CN116700296A

CN116700296A - Intelligent planning method and system for autonomous charging unmanned electric vehicle based on deep learning

Info

Publication number: CN116700296A
Application number: CN202310921701.8A
Authority: CN
Inventors: 杜咏峻
Original assignee: Individual
Current assignee: Individual
Priority date: 2023-07-26
Filing date: 2023-07-26
Publication date: 2023-09-05

Abstract

The invention discloses an intelligent planning method of an autonomous charging unmanned electric vehicle based on deep learning, which relates to the field of intelligent transportation and comprises the steps of acquiring basic data of a stylized unmanned electric vehicle and preprocessing; establishing a vehicle comprehensive state evaluation model based on the preprocessed data; establishing an environment recognition model based on the vehicle comprehensive state evaluation model; constructing a charging demand prediction model and an environment constraint recognition model, and integrating multi-source information planning to generate an optimal return charging path by adopting an ant colony algorithm and a deep reinforcement learning method; the vehicle is driven along the optimal return charging path through an automatic driving technology, and is charged after reaching a charging station; and monitoring the vehicle state and the model prediction result through a remote monitoring system. According to the method, the ant colony algorithm and the deep reinforcement learning are combined to conduct the optimal planning of the charging path, so that the charging constraint is considered, the closed-loop control of the vehicle energy state is realized, and the path generation is more intelligent.

Description

Intelligent planning method and system for autonomous charging unmanned electric vehicle based on deep learning

Technical Field

The invention relates to the field of intelligent transportation, in particular to an intelligent planning method and system for an autonomous charging unmanned electric vehicle based on deep learning.

Background

With the development of unmanned electric vehicle technology, the application of the unmanned electric vehicle in a fixed field is mature, and simple and repeated operation tasks can be realized. However, such unmanned electric vehicles generally employ fixed routes and charging strategies such that the duration and range of autonomous operation is still very limited. Under a complex environment, a fixed charging strategy is difficult to adapt to various environmental changes and driving requirements, so that the unmanned electric vehicle is difficult to complete a complex mission.

The core problem faced by the current unmanned electric vehicle is short endurance mileage. Most unmanned electric vehicles rely on simple fixed charging strategies, which are difficult to cope with complex and changeable environmental conditions in actual operation, such as complex terrains, road conditions, vehicle state changes, and the like. This makes it difficult for existing unmanned electric vehicles to accommodate the need to perform complex tasks. To achieve continuous autonomous operation of unmanned electric vehicles, an intelligent system is required to actively determine optimal charging opportunities and charging stations based on vehicle conditions and environmental changes. Therefore, the unmanned electric vehicle can realize active and intelligent charging behavior planning, and gets rid of the limitation of a fixed charging mode, so that the application range and the task complexity of the unmanned electric vehicle are enlarged.

Disclosure of Invention

This section is intended to outline some aspects of embodiments of the application and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section as well as in the description of the application and in the title of the application, which may not be used to limit the scope of the application.

The application is provided in view of the problems of short endurance mileage and low energy utilization efficiency of the above or existing unmanned electric vehicle.

Therefore, the application aims to solve the problems of realizing the prediction of the charging demand and the optimization of the charging plan of the unmanned electric vehicle so as to prolong the endurance mileage and improve the energy utilization efficiency.

In order to solve the technical problems, the application provides the following technical scheme:

in a first aspect, an embodiment of the present application provides an intelligent planning method for an autonomous charging unmanned electric vehicle based on deep learning, which includes acquiring basic data of a programmed unmanned electric vehicle, and preprocessing the acquired basic data of the programmed unmanned electric vehicle; establishing a vehicle comprehensive state evaluation model based on the preprocessed data; establishing an environment recognition model based on the vehicle comprehensive state evaluation model, and outputting environment condition information; constructing a charging demand prediction model and an environment constraint recognition model, and integrating multi-source information planning to generate an optimal return charging path by adopting an ant colony algorithm and a deep reinforcement learning method; the vehicle is driven along the optimal return charging path through an automatic driving technology, and is charged after reaching a charging station; and monitoring the vehicle state and the model prediction result through a remote monitoring system, and starting an abnormal processing flow in time after the abnormal situation is found.

As a preferable scheme of the intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning, the invention comprises the following steps: the vehicle comprehensive state evaluation model is established based on the preprocessed data, and comprises the following steps: respectively establishing a battery health evaluation model and a motor state evaluation model based on the preprocessed data; taking the battery state evaluation model and the motor state evaluation model as sub-models for input, establishing a vehicle comprehensive state evaluation model based on deep learning, and training the model; outputting a comprehensive state evaluation result of the vehicle, and realizing monitoring and evaluation of the states of the battery and the motor; the battery state evaluation model comprises good battery state, initial decline of the battery state and serious decline of the battery state, when the battery state is good, the internal resistance of the battery is low, the charge-discharge curve is stable, the capacity decline of the battery is slow, and the charge-discharge response time of the battery is short; when the state of health of the battery begins to decline, the internal resistance of the battery begins to slowly increase, the fluctuation of a charge-discharge curve is not obvious, the capacity decay rate of the battery begins to increase, and the charge-discharge response time of the battery is prolonged; when the state of health of the battery is seriously declined, the internal resistance of the battery is greatly increased, the fluctuation of a charge-discharge curve is increased, the capacity of the battery is rapidly attenuated, and the charge-discharge response time of the battery is obviously prolonged; the motor state in the motor state evaluation model comprises a motor normal state, a motor rotor fault state and a motor insulation fault state, under the motor normal state, the rotating speed is kept in a stable range, the current fluctuation is small, the temperature is controlled in an allowable range, and the vibration amplitude is low; in the motor rotor fault state, the rotation speed fluctuation is increased, rapid rise or fall occurs, the current fluctuation is severe, the peak value is intermittently and greatly increased, the temperature rise is slow, the vibration overall is increased, and the vibration of each bearing is obviously different from that of the normal bearing; in the insulation fault state of the motor, the rotating speed is normal, the current has no obvious fluctuation, the temperature rises rapidly, and the vibration is slightly increased.

