation coefficient, the carrier phase offset, and the 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.

[0062] The specific implementation of step S04 involves demodulating and analyzing the acquired carrier signal to extract abnormal features of the carrier modulation parameters. First, a digital phase-locked loop (PLL) is used to synchronously demodulate the carrier signal. The PLL bandwidth 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. The amplitude attenuation coefficient is calculated using the logarithmic ratio method, the phase offset using the phase difference method, and the frequency drift using the frequency difference method. Next, the calculated parameter deviation values ​​are arranged in chronological order to construct a carrier demodulation anomaly matrix. Finally, the anomaly matrix is ​​subjected to feature statistical analysis to calculate the mean, variance, and rate of change of each parameter. The purpose of this step is to analyze the abnormal modulation characteristics of the carrier signal and provide demodulation feature data for detecting electricity theft.

[0063] The specific implementation of step S05 is to perform probability analysis of electricity theft behavior based on a pre-trained electricity theft behavior identification model. First, the carrier attenuation matrix and carrier demodulation anomaly matrix are used as model inputs. Then, a carrier feature network graph is constructed using the carrier amplitude attenuation coefficient, carrier phase offset, and carrier frequency drift value in the carrier attenuation matrix. The nodes of the network graph represent the carrier features at different time points, and the weights of the edges represent the degree of difference 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 electricity theft behavior feature index, which reflects the degree of temporal change of the carrier features. Finally, combined with the anomaly parameters in the carrier demodulation anomaly matrix, the probability of electricity theft behavior is calculated using a weighted summation method. The weight coefficients are determined through model training. The purpose of this step is to make a preliminary judgment on electricity theft behavior using machine learning methods.

[0064] The specific implementation of step S06 involves calculating the electricity theft behavior judgment index based on the set of equations. First, the power loss anomaly index is calculated based on the power characteristic equation. This equation uses the active power and reactive power at the input and output ends as input variables, and 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 characteristic 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 time-series correlation index of electricity theft behavior is calculated based on the time-series 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 adopts an exponential decay function, with the weight of the most recent time point being 1, and the weight of historical time points decreasing as the time interval increases. Finally, the calculation results of the three equations are comprehensively evaluated to obtain a comprehensive judgment index of electricity theft behavior. The purpose of this step is to quantitatively evaluate electricity theft behavior from multiple perspectives.

[0065] The specific implementation of step S07 involves using a data storage device to record system operating parameters and historical data to establish a database of electricity theft characteristics. First, the system operating parameters are stored in chronological order, including basic data such as voltage, current, and power, as well as derived data such as carrier characteristics and abnormal indicators. The data storage period is 1 minute. Then, the historical data is compressed and stored using a lossless compression algorithm to reduce storage space usage, with a compression ratio of no less than 5:1. Next, a data index structure is established, using a B-tree index to improve 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 electricity theft analysis.

[0066] The specific implementation of step S08 is to realize the alarm function for electricity theft. First, it is determined whether the time-series correlation index of electricity theft exceeds a preset threshold. The preset threshold is set to 0.8. When the index exceeds the threshold, the alarm program is activated. Then, alarm information is sent to the monitoring center through a real-time communication device. The alarm information includes timestamp, location information, abnormal index value, etc. Encrypted transmission is used to ensure data security. Next, the alarm device is activated 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 cause, and processing result. The purpose of this step is to detect and handle electricity theft in a timely manner.

[0067] The specific implementation of the electricity theft detection model adopts an improved residual neural network structure. First, in the input layer, a standardization processing module is used to normalize the input data. Batch normalization is used to calculate the mean and variance of the data, and then a standardization transformation is performed. Next, multiple improved residual blocks are set in the feature extraction layer. Each residual block contains two 3×3 convolutional layers and one skip connection. Channel attention and spatial attention mechanisms are added to the standard residual structure. Channel attention adopts a squeezed excitation network structure, and spatial attention adopts an adaptive pooling method. Then, a feature fusion module is added between the feature extraction layer and the classification layer. An adaptive weight method is used to fuse power line carrier features and power loss features. The weight coefficients are automatically learned through the backpropagation algorithm. Finally, a fully connected layer is used in the classification layer to reduce the dimensionality of the feature vector and output the probability of electricity theft.

