A deep learning-based charging pile abnormal electricity use behavior identification method

Abstract

The present application relates to the technical field of intelligent power system and deep learning, and particularly relates to a charging pile abnormal power consumption behavior identification method based on deep learning. The content includes: introducing dynamic time correlation modeling, analyzing power consumption time series data to obtain time dynamic characteristics; based on power consumption time series data, capturing the spatial relationship and mutual influence between different charging piles to obtain spatial correlation characteristics; fusing the time dynamic characteristics and the spatial correlation characteristics to generate multi-scale fused spatio-temporal characteristics; based on the multi-scale fused spatio-temporal characteristics, performing nonlinear mapping and classification to determine the abnormal type. The problems that the traditional charging pile abnormal power consumption behavior identification method mainly relies on simple rule judgment, cannot handle complex spatio-temporal dependence relationship, is prone to false positives or false negatives, mostly ignores the spatial correlation between charging piles, cannot effectively utilize the collaborative information of multiple charging piles for abnormal detection, and has low abnormal detection precision are solved.

Description

Technical Field

[0001] This invention relates to the fields of intelligent power systems and deep learning technology, and in particular to a method for identifying abnormal electricity consumption behavior of charging piles based on deep learning. Background Technology

[0002] With the increasing popularity of electric vehicles, charging stations, as a crucial infrastructure for electric vehicle charging, have been widely deployed across various regions. The stability and safety of charging stations directly affect the normal use of electric vehicles and the safety of users. However, during actual operation, charging stations may exhibit abnormal power consumption behaviors due to equipment malfunctions, improper user operation, or changes in environmental factors, such as current overload, voltage instability, and sudden power surges. These abnormal behaviors not only affect charging efficiency but may also pose safety hazards, such as battery damage, equipment damage, and even fire risks.

[0003] Therefore, timely and accurate detection and identification of abnormal power consumption behavior in charging piles to ensure the safety of the charging process and the reliability of equipment has become an important task in current charging pile management. With the increasing number of charging piles and continuous technological development, how to improve the accuracy and efficiency of anomaly detection through more intelligent methods has become a pressing technical problem. Based on this, using deep learning technology to achieve intelligent identification of abnormal power consumption behavior in charging piles has significant practical implications and application value.

[0004] Traditional methods for identifying abnormal electricity consumption behavior in charging piles have the following technical problems: they mainly rely on simple rule judgments or threshold-based monitoring methods, which cannot handle complex spatiotemporal dependencies, resulting in poor sensitivity to minor fluctuations or sudden anomalies, and are prone to false alarms or missed alarms; most of them ignore the spatial correlation between charging piles, cannot effectively utilize the collaborative information of multiple charging piles for anomaly detection, are difficult to cope with the complex behavior patterns of charging piles under different environmental conditions, and have low anomaly detection accuracy. Summary of the Invention

[0005] This invention provides a deep learning-based method for identifying abnormal power consumption behavior in charging piles. This addresses the shortcomings of traditional methods that rely primarily on simple rule-based judgments or threshold-based monitoring, which are unable to handle complex spatiotemporal dependencies. These methods suffer from poor sensitivity to minor fluctuations or sudden anomalies, leading to false alarms or missed alarms. Furthermore, they often neglect the spatial relationships between charging piles, failing to effectively utilize the collaborative information of multiple charging piles for anomaly detection. This makes it difficult to address the complex behavior patterns of charging piles under different environmental conditions, resulting in low anomaly detection accuracy.

[0006] The present invention provides a method for identifying abnormal electricity consumption behavior of charging piles based on deep learning, which specifically includes the following technical solutions:

[0007] A deep learning-based method for identifying abnormal electricity consumption behavior in charging stations includes the following steps:

[0008] S1. Acquire the electricity consumption data of charging piles and record it in the form of time series to generate electricity consumption time series data; introduce dynamic time correlation modeling to analyze the electricity consumption time series data and obtain time dynamic characteristics; based on the electricity consumption time series data, introduce spatial correlation modeling to capture the spatial relationship and mutual influence between different charging piles and obtain spatial correlation characteristics.