As a preferable scheme of the intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning, the invention comprises the following steps: the method for establishing the environment recognition model based on the vehicle comprehensive state evaluation model comprises the following steps of: establishing an environment recognition model based on a multi-mode data fusion and deep learning method; taking the data obtained by fusing the output of the vehicle state evaluation model and the multi-mode data as the input of the environment recognition model, and enabling the vehicle state evaluation model to learn and distinguish different environment conditions through the supervised learning and training model; predicting the test set data by using the trained environment recognition model, calculating the evaluation index of environment recognition, and verifying the model effect; applying an environment recognition model on the newly acquired real-time data, outputting the environment state, and providing a corresponding adaptive strategy for vehicle operation according to the environment state; the corresponding adaptive strategy for proposing the vehicle running according to the environment state comprises the steps of dynamically adjusting the vehicle speed according to the speed limit sign of the urban road section in the urban environment, planning an optimal navigation path according to the real-time traffic condition, avoiding the traffic jam road section, and enhancing the monitoring capability of pedestrians and non-motor vehicles to cope with the emergency in the urban environment; under suburban environment, the cruising speed of the vehicle is optimized and regulated in real time according to the bending degree of the road, the speed of the vehicle is intelligently regulated according to the distance between the front vehicle and the rear vehicle, and the change of the road quality and the gradient is monitored so as to make adjustment for a vehicle system in advance; under a high-speed environment, selecting the optimal lane changing time and position according to a preset navigation route and traffic flow, judging the far-near light switching time according to the distance and speed of a front vehicle, monitoring whether the current road traffic flow direction is deviated, and preventing from entering other road sections by mistake; under the parking environment, the vehicle enters a parking space and completes parking operation, whether moving barriers exist around the vehicle or not is continuously monitored, and when the vehicle door is unlocked/locked is intelligently judged.

As a preferable scheme of the intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning, the invention comprises the following steps: enabling the learning and training of models by supervised learning and differentiation of different environmental conditions comprises the steps of: taking the characteristics obtained by fusing the output of the vehicle state evaluation model and the multi-mode data as the input of the environment recognition model; the multi-mode data under different environmental conditions and the corresponding environmental classifications are used as label pairing to form a training data set with labels; building a structure of an environment recognition model based on a deep learning method; training the environment recognition model by using a training data set with a label; evaluating the trained environment recognition model through verification set data; and adjusting the model structure and parameters according to the verification result, and retraining until a preset performance index is reached.

As a preferable scheme of the intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning, the invention comprises the following steps: the method for generating the optimal return charging path by integrating the multisource information planning comprises the following steps of: establishing a charging demand prediction model, and predicting the electric quantity required by return; according to the length of the return route, calculating the predicted total electricity consumption in the return process by combining the parameters of electricity consumption per unit mileage; comparing the current state of charge of the battery with the predicted electric quantity required by the return voyage and the total electric quantity consumption, and judging whether the current electric quantity can meet the return voyage requirement; when the electric quantity meets the requirement, calculating an optimal return charging path with minimum time and energy consumption as an ideal path; continuously acquiring real-time positioning in the return voyage, monitoring the deviation between the actual route and the ideal route, and triggering a correction model to enable the actual route to be in line with the ideal route again if the deviation exceeds a preset threshold value; in path deviation monitoring and correction, identifying environmental conditions and constraint information, and dynamically adjusting a correction path according to environmental influence; and outputting an optimal return charge path meeting the charge and environment constraints.

As a preferable scheme of the intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning, the invention comprises the following steps: the method for driving the vehicle along the optimal return charging path planned in advance by the automatic driving technology comprises the following steps: acquiring the current position of a vehicle, and calling an optimal return charging path; controlling the vehicle to return according to the optimal return charging path by using an automatic driving control algorithm; when a vehicle enters a charging station, calculating charging strategies of different electricity price time periods by using a charging model; connecting charging equipment and starting charging, and monitoring and dynamically adjusting a charging strategy in real time in the charging process; evaluating the charging effect and updating the battery parameters; and planning a next-round on-duty plan according to the new-round patrol and overhaul task demands and the environmental assessment result, and entering a next preset unmanned supervision period.

As a preferable scheme of the intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning, the invention comprises the following steps: the charging strategy for calculating different electricity price time periods by using the charging model comprises the following steps: an electric automobile charging model is constructed, wherein the charging process is represented as follows:

Wherein, the soc is the battery power, and the soc ₀ For initial charge of battery, soc _up At the upper limit of battery capacity, p _i For the output power of the charging pile eta _i For charging efficiency, P _i And the power grid output power is obtained, and t is the charging time.

The process of discharging the electric automobile to the power grid is represented as follows:

wherein, the soc is the battery power, and the soc ₀ For initial charge of battery, soc _low Is the lowest capacity limit of the battery, p ₀ For discharging power of battery, eta ₀ For discharging inverter efficiency, P ₀ The power is the AC power output by the inverter, and t is the charging time.

The specific formula aimed at minimizing the electricity charge is as follows:

wherein C is _j For the electricity price of each period, P _i /η _i For grid side power, t _j Is shown inWorking time of charging pile at each time interval, soc _set For program demand, soc ₀ For initial charge of battery, p _i For charging pile output power, soc _low Is the lowest capacity limit of the battery, soc _up Is the upper limit of the battery capacity, t is the charging time, t ₀ For start-up time, t _end To end time, t _k For the working time of the charging pile in the period j=k, namely the working time of the charging pile in the last period, t ₁ And j=1 time periods are the working time of the charging pile, namely the working time of the charging pile in the first time period, and k is the total time period number of the planning process.

The energy transfer through different electricity price time periods earns a gap, and the corresponding charge and discharge optimization model is as follows:

Wherein delta is a charge-discharge sign, C _j For the electricity price of each period, P _i /η _i For grid side power, t _j The working time of the charging pile in each period is represented, the soc is the battery power, and the soc _set For program demand, soc ₀ For initial charge of battery, p ₀ For discharging power of battery, eta ₀ For discharging inverter efficiency, P ₀ For ac power output by inverter, soc _low Is the lowest capacity limit of the battery, soc _up Is the upper limit of the battery capacity, t is the charging time, t ₀ For start-up time, t _end To end time, t _k For the working time of the charging pile in the period j=k, namely the working time of the charging pile in the last period, t ₁ And j=1 time periods are the working time of the charging pile, namely the working time of the charging pile in the first time period, and k is the total time period number of the planning process.

In a second aspect, the embodiment of the invention provides an intelligent planning system for an autonomous charging unmanned electric vehicle based on deep learning, which comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring basic data of a stylized unmanned electric vehicle and preprocessing the data; the vehicle state evaluation module is used for establishing a vehicle comprehensive state evaluation model and monitoring and evaluating the states of the battery and the motor; the environment recognition module is used for building an environment recognition model, outputting environment condition information and providing an adaptive strategy for vehicle operation; the path planning module is used for predicting the charging requirement, planning the optimal return charging path and performing a path correction function; and the automatic driving module is used for returning and charging according to the planned path by utilizing an automatic driving technology and calculating a charging strategy.

In a third aspect, embodiments of the present invention provide a computer apparatus comprising a memory and a processor, the memory storing a computer program, wherein: the computer program instructions, when executed by a processor, implement the steps of the deep learning-based intelligent planning method for the autonomous charging unmanned electric vehicle according to the first aspect of the invention.

In a fourth aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: the computer program instructions, when executed by a processor, implement the steps of the deep learning-based intelligent planning method for the autonomous charging unmanned electric vehicle according to the first aspect of the invention.