[0068] The training dataset for the electricity theft identification model was established by combining laboratory simulation and on-site data collection. First, different types of electricity theft were simulated in a laboratory environment, and corresponding carrier and power feature data were collected. Then, data on normal operation status were collected at actual charging pile sites. Next, the collected data were labeled to clearly identify the types of electricity theft. Finally, the dataset was divided into training, validation, and test sets in a ratio of 8:1:1.

[0069] The model training process first employs the mini-batch stochastic gradient descent algorithm with a batch size of 32 and an initial learning rate of 0.001. Cosine annealing is used to dynamically adjust the learning rate. Then, the cross-entropy loss function is used to calculate the error between the predicted result and the true label. Next, the gradient is calculated and the network parameters are updated using the backpropagation algorithm. Gradient pruning is used to prevent gradient explosion. Finally, the training effect is evaluated using a validation set. Training is stopped when the validation accuracy reaches 95%. The trained model is then tested using a test set with a test accuracy requirement of no less than 93%.

[0070] This invention involves multiple calculation processes and matrix construction, each of which is described in detail below:

[0071] The voltage data matrix and current data matrix are constructed as follows:

[0072] ;

[0073] ;

[0074] In the formula, and They represent the first The input and output voltages at each sampling time, in volts; and They represent the first The input and output currents at each sampling time, in amperes; This represents the number of sampling points, with a value of 1000.

[0075] The construction process of voltage and current data matrices: First, the sampling frequency requirement is considered. According to the Nyquist sampling theorem, the sampling frequency is set to 1000Hz, which can effectively capture signal changes below 500Hz. The number of rows in the matrix corresponds to the sampling time point, and the number of columns corresponds to the measurement parameters. Two columns are used to store the input and output data respectively, which facilitates subsequent calculation and processing. Each matrix element is obtained by sampling through a 16-bit ADC with a sampling accuracy of 0.1%, which meets the measurement requirements.

[0076] The calculations for active power and reactive power are expressed as follows:

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] In the formula, and These represent the active power at the input and output terminals, respectively, in watts. and These are the reactive power at the input and output terminals, respectively, in volts (var). and These are the phase angles at the input and output terminals, respectively, in radians; The value is 50 to represent the number of sampling points within the calculation period; , , , This is the measurement error correction factor, ranging from 0.01 to 0.05.

[0082] The derivation process for calculating active power and reactive power is described in detail below: Based on the definition of power, active power is the average of the product of the instantaneous values ​​of voltage and current; considering the phase difference in actual measurements, a phase angle term is introduced. and To improve calculation accuracy, a 50-point digital integration algorithm is used; a measurement error correction coefficient is added. to The results were obtained through experimental calibration; this calculation method can simultaneously obtain active power and reactive power, and has high calculation efficiency.

[0083] The calculation representation of carrier feature extraction is as follows:

[0084] ;

[0085] ;

[0086] ;

[0087] In the formula, This is the carrier amplitude, measured in volts. The carrier phase is expressed in radians. The carrier frequency is measured in Hertz (Hz). and These are in-phase components and quadrature components, respectively; , , This is a noise correction term, ranging from 0.001 to 0.01.

[0088] The calculation derivation process for carrier feature extraction is described in detail below: Using the principle of quadrature demodulation, the carrier signal is decomposed into in-phase components. and orthogonal components Amplitude calculation uses the vector magnitude method, phase calculation uses the arctangent function, and frequency calculation uses the phase derivative method, which can track frequency changes in real time. A noise correction term is introduced. , , The results were obtained through experimental statistics; this method has high noise resistance.

[0089] The carrier attenuation matrix is ​​constructed as follows:

[0090] ;

[0091] In the formula, For the first The amplitude attenuation coefficient at each sampling time, dimensionless, ranging from 0 to 1; For the first Phase offset at each sampling moment, in radians; For the first The frequency drift value at each sampling time, in Hertz; This represents the number of sampling points, with a value of 100.

[0092] The process of constructing the carrier attenuation matrix is ​​described in detail below: the three column vectors of the matrix correspond to amplitude attenuation, phase shift, and frequency drift, respectively; each parameter is represented by a relative value for easy comparison and analysis; the number of sampling points is set to 100 to meet the requirements of feature extraction; the matrix comprehensively reflects the transmission characteristics of the carrier signal.