[0009] S2. By using a spatiotemporal scale transformation model, the temporal dynamic features and spatial correlation features are fused to generate multi-scale fused spatiotemporal features. Based on the multi-scale fused spatiotemporal features, nonlinear mapping and classification are performed to obtain a multi-dimensional probability vector. Based on the multi-dimensional probability vector, the anomaly type is determined and abnormal electricity consumption behavior of the charging pile is identified.

[0010] Preferably, S1 specifically includes:

[0011] In the process of dynamic time correlation modeling, transform convolution operation is performed on electricity consumption time series data to extract time series features.

[0012] Preferably, S1 specifically includes:

[0013] Based on time series characteristics, the time dynamic characteristics are calculated by weighted summation combined with a time decay term.

[0014] Preferably, S1 specifically includes:

[0015] In the process of spatial correlation modeling, spatial features are extracted from electricity consumption time series data through spatial convolution operations.

[0016] Preferably, S1 specifically includes:

[0017] Based on spatial characteristics, spatial weight coefficients are introduced, and combined with spatiotemporal joint weight terms, spatial correlation characteristics are calculated.

[0018] Preferably, S2 specifically includes:

[0019] The spatiotemporal scale transformation model introduces different scale angles. At each scale, a transformation matrix is ​​introduced to weight and fuse the temporal dynamic features and spatial correlation features to obtain spatiotemporal features.

[0020] Preferably, S2 specifically includes:

[0021] The spatiotemporal features at all scales are weighted and fused after linear transformation to obtain the spatiotemporal features after multi-scale fusion.

[0022] Preferably, S2 specifically includes:

[0023] The spatiotemporal features fused from multiple scales are used as input to a multilayer perceptron model for nonlinear mapping and discrimination to obtain a multidimensional probability vector. The maximum value of the multidimensional probability vector is determined to identify abnormal behavior of the charging pile.

[0024] The beneficial effects of the technical solution of the present invention are:

[0025] 1. Through multi-level and multi-scale deep learning processing such as dynamic time correlation modeling, spatial correlation modeling, and spatiotemporal scale transformation model, this invention can accurately capture various minute fluctuations and sudden anomalies in the power consumption process of charging piles, and identify various abnormal behaviors of charging piles in complex environments, such as overload, unstable current, and sudden power surge.

[0026] 2. By integrating spatiotemporal features and multi-scale modeling, it can fully utilize electricity consumption time-series data, spatial information, and spatiotemporal dependencies at different scales, thereby improving the accuracy and robustness of anomaly identification. By using deep learning models, it can automatically learn and optimize the potential patterns in charging pile electricity consumption data, reducing the need for manual intervention. It can also make adaptive adjustments based on actual data, continuously optimize identification performance, and adapt to changes in different charging piles and usage environments.

[0027] 3. The deep learning method of the present invention is not only applicable to charging piles of different models and brands, but can also adapt to the abnormal behavior recognition needs in different scenarios according to the ever-changing power demand and external environment; through continuous training of the deep learning model, it can adapt to more types of abnormal behavior patterns and has good scalability. Attached Figure Description

[0028] Figure 1 This is a flowchart of a deep learning-based method for identifying abnormal electricity consumption behavior in charging piles, as described in this invention. Detailed Implementation

[0029] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] The following description, in conjunction with the accompanying drawings, details a specific scheme for a deep learning-based method for identifying abnormal electricity consumption behavior in charging piles provided by this invention.

[0032] See attached document Figure 1 The diagram illustrates a flowchart of a deep learning-based method for identifying abnormal electricity consumption behavior in charging piles, according to an embodiment of the present invention. The method includes the following steps:

[0033] S1. Acquire the electricity consumption data of charging piles and record it in the form of time series to generate electricity consumption time series data; introduce dynamic time correlation modeling to analyze the electricity consumption time series data and obtain time dynamic characteristics; based on the electricity consumption time series data, introduce spatial correlation modeling to capture the spatial relationship and mutual influence between different charging piles and obtain spatial correlation characteristics.

[0034] By integrating smart meters or sensors into the charging piles, electrical parameters such as voltage, current, and power are collected in real time to obtain the charging pile's power consumption data. Specifically, each charging pile is equipped with power monitoring equipment that can record electrical parameters such as voltage, current, and power during the charging process at a frequency of seconds or even milliseconds. The frequency and granularity of data collection can be adjusted according to actual needs to ensure that minute fluctuations and sudden anomalies during the charging process can be captured.

