Method and system for predicting residual endurance mileage of lead storage battery of electric bicycle

Abstract

The invention discloses an electric bicycle lead storage battery remaining endurance mileage prediction method and system, and relates to the technical field of electric bicycles, and the method comprises the steps: carrying out the iterative training of a cloud mileage model based on the riding historical data at a cloud end, and generating a bicycle end model parameter; at the bicycle end, the bicycle end model generates a residual endurance mileage prediction value according to the battery SOC and bicycle end model parameters output by the cloud end, and adjusts the residual endurance mileage prediction value in real time in combination with the real-time riding state and environmental factors; wherein the riding historical data comprises vehicle message data and battery message data; the cloud mileage model comprises a single equipment regression model, a multi-equipment regression model, a single equipment electric quantity integral model and a sample equipment electric quantity integral model. According to the method, cloud computing and vehicle end edge computing are combined, real-time data can be fully utilized, and the method can adapt to complex and changeable use environments, so that the accuracy of a residual endurance mileage prediction result is improved.

Description

Technical Field

[0001] The invention relates to the technical field of electric bicycles, and in particular to a method and system for predicting the remaining cruising range of a lead-acid battery of an electric bicycle. Background Art

[0002] In the actual application of electric bicycles, it is very important for users to accurately calculate the remaining range of lead-acid batteries. At present, the calculation of the range of electric bicycles mainly relies on simple mathematical formulas. Although this method has certain guiding significance in theory, it often has large errors in practical applications.

[0003] In addition, traditional machine learning methods are used to predict the remaining range of electric bicycles, such as by collecting historical battery discharge data and using decision tree and other algorithms to establish prediction models. On the one hand, these methods often rely on a large amount of historical data and cannot adapt to changes in real-time riding status and environmental changes, resulting in a decrease in the accuracy of the prediction results of the remaining range. On the other hand, the prediction of the remaining range of lead-acid batteries for electric bicycles involves a comprehensive consideration of multiple factors, such as battery temperature, battery aging, riding habits, etc. Traditional machine learning algorithms often need to build complex models when dealing with these factors, which not only increases the complexity of the calculation, but also reduces the generalization ability of the model. Summary of the invention

[0004] In view of the above problems and technical requirements, the inventors have proposed a method and system for predicting the remaining cruising range of a lead-acid battery of an electric bicycle, which can provide accurate prediction of the remaining cruising range and reduce the complexity of the model. The technical solution of the invention is as follows:

[0005] In a first aspect, the present application provides a method for predicting the remaining cruising range of a lead-acid battery of an electric bicycle, comprising the following steps:

[0006] In the cloud, the cloud mileage model is iteratively trained based on historical riding data to generate vehicle-side model parameters;

[0007] On the vehicle side, the vehicle side model generates a predicted value of the remaining cruising range based on the battery SOC and the vehicle side model parameters output by the cloud;

[0008] Combined with the real-time riding status and environmental factors during riding, the remaining mileage prediction value is adjusted in real time;

[0009] Among them, riding history data includes vehicle message data and battery message data; cloud mileage models include single device regression model, multi-device regression model, single device power integration model, and sample device power integration model.

[0010] The further technical solution is to iteratively train the cloud mileage model based on the riding history data to generate the vehicle-side model parameters, including:

[0011] Clean and correct the vehicle message data and battery message data obtained from various devices;

[0012] Using the corresponding vehicle message data and battery message data, the single device regression model, multi-device regression model, single device power integral model, and sample device power integral model are iteratively trained to obtain the prediction results of each model;

[0013] The prediction results of the four models are integrated and learned, and the integrated results are output as the vehicle-side model parameters;

[0014] Among them, the prediction result is the vehicle speed related to SOC and mileage.

[0015] Its further technical solution is to clean and correct the vehicle message data and battery message data obtained from various devices, including:

[0016] ETL is used to clean vehicle message data and battery message data;

[0017] The vehicle message data includes the remaining power SOC of the main and auxiliary batteries and the corresponding mileage. The battery message data includes the remaining power SOC of the main and auxiliary batteries. The abnormal values ​​are filtered out and the SOC is arranged in order.