[0093] The carrier demodulation anomaly matrix is ​​constructed as follows:

[0094] ;

[0095] In the formula, This indicates an abnormal value in amplitude attenuation; This indicates an abnormal phase shift value; This indicates an outlier in the frequency drift. , , These represent the amplitude attenuation coefficient, phase offset, and frequency drift value under standard operating conditions.

[0096] The weight calculation representation of the network graph of electricity theft behavior characteristics is as follows:

[0097] ;

[0098] In the formula, For nodes in the network graph and nodes Edge weights between them; and Representing nodes respectively and nodes The Each feature component; This is the smoothing factor, with a value of 0.01.

[0099] The approach to calculating the weights of the graph in the feature space for electricity theft behavior is as follows: Euclidean distance is used to measure the similarity between nodes in the feature space; a smoothing factor is introduced. This method avoids zero weights and can effectively describe the spatiotemporal correlation of features.

[0100] The expression for the power characteristic equation is as follows:

[0101] ;

[0102] In the formula, This is a dimensionless index for power loss anomaly. The standard active power loss rate is set to 0.05. The standard reactive power loss rate is set to 0.08. and Let be the weight coefficient, and satisfy... ; This is a correction term, ranging from 0 to 0.1.

[0103] The power characteristic equation is established as follows: the relative deviation between input and output power is used to represent the degree of anomaly; the influence of active power and reactive power are considered separately; the weighting coefficients are determined through experimental optimization; and correction terms are used. This equation is used to compensate for measurement errors and can accurately reflect power transmission anomalies.

[0104] The expression for the carrier attenuation characteristic equation is as follows:

[0105] ;

[0106] In the formula, This is a dimensionless index for carrier attenuation anomaly. , , Let be the feature weight coefficients, and satisfy . ; This is a correction term, ranging from 0 to 0.1.

[0107] The carrier attenuation characteristic equation is established as follows: a multi-feature weighted average method is used; the feature weight coefficients are optimized through a large amount of experimental data; and a correction term is used. This is used to improve the robustness of the equation; the equation comprehensively assesses the degree of anomaly in the carrier signal.

[0108] The expression for the time series correlation equation is as follows:

[0109] ;

[0110] In the formula, This is a time-series correlation indicator for electricity theft, and is dimensionless. and The first Power loss anomaly index and carrier attenuation anomaly index at each historical moment. For time weighting coefficients, This is the attenuation coefficient, with a value of 0.1; This represents the length of the historical data, with a value of 10. This is a correction term, ranging from 0 to 0.1.

[0111] The approach to establishing the time-series correlation equation is as follows: Exponentially decaying weights are used to represent time correlation; the decay coefficient... Experiments confirmed that the historical data length was set to 10 time points; correction terms... This equation is used to compensate for accumulated errors; it effectively describes the temporal characteristics of electricity theft.

[0112] The construction principles and significance of the above equations are as follows: The power characteristic equation considers the anomalies in active and reactive power losses, describes the degree of anomaly in the form of relative deviation, introduces weighting coefficients to achieve feature fusion, and adds correction terms to improve robustness; The carrier attenuation characteristic equation uses a multi-feature weighted average method to comprehensively evaluate the degree of carrier signal anomaly, and the feature weighting coefficients are determined through experimental optimization; The time-series correlation equation uses exponential decay weights, which reflects the influence of historical data on the current judgment. The longer the time span, the smaller the weight, which is more in line with the actual situation.

[0113] Optionally, the computational representation in the improved residual neural network is as follows:

[0114] ;

[0115] In the formula, To improve the output of the residual block; For input features; It is a mapping function for two 3×3 convolutions; Channel attention weights; For spatial attention weights.

[0116] Optionally, the computation of the feature fusion module is represented as follows:

[0117] ;

[0118] In the formula, The fused feature vector; This is the characteristic vector of the power line carrier. This is the power loss feature vector; and The adaptive weight coefficients satisfy the following conditions: .

