CN108177648A - A kind of energy management method of the plug-in hybrid vehicle based on intelligent predicting - Google Patents

Abstract

The present invention provides an energy management method for a plug-in hybrid electric vehicle based on intelligent prediction, utilizes driving condition information with multi-source and mixed characteristics, online learning optimizes PHEV global energy distribution, and improves the intelligence of the hybrid power system level; establishing a real-time learning and prediction model, improving the prediction accuracy of operating conditions in the control time domain, and realizing multi-objective optimized energy management in the rolling time domain are of great significance for deeply tapping the energy-saving potential of PHEVs and have many beneficial effects.

Description

An energy management method for plug-in hybrid electric vehicles based on intelligent prediction

technical field

The invention relates to the technical field of energy management of a plug-in hybrid vehicle, in particular to a method for realizing energy management of a plug-in hybrid vehicle by means of intelligent prediction.

Background technique

Due to the characteristics of multiple sources of energy input, diversity of working modes, and strong coupling of various forms of energy flow in the power system of plug-in hybrid electric vehicles, its comprehensive control and energy management have always been relatively complex in this field. Technical Difficulties. Its energy management is closely related to various driving conditions such as traffic environment, road pedestrians, weather, driver's driving style, vehicle's own state, road gradient, etc. However, how to use these driving conditions with multi-source and mixed characteristics , realize online learning to optimize the global energy distribution of plug-in hybrid vehicles, and improve the intelligence level of the hybrid system, which is an urgent problem to be solved in this field.

Contents of the invention

Aiming at the above-mentioned technical problems in this field, the present invention provides an energy management method for a plug-in hybrid electric vehicle, which specifically includes the following steps:

Step 1. Online extraction of multi-dimensional driving condition information corresponding to the target driving route, and establishment of a global driving condition reconstruction model for the target driving route based on a deep learning algorithm. In this way, the pre-dynamic reconstruction of the target driving route is realized.

Step 2, based on the reconstruction model of the global driving conditions established in the step 1, establish a reinforcement learning network model to obtain the optimal energy trajectory of the power battery of the plug-in hybrid electric vehicle;

Step 3. Construct a driver style deep convolutional neural network model and a traffic information deep convolutional neural network model respectively according to the vehicle's own state information and traffic information, and extract the corresponding driver style features and traffic information features. The learning algorithm establishes a real-time prediction model of the future short-term operating conditions of the vehicle;

Step 4. According to the life model of the power battery, the optimal energy trajectory of the power battery in the step 2 is used as the final value constraint during rolling, and combined with the future short-term working conditions established in the step 3, real-time A predictive model is used to establish a control strategy for the power battery.

Further, the multi-dimensional driving condition information extracted online in step 1 includes information extracted from open source map service providers, traffic monitoring platforms, vehicle flow information, signal light information, pedestrian information and weather information from the vehicle vision system. The above various information is mapped to standard driving condition information through a standard mapping method.

Further, the establishment of the reinforcement learning network model in the step 2 to obtain the optimal energy trajectory of the power battery of the plug-in hybrid electric vehicle specifically includes: taking the least global energy consumption as the reinforcement learning network model Enhanced rewards.

Further, the vehicle's own state information in step 3 includes information related to steering, accelerator pedal, brake pedal, and the like. Different drivers have great psychological differences in the judgment of vehicle following, lane change behavior and signal lights during driving, which leads to different driving behaviors. In addition to the driving style of the driver, there are still some difficult-to-measure traffic environment information that will affect the future working conditions, so the traffic information includes such as the speed of the environment vehicle, the change of signal lights, and the random walking of pedestrians on the road. Databases are established for the vehicle's own state information and traffic information, and the deep convolutional neural network models of the driver's style and traffic information are respectively constructed using the samples in the database.

Each network layer of the neural network is added to the neural network at the end of a deep belief network composed of a plurality of Gauss-Bernoulli restricted Boltzmann machines superimposed, and the future short-term prediction of the vehicle is performed using the features generated by the deep belief network. Real-time forecasting of working conditions. The training process of the entire network is mainly composed of two parts, one is to train multiple Gauss-Bernoulli restricted Boltzmann machines layer by layer, and each Gauss-Bernoulli restricted Boltzmann machine model uses energy-based Joint probability expression. Another part of the model training process is parameter fine-tuning under supervised training. The specific operation is to introduce the neural network regression layer at the end to form a deep convolutional neural network structure, and use the back propagation algorithm to fine-tune the pre-trained parameters. The parameter update formula is as follows:

Where J(W, b; x, y) is the loss function of the model, which is determined by the parameters W, b, input x and output y, δ is the residual, 1 represents the number of layers, w ⁽¹⁾ is the weight parameter, b ⁽¹⁾ is the bias parameter, a ⁽¹⁾ is the activation value, m is the number of sample values, α is the learning rate, w ⁽¹⁾ is the regularization term, and λ is the regularization term coefficient.