The invention has the beneficial effects that: according to the invention, by constructing a vehicle state evaluation and environment recognition model driven by deep learning, adopting an integrated and optimized charging path planning and automatic driving return technology, and establishing a remote monitoring mechanism, the electric vehicle can realize closed-loop autonomous charging management, and has the capabilities of state accurate prediction, environment perception adaptation, efficient charging planning, automatic return charging and remote exception handling, and the intelligent level and unmanned operation reliability of the electric vehicle are improved obviously as a whole.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. Wherein:

fig. 1 is a flow chart of an intelligent planning method of an autonomous charging unmanned electric vehicle based on deep learning.

Fig. 2 is an internal structural diagram of a computer device of an intelligent planning method of an autonomous charging unmanned electric vehicle based on deep learning.

Detailed Description

In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.

Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

Example 1

Referring to fig. 1 and 2, for a first embodiment of the present invention, there is provided an intelligent planning method for an autonomous charging unmanned electric vehicle based on deep learning, including,

s1: and acquiring the basic data of the programmed unmanned electric vehicle, and preprocessing the acquired basic data of the programmed unmanned electric vehicle.

Specifically, the programmed unmanned electric vehicle basic data comprise real-time data, electric vehicle programming information, site information and driving data, wherein the real-time data comprise the current State of charge (SOC), electric vehicle real-time power information and electric vehicle position information of the electric vehicle, and the information can be obtained in real time through a sensor and data acquisition equipment; the electric automobile programming information comprises current on-duty period information (period starting and ending time) and optimal charging position information when the on-duty period is ended, wherein the on-duty period refers to that the vehicle executes specific tasks such as patrol, inspection, patrol and the like in a specific period of time, and a system administrator or dispatcher plans and sets the information in advance and issues and manages the information through the system; the method comprises the steps of site information, site length, site width, basic path information and charging pile distribution position information, wherein the basic path information refers to a basic path of a vehicle running in the site and comprises a straight path, a curve path, a circular path and the like, and the charging pile distribution position information refers to a specific position of a charging pile in the site and can be identified and managed through a map and a mark; the driving data includes information such as driving route, environmental condition, vehicle state, energy consumption, etc., and can be collected by GPS and vehicle-mounted equipment.

Further, preprocessing the obtained basic data of the stylized unmanned electric vehicle comprises data flushing (deleting missing data, repairing error data, converting data format and merging data), data denoising (removing noise and abnormal values in the data through filtering, smoothing or interpolation and other technologies) and data normalization processing.

S2: and establishing a vehicle comprehensive state evaluation model based on the preprocessed data.

Specifically, the method comprises the following steps:

s2.1: and respectively establishing a battery health evaluation model and a motor state evaluation model based on the preprocessed data.

The method specifically comprises the following steps:

s2.1.1: and constructing a battery health assessment model, designing uniform input and output interfaces, and returning a battery health state prediction result.

Specifically, the method comprises the following steps:

s2.1.1.1: and extracting relevant characteristic parameters of the battery according to the preprocessed data.

Specifically, the extracted battery-related characteristic parameters include battery voltage, current, temperature, and internal resistance.

S2.1.1.2: and analyzing the change rule of the characteristic parameters under different battery health states.

Preferably, the method comprises the steps of good battery health state, initial decline of the battery health state and serious decline of the battery health state, when the battery health state is good, the internal resistance of the battery is low, the charge-discharge curve is stable, the capacity decline of the battery is slow, and the charge-discharge response time of the battery is short; when the state of health of the battery begins to decline, the internal resistance of the battery begins to slowly increase, the fluctuation of a charge-discharge curve is not obvious, the capacity decay rate of the battery begins to increase, and the charge-discharge response time of the battery is prolonged; when the state of health of the battery is severely declined, the internal resistance of the battery is greatly increased, the fluctuation of a charge-discharge curve is increased, the capacity of the battery is rapidly attenuated, and the charge-discharge response time of the battery is obviously prolonged.

S2.1.1.3: key feature parameters that reflect the state of health of the battery are selected.

Specifically, through the analysis step S2.1.1.2, the change rule of the characteristic parameters under different battery health states is already defined; according to the selection principle, we need to select key characteristic parameters which can reflect the state of health of the battery, and these parameters need to meet the following conditions: first, the trend of the parameter change must be significant and highly correlated with the battery state of health; secondly, the parameters should be easy to collect, and the data acquisition is convenient; in addition, the selected parameters have a great influence on the practical application and safety performance of the battery; by taking these factors into account, we can choose the most representative characteristic parameters to accurately reflect the state of health of the battery.

Comprehensively considering, the internal resistance, capacity, charging curve and the like of the battery are selected as key characteristic parameters.

S2.1.1.4: and constructing a battery health state evaluation index.

Specifically, the principle of constructing the battery state of health evaluation index comprises that the index is highly sensitive and can reflect the tiny change of the battery state of health; the index is convenient to calculate and is based on easily-obtained characteristic parameters; the index should be quantifiable and capable of being represented numerically.

And comprehensively considering, and finally selecting the increase rate of the battery charge and discharge internal resistance, the attenuation rate of the battery charge and discharge capacity and the deviation degree of the battery charge and discharge curve as evaluation indexes.

S2.1.1.5: and establishing a regression model between the battery health state evaluation index and the characteristic parameters.

Specifically, characteristic parameters of the battery voltage, current, temperature and internal resistance are extracted according to step S2.1.1.1, the relation between the characteristic parameters and the battery health state is analyzed according to step S2.1.1.2, an internal resistance, capacity and charge-discharge curve are selected as key characteristic parameters according to step S2.1.1.3, and an internal resistance increase rate, a capacity decay rate and a curve deviation degree are selected as evaluation indexes according to step S2.1.1.4.

Further, collecting battery test data comprising the characteristic parameters and indexes; establishing a regression model between the internal resistance increasing rate and the internal resistance value, a regression model between the capacity attenuation rate and the capacity and a regression model between the curve deviation degree and the charge-discharge curve through a linear regression algorithm; evaluating error indexes of each model, training the regression model by using a training data set, and optimizing the model by adjusting parameters; repeating the above process until an accurate relation model of the evaluation index and the characteristic parameter is established; from these sub-models, a final battery health assessment model is constructed.

S2.1.1.6: and returning a battery health state prediction result.

S2.1.2: and establishing a motor state evaluation model, designing unified input and output interfaces, and returning a motor fault state prediction result.

Specifically, the method comprises the following steps:

s2.1.2.1: and extracting relevant characteristic parameters of the motor according to the preprocessed data.

Specifically, it is proposed that the motor-related characteristic parameters include motor rotation speed, current, temperature, and vibration.

S2.1.2.2: and analyzing the change rule of the characteristic parameters under different motor fault states.