[0035] The electricity consumption data (electrical parameters such as current, voltage, and power) of charging piles is continuously recorded in time series form to obtain electricity consumption time series data. To fully explore the dependencies between different time points in the electricity consumption time series data, dynamic temporal correlation modeling is introduced. Based on a long short-term memory network, a transform-domain convolutional network is used to perform transform convolution operations on the electricity consumption time series data, outputting time series features, and generating dynamic time features based on the time series dependencies. Specifically, the electricity consumption time series data of the charging piles is set as follows: ,in Representing a point in time Electricity consumption data (such as current, voltage, or power values). It refers to the length of the time series; time points. The dynamic characteristics need to comprehensively consider the correlation between the current moment and past moments. Therefore, a weighted sum combined with a decay function is used to express the temporal dependency. The formula for the temporal dynamic characteristics at each moment is:

[0036]

[0037] in, Indicates the current time point Its time dynamic characteristics reflect the dynamic relationship between the current moment and the past moment; It is the length of the time series, which includes all points in time from the past to the present; Indicates a historical point in time. At the current time point The intensity of the impact; It is a time decay term, used to quantify the timeliness of historical impact; It is a historical point in time. The correlation weight coefficients are optimized using gradient descent, and their values ​​range from [0,1]. It is an index of historical time points; It is the current time point; It is a historical point in time. The time decay coefficient is used to control the degree of decay of the dependency relationship between time points. It is initialized by statistical characteristics of the electricity consumption time series data of the charging pile and optimized by gradient descent method. The value range is [0.01, 0.1]. The statistical characteristics refer to the analysis results of factors such as the changing trend, periodicity, degree of lag effect, and duration of abnormal response in the historical electricity consumption time series data. It is used to help judge the dependency strength and decay rate between time points. The statistical method is a technical means well known to those skilled in the art and will not be described in detail here. By transforming the convolution operation from historical time points Electricity consumption data of charging piles The extracted time-series features have already undergone normalization of the electricity consumption data during feature extraction, therefore there are no dimensional issues. The normalization method is a well-known technique in the art and will not be elaborated here. These are the convolution kernel weights; It is a point in time. The bias term is used to correct temporal features and serves as the parameter of the transform domain convolutional network and the weight of the convolution kernel. The training method for the transform domain convolutional network obtained through synchronous training is a well-known technique in the art and will not be elaborated here; the gradient descent method is also a well-known technique in the art and will not be elaborated here.

[0038] Because charging stations are distributed in different spatial locations, their electricity consumption behaviors may be correlated. For example, adjacent charging stations may be affected by the same external environmental factors, exhibiting similar electricity consumption patterns. Therefore, based on convolutional neural networks, spatial correlation modeling is introduced to capture the spatial relationships and mutual influences between different charging stations. By considering the spatial dependencies between charging stations, the interaction between different charging stations can be understood, and abnormal electricity consumption behaviors can be effectively identified. Let be the set of location coordinates of the charging piles, where It refers to the number of charging stations. This is the location index of the charging station, ranging from 1 to... The spatial correlation characteristics of each charging station are calculated using the following formula:

[0039]

[0040] in, Indicates the first The spatial correlation characteristics of each charging pile reflect the spatial relationship between the current charging pile and other charging piles, as well as the correlation of power behavior caused by spatial location. For the first The charging pile and the first The spatial weighting coefficient between charging piles ranges from [0,1] and is calculated using a distance metric (such as Euclidean distance). The calculation method is a well-known technique in the field and will not be elaborated here. It is a spatiotemporal joint weight term used to quantify the first... At a historical point in time, the charging station Electricity consumption data for the first At the current time, the charging station The intensity of the dynamic influence of spatial correlation characteristics; It is the time weighting coefficient, indicating the weighting factor when considering the first time step. The first charging pile and the first When considering the spatial relationship of each charging pile, the importance of each time point is obtained through joint optimization of attention mechanism and gradient descent, with a value range of [0,1]. The attenuation coefficient is the spatial correlation coefficient, reflecting the first... The charging pile is for the first The impact of each charging pile was optimized using the gradient descent method, with a value range of [0.001, 5]. To achieve historical points through spatial convolution operations From the Electricity consumption data of each charging station Spatial features extracted from them, These are the convolutional kernel weights, obtained through training and optimization. For the first The spatial bias term of each charging pile is used to adjust the spatial correlation features of each charging pile. It is initialized as a zero vector and obtained through training optimization. The attention mechanism and gradient descent method, i.e. the model training optimization method, are all technical means well known to those skilled in the art and will not be described in detail here.