[0119] Optionally, the probability of electricity theft can be calculated as follows:

[0120] ;

[0121] In the formula, This represents the probability of electricity theft occurring. It is a time-series correlation indicator; This is the sum of the weights of the minimum spanning tree; and These are model parameters; This is a bias term.

[0122] Specifically, the principle of this invention is based on the following scientific understanding: electricity theft simultaneously affects both the power line carrier signal transmission characteristics and the system power loss characteristics. Regarding the power line carrier signal, electricity theft alters the circuit topology, causing abnormal attenuation, phase shift, and frequency drift during transmission. By establishing a power line carrier attenuation matrix, the system can quantify these abnormal characteristics, providing reliable feature indicators for electricity theft detection. Regarding power loss characteristics, electricity theft causes a power imbalance between the input and output of the charging pile, manifested as abnormal changes in active and reactive power. By constructing a power feature vector, the system can capture these abnormal changes.

[0123] This invention employs an improved residual neural network structure for feature learning and pattern recognition, a design based on the fundamental principles of deep learning. First, residual connections effectively mitigate the vanishing gradient problem in deep networks, ensuring effective model training. Second, the introduction of channel attention and spatial attention mechanisms enables the model to adaptively focus on important features, improving feature extraction efficiency. Finally, the feature fusion module learns the optimal feature combination weights, achieving intelligent fusion of power line carrier features and power loss features.

[0124] The technical solution of this invention follows a logical chain of "multi-source sensing - feature extraction - feature fusion - intelligent recognition," with each link closely connected and mutually supporting. The hardware module design ensures the comprehensiveness and accuracy of data acquisition, while the software module design guarantees the intelligence and reliability of data processing and decision-making. Through systematic design and optimization, this invention achieves high efficiency and high accuracy in detecting electricity theft.

[0125] 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 below.

[0126] In Embodiment 1 of the present invention, the control chip adopts a high-performance 32-bit processor STM32H750VBT6 with a main frequency of 480MHz, and has 2MB of flash memory and 1MB of RAM built-in. It is used to execute system control programs, data processing and communication management. The system control module interacts with each functional module and transmits control signals through an internal bus with a bus clock frequency of 240MHz.

[0127] The voltage acquisition device includes two sets of high-precision Hall voltage sensors (LEM LV25P) at the input and output ends, with a range of 0 to 1000V, an accuracy of 0.1%, and a sampling frequency of 1000Hz. Data acquisition is performed via a 16-bit ADC to achieve real-time monitoring of the input and output voltages of the charging pile. The sampled data conforms to the following voltage data matrix format:

[0128] ;

[0129] In the formula, and They represent the first The input and output voltages at each sampling time, in volts.

[0130] The current acquisition device uses a high-precision Hall current sensor, LEM LAH25NP, with a range of 0 to 100A, an accuracy of 0.1%, and a sampling frequency of 1000Hz. Data acquisition is performed via a 16-bit ADC to achieve real-time monitoring of the input and output currents of the charging pile. The sampled data meets the following current data matrix format:

[0131] ;

[0132] In the formula, and They represent the first The input and output currents at each sampling time are measured in amperes.

[0133] The power analysis device uses a dedicated digital signal processor TMS320F28335 with a main frequency of 150MHz. It has a built-in hardware multiplier and floating-point unit, and calculates active and reactive power through a digital integration algorithm with a calculation accuracy better than 0.2%. It can simultaneously process power data from both the input and output terminals. The power calculation satisfies the following equation:

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] In the formula, and These represent the active power at the input and output terminals, respectively, in watts. and These represent the reactive power at the input and output terminals, respectively, in vars. and These are the phase angles at the input and output terminals, respectively, in radians; The value is 50 to represent the number of sampling points within the calculation period; , , , This is the measurement error correction factor, ranging from 0.01 to 0.05.

[0139] The carrier feature extraction device uses the high-speed data acquisition chip AD9226 with a sampling rate of 100Hz and a resolution of 12 bits. Combined with a bandpass filter and a digital phase-locked loop circuit, it extracts the amplitude, phase, and frequency characteristics of the power line carrier signal. The filter's center frequency is 100kHz and its bandwidth is 20kHz. The carrier feature extraction satisfies the following equation:

[0140] ;

[0141] ;

[0142] ;

[0143] In the formula, This is the carrier amplitude, measured in volts. The carrier phase is expressed in radians. The carrier frequency is measured in Hertz (Hz). and These are in-phase components and quadrature components, respectively; , , This is a noise correction term, ranging from 0.001 to 0.01.