Preferably, the method comprises a motor normal state, a motor rotor fault state and a motor insulation fault state, wherein in the motor normal state, the rotating speed is kept in a stable range, the current fluctuation is small, the temperature is controlled in an allowable range, and the vibration amplitude is low; in the motor rotor fault state, the rotation speed fluctuation is increased, rapid rise or fall occurs, the current fluctuation is severe, the peak value is intermittently and greatly increased, the temperature rise is slow, the vibration overall is increased, and the vibration of each bearing is obviously different from that of the normal bearing; in the insulation fault state of the motor, the rotating speed is basically normal, the current has no obvious fluctuation, the temperature rises rapidly, and the vibration is slightly increased.

S2.1.2.3: key characteristic parameters capable of reflecting motor faults are selected.

Comprehensively considering, selecting the fluctuation degree of the motor rotation speed, the peak factor of the current waveform, the speed change of temperature rise and the difference of vibration frequency spectrums as key characteristic parameters.

S2.1.2.4: and constructing a motor fault state evaluation index.

And comprehensively considering the rotation speed fluctuation index, the current peak factor, the temperature rise rate and the vibration spectrum difference as evaluation indexes.

S2.1.2.5: and establishing a regression model between the motor fault evaluation index and the characteristic parameters.

Specifically, collecting operation data comprising motor characteristic parameters and fault states; determining the relationship between the motor fault evaluation index and the characteristic parameters by utilizing data visualization and correlation analysis; establishing an initialized linear regression model comprising a linear regression model between a rotational speed fluctuation index and rotational speed fluctuation, a parameter regression model of a current peak factor and a current waveform peak factor, a parameter regression model of a temperature rise rate and a temperature rise rate, and a parameter regression model of vibration spectrum difference and vibration spectrum difference; training the initialized linear regression model by using training set data to obtain model parameters; testing the effect of the linear regression model on the validation set; if the verification effect is poor, adjusting the form of the regression model, and trying a nonlinear regression model (such as polynomial regression and SVM regression); repeating the training and verifying process for the nonlinear regression model until the model reaches a predetermined performance index; obtaining a regression relation model between the final four sets of characteristic parameters and the evaluation indexes; and integrating the four sub-regression models to construct a final motor fault assessment model.

S2.1.2.6: and returning a motor fault state prediction result.

S2.1.3: and summarizing the battery health evaluation model and the motor state evaluation model, and carrying out standardization and encapsulation.

S2.2: and taking the battery state evaluation model and the motor state evaluation model as sub-models for input, establishing a vehicle comprehensive state evaluation model based on deep learning, and training the model.

Specifically, the method comprises the following steps:

s2.2.1: and determining input data, and establishing a vehicle comprehensive state evaluation model by adopting a deep neural network.

Preferably, the battery state evaluation model and the motor state evaluation model are input as sub-models; and establishing a vehicle comprehensive state evaluation model by adopting a deep neural network, wherein an input layer corresponds to each input characteristic, an implicit layer comprises extraction characteristics such as a plurality of convolution layers, a recursion layer and the like, and an output layer gives a vehicle comprehensive state evaluation result.

S2.2.2: and training a vehicle comprehensive state evaluation model by using training set data, selecting a mean square error as a loss function, and continuously optimizing model parameters through back propagation.

Specifically, the collected battery state prediction data, motor state prediction data and vehicle operation data are divided into a training set and a verification set; adopting the deep learning model established in the step S2.2.1 as a vehicle comprehensive state evaluation model; and selecting the mean square error as a loss function, defining an evaluation index (such as accuracy and F1 score), and carrying out multiple iterations on the training set data, and continuously optimizing model parameters until the model reaches a preset performance index on a verification set.

The specific formula for the mean square error MSE is as follows:

wherein y is _i To be a true value of the value,n is the number of samples, which is the predicted value of the model.

S2.2.3: and performing effect verification on the vehicle comprehensive state evaluation model.

Specifically, the test set data is input into a trained comprehensive vehicle state evaluation model, a vehicle state prediction result is output, the prediction accuracy and reliability of the evaluation model are compared with the actual vehicle state, and model parameters and structures are adjusted according to the verification result until expected performance is achieved.

S2.3: and outputting a comprehensive state evaluation result of the vehicle to realize the monitoring and evaluation of the states of the battery and the motor.

Specifically, the real-time monitored vehicle operation data is input into a trained vehicle comprehensive state evaluation model, a vehicle comprehensive state evaluation result is output, and the real-time monitoring and evaluation of the states of the battery and the motor are realized according to the evaluation result, so that the operation and maintenance of the vehicle are further guided.

S3: and establishing an environment recognition model based on the vehicle comprehensive state evaluation model, and outputting environment condition information.

Specifically, the method comprises the following steps:

s3.1: and establishing an environment recognition model based on the multi-mode data fusion and the deep learning method.

Specifically, the multi-mode data required by building the environment recognition model, including data such as images, videos, voices and distance sensors, are collected, and preprocessing is carried out on the data of different modes, including data cleaning, synchronization and standardization, so that the quality and consistency of the data are ensured.

Further, an environment recognition model is established based on a multi-mode data fusion and deep learning method; extracting features of the data of each mode, designing a multi-mode feature fusion structure, and effectively fusing the features of different modes together; and constructing a deep neural network, wherein the deep neural network comprises a convolution layer, a full connection layer and the like for extracting environmental characteristics, an output layer is used for classifying, and the characteristics after multi-mode data fusion are subjected to environmental condition classification.

S3.2: the data obtained by fusing the vehicle state evaluation model output and the multi-mode data is used as the input of the environment recognition model, and different environment conditions can be learned and distinguished through the supervised learning and training model.

Specifically, the method comprises the following steps:

s3.2.1: and taking the characteristics obtained by fusing the output of the vehicle state evaluation model and the multi-mode data as the input of the environment recognition model.

Specifically, the output of the vehicle state evaluation model and the multi-mode data are subjected to feature fusion to obtain a comprehensive feature vector.

S3.2.2: and matching the multi-mode data under different environmental conditions and the corresponding environmental classifications as labels to form a training data set with the labels.

Specifically, multi-modal data (including images, audio, sensor data, etc.) under different environmental conditions is collected; assigning a corresponding environmental classification label to each data sample (e.g., marking data in a scene as an environmental classification of city, highway, country, etc.); and matching the multi-mode data of each data sample with the corresponding environment classification label to form a training data set with the label.

S3.2.3: and building a structure of the environment recognition model based on a deep learning method.

Preferably, factors such as a detection target, scene complexity, training data scale and the like are comprehensively considered, a convolutional neural network VgNet is selected as a structure of an environment recognition model, and a corresponding network hierarchical structure is designed to comprise an input layer, a hidden layer, an output layer, an activation function, a loss function and the like.

S3.2.4: the training data set with the labels is used for training the environment recognition model.