[0018] Traverse each SOC, calculate the number of kilometers that can be ridden for every 1% of power consumed based on the current SOC and the SOC in the interval of the previous and next n SOCs and the corresponding mileage, and then infer the stable vehicle speed corresponding to the current SOC based on the relationship between vehicle speed, SOC and mileage, so as to correct the vehicle speed in the vehicle message data;

[0019] Here, n is an even number.

[0020] Its further technical solution is to integrate the prediction results of the four models and output the integrated results as the vehicle-side model parameters, including:

[0021] Filter out the outliers in the prediction results of the four models, calculate the mean of the remaining prediction results, and find the prediction result closest to the reference value from the remaining prediction results;

[0022] If the difference between the found prediction result and the mean exceeds the first threshold, the mean of the remaining prediction results is output as the final vehicle-side model parameter, otherwise the found prediction result is output as the final vehicle-side model parameter.

[0023] A further technical solution is that a method for obtaining a reference value includes:

[0024] The average riding speed of the current vehicle in the recent period is used as a reference value.

[0025] Its further technical solution is to adjust the remaining range prediction value in real time based on the real-time riding status and environmental factors during riding, including:

[0026] Determine the vehicle mode based on real-time riding status and environmental factors;

[0027] Multiply the remaining cruising range prediction value output by the vehicle-side model by the coefficient corresponding to the current mode to obtain the adjusted remaining cruising range prediction value.

[0028] A further technical solution is to determine the vehicle mode according to the real-time riding status and environmental factors, including:

[0029] Obtain the real-time riding status and environmental factors during riding. The real-time riding status includes the vehicle load and average vehicle speed. Environmental factors include the vehicle uphill and downhill angles.

[0030] When it is continuously detected within a certain period of time that the vehicle load is not less than the first weight threshold, or the average speed of the vehicle is not less than the first speed threshold, or the vehicle's uphill and downhill angle is greater than the first angle threshold, the vehicle is considered to be in power consumption mode;

[0031] When it is continuously detected within a certain period of time that the vehicle load is not greater than the second weight threshold, the vehicle average speed is not greater than the second speed threshold, and the vehicle uphill and downhill angles are less than the second angle threshold, the vehicle is considered to be in economic mode;

[0032] The rest of the cases are considered to be in normal mode;

[0033] Among them, the first weight threshold is greater than the second weight threshold, the first speed threshold is greater than the second speed threshold, and the first angle threshold is greater than the second angle threshold.

[0034] A further technical solution is that the coefficient corresponding to the normal mode is 1, the coefficient corresponding to the power consumption mode is less than 1, and the coefficient corresponding to the economic mode is greater than 1.

[0035] Its further technical solution is that the vehicle message data includes the remaining power SOC of the main and auxiliary batteries and the corresponding mileage, the number of charging cycles of the main and auxiliary batteries, the riding status, the total mileage, the vehicle speed, and the battery voltage at the vehicle end;

[0036] The battery message data includes the remaining power SOC of the main and auxiliary batteries, the main battery voltage, the main and auxiliary battery ambient temperature, and the number of main and auxiliary battery charging cycles.

[0037] In the second aspect, the present application also provides an electric bicycle lead-acid battery remaining range prediction system, including an interactive cloud server and a vehicle-side central controller, the cloud server and the vehicle-side central controller contain computer programs, and when the computer programs are executed by both ends, the steps of the method described in the first aspect are implemented.

[0038] The beneficial technical effects of the present invention are:

[0039] In the above method and system, the vehicle-side model generates a remaining cruising range prediction value based on the battery SOC and the vehicle-side model parameters output by the cloud, and makes real-time adjustments based on the real-time riding status and environmental factors. By combining cloud computing with vehicle-side edge computing, it can not only make full use of real-time data, but also adapt to complex and changeable usage environments, thereby improving the accuracy of the remaining cruising range prediction results.