[0144] The signal conditioning device includes a two-stage operational amplifier circuit and a Butterworth low-pass filter. The operational amplifier uses an OP27 with an adjustable gain range of 1 to 100 times and a cutoff frequency of 10kHz. It is used to amplify and filter the acquired signal to improve the signal-to-noise ratio. The first stage has a gain of 10 times, and the second stage has an adjustable gain. The filter uses a fourth-order Butterworth low-pass filter. The operational amplifier is a low-noise, low-drift OP27. The circuit gain and bandwidth can be adjusted by a programmable potentiometer AD8400.

[0145] The data storage device uses 32GB industrial-grade eMMC flash memory, supports the FAT32 file system, and uses a circular write method to store data. When the storage space is insufficient, the oldest data is automatically deleted. It has power failure protection and is used to store system configuration parameters, operation logs and historical data. The data retention period is 3 months.

[0146] The real-time communication device uses the SIM7600CE 4G wireless communication module, supports the TCP / IP protocol, has a communication rate of up to 100Mbps, uses the RSA encryption algorithm to encrypt transmitted data, has a key length of 2048 bits, and has a data transmission delay of less than 100ms.

[0147] 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 show the system's working status: green indicates normal, yellow indicates warning, and red indicates malfunction.

[0148] 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. It has overvoltage protection threshold of 36V, undervoltage protection threshold of 9V, and overcurrent protection threshold of 10A. It can provide stable operating voltage for each module and has a conversion efficiency of up to 95%.

[0149] The above devices communicate with each other via an RS485 bus at a rate of 115200bps using the Modbus RTU protocol. Each device has an independent slave address, and the control chip acts as the master to manage data transmission. The communication frame format includes the device address, function code, data area, and check bit.

[0150] All modules use industrial-grade components, with an operating temperature range of -40 degrees Celsius to 85 degrees Celsius and a relative humidity range of 5% to 95%. The protection level reaches IP65, meeting the requirements for outdoor installation and use. The shell is made of aluminum alloy material with a powder-coated surface, providing excellent corrosion resistance.

[0151] The system uses a switching power supply with an input voltage of 220V AC and multiple output DC voltages, including 3.3V, 5V, and 12V, with a total power of 200W. It has a complete electromagnetic compatibility design and meets national standards in terms of radiated emissions, conducted emissions, and immunity.

[0152] In this embodiment 1, the specific implementation methods of steps S01 to S08 are as follows:

[0153] The specific implementation of step S01 involves using a high-speed data acquisition scheme to collect voltage and current data at the input and output terminals of the charging pile. A voltage acquisition device collects voltage data from the input and output terminals of the charging pile at a sampling frequency of 1000Hz. The sampled data is then converted by a 16-bit ADC and stored in a voltage data buffer. Simultaneously, a current acquisition device collects current data from the input and output terminals of the charging pile at a sampling frequency of 1000Hz. The sampled data is also converted by a 16-bit ADC and stored in a current data buffer. The collected voltage and current data are respectively constructed into a voltage data matrix and a current data matrix, with the matrix format satisfying:

[0154] ;

[0155] ;

[0156] In the formula, and They represent the first The input and output voltages at each sampling time, in volts; and They represent the first The input and output currents at each sampling time, in amperes; The number of sampling points is 1000. The sampled data is processed by a moving average filtering algorithm with a filter window length of 16 points to eliminate sampling noise and improve data quality.

[0157] 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 terminals of the charging pile. The calculation equation is as follows:

[0158] ;

[0159] ;

[0160] ;

[0161] ;

[0162] In the formula, and These represent the active power at the input and output terminals, respectively, in watts. and These represent the reactive power at the input and output terminals, respectively, in vars. and These are the phase angles at the input and output terminals, respectively, in radians; The value is 50 to represent the number of sampling points within the calculation period; , , , The measurement error correction coefficient ranges from 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.