Specifically, a training data set (multi-modal data and corresponding environment classification labels) with labels is input into an environment recognition model; the model carries out forward propagation through a training data set to generate a prediction result of environmental classification; calculating a loss function between a prediction result and a real label by using a supervised learning method; calculating the gradient of the loss function to the model parameters using a back propagation algorithm; updating parameters of the model to reduce the value of the loss function by means of gradient descent; and repeating the steps repeatedly, and training all training data sets for a plurality of times to gradually adjust the parameters of the model, so that the model can accurately learn and distinguish different environmental conditions.

S3.2.5: and evaluating the trained environment recognition model through the verification set data.

Further, using the verification set data, inputting the verification set data into the trained environment recognition model; generating a prediction result of the environmental classification by the model through forward propagation; comparing the prediction result with the real environment classification of the verification set data; the performance of the model on the environment recognition task is measured by using evaluation indexes (such as accuracy, recall, F1 value and the like).

S3.2.6: and adjusting the model structure and parameters according to the verification result, and retraining until a preset performance index is reached.

Specifically, if the performance of the model on the verification set does not reach the standard, the structure and parameters of the model are adjusted, the number of layers of the network and the number of neurons are modified or a regularization method is introduced to improve the performance of the model; training the adjusted model again, and repeating the steps until the model reaches a preset performance index.

S3.3: and predicting the test set data by using the trained environment recognition model, calculating the evaluation index of environment recognition, and verifying the model effect.

S3.4: and applying an environment recognition model to the newly acquired real-time data, outputting the environment state, and providing a corresponding adaptive strategy for vehicle operation according to the environment state.

Specifically, under the urban environment, the vehicle speed is dynamically adjusted according to the speed limit sign of the urban road section; planning an optimal navigation path according to real-time traffic conditions, and avoiding traffic congestion road sections; the monitoring capability of pedestrians and non-motor vehicles is enhanced to cope with emergency situations in urban environments. Under suburban environment, the cruising speed of the vehicle is optimized and adjusted in real time according to the bending degree of the road; the speed of the vehicle is intelligently adjusted according to the distance between the front vehicle and the rear vehicle; changes in road mass and grade are monitored to make adjustments to the vehicle system in advance. Under a high-speed environment, selecting the optimal channel switching time and position according to a preset navigation route and traffic flow; judging the switching time of the high beam and the low beam according to the distance and the speed of the front vehicle; and monitoring whether the current road traffic direction is deviated or not, and preventing from entering other road sections by mistake. Entering a parking space and completing parking operation in a parking environment; continuously monitoring whether a moving obstacle exists around the vehicle; intelligent determination of when to unlock/lock the door.

S4: and constructing a charging demand prediction model and an environment constraint recognition model, and integrating multi-source information planning to generate an optimal return charging path by adopting an ant colony algorithm and a deep reinforcement learning method.

Specifically, the method comprises the following steps:

s4.1: and (5) establishing a charging demand prediction model, and predicting the electric quantity required by the return journey.

Specifically, the preprocessed running data is used as a training set, and a deep learning model is selected to predict the electric quantity required by return voyage; inputting the preprocessed driving data into a model for training; the model is used for prediction, and by inputting information such as the current position, the destination, the route and the like of the returning vehicle into the model, the model outputs a predicted value which represents the electric quantity required by returning.

S4.2: and according to the length of the return route, calculating the predicted total electric quantity consumption in the return process by combining the parameters of the electric quantity consumption per unit mileage.

Specifically, the length of the return path is calculated according to the length of the actual path, and the parameters of electricity consumption per unit mileage are used for calculation by combining the electricity demand prediction and the path length, so that the specific formula of the predicted total electricity consumption TE in the return process is obtained as follows:

TE＝ED×PL×EC+CL

where ED is the electricity demand forecast, PL is the path length, EC is the electricity consumption parameter per unit mileage, and CL is the charging loss.

S4.3: and comparing the current state of charge of the battery with the predicted electric quantity required by the return and the total electric quantity consumption, and judging whether the current electric quantity can meet the return requirement.

Specifically, if the current state of charge is greater than or equal to the predicted required electric quantity for return and the current state of charge is greater than or equal to the total electric quantity consumption, then the current electric quantity can be considered to be sufficient to meet the return demand; if the current state of charge is less than the predicted amount of power required for the return trip or the current state of charge is less than the total power consumption, then the current amount of power may be insufficient to complete the return trip, requiring corresponding actions such as charging or adjusting the travel plan, etc.

S4.4: and when the electric quantity meets the requirement, calculating an optimal return charging path with minimum time and energy consumption as an ideal path.

Specifically, determining the current state of charge and the predicted electric quantity required for the return, and analyzing the available charging stations and the information of the charging speeds, the positions and the like of the charging stations; constructing a driving road network diagram of the electric vehicle based on Dijkstra algorithm, wherein nodes represent charging station positions, and edges represent driving paths among the nodes; setting a Cost function, and comprehensively considering the path running time and the energy consumption to calculate a path Cost; iteratively searching the minimum Cost value of all feasible paths through Dijkstra algorithm to obtain an optimal path; and planning a charging point and a charging amount according to the position and the charging speed of the charging station on the path, so as to ensure that the electric quantity can sufficiently support the return journey.

Specifically, the specific formula of the Cost function is as follows:

Cost(u,v)＝αTime(u,v)+βEnergy(u,v)

where u, v are two nodes in the path diagram, time (u, v) is the travel Time from node u to node v, energy (u, v) is the Energy consumption from node u to node v, and α, β are weight coefficients.

It should be noted that Time (u, v) may be estimated according to the path distance and the vehicle speed, and Energy (u, v) may be estimated according to the path length and hundred kilometers of Energy consumption of the vehicle; by adjusting the sizes of alpha and beta, the weight of time and energy consumption in a Cost function can be controlled; if the time is more important, alpha can be increased; if one looks more at the energy consumption, β can be increased.

S4.5: and continuously acquiring real-time positioning in the return voyage, monitoring the deviation between the actual path and the ideal path, and triggering a correction model to enable the actual path to be in line with the ideal path again if the deviation exceeds a preset threshold value.

Preferably, the method comprises the following steps:

s4.5.1: the real-time position and speed of the vehicle is acquired using GPS.

S4.5.2: and establishing a path correction model by adopting a deep reinforcement learning method.

S4.5.3: and calculating the deviation between the actual path and the ideal path, and triggering a correction model if the deviation exceeds a preset threshold value.

In particular, acquiring a real-time position (x _t ,y _t ) And velocity v _t Predicting the next time position (x) according to the preset time interval deltat _t+1 ,y _t+1 ) Finding the nearest point (x _p ,y _p ) The Euclidean distance d between two points is calculated, and the specific formula is as follows:

wherein, (x) _t+1 ,y _t+1 ) For the next time position, (x) _p ,y _p ) To find the closest point to the predicted position on the optimal path.