Finally, the process of predicting working conditions based on the deep convolutional neural network model can be expressed by the following mapping relationship and forward propagation formula:

f(traffic information, driving style information; W, b)=v

z ^(l+1) = W ^(l) a ^(l) + b ^(l)

a ^(l+1) = f(z ^(l+1) )

where v is the vehicle speed, 1 represents the number of layers, w ⁽¹⁾ is the weight parameter, b ⁽¹⁾ is the bias parameter, a ⁽¹⁾ is the activation value, z ⁽¹⁺¹⁾ is the unit input weighted sum.

Further, the life model of the power battery in the step 4 is an empirical model of the cycle life of the lithium-ion power battery, and the capacity loss of the power battery is introduced into the objective function of energy consumption optimization to achieve multi-objective optimal management, and the power battery The empirical model of capacity change with charge and discharge current is:

Among them, Q _loss is the loss capacity of the power battery, B _exp is the prefactor, which is inversely proportional to the C _rate , R is the gas constant, T _batt is the average absolute temperature of the battery, and _Ah is the accumulated charging and discharging ampere hours of the power battery.

Further, the step 4 also includes establishing a cost function related to power battery life and driving energy consumption.

According to the energy management method of the plug-in hybrid vehicle provided by the present invention, the driving condition information with multi-source and mixed characteristics is utilized, online learning optimizes the global energy distribution of PHEV, and improves the intelligence level of the hybrid system; Real-time learning and forecasting models, improving the accuracy of operating condition prediction in the control time domain, and realizing multi-objective optimized energy management in the rolling time domain are of great significance for deeply tapping the energy-saving potential of PHEVs, and have many beneficial effects.

Description of drawings

Figure 1 is a schematic diagram of online extraction of multi-dimensional driving condition information corresponding to the target driving route

Figure 2 is a schematic diagram of establishing a real-time prediction model for future short-term operating conditions of vehicles

Figure 3 is a schematic diagram of the construction process of a deep convolutional neural network

Fig. 4 is the overall flow diagram of the method provided by the present invention

Detailed ways

The technical solutions of the present invention will be further explained in detail below in conjunction with the accompanying drawings.

An energy management method for a plug-in hybrid electric vehicle provided by the present invention, as shown in FIG. 4 , specifically includes the following steps:

Step 1. As shown in FIG. 1 , online extraction of multi-dimensional driving condition information corresponding to the target driving route, and establishment of a global driving condition reconstruction model for the target driving route based on a deep learning algorithm. In this way, the pre-dynamic reconstruction of the target driving route is realized.

Step 2, based on the reconstruction model of the global driving conditions established in the step 1, establish a reinforcement learning network model to obtain the optimal energy trajectory of the power battery of the plug-in hybrid electric vehicle;

Step 3. Construct a driver style deep convolutional neural network model and a traffic information deep convolutional neural network model respectively according to the vehicle's own state information and traffic information, and extract the corresponding driver style features and traffic information features. The learning algorithm establishes a real-time prediction model of the future short-term operating conditions of the vehicle;

Step 4. According to the life model of the power battery, the optimal energy trajectory of the power battery in the step 2 is used as the final value constraint during rolling, and combined with the future short-term working conditions established in the step 3, real-time A predictive model is used to establish a control strategy for the power battery.

In a preferred embodiment of the present application, the multi-dimensional driving condition information extracted online in step 1 includes information extracted from open source map service providers, traffic monitoring platforms, vehicle flow information, signal light information, pedestrian information and weather information.

In a preferred embodiment of the present application, the establishment of the reinforcement learning network model in the step 2 to obtain the optimal energy trajectory of the power battery of the plug-in hybrid electric vehicle specifically includes: taking the minimum global energy consumption as Reinforcement rewards for the reinforcement learning network model.

In a preferred embodiment of the present application, as shown in FIG. 2 , the vehicle's own state information in step 3 includes information related to steering, accelerator pedal, brake pedal, and the like. Different drivers have great psychological differences in the judgment of vehicle following, lane change behavior and signal lights during driving, which leads to different driving behaviors. In addition to the driving style of the driver, there are still some difficult-to-measure traffic environment information that will affect the future working conditions, so the traffic information includes such as the speed of the environment vehicle, the change of signal lights, and the random walking of pedestrians on the road. Databases are established for the vehicle's own state information and traffic information, and the deep convolutional neural network models of the driver's style and traffic information are respectively constructed using the samples in the database.