[0163] The specific implementation of step S03 involves acquiring the amplitude, phase, and frequency characteristics of the power line carrier signal using a carrier feature extraction device. First, a bandpass filter is used to preprocess the carrier signal; the filter's center frequency is 100kHz and its bandwidth is 20kHz. Then, orthogonal demodulation is used to extract the carrier signal features.

[0164] ;

[0165] ;

[0166] ;

[0167] In the formula, This is the carrier amplitude, measured in volts. The carrier phase is expressed in radians. The carrier frequency is measured in Hertz (Hz). and These are in-phase components and quadrature components, respectively; , , The noise correction term was measured, ranging from 0.001 to 0.01; the sampling frequency was 100Hz, and data was acquired using an AD9226. The acquired feature data was used to construct a carrier attenuation matrix.

[0168] ;

[0169] In the formula, For the first The amplitude attenuation coefficient at each sampling time, dimensionless, ranging from 0 to 1; For the first Phase offset at each sampling moment, in radians; For the first The frequency drift value at each sampling time, in Hertz; This represents the number of sampling points, with a value of 100.

[0170] The specific implementation of step S04 involves demodulating and analyzing the acquired carrier signal, extracting abnormal features of the carrier modulation parameters, and using a digital phase-locked loop (PLL) to synchronously demodulate the carrier signal. The PLL bandwidth is 1 kHz, the locking time is less than 1 millisecond, and the deviation between the actual demodulation parameters and the theoretical parameters is calculated to construct a carrier demodulation anomaly matrix.

[0171] ;

[0172] In the formula, This indicates an abnormal value in amplitude attenuation; This indicates an abnormal phase shift value; This indicates an outlier in the frequency drift. , , These represent the amplitude attenuation coefficient, phase offset, and frequency drift value under standard operating conditions; characteristic statistical analysis is performed on the anomaly matrix to calculate the mean, variance, and rate of change of each parameter.

[0173] The specific implementation of step S05 is to perform probability analysis of electricity theft behavior based on a pre-trained electricity theft behavior identification model. The carrier attenuation matrix and carrier demodulation anomaly matrix are used as model inputs. A carrier feature network graph is constructed using features from the carrier attenuation matrix. The edge weights of the network graph are calculated to satisfy the following:

[0174] ;

[0175] In the formula, For nodes in the network graph and nodes Edge weights between them; and Representing nodes respectively and nodes The Each feature component; The smoothing factor is set to 0.01. The Kruskal algorithm is used to calculate the minimum spanning tree of the network graph, with a time complexity of 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 a characteristic index of electricity theft. Combined with the anomaly parameters in the carrier demodulation anomaly matrix, the probability of electricity theft is calculated using the following equation:

[0176] ;

[0177] In the formula, This represents the probability of electricity theft occurring. It is a time-series correlation indicator; This is the sum of the weights of the minimum spanning tree; and These are model parameters; This is a bias term.

[0178] The specific implementation of step S06 is to calculate the electricity theft behavior judgment index based on the electricity theft behavior judgment equation set. The power characteristic equation is:

[0179] ;

[0180] In the formula, This is a dimensionless index for power loss anomaly. The standard active power loss rate is set to 0.05. The standard reactive power loss rate is set to 0.08. and Let be the weight coefficient, and satisfy... ; This is a correction term, ranging from 0 to 0.1; the carrier attenuation characteristic equation is:

[0181] ;

[0182] In the formula, This is a dimensionless index for carrier attenuation anomaly. , , Let be the feature weight coefficients, and satisfy . ; The correction term ranges from 0 to 0.1; the time series correlation equation is:

[0183] ;

[0184] In the formula, This is a time-series correlation indicator for electricity theft, and is dimensionless. and The first Power loss anomaly index and carrier attenuation anomaly index at each historical moment. For time weighting coefficients, This is the attenuation coefficient, with a value of 0.1; This represents the length of the historical data, with a value of 10. This is a correction term, ranging from 0 to 0.1.

[0185] The specific implementation of step S07 is to use a data storage device to record system operating parameters and historical data, establish a database of electricity theft behavior characteristics, store system operating parameters in chronological order, including basic data such as voltage, current, and power, as well as derived data such as carrier characteristics and abnormal indicators, with a data storage period of 1 minute, compress and store historical data, use a lossless compression algorithm to reduce storage space occupation, with a compression ratio of not less than 5:1, establish a data index structure, and use a B-tree index to improve data query efficiency.

