An electric vehicle charging detection algorithm based on multi-window cumulative sum algorithm and one-dimensional residual network

Abstract

The invention discloses an electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network, and relates to the field of load identification technology. The algorithm includes a training phase and an application phase. In the training phase, the current and voltage data of household appliances are collected, offline training data is generated, and the data is standardized to generate feature vectors, and the one-dimensional residual network model is trained using sample data marked as positive and negative examples. In the application phase, STM32F103 is used to collect current and voltage data, which are sent to the processing end after AD conversion, and the multi-window cumulative sum algorithm is used to perform event detection. After the event is detected, the current and voltage data after the event is subtracted from the current and voltage data before the event occurs to obtain the current and voltage data of the switching appliance, and then the one-dimensional residual network is used to extract the characteristic value of the switching appliance from the current and voltage data of the switching appliance, and it is judged whether the event is an electric vehicle charging event.

Description

Technical Field

[0001] The present application relates to the field of electricity technology, and in particular to an electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network. Background Art

[0002] At present, the construction of the user-side perception system of the power grid is relatively lagging behind. Problems such as unknown user load types and unclear grid load composition have seriously restricted the realization of smart power consumption and efficient energy consumption on the user side. Non-intrusive load identification technology (NILM) does not need to invade the user's internal, and only needs to process and analyze the total load power consumption data to obtain detailed power consumption information of the user's load, such as load type, load quantity, load working status, etc. The analysis results obtained by non-intrusive load identification are of great practical value to ordinary users and power companies. If ordinary users can obtain the current operating conditions of various household electrical equipment, it will help users adjust their electricity consumption habits and use electricity reasonably, thereby achieving the purpose of saving energy consumption and electricity bills. Therefore, this efficient and convenient monitoring technology has broad application prospects.

[0003] Non-intrusive load identification, the difference between different non-intrusive load monitoring schemes lies mainly in the differences in the extracted load characteristics and load decomposition algorithms. According to the load characteristics selected by the monitoring scheme and the difference in the load decomposition algorithm, the non-intrusive load monitoring scheme can be divided into the following three types of current non-intrusive identification: 1. Non-intrusive load identification method based on expert system: Load identification is based on the characteristic values ​​selected by the experts themselves and the rules designed by themselves. The advantages are strong pertinence and robustness. If a unique characteristic value for identifying each electrical device can be found, it is easy to identify the switching status and operating status of the target load. But the reality is that this is impossible. First, it is difficult to find the exclusive electrical characteristics of many electrical devices, and some electrical equipment does not even have its own exclusive electrical characteristics. Second, even if you want to find it, the manpower and material resources required are too huge, and the time required is too long. At the same time, electrical equipment is constantly updated, and the expert system must be updated along with the update of electrical equipment. First, the original electrical characteristics may become invalid. Second, the maintenance cost of the expert system is too high. 2. Non-intrusive load identification method based on traditional machine learning algorithm: The calculation result is obtained based on the feature values ​​selected by the experts themselves and the rules obtained by machine learning algorithm training; 3. The non-intrusive load identification method based on deep learning is not based on specific eigenvalues. It extracts high-order eigenvalues ​​from the original current and voltage data through a deep learning algorithm, and uses the rules trained by the deep learning algorithm on these high-order eigenvalues ​​for calculation. Summary of the invention

[0004] The following is an overview of the topics described in detail in this article.

[0005] The embodiment of the present application is based on an electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network, which can at least ensure and improve the accuracy of electric vehicle charging detection.

[0006] The present invention is based on load electrical data sampling, and the fused load electrical data sampling includes a total electrical signal sampling terminal and more than or equal to one sub-electrical signal sampling terminal. Through fused data sampling, multiple sampling terminals can be effectively utilized to expose the electrical characteristics of the load.

[0007] An embodiment of the present application provides a multi-eigenvalue load identification method.

[0008] The embodiment of the present application provides an electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network, which can at least ensure a good detection effect on electric bicycle charging events in daily life. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a flow chart of an electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network provided by an embodiment of the present invention;

[0010] Figure 2 is the training and testing process of the present invention;

[0011] Figure 3 It is the network structure of the one-dimensional residual network of the present invention. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the characteristics, operations or features described in the specification can be combined in any appropriate manner to form various implementation methods. At the same time, the steps or actions in the method description can also be replaced or adjusted in order in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the accompanying drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a necessary sequence, unless otherwise specified that a certain sequence must be followed.