[0040] The cloud can collect and process the riding data and battery data of electric bicycles in real time, and use the integrated learning algorithm to optimize the prediction results of multiple simple cloud mileage models. While reducing the complexity of the model, it can accurately predict the remaining range of the electric bicycle lead-acid battery and improve computing efficiency.

[0041] Accurate remaining range prediction can help users better plan their travel routes and avoid the inconvenience caused by low battery. In addition, users can also adjust their driving habits based on the prediction results to extend the battery life. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a calculation data flow diagram for predicting the remaining cruising range of an electric bicycle lead-acid battery provided in this application.

[0043] Figure 2 It is a schematic diagram of the cloud computing technology architecture provided by this application.

[0044] Figure 3 It is an overall schematic diagram of the remaining cruising range prediction method of the lead-acid battery of an electric bicycle provided in the present application.

[0045] Figure 4 This is the cloud mileage model calculation data flow diagram provided by this application.

[0046] Figure 5 It is the logic diagram of the integrated learning algorithm design provided in this application. DETAILED DESCRIPTION

[0047] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings.

[0048] An embodiment of the present application provides a system for predicting the remaining cruising range of a lead-acid battery of an electric bicycle, including an interactive cloud server and a vehicle-side central controller. The cloud server and the vehicle-side central controller contain computer programs. When the computer programs are executed by both ends, the steps of a method for predicting the remaining cruising range of a lead-acid battery of an electric bicycle provided in another embodiment are implemented. The operating principle of the system is first introduced below.

[0049] like Figure 1 As shown in the system calculation data flow diagram, the riding history data, including vehicle message data and battery message data, is obtained from the vehicle bus and sensors on the vehicle side (1). These data are collected through cloud data collection (2) and arrive at the data warehouse (3). Data cleaning ETL (Extract-Transform-Load) and other tasks are completed in the warehouse. In the library, the travel data of the previous day is calculated in the form of partitions through the fact table every day to prepare data for model training and calculation (4). In the process of model training and calculation (4), compared with traditional complex models, this application sets up multiple simple models, mainly including a single device regression model, a multi-device regression model, a single device power integration model and a sample device power integration model. A single device can be understood as a current electric bicycle, and multiple devices can be understood as a variety of electric bicycles provided by the manufacturer. The sample device can be understood as an electric bicycle that is not shipped out of the factory and is mostly used for testing. When the new device has no recent riding data during model training, the corresponding data can be matched from the sample device database. The prediction results of each model enter the data result (5). The use of integrated learning technology can improve the accuracy of each model prediction and obtain higher calculation accuracy. The integrated result is used as a model library file for online reasoning in the cloud and sent to the vehicle edge side (6). The edge side (6) further optimizes and adjusts the remaining cruising range prediction value based on the real-time riding status and environmental factors, thereby obtaining an accurate real-time remaining cruising range prediction value.

[0050] like Figure 2 As shown in the diagram of the cloud-side computing technology architecture, the cloud-based mileage model is based on the historical riding data for cloud computing. The historical riding data recorded by the Internet of Things is placed in the data warehouse through ETL data processing, and then the model is iteratively trained on the cloud to generate model results calculated by different models for different devices. This cloud-based technical solution completes: (1) daily device incremental data processing and feature engineering, completes cloud-based data correction and provides training data for cloud-based mileage model training; (2) the cloud-based mileage model completes iterative training every day; (3) the training model result library is stored and managed; (4) the remaining cruising range is calculated on the cloud; and (5) the vehicle-side model parameters are provided to the vehicle-side in the form of parameters.

[0051] Based on the same inventive concept, another embodiment of the present application provides a method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle. Figure 3As shown, it specifically includes: in the cloud, iteratively training the cloud mileage model based on the riding history data to generate the vehicle-side model parameters. On the vehicle side, the vehicle-side model generates the remaining cruising range prediction value based on the battery SOC and the vehicle-side model parameters output by the cloud, and adjusts the remaining cruising range prediction value in real time based on the real-time riding status and environmental factors during riding, so as to obtain a more accurate remaining cruising range prediction value that matches the current riding status.