[0186] The specific implementation of step S08 is to realize the alarm function for electricity theft, determine whether the time-series correlation index of electricity theft exceeds the preset threshold of 0.8, and when the index exceeds the threshold, send alarm information to the monitoring center through the real-time communication device. The alarm information includes timestamp, location information, abnormal index value, etc. The data is encrypted using the RSA encryption algorithm with a key length of 2048 bits. At the same time, the alarm device is activated to emit an audible and visual alarm signal with a sound pressure level of 95 decibels and a flash frequency of 2Hz.

[0187] The electricity theft detection model is based on an improved residual neural network, and its computation satisfies:

[0188] ;

[0189] In the formula, To improve the output of the residual block; For input features; It is a mapping function for two 3×3 convolutions; Channel attention weights; For spatial attention weights; the calculation of the feature fusion module satisfies:

[0190] ;

[0191] In the formula, The fused feature vector; This is the characteristic vector of the power line carrier. This is the power loss feature vector; and The adaptive weight coefficients satisfy the following conditions: .

[0192] The training process of the model adopts the mini-batch stochastic gradient descent algorithm with a batch size of 32 and an initial learning rate of 0.001. The learning rate is dynamically adjusted using a cosine annealing strategy. The loss function is the cross-entropy function. The gradient is calculated and the network parameters are updated through the backpropagation algorithm. Gradient pruning is used to prevent gradient explosion. Training is stopped when the validation accuracy reaches 95%, and the testing accuracy is required to be no less than 93%.

[0193] The training dataset was established by combining laboratory simulation and on-site data collection. In the laboratory environment, different types of electricity theft were simulated and corresponding carrier and power characteristic data were collected. In the actual charging pile site, data on normal operation status was collected. The collected data were labeled to clearly identify the types of electricity theft. The dataset was divided into training set, validation set and test set in a ratio of 8:1:1.

[0194] To better understand and implement this invention, the following is a specific application scenario of this invention, Example 2: The R&D team first conducted calibration tests on the voltage and current acquisition device of the charging pile, and the test results are shown in Table 1:

[0195] Table 1 Calibration Test Results of Voltage and Current Acquisition Device

[0196]

[0197] Figure 3 The comparison of voltage and current calibration test results is shown. After the calibration test was completed, the R&D team conducted a calibration experiment on the power analysis device using a standard power source. The experimental results are shown in Table 2.

[0198] Table 2. Results of the calibration experiment for the power analysis device

[0199]

[0200] During actual operation, a certain charging pile experienced an abnormal situation on January 15, 2025. The voltage and current data recorded by the system are shown in Table 3:

[0201] Table 3 Voltage and current data under abnormal conditions

[0202]

[0203] Based on the collected data, the system calculates the power feature vector, and the calculation results are shown in Table 4:

[0204] Table 4 Power characteristic calculation results

[0205]

[0206] Figure 4 The trend of power variation over time is shown. Multiple line graphs are used to illustrate the changes in input and output active and reactive power, clearly demonstrating the dynamic change process of power parameters. Meanwhile, the carrier characteristic data collected by the system are shown in Table 5.

[0207] Table 5 Carrier characteristic data

[0208]

[0209] Figure 5 The changing trends of carrier characteristics are shown. Multiple axes are used to display the changes in amplitude attenuation coefficient, phase shift, and frequency drift over time, highlighting abnormal changes in carrier characteristics. The results calculated by the system based on the equations for judging electricity theft are shown in Table 6.

[0210] Table 6 Calculation Results of Indicators for Judging Electricity Theft

[0211]

[0212] Figure 6 The system demonstrates the changing trends of indicators used to determine electricity theft. Line graphs are used to show the changes in power loss anomalies, carrier attenuation anomalies, and time-series correlation indicators, with warning threshold lines clearly marked, providing a visual representation of the electricity theft detection process. When the time-series correlation indicator exceeds a preset threshold of 0.8, the system determines that electricity theft has occurred, sends an alarm message to the monitoring center via real-time communication, and simultaneously activates an audible and visual alarm.