Further, d is compared with a preset threshold delta, and if d is less than or equal to delta, the actual path approaches the optimal path without correction; if d is more than delta, triggering a correction model; if d is larger than delta for N times continuously, delta is increased, and the correction range is enlarged; if d is less than or equal to delta for M times continuously, delta is reduced, and the correction range is narrowed.

Furthermore, whether the corrected Boolean value and the delta of the correction model are finally output, wherein the Boolean value is TRUE and is a trigger path correction model, the Boolean value is FALSE and is a non-trigger path correction model, and the delta value is output so that the correction model can adjust the correction range and the correction strength as required, and dynamic and intelligent path correction is realized.

S4.5.4: and carrying out smoothing correction operation by combining the vehicle kinematic model.

S4.5.5: and continuously monitoring the path, identifying the road condition change, and dynamically adjusting the correction model.

S4.5.6: and when the corrected path meets the precision requirement, outputting the corrected path as a final actual return path.

S4.6: in path deviation monitoring and correction, environmental conditions and constraint information are identified, and the corrected path is dynamically adjusted according to environmental impact.

Specifically, when an obstacle is detected in front, taking evading measures, and selecting a proper detour path according to the environment and the vehicle capacity so as to avoid collision or trapping in the obstacle; according to the wet and slippery degree of the environment, the road surface condition and the vehicle performance, the speed and the strength are adjusted to ensure the safety and the feasibility of the correction path; the optimal charging location is dynamically selected based on the energy status and availability of charging facilities to ensure that the revised path matches the energy demand.

S4.7: and outputting an optimal return charge path meeting the charge and environment constraints.

S5: and driving the vehicle along the optimal return charging path through an automatic driving technology, and charging after reaching a charging station.

Specifically, the method comprises the following steps:

s5.1: and (4) acquiring the current position of the vehicle, and calling the optimal return charging path output in the step (S4).

S5.2: and controlling the vehicle to return according to the optimal return charging path by using an automatic driving control algorithm.

S5.3: when the vehicle enters the charging station, charging strategies of different electricity price periods are calculated by using the charging model.

Further, an electric vehicle charging model is constructed, wherein the charging process is represented as follows:

wherein, soc is the battery power; soc ₀ For initial charge of battery, soc _up At the upper limit of battery capacity, p _i For the output power of the charging pile eta _i For charging efficiency, P _i And the power grid output power is obtained, and t is the charging time.

Further, the discharging process of the electric automobile to the power grid is represented as follows:

Preferably, it is assumed that the programmed unmanned electric vehicle optimizes the full cycle (which may be 1 day, or 6 hours, or other time period value, depending on the programmed vehicle operating cycle, and the patent considers the ending time period, the soc value, to be settable, here set to the soc _set ) In total, k electricity price time periods are separated by each time period interval t. The start time and the end time are not necessarily at the end of the period, so the two periods are less than t in length and are respectively marked as t ₀ And t _end Therefore, for the G2V car, a specific formula aimed at minimizing the electric charge is as follows:

Wherein C is _j For the electricity price of each period, P _i /η _i For grid side power, t _j Indicating the working time of the charging pile at each time interval, the soc _set For program demand, soc ₀ For initial charge of battery, p _i For charging pile output power, soc _low Is the lowest capacity limit of the battery, soc _up Is the upper limit of the battery capacity, t is the charging time, t ₀ For start-up time, t _end To end time, t _k For the working time of the charging pile in the period j=k, namely the working time of the charging pile in the last period, t ₁ And j=1 time periods are the working time of the charging pile, namely the working time of the charging pile in the first time period, and k is the total time period number of the planning process.

Furthermore, the electric automobile charging time schedule can be obtained by solving the formula.

Preferably, for the electric automobile, besides selecting to charge in the lowest electricity price period, the energy transfer in different electricity price periods can earn a gap, so that the economic benefit is remarkable, and the corresponding charge and discharge optimization model is as follows:

Wherein, delta is a charge-discharge sign, the value is 0 or 1, delta is 1 to indicate the charge in the period, delta is 0 to indicate the discharge in the period, and the same period cannot contain two processes due to the conversion efficiency of charge and discharge; equation constraint indicates that the final battery residual capacity meets the set requirement, soc _low ＜soc＜soc _up Indicating upper and lower limits of battery energy constraints during battery energy handling.

S5.4: and connecting charging equipment and starting charging, and monitoring and dynamically adjusting a charging strategy in real time in the charging process.

S5.5: and evaluating the charging effect and updating the battery parameters.

S5.6: and planning a next-round on-duty plan according to the new-round patrol and overhaul task demands and the environmental assessment result, and entering a next preset unmanned supervision period.

S6: and monitoring the vehicle state and the model prediction result through a remote monitoring system, and starting an abnormal processing flow in time after the abnormal situation is found.

Furthermore, the embodiment also provides an intelligent planning system of the autonomous charging unmanned electric vehicle based on deep learning, which comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring basic data of the programmed unmanned electric vehicle and preprocessing the data; the vehicle state evaluation module is used for establishing a vehicle comprehensive state evaluation model and monitoring and evaluating the states of the battery and the motor; the environment recognition module is used for building an environment recognition model, outputting environment condition information and providing an adaptive strategy for vehicle operation; the path planning module is used for predicting the charging requirement, planning the optimal return charging path and performing a path correction function; and the automatic driving module is used for returning and charging according to the planned path by utilizing an automatic driving technology and calculating a charging strategy.

The embodiment also provides computer equipment which is suitable for the situation of the intelligent planning method of the autonomous charging unmanned electric vehicle based on deep learning, and comprises a memory and a processor; the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the autonomous charging unmanned electric vehicle intelligent planning method based on deep learning, which is provided by the embodiment.

The computer device may be a terminal comprising a processor, a memory, a communication interface, a display screen and input means connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

The present embodiment also provides a storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of: acquiring basic data of the programmed unmanned electric vehicle, and preprocessing the acquired basic data of the programmed unmanned electric vehicle; establishing a vehicle comprehensive state evaluation model based on the preprocessed data; establishing an environment recognition model based on the vehicle comprehensive state evaluation model, and outputting environment condition information; constructing a charging demand prediction model and an environment constraint recognition model, and integrating multi-source information planning to generate an optimal return charging path by adopting an ant colony algorithm and a deep reinforcement learning method; the vehicle is driven along the optimal return charging path through an automatic driving technology, and is charged after reaching a charging station; and monitoring the vehicle state and the model prediction result through a remote monitoring system, and starting an abnormal processing flow in time after the abnormal situation is found.

In summary, by constructing the vehicle state evaluation and environment recognition model driven by deep learning, adopting the integrated and optimized charging path planning and automatic driving return technique and establishing the remote monitoring mechanism, the electric vehicle can realize closed-loop autonomous charging management, has the capabilities of state accurate prediction, environment perception adaptation, efficient charging planning, automatic return charging and remote exception handling, and improves the intelligent level of the electric vehicle and the reliability of unmanned operation and maintenance obviously as a whole.

