2.1

New Energy Vehicles

According to the types of sensors required, new energy vehicle operation monitoring systems can be divided into three categories. Based on the virtual interface technology, real-time tracking of the vehicle’s driving status is realized, so as to obtain a good driving experience and comfort. The monitoring system can be used in a variety of vehicle models as an auxiliary means to meet the user’s requirements for the performance parameters and environmental requirements of various complex components such as batteries and electric vehicles. It is also suitable for obtaining effective protection measures by detecting the working conditions of the vehicle and analyzing and drawing conclusions under the actual operating conditions of the vehicle. Based on artificial neural network technology, real-time collection of vehicle driving status information is realized. The hardware equipment part, the core processor, sensor and other devices of the whole new energy vehicle are all controlled by MCU. Among them, the temperature controller is one of the most critical components of the entire control system, which plays a vital role in the performance of the entire vehicle. The battery charging module can directly provide the required energy and a short-lived power source for the electric passenger car. The motor drive circuit is used to provide power to the motor and realize the speed and direction adjustment and speed control functions [1].

The operation monitoring of new energy vehicles is a complex and huge process, which involves many aspects, such as: battery voltage, vehicle speed and other parameters. In the monitoring system, intelligent control technology is critical. With the continuous improvement of computer software and hardware development capabilities, network communication and sensor equipment performance, and automotive electronic control systems, related requirements continue to increase. At the same time, because traditional traffic management methods can no longer adapt to the current development, there is a need for a real-time monitoring of the current environmental conditions of vehicles. And can automatically adjust the vehicle speed according to the driving state of the driver to achieve the purpose of energy saving, emission reduction and reducing traffic accidents [4].

2.2

Design Process of Convolutional Neural Network

The convolutional neural network starts at the exit node when propagating backward. First get the standard node through the output layer. During this time, this is equivalent to getting the deviation value of each data point. Then start back propagation from this point. Backpropagation is a fully connected network layer from the beginning. This is the third folded layer, directly connected to the fully connected layer. Because the third layer of convolution does not use padding, the feature size of the standard node is smaller than before the third layer of convolution. An operation must be added to make the size of the feature after backpropagation equal to the size before the third convolutional layer. Further forward propagation forms the second layer of pooling. When the pooling layer propagates backwards, the problem of the feature dimension becoming smaller due to the forward pooling process should be corrected [6].

The loss function is used to estimate the degree of inconsistency between the predicted value of the neural network model and the actual value. It is a function with non-negative real values. The back propagation part of the neural network learning process is also based on the loss function, and the calculated loss value is changed for the weighting matrix W and the biasing matrix y. The smaller the loss value of the loss function, the closer to the expected optimal model. The formula of the quadratic cost function is:

Among them, b represents the output value of the network, x represents the output value, a represents the input sample, D represents the cost value, and m represents the total number of samples. When there is one sample, the quadratic cost function is:

Currently, artificial neural networks are usually trained using backpropagation algorithms. Learning artificial neural networks is to reduce D as much as possible, and propagate back to D in this way to change the parameters.

In the process of monitoring the operation of new energy vehicles, the convolutional neural network is applied to the control system of new energy vehicles. It mainly responds to fault signals and environmental disturbances generated during electric driving. In the traditional sense, the system itself has no reliability problems. The reliability of the new energy vehicle operation monitoring system based on convolutional neural network mainly refers to whether the problems existing in the actual production and life process can be effectively solved. From design to testing to application, each link should have corresponding personnel responsible and implement the relevant responsibility system to ensure the normal and stable operation of the entire control system. It is also a crucial part for fault diagnosis.