[0013] In the description of the present invention, the above is understood to include the number. If there is a description of the first and the second, it is only for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0014] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0015] There are three key steps in non-intrusive load identification: event detection, load characteristic data generation, and load identification. When an electrical device starts working or changes its working state, it is usually accompanied by changes in current and power, which causes the bus to generate corresponding state changes. The purpose of event detection is to capture the state transition events of the device, record the load information before and after the change, and prepare for the next step of load characteristic data generation and load identification. The current mainstream event detection algorithms are mainly divided into two categories. One category uses end-to-end machine learning algorithms to achieve event detection and load identification at the same time. This type of algorithm requires a large amount of data sets for training, and it is unrealistic to collect data sets from all electrical devices. Therefore, it is easy to produce overfitting on a limited data set and has poor robustness. At the same time, this type of algorithm also has large time requirements and computing power requirements for model training. The other category does not require additional data sets for training, and only relies on simple statistical rules for event detection, such as the cumulative sum algorithm. At the same time, the cumulative sum algorithm can produce an amplification effect by accumulating small changes, and has good sensitivity for detecting low-power electrical appliances such as electric vehicles. However, the cumulative sum algorithm also has some problems, such as the inability to obtain the changing average electrical power, and false alarms caused by the complex transient characteristics of each electrical appliance.

[0016] Based on the above situation, an embodiment of the present invention provides an electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network. Using the multi-window cumulative sum algorithm for event detection has at least the following beneficial effects: compared with a simple cumulative sum algorithm, the electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network described in the present invention has a high accuracy rate in identifying electric bicycle charging events in daily life.

[0017] The embodiments of the present application are further described below in conjunction with the accompanying drawings.

[0018] Figure 2 It is a flow chart of training and application of the method provided by one embodiment of the present invention.

[0019] The training phase of the method is as follows: Data collection: Collect current and voltage data of various common household appliances to generate offline training data. Offline training data will be used to generate feature vectors and train classifiers; Feature vector generation: Since the one-dimensional convolutional neural network has strong feature extraction capabilities, if the VI trajectory graph or power value and other features are directly used for separate training, it is easy to produce overfitting and poor robustness. Therefore, for each sample in the offline training data, the original current and voltage data are used for standardization to obtain the feature vector, which is used for the next step of training; Model training: In model training, the labels of the current and voltage data collected when the electric vehicle is charging are marked as positive examples, and the current and voltage data labels of other electrical appliances are marked as negative examples.

[0020] The algorithm application phase includes the following processes: Data collection stage: Use stm32F103 as the collection chip to collect current and voltage data, and after AD conversion, send the converted data to the processing end through Socket; Switching event detection: When the appliance is turned on, the multi-window sliding algorithm can identify whether an event has occurred by calculating the difference between the power average in the current detection window and the power average in the sliding window, and judging whether the difference is greater than threshold 1. If an event has occurred, it will receive 6 frames of data after the sliding window, and intercept the complete current and voltage data to reconfirm the event. If it exceeds threshold 2, it is determined that the event has occurred; Characteristic vector generation: subtract the current and voltage data before the switching event from the current and voltage data after the switching event to obtain the current and voltage data of the switching device, and standardize the data to generate a characteristic vector; Model prediction: The acquired features are sent to the model, and the model is used to determine whether the appliance that triggers the switching event is an electric vehicle.

Claims

1. An electric vehicle charging detection algorithm based on a multi-window cumulative sum algorithm and a one-dimensional residual network, characterized in that Use multi-window cumulative sum algorithm for event detection, set up multiple sliding windows to calculate real-time power, realize event detection and prevent misjudgment. Use the original current and voltage data as load feature data after the event is detected. Use one-dimensional residual network as the feature extractor of the original data. Through the deep network structure, extract abstract feature representation from the current and voltage data layer by layer, and perform load identification after extracting high-level feature representation.

2. The method of claim 1 wherein the multi-window cumulative sum algorithm is used for event detection, characterized in that The algorithm selects the power value as the parameter to determine whether there is a switching event, and includes three sliding windows: average power window, cumulative sum window, and reconfirmation window, which are used to perform different tasks respectively.

3. According to the average power window, cumulative sum window, and reconfirmation window required by claim 2, it is characterized in that The average power window is used to calculate the average value of the current total power. The cumulative sum window follows the average power window. As the window slides, the cumulative sum array is continuously updated and compared with threshold 1 (a threshold set manually) to determine whether an event has occurred. When the cumulative sum window detects an event, it compares and confirms whether the difference between the average power in the window and the average power window exceeds threshold 2 (same meaning as threshold 1). If it exceeds, an alarm is issued. If it does not exceed, the event detected by the cumulative sum window is determined to be a false alarm.

Images