[0041] By leveraging spatial relationships, the spatial association model can be enhanced to learn the similar electricity consumption behaviors among charging piles, thus helping to identify abnormal electricity consumption behaviors caused by similar spatial layouts.

[0042] S2. By using a spatiotemporal scale transformation model, the temporal dynamic features and spatial correlation features are fused to generate multi-scale fused spatiotemporal features. Based on the multi-scale fused spatiotemporal features, nonlinear mapping and classification are performed to obtain a multi-dimensional probability vector. Based on the multi-dimensional probability vector, the anomaly type is determined and abnormal electricity consumption behavior of the charging pile is identified.

[0043] The electricity consumption behavior of charging piles may exhibit different patterns at different time and spatial scales. To better extract spatiotemporal features and fuse information from different scales, a spatiotemporal scale transformation model is introduced. This model fuses spatiotemporal features from different scale perspectives to enhance the ability to identify abnormal behavior. Specifically, multiple scales are introduced. At each scale, spatial correlation features and temporal dynamic features are weighted using corresponding transformation matrices to obtain spatiotemporal features. By fusing spatial and temporal information, it is ensured that the spatiotemporal features at different scales can fully reflect the abnormal behavior of charging piles, thereby capturing multi-level spatiotemporal relationships and improving the ability to identify abnormal behavior. The spatiotemporal features at each scale are calculated using the following formula:

[0044]

[0045] in, For the first The spatiotemporal characteristics at various scales reflect the spatiotemporal correlation characteristics of charging piles at different scales; and The first The transformation matrix of spatial correlation features and temporal dynamic features at each scale is used to control feature fusion at different scales. It is obtained by gradient descent, which is a well-known technique in the art and will not be described in detail here. For the first The scale bias term is used to adjust the spatiotemporal features at each scale. The gradient is calculated by the backpropagation algorithm and updated using an optimizer (such as SGD or Adam). The backpropagation algorithm and optimizer update method are well known to those skilled in the art and will not be described in detail here. The spatiotemporal scale transformation model captures the spatiotemporal features of the charging pile from different scale perspectives.

[0046] The spatiotemporal features at all output scales are weighted and fused after linear transformation, and a classification bias term is added to obtain the multi-scale fused spatiotemporal features. These multi-scale fused spatiotemporal features are then used as input to a multilayer perceptron (MLP) model, which is subsequently processed through multiple fully connected layers and activation functions for nonlinear mapping and discrimination to obtain the final classification output. The multilayer perceptron (MLP) model is a well-known technique and will not be described in detail here. The final classification output is a multidimensional probability vector representing the probability distribution of abnormal states. Each dimension of the multidimensional probability vector represents the prediction confidence of the multilayer perceptron model for a certain type of abnormality, such as overload, power surge, or charging interruption. The number of dimensions is consistent with the number of abnormality types. The formula for calculating the multidimensional probability vector is as follows:

[0047]

[0048] in, It is a multidimensional probability vector that represents the probability distribution of each possible abnormal electricity consumption behavior, used to identify whether the charging pile has abnormal behavior and its abnormality type (such as overload, short circuit, etc.). Indicates the first A weight matrix of several scales is used to map spatiotemporal features to the final classification space. It is initialized using Xavier and automatically learned through the backpropagation algorithm. The Xavier method and the backpropagation algorithm are well known to those skilled in the art and will not be described in detail here. It is a classification bias term used to shift the spatiotemporal features after linear weighted summation, enabling the multilayer perceptron model to have stronger representation capabilities. Its value range is [-1, 1].