Example 2

Referring to fig. 1 and 2, for a second embodiment of the present invention, the embodiment provides an intelligent planning method for an autonomous charging unmanned electric vehicle based on deep learning, and in order to verify the beneficial effects of the present invention, scientific demonstration is performed through economic benefit calculation and simulation experiments.

Specifically, the programmed unmanned electric vehicle base data is shown in table 1.

Table 1 Programming unmanned electric vehicle base data

Further, preprocessing the data includes format conversion, filtering, normalization and the like, so as to obtain high-quality structured data.

Preferably, a battery health evaluation sub-model is constructed to evaluate the battery state, a motor state evaluation sub-model is constructed to evaluate the motor state, the sub-models are summarized for standardization and encapsulation, the battery and the motor state evaluation model are used as input, and a vehicle comprehensive state evaluation model based on deep learning is established.

Further, multi-modal data (such as images and audio) are collected first, and the data are processed by using a convolutional neural network, and the fused feature vectors are taken as input at an input layer; the convolution layer is then used to extract features and associate these features with the tag data samples through the full connection layer; finally, carrying out classification prediction through an output layer; the accuracy of the verification set reaches 92%.

Specifically, the charging demand prediction model predicts that 25kW of electric quantity is required for the return voyage, predicts that the total electric quantity consumes 0.896kWh according to the path length of 3.2 km and the electric consumption of 280 Wh/km, and the current SOC is 75%, and the current electric quantity can meet the return voyage demand, so that the Dijkstra algorithm is used for searching the optimal return voyage charging path with the minimum time and energy consumption, and if the deviation is overlarge, the model is corrected to enable the path to be in line with the optimal path again.

Further, the current position is obtained, the charging station is reached according to the optimal return charging path, the charging station position is 39 degrees 54 '20' N,116 degrees 23 '45' E according to the field information, a time-interval charging strategy is formulated, and the peak time interval (0.5 yuan/degree): charging 6kWh, flat period (0.3 yuan/degree): charging 10kWh, connecting charging equipment to monitor charging effect, detecting state through a remote monitoring system, if the motor is found to be overheated, starting cooling treatment.

Further, a comparison table of the present invention with the prior art is shown in table 2.

TABLE 2 comparison of technical indicators of the present invention with the prior art

Technical index	The method of the invention	Existing solutions
			Accuracy of vehicle state assessment	95％	85％
Environmental recognition accuracy	96％	90％
			Charging demand prediction error	±3％	±8％
Charging price	0.48 yuan/degree	0.45 yuan/degree
			Optimal path planning time	For 5 minutes	For 10 minutes
Path deviation tolerance	10 meters	15 m
			Abnormality detection reaction time	2 seconds	5 seconds

Specifically, compared with the existing scheme, the method provided by the invention has more excellent results in the aspects of vehicle state evaluation accuracy, environment recognition accuracy and charging demand prediction error. The vehicle state evaluation accuracy rate of the method reaches 95%, is higher than 85% of that of the existing scheme, and can evaluate the state of the vehicle more accurately; in the aspect of the environment recognition accuracy, the method reaches 96 percent, and compared with 90 percent of the existing scheme, the method has higher accuracy, and can more accurately recognize the environment conditions around the vehicle; in the aspect of the prediction error of the charging demand, the error range of the method is +/-3%, and the error range of the existing scheme is +/-8%, so that the charging demand of the vehicle can be predicted more accurately.

Furthermore, in terms of charging price, the charging price of the method is 0.48 yuan/degree which is slightly higher than 0.45 yuan/degree of the existing scheme, and a user can select a proper charging scheme according to own requirements and budget; in the aspect of optimal path planning time, the method only needs 5 minutes, but the existing scheme needs 10 minutes, so that the optimal path can be planned more quickly; in terms of path deviation tolerance, the tolerance of the method is 10 meters, which is slightly smaller than 15 meters of the existing scheme, so that the path can be planned more accurately; in terms of path deviation tolerance, the tolerance of the method is 10 meters, which is slightly less than 15 meters of the existing scheme, and the path can be planned more accurately.

It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention may be modified or substituted without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered in the scope of the claims of the present invention.

Claims

1. The intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning is characterized by comprising the following steps of: comprising the steps of (a) a step of,

Acquiring basic data of the programmed unmanned electric vehicle, and preprocessing the acquired basic data of the programmed unmanned electric vehicle;

establishing a vehicle comprehensive state evaluation model based on the preprocessed data;

establishing an environment recognition model based on the vehicle comprehensive state evaluation model, and outputting environment condition information;

constructing a charging demand prediction model and an environment constraint recognition model, and integrating multi-source information planning to generate an optimal return charging path by adopting an ant colony algorithm and a deep reinforcement learning method;

the vehicle is driven along the optimal return charging path through an automatic driving technology, and is charged after reaching a charging station;

and monitoring the vehicle state and the model prediction result through a remote monitoring system, and starting an abnormal processing flow in time after the abnormal situation is found.

2. The intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning of claim 1, wherein the intelligent planning method comprises the following steps: the vehicle comprehensive state evaluation model is built based on the preprocessed data, and comprises the following steps of:

respectively establishing a battery health evaluation model and a motor state evaluation model based on the preprocessed data;

taking the battery state evaluation model and the motor state evaluation model as sub-models for input, establishing a vehicle comprehensive state evaluation model based on deep learning, and training the model;

Outputting a comprehensive state evaluation result of the vehicle, and realizing monitoring and evaluation of the states of the battery and the motor;

the battery state of health in the battery state assessment model includes good battery state of health, initial decline in battery state of health, severe decline in battery state of health,

when the battery is in a good state, the internal resistance of the battery is low, the charge-discharge curve is stable, the capacity of the battery is slowly attenuated, and the charge-discharge response time of the battery is short;

when the state of health of the battery begins to decline, the internal resistance of the battery begins to slowly increase, the fluctuation of a charge-discharge curve is not obvious, the capacity decay rate of the battery begins to increase, and the charge-discharge response time of the battery is prolonged;

when the state of health of the battery is seriously declined, the internal resistance of the battery is greatly increased, the fluctuation of a charge-discharge curve is increased, the capacity of the battery is rapidly attenuated, and the charge-discharge response time of the battery is obviously prolonged;

the motor states in the motor state evaluation model include a motor normal state, a motor rotor fault state and a motor insulation fault state,

in a normal state of the motor, the rotating speed is kept in a stable range, current fluctuation is small, the temperature is controlled in an allowable range, and the vibration amplitude is low;

in the motor rotor fault state, the rotation speed fluctuation is increased, rapid rise or fall occurs, the current fluctuation is severe, the peak value is intermittently and greatly increased, the temperature rise is slow, the vibration overall is increased, and the vibration of each bearing is obviously different from that of the normal bearing;

In the insulation fault state of the motor, the rotating speed is normal, the current has no obvious fluctuation, the temperature rises rapidly, and the vibration is slightly increased.