[0052] Among them, the riding history data includes vehicle message data and battery message data. The vehicle message data includes the remaining power SOC of the main and auxiliary batteries and the corresponding mileage, the number of main and auxiliary battery charging cycles, riding status, total mileage, vehicle speed, and vehicle-end battery voltage. The battery message data includes the remaining power SOC of the main and auxiliary batteries, the main battery voltage, the main and auxiliary battery ambient temperature, and the number of main and auxiliary battery charging cycles. The cloud mileage model includes a single device regression model and a multi-device regression model built based on the existing machine learning algorithm, and also includes a single device power integration model and a sample device power integration model built based on the existing power integration method. Figure 4 As shown in the cloud-based mileage model calculation data flow diagram, the corresponding model is iteratively trained using vehicle message data and battery message data, and then the vehicle-side model parameters are output to the vehicle-side through integrated learning technology for the calculation of the remaining mileage prediction value.

[0053] Among them, the cloud mileage model is iteratively trained based on the riding history data to generate the vehicle-side model parameters, which specifically includes: cleaning and correcting the vehicle message data and battery message data obtained from various devices, and then using the corresponding vehicle message data and battery message data to iteratively train the single device regression model, multi-device regression model, single device power integration model, and sample device power integration model to obtain the prediction results of each model, which are the vehicle speed V related to SOC and mileage. 里程-soc值 Finally, the prediction results of the four models are integrated and learned, and the integrated results are output as the vehicle-side model parameters. The advantage of this step is that this application takes into account that the data volume of new equipment or a certain equipment in the near future does not support model training, so through the superposition calculation of multiple simple models and the integration of the results of each model, higher calculation accuracy can be obtained.

[0054] Due to the physical and chemical properties of lead-acid batteries, the SOC power fluctuation range is relatively large. Therefore, this application adopts a stable V calculation method when correcting the vehicle message data and battery message data obtained from various types of equipment. The data cleaning and correction process for the vehicle message data and battery message data obtained by each type of equipment is the same, including: using ETL to complete data cleaning for the vehicle message data and battery message data, filtering out abnormal values ​​in the remaining power SOC of the main and auxiliary batteries and the corresponding mileage in the message data, and arranging the SOC in order of size, such as SOC1, SOC2, SOC3, ... SOC m , SOC m+1 , SOC m+2 , SOC m+3 , SOC m+4 、…SOC n . Traverse each SOC, and compare the current SOC with the previous and next n SOCs (for example, the current SOC m+2 , n=2 time interval range is [SOC m ,SOC m+4 ]) and the corresponding mileage, the average SOC value in the range can be obtained as the current SOC, and the mileage that can be ridden for every 1% of power consumed can be calculated. Then, based on the relationship between vehicle speed, SOC and mileage (for example, cruising range = battery voltage × battery capacity × vehicle speed ÷ motor power), the stable vehicle speed V corresponding to the current SOC can be inferred. 里程-soc值 , to correct the vehicle speed in the vehicle message data. By traversing each SOC, the curve relationship between vehicle speed, SOC and mileage can be obtained. Let the horizontal coordinate of the curve be SOC, the vertical coordinate be vehicle speed, and the area enclosed by a point on the curve and the horizontal and vertical coordinates be the corresponding riding kilometers (i.e. mileage). Optionally, n is an even number, preferably 2, 4, 6 to divide the interval.

[0055] Combination Figure 5 As shown, the prediction results of the four models are integrated and learned, and the integrated results are output as the vehicle-side model parameters, which specifically include: taking the average riding speed of the current vehicle in the recent period (which can be obtained through the curve relationship) as a reference value, filtering out the abnormal values ​​in the prediction results of the four models, calculating the mean of the remaining prediction results, and finding the prediction result closest to the reference value from the remaining prediction results. If the difference between the found prediction result and the mean exceeds the first threshold (such as 0.3), the mean of the remaining prediction results is output as the final vehicle-side model parameter, otherwise the found prediction result is output as the final vehicle-side model parameter.