[0049] Assume there is For each anomaly type (e.g., overload, short circuit, unstable current, etc.), a multidimensional probability vector is generated. It is A vector of dimension, where: This represents the probability of an "overload" occurring. This represents the probability of a "short circuit" occurring; and so on. This represents the probability of the last type of abnormal behavior occurring;

[0050] By analyzing multidimensional probability vectors The maximum value is used to determine and identify the current abnormal behavior of the charging pile. For example, if If the value is the maximum value and represents "unstable current", then it is determined that the current charging pile has abnormal behavior of unstable current, thus providing a basis for subsequent management and maintenance.

[0051] In summary, a deep learning-based method for identifying abnormal electricity consumption behavior in charging piles has been developed.

[0052] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0053] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0054] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for identifying abnormal electricity consumption behavior of charging piles based on deep learning, characterized in that, Includes the following steps: S1. Obtain the electricity consumption data of charging piles and record it in time series form to generate electricity consumption time series data; introduce dynamic time correlation modeling, perform transform convolution operation on the electricity consumption time series data to extract time series features, and based on the time series features, obtain the time dynamic features by weighted summation and combining the time decay term; the specific calculation formula of the time dynamic features is as follows: in, Indicates the current time point Temporal dynamic characteristics; It is the length of the time series; Indicates a historical point in time. At the current time point The intensity of the impact; It is a time decay term; It is a historical point in time. The correlation weight coefficient; It is an index of historical time points; It is the current time point; It is a historical point in time. The time decay coefficient; By transforming the convolution operation from historical time points Electricity consumption data of charging piles The extracted time-series features, These are the convolution kernel weights; It is a point in time. The bias term; Based on electricity consumption time-series data, spatial correlation modeling is introduced. Spatial features are extracted from the data through spatial convolution operations. Based on these features, spatial weight coefficients are introduced, combined with a spatiotemporal joint weight term, to capture the spatial relationships and mutual influences between different charging piles, thus obtaining spatial correlation features. The specific calculation formula for these spatial correlation features is as follows: in, Indicates the first Spatial correlation characteristics of individual charging piles; It refers to the number of charging stations; For the first The charging pile and the first Spatial weighting coefficients between charging piles; It is a spatiotemporal joint weight term; It is the time weighting coefficient, indicating the weighting factor when considering the first time step. The first charging pile and the first When considering the spatial relationships of charging stations, the importance of each point in time; The attenuation coefficient is the spatial correlation coefficient, reflecting the first... The charging pile is for the first The extent of the impact on each charging station; To achieve historical points through spatial convolution operations From the Electricity consumption data of each charging station Spatial features extracted from them, These are the convolution kernel weights; For the first Spatial offset of each charging station; S2. Through a spatiotemporal scale transformation model, different scale angles are introduced. At each scale, a transformation matrix is ​​introduced to weight and fuse the temporal dynamic features and spatial correlation features to obtain spatiotemporal features. The spatiotemporal features of all scales are then weighted and fused after linear transformation to generate multi-scale fused spatiotemporal features. Based on the multi-scale fused spatiotemporal features, nonlinear mapping and classification are performed to obtain a multi-dimensional probability vector. Based on the multi-dimensional probability vector, the anomaly type is determined and abnormal electricity consumption behavior of charging piles is identified.

2. The method for identifying abnormal electricity consumption behavior of charging piles based on deep learning according to claim 1, characterized in that, S2 specifically includes: The specific formula for calculating spatiotemporal features is as follows: in, For the first Spatiotemporal characteristics at various scales; It refers to the number of charging stations; Indicates the first Spatial correlation characteristics of individual charging piles; It is the length of the time series; and The first Transformation matrices of spatial correlation features and temporal dynamic features at various scales; Indicates the current time point Temporal dynamic characteristics; For the first Scale bias term for each scale.

3. The method for identifying abnormal electricity consumption behavior of charging piles based on deep learning according to claim 2, characterized in that, S2 specifically includes: The spatiotemporal features fused from multiple scales are used as input to a multilayer perceptron model for nonlinear mapping and discrimination to obtain a multidimensional probability vector. The maximum value of the multidimensional probability vector is determined to identify abnormal behavior of the charging pile.

Images