The reliability of the new energy vehicle operation monitoring system based on convolutional neural network is mainly reflected in the following aspects: The hardware platform is reliable. Corresponding testing must be done before circuit and software development. Including the analysis and evaluation of performance parameters and abnormal conditions of various sensor signals and fault data through the simulator or video recorder. For different types of fault samples and input and output signals, accurate judgments must be made and corresponding treatment methods must be adopted to reduce the system failure rate. Network reliability requires that a hybrid mode is used to store and transmit data samples. The fault diagnosis and maintenance can use artificial neural network methods to realize the functions of fault diagnosis and maintenance work automation. When designing the network topology, it is necessary to ensure the reliability of the system and ensure the safety and reliability of the entire system.

2.3

New Energy Vehicle Intelligent Power Monitoring System Architecture

The power of a new energy vehicle includes a battery and a hydrogen reactor. The structure of the power system is shown in Figure 1:

Figure 1:

New Energy Vehicle Power System Structure

The compressed air system or hydrogen system provides compressed air and hydrogen to the chimney. Hydrogen and compressed air react in the chimney to produce electricity. The intake speed can be controlled by adjusting the duty cycle and pressure of the air compressor and hydrogen switch. It can be controlled by adjusting the appropriate valve.

The intelligent power control system is mainly connected to the car electric system through the CAN bus, and exchanges information with the car electric system power control device driver through the CANOPEN communication module of the car control terminal. The information collected by the vehicle power supply unit is transmitted to the vehicle control terminal through the CAN bus, and the data display module displays the relevant information on the screen. The control module of the device can convert some control signals through the CANOPEN communication module, and then transmit it to the vehicle power supply through the CAN bus to achieve the purpose of remote control.

3.1

Construction of the Test Platform

Before starting the formal test, first build a system test platform. Due to various limitations of the experimental conditions, it is impossible to use vehicle control, engine control and battery management systems for comprehensive testing. This article uses an independent battery management system. The development of the laboratory is to supplement the system verification experience. The system test platform is mainly composed of a smart phone, a vehicle control unit (VCU) using F28M35 as a single-chip microcomputer, and a battery management system (BMS). The vehicle control based on the F28M35 multi-core microcontroller is the control center of the entire system. The battery management system is the object of the whole system monitoring. The smartphone is the monitoring center of the entire system.

3.2

Function Test

3.3

Experimental Process

The GPU outputs the updated information in each iteration, continuously iteratively calculates and updates the model, and stores the current weights in the backup directory every 100 iterations. In each iteration, we can see that the IOU value is constantly changing, and the average value of loss is also constantly decreasing. When the loss value reaches a certain standard or the number of iterations reaches a predetermined value, the learning process can be terminated.

In the experimental setup of this article, 100 iterations all lasted 5 minutes, and a total of 500 iterations were carried out for a total of 25 minutes. For better quality GPUs, this value will be lower.

4.1

Changes in Loss and IOU Values in Different Iterations

During the first 40 iterations, the loss value decreased exponentially, and after 100 iterations, the loss value dropped to about 0.17. However, the IOU value has shown an upward trend. The details are shown in Table 1:

Table 1.

Change of Loss and IOU Value for the Number of Iterations within 100

As shown in Figure 2, it can be seen from the two parameter change graphs that the IOU parameter gradually increases as the number of iterations increases, and finally reaches a certain stable position after 400 iterations and remains unchanged. The loss data of 100-400 iterations has been declining, but the drop value is already very small. After 500 iterations, the loss value is basically stable at about 0.01.

Figure 2:

Changes in Loss and IOU Values in Different Iterations

4.2

Alarm lamp Recognition Result

Even after 500 iterations, a curve will still be drawn based on the IOU parameter and loss value. In the first 40 iterations, the loss value decreased exponentially, and after 100 iterations, the loss value dropped to approximately 0.10. In the process of 100 to 500 iterations, the loss data is continuously reduced, but the reduction value is already small. In the first 100 iterations, IOU first rises, then falls, and then rises. The value has always been an upward trend in 100-500 iterations. The specific changes are shown in Table 2:

Table 2.

Alarm Lamp Recognition Loss and IOU Results

These data results show that the convolutional neural network has a good recognition effect and recognition speed, and can also meet the actual application requirements.