[0056] The training method of the vehicle-side model is the same as that of the cloud-side mileage model. The vehicle-side model stores the relationship between vehicle speed, SOC and mileage. Therefore, according to the battery SOC and the vehicle-side model parameter V 里程-soc值 The remaining mileage prediction value can be calculated. Combined with the real-time riding status and environmental factors during riding, the remaining mileage prediction value is adjusted in real time, including: determining the vehicle mode according to the real-time riding status and environmental factors, where the real-time riding status includes the vehicle load and the average speed of the vehicle, and the environmental factors include the vehicle uphill and downhill angles. According to the above factors, the vehicle mode can be divided into power consumption, normal, and economic mode. Multiply the remaining mileage prediction value output by the vehicle-side model by the coefficient corresponding to the current mode to obtain the adjusted remaining mileage prediction value.

[0057] Among them, the mode of the vehicle is determined according to the real-time riding state and environmental factors, specifically including: obtaining the real-time riding state and environmental factors during riding, when the vehicle load is continuously detected to be not less than the first weight threshold (such as 90kg) for a certain period of time, or the average speed of the vehicle is not less than the first speed threshold (such as 25km / h), or the vehicle uphill and downhill angle is greater than the first angle threshold (such as 5 degrees), the vehicle is considered to be in power consumption mode. When the vehicle load is continuously detected to be not greater than the second weight threshold (such as 60kg) for a certain period of time, and the average speed of the vehicle is not greater than the second speed threshold (such as 15km / h), and the vehicle uphill and downhill angle is less than the second angle threshold (such as 2 degrees, it can be assumed that the vehicle has no continuous uphill and downhill), the vehicle is considered to be in economic mode. In other cases, it is considered to be in normal mode, the coefficient corresponding to the normal mode is 1, the coefficient corresponding to the power consumption mode is less than 1, and the power consumption mode is 0.25 less than normal riding according to actual statistics, so the coefficient corresponding to the power consumption mode is 0.75, and the coefficient corresponding to the economic mode is greater than 1, which can be set to 1.1.

[0058] Combination Figure 1 , Figure 4 As shown, the method for predicting the remaining cruising range of a lead-acid battery of an electric bicycle corresponds to the process of calculating the data flow of the system, obtaining the vehicle message data and battery message data of the ride, and storing the data in the cloud. The cloud performs data ETL processing and feature engineering to complete cloud data correction and training data preparation. Through the integrated learning algorithm solution, the cloud mileage model training is completed to generate the vehicle-side model parameters. At this time, the process of autonomous learning in the cloud to obtain the training parameters is completed. Then the model parameter strategy is output to the vehicle side, and the remaining cruising range prediction value is calculated in real time in combination with the real-time riding status and environmental factors during the riding process. At the same time, after the cloud data correction, the cloud performs state correction work for the accurate battery SOC / SOH (this part is not the focus of this application, so it will not be introduced in detail), obtains the change curve of the battery SOC in each stage, and improves and supplements the calculation of the relationship between SOC and mileage in the real-time process.

[0059] The above is only a preferred embodiment of the present application, and the present invention is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present invention should be considered to be included in the protection scope of the present invention.

Claims

1. A method for predicting the remaining cruising range of a lead-acid battery of an electric bicycle, characterized in that: The method comprises: In the cloud, the cloud mileage model is iteratively trained based on historical riding data to generate vehicle-side model parameters; On the vehicle side, the vehicle side model generates a predicted value of the remaining cruising range based on the battery SOC and the vehicle side model parameters output by the cloud; In combination with the real-time riding status and environmental factors during riding, the remaining cruising range prediction value is adjusted in real time; Among them, the riding history data includes vehicle message data and battery message data; the cloud mileage model includes a single device regression model, a multi-device regression model, a single device power integration model, and a sample device power integration model.

2. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 1, characterized in that: The iterative training of the cloud mileage model based on the riding history data to generate the vehicle-side model parameters includes: Clean and correct the vehicle message data and battery message data obtained from various devices; Using corresponding vehicle message data and battery message data, iteratively train the single device regression model, the multi-device regression model, the single device power integral model, and the sample device power integral model to obtain prediction results for each model; Performing integrated learning on the prediction results of the four models, and outputting the integrated results as the vehicle-side model parameters; The prediction result is the vehicle speed related to the SOC and the mileage.

3. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 2, characterized in that: The data cleaning and correction of vehicle message data and battery message data obtained from various devices include: ETL is used to clean the vehicle message data and battery message data; The vehicle message data includes the remaining power SOC of the main and auxiliary batteries and the corresponding mileage, and the battery message data includes the remaining power SOC of the main and auxiliary batteries, and the abnormal values ​​are filtered out and the SOC is arranged in order; Traversing each SOC, calculating the number of kilometers that can be ridden for every 1% of power consumed based on the current SOC and the SOCs in the interval of the previous and next n SOCs and the corresponding mileage, and deducing the stable vehicle speed corresponding to the current SOC based on the relationship between the vehicle speed, SOC and the mileage, so as to correct the vehicle speed in the vehicle message data; Here, n is an even number.

4. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 2, characterized in that: The prediction results of the four models are integrated and learned, and the integrated results are output as the vehicle-side model parameters, including: Filter out the outliers in the prediction results of the four models, calculate the mean of the remaining prediction results, and find the prediction result closest to the reference value from the remaining prediction results; If the difference between the found prediction result and the mean exceeds a first threshold, the mean of the remaining prediction results is output as the final vehicle-side model parameter; otherwise, the found prediction result is output as the final vehicle-side model parameter.

5. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 4, characterized in that: The method for obtaining the reference value includes: The average riding speed of the current vehicle in the recent period is used as the reference value.

6. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 1, characterized in that: In combination with the real-time riding status and environmental factors during riding, the remaining cruising range prediction value is adjusted in real time, including: Determine the vehicle mode based on real-time riding status and environmental factors; The remaining cruising range prediction value output by the vehicle-side model is multiplied by the coefficient corresponding to the current mode to obtain an adjusted remaining cruising range prediction value.

7. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 6, characterized in that: Determining the mode of the vehicle according to the real-time riding status and environmental factors includes: Acquire the real-time riding status and environmental factors during riding, wherein the real-time riding status includes the vehicle load and the average speed of the vehicle, and the environmental factors include the vehicle uphill and downhill angles; When it is continuously detected within a certain period of time that the vehicle load is not less than a first weight threshold, or the average speed of the vehicle is not less than a first speed threshold, or the uphill and downhill angles of the vehicle are greater than a first angle threshold, the vehicle is considered to be in power consumption mode; When it is continuously detected within a certain period that the vehicle load is not greater than the second weight threshold, the average speed of the vehicle is not greater than the second speed threshold, and the uphill and downhill angles of the vehicle are less than the second angle threshold, the vehicle is considered to be in the economic mode; The rest of the cases are considered to be in normal mode; Among them, the first weight threshold is greater than the second weight threshold, the first speed threshold is greater than the second speed threshold, and the first angle threshold is greater than the second angle threshold.

8. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 7, characterized in that: The coefficient corresponding to the normal mode is 1, the coefficient corresponding to the power consumption mode is less than 1, and the coefficient corresponding to the economic mode is greater than 1.

9. The method for predicting the remaining cruising range of a lead-acid battery for an electric bicycle according to claim 1, characterized in that: The vehicle message data includes the remaining power SOC of the main and auxiliary batteries and the corresponding mileage, the number of charging cycles of the main and auxiliary batteries, the riding status, the total mileage, the vehicle speed, and the battery voltage at the vehicle end; The battery message data includes the remaining power SOC of the main and auxiliary batteries, the voltage of the main battery, the ambient temperature of the main and auxiliary batteries, and the number of charging cycles of the main and auxiliary batteries.

10. A system for predicting the remaining cruising range of a lead-acid battery for an electric bicycle, characterized in that: It comprises an interactive cloud server and a vehicle-side central controller, wherein the cloud server and the vehicle-side central controller contain computer programs, and when the computer programs are executed by both ends, the steps of the method described in any one of claims 1 to 9 are implemented.

Images