3. The intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning of claim 1, wherein the intelligent planning method comprises the following steps: the method for establishing the environment recognition model based on the vehicle comprehensive state evaluation model comprises the following steps of:

establishing an environment recognition model based on a multi-mode data fusion and deep learning method;

taking the data obtained by fusing the output of the vehicle state evaluation model and the multi-mode data as the input of the environment recognition model, and enabling the vehicle state evaluation model to learn and distinguish different environment conditions through the supervised learning and training model;

predicting the test set data by using the trained environment recognition model, calculating the evaluation index of environment recognition, and verifying the model effect;

applying an environment recognition model on the newly acquired real-time data, outputting the environment state, and providing a corresponding adaptive strategy for vehicle operation according to the environment state;

the corresponding adaptive strategy for suggesting vehicle operation based on environmental conditions includes,

in urban environment, the speed is dynamically adjusted according to the speed limit sign of the urban road section, an optimal navigation path is planned according to the real-time traffic condition, the traffic jam road section is avoided, and the monitoring capability of pedestrians and non-motor vehicles is enhanced to cope with emergency in the urban environment;

Under suburban environment, the cruising speed of the vehicle is optimized and regulated in real time according to the bending degree of the road, the speed of the vehicle is intelligently regulated according to the distance between the front vehicle and the rear vehicle, and the change of the road quality and the gradient is monitored so as to make adjustment for a vehicle system in advance;

under a high-speed environment, selecting the optimal lane changing time and position according to a preset navigation route and traffic flow, judging the far-near light switching time according to the distance and speed of a front vehicle, monitoring whether the current road traffic flow direction is deviated, and preventing from entering other road sections by mistake;

under the parking environment, the vehicle enters a parking space and completes parking operation, whether moving barriers exist around the vehicle or not is continuously monitored, and when the vehicle door is unlocked/locked is intelligently judged.

4. The intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning of claim 3, wherein the intelligent planning method comprises the following steps: the method for learning and training the model through supervision type comprises the following steps of:

taking the characteristics obtained by fusing the output of the vehicle state evaluation model and the multi-mode data as the input of the environment recognition model;

the multi-mode data under different environmental conditions and the corresponding environmental classifications are used as label pairing to form a training data set with labels;

Building a structure of an environment recognition model based on a deep learning method;

training the environment recognition model by using a training data set with a label;

evaluating the trained environment recognition model through verification set data;

and adjusting the model structure and parameters according to the verification result, and retraining until a preset performance index is reached.

5. The intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning of claim 1, wherein the intelligent planning method comprises the following steps: the method for generating the optimal return charging path by integrating the multisource information planning comprises the following steps of:

establishing a charging demand prediction model, and predicting the electric quantity required by return;

according to the length of the return route, calculating the predicted total electricity consumption in the return process by combining the parameters of electricity consumption per unit mileage;

comparing the current state of charge of the battery with the predicted electric quantity required by the return voyage and the total electric quantity consumption, and judging whether the current electric quantity can meet the return voyage requirement;

when the electric quantity meets the requirement, calculating an optimal return charging path with minimum time and energy consumption as an ideal path;

continuously acquiring real-time positioning in the return voyage, monitoring the deviation between the actual route and the ideal route, and triggering a correction model to enable the actual route to be in line with the ideal route again if the deviation exceeds a preset threshold value;

In path deviation monitoring and correction, identifying environmental conditions and constraint information, and dynamically adjusting a correction path according to environmental influence;

and outputting an optimal return charge path meeting the charge and environment constraints.

6. The intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning of claim 1, wherein the intelligent planning method comprises the following steps: the automatic driving technology for driving the vehicle along the optimal return charging path planned in advance comprises the following steps:

acquiring the current position of a vehicle, and calling an optimal return charging path;

controlling the vehicle to return according to the optimal return charging path by using an automatic driving control algorithm;

when a vehicle enters a charging station, calculating charging strategies of different electricity price time periods by using a charging model;

connecting charging equipment and starting charging, and monitoring and dynamically adjusting a charging strategy in real time in the charging process;

evaluating the charging effect and updating the battery parameters;

and planning a next-round on-duty plan according to the new-round patrol and overhaul task demands and the environmental assessment result, and entering a next preset unmanned supervision period.

7. The intelligent planning method for the autonomous charging unmanned electric vehicle based on deep learning of claim 6, wherein the intelligent planning method comprises the following steps: the charging strategy for calculating different electricity price time periods by using the charging model comprises the following steps:

An electric automobile charging model is constructed, wherein the charging process is represented as follows:

wherein, the soc is the battery power, and the soc ₀ For initial charge of battery, soc _up At the upper limit of battery capacity, p _i For the output power of the charging pile eta _i For charging efficiency, P _i The power grid output power is obtained, and t is the charging time;

wherein, the soc is the battery power, and the soc ₀ For initial charge of battery, soc _low Is the lowest capacity limit of the battery, p ₀ For discharging power of battery, eta ₀ For discharging inverter efficiency, P ₀ The alternating current power is output by the inverter, and t is the charging time;

the specific formula aimed at minimizing the electricity charge is as follows:

wherein C is _j For the electricity price of each period, P _i /η _i For grid side power, t _j Indicating the working time of the charging pile at each time interval, the soc _set For program demand, soc ₀ For initial charge of battery, p _i For charging pile output power, soc _low Is the lowest capacity limit of the battery, soc _up Is the upper limit of the battery capacity, t is the charging time, t ₀ For start-up time, t _end To end time, t _k For the working time of the charging pile in the period j=k, namely the working time of the charging pile in the last period, t ₁ The working time of the charging pile in the j=1 period, namely the working time of the charging pile in the first period, wherein k is the total period number of the planning process;

8. The intelligent planning system of the autonomous charging unmanned electric vehicle based on deep learning is based on the intelligent planning method of the autonomous charging unmanned electric vehicle based on deep learning, which is characterized in that: also included is a method of manufacturing a semiconductor device,

the data acquisition module is used for acquiring basic data of the stylized unmanned electric vehicle and preprocessing the data;

The vehicle state evaluation module is used for establishing a vehicle comprehensive state evaluation model and monitoring and evaluating the states of the battery and the motor;

the environment recognition module is used for building an environment recognition model, outputting environment condition information and providing an adaptive strategy for vehicle operation;

the path planning module is used for predicting the charging requirement, planning the optimal return charging path and performing a path correction function;

and the automatic driving module is used for returning and charging according to the planned path by utilizing an automatic driving technology and calculating a charging strategy.

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: the processor executes the computer program to implement the method for intelligent planning of the autonomous charging unmanned electric vehicle based on deep learning of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized by: the computer program when executed by a processor realizes the steps of the autonomous charging unmanned electric vehicle intelligent planning method based on deep learning as set forth in any one of claims 1 to 7.