[0213] An inspection of the charging station revealed that someone had illegally connected a shunt device between the charging station's output and the electric vehicle, resulting in the theft of some electricity. This incident demonstrates that the system of this invention can effectively detect electricity theft and issue timely alarms.

[0214] In this embodiment 2, traditional electricity theft detection methods mainly rely on the difference between input and output power, which is easily affected by measurement errors and load fluctuations, resulting in a high false alarm rate. This invention, however, employs a method combining power line carrier characteristics and power characteristics, achieving feature fusion through an improved residual neural network and introducing time-series correlation analysis, significantly improving detection accuracy. Specifically, this is reflected in the following aspects: 1. Detection accuracy increases from 85% to 93% compared to traditional methods. 2. False alarm rate decreases from 15% to 5%. 3. Detection response time is shortened from 10 seconds to 1 second. 4. It can identify multiple types of electricity theft behavior, making it more adaptable. 5. It possesses adaptive learning capabilities, allowing for continuous optimization of the detection model.

[0215] In this second embodiment, the training process of the electricity theft detection model used 15,000 sets of normal data and 5,000 sets of electricity theft data, divided into training, validation, and test sets in an 8:1:1 ratio. During training, the learning rate started from 0.001 and was adjusted using a cosine annealing strategy. By the 1000th iteration, the validation accuracy reached 95.2%, and the test accuracy reached 93.8%, meeting the system requirements.

[0216] It should be noted that the variables involved in this invention are explained in detail in Table 7 below.

[0217] Table 7 Variable Explanation Table

[0218]

[0219] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A DC charging pile system supporting rapid detection of electricity theft, 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 contains a system control module, which is used to collect voltage and current data from the input and output terminals of the charging pile to establish a matrix; calculate active and reactive power to construct a power feature vector; collect carrier signal features to establish a carrier attenuation matrix; and demodulate and analyze the carrier signal to construct a carrier demodulation anomaly matrix. The probability of electricity theft is calculated based on a pre-trained electricity theft identification model; the electricity theft judgment index is calculated based on the electricity theft judgment equation set; system operating parameters and historical data are recorded; when the time-series correlation index of electricity theft exceeds the preset threshold, alarm information is sent and an audible and visual alarm signal is issued. The set of equations for judging electricity theft 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 anomaly index, the carrier attenuation characteristic equation is used to calculate the carrier attenuation anomaly index, and the time-series correlation equation is used to calculate the time-series correlation index of electricity theft. The inputs to the power characteristic equation include the active power at the charging pile input end, the active power at the charging pile output end, the reactive power at the charging pile input end, the reactive power at the charging pile output end, and the power loss threshold. The inputs to 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, where the standard attenuation threshold refers to the standard value of carrier signal attenuation under normal operating conditions. The inputs to the time-series correlation equation include the power loss anomaly index, the carrier attenuation anomaly index, historical data, and the time weight coefficient, where the time weight coefficient refers to the weight of electricity theft occurring in different time periods. The electricity theft identification model employs an improved residual neural network structure, and adds a feature fusion module between the feature extraction layer and the classification layer of the improved residual neural network structure. The feature fusion module is used to fuse power line carrier features and power loss features.

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

3. A DC charging pile system supporting rapid detection of electricity theft 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 to record the power fluctuation characteristics.

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

5. A DC charging pile system supporting rapid detection of electricity theft according to claim 4, characterized in that, The signal conditioning device is used to filter and amplify the acquired signal; the data storage device is used to store system operating parameters and historical data; the real-time communication device is used to transmit data to the monitoring center; the alarm device is used to issue audible and visual alarm signals; and the power management device is used to monitor the system power supply voltage.

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

7. A DC charging pile system supporting rapid detection of electricity theft 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, the carrier phase offset, and the carrier frequency drift value.

8. A DC charging pile system supporting rapid detection of electricity theft according to claim 7, characterized in that, The steps for establishing the training dataset for the electricity theft identification model include collecting normal operation status data and electricity theft status data, labeling the electricity theft behavior tags corresponding to the carrier attenuation matrix and power feature vector, and dividing the dataset into training dataset, validation dataset and test dataset in an 8:1:1 ratio.

Images