In a preferred embodiment of the present application, as shown in FIG. 3 , each network layer of the neural network is added at the end of a deep belief network composed of a plurality of Gauss-Bernoulli restricted Boltzmann machines. The network uses the features generated by the deep belief network to predict the future short-term working conditions of the vehicle in real time. The training process of the entire network is mainly composed of two parts, one is to train multiple Gauss-Bernoulli restricted Boltzmann machines layer by layer, and each Gauss-Bernoulli restricted Boltzmann machine model uses energy-based Joint probability expression. Another part of the model training process is parameter fine-tuning under supervised training. The specific operation is to introduce the neural network regression layer at the end to form a deep convolutional neural network structure, and use the back propagation algorithm to fine-tune the pre-trained parameters. The parameter update formula is as follows:

Where J(W, b; x, y) is the loss function of the model, which is determined by the parameters W, b, input x and output y, δ is the residual, 1 represents the number of layers, w ⁽¹⁾ is the weight parameter, b ⁽¹⁾ is the bias parameter, a ⁽¹⁾ is the activation value, m is the number of sample values, a is the learning rate, w ⁽¹⁾ is the regularization term, and λ is the regularization term coefficient.

In a preferred embodiment of the present application, the process of realizing the forecasting working conditions based on the deep convolutional neural network model can be expressed by the following mapping relationship and forward propagation formula:

f(traffic information, driving style information; W, b)=v

z ^(l+1) = W ^(l) a ^(l) + b ^(l)

a ^(l+1) = f(z ^(l+1) )

where v is the vehicle speed, 1 represents the number of layers, w ⁽¹⁾ is the weight parameter, b ⁽¹⁾ is the bias parameter, a ⁽¹⁾ is the activation value, z ⁽¹⁺¹⁾ is the unit input weighted sum.

In a preferred embodiment of the present application, the life model of the power battery in the step 4 is an empirical model of the cycle life of lithium-ion power batteries, and the power battery capacity loss is introduced into the objective function of energy consumption optimization to achieve Multi-objective optimization management, the empirical model of power battery capacity changing with charge and discharge current is:

Among them, Q _loss is the loss capacity of the power battery, D _exp is the prefactor, which is inversely proportional to the C _rate , R is the gas constant, T _batt is the average absolute temperature of the battery, and _Ah is the accumulated charging and discharging ampere hours of the power battery.

In a preferred embodiment of the present application, the step 4 further includes establishing a cost function related to power battery life and driving energy consumption.

Although the embodiments of the present invention have been shown and described, those skilled in the art can understand that various changes, modifications and substitutions can be made to these embodiments without departing from the principle and spirit of the present invention. and modifications, the scope of the invention is defined by the appended claims and their equivalents.

Claims (9)

1. An energy management method for a plug-in hybrid electric vehicle based on intelligent prediction, characterized in that: it specifically comprises the following steps: Step 1. Online extraction of multi-dimensional driving condition information corresponding to the target driving route, and establishment of a global driving condition reconstruction model for the target driving route based on a deep learning algorithm. Step 2, based on the reconstruction model of the global driving conditions established in step 1, establish a reinforcement learning network model to obtain the optimal energy trajectory of the power battery of the plug-in hybrid vehicle; Step 3. Construct a driver style deep convolutional neural network model and a traffic information deep convolutional neural network model respectively according to the vehicle's own state information and traffic information, and extract the corresponding driver style features and traffic information features. The learning algorithm establishes a real-time prediction model of the future short-term operating conditions of the vehicle; Step 4. According to the life model of the power battery, the optimal energy trajectory of the power battery in the step 2 is used as the final value constraint during rolling, and combined with the future short-term working conditions established in the step 3, real-time A predictive model is used to establish a control strategy for the power battery. 2. The method according to claim 1, characterized in that: the multi-dimensional driving condition information extracted online in the step 1 includes: extracted from open source map service providers, traffic monitoring platform, vehicle flow information and signal lights of the vehicle vision system information, pedestrian information and weather information. 3. The method according to claim 1, characterized in that: the establishment of a reinforcement learning network model in the step 2 to obtain the optimal energy trajectory of the power battery of the plug-in hybrid vehicle specifically includes: The least global energy consumption is used as the reinforcement reward of the reinforcement learning network model. 4. The method according to claim 1, characterized in that: the self state information of the vehicle in the step 3 includes information related to steering, accelerator pedal, and brake pedal; the traffic information includes information related to environmental vehicle speed, Information related to signal light conversion and random walking of pedestrians on the road; databases are respectively established for the vehicle's own state information and traffic information, and the samples in the database are used to construct the driver's style deep convolutional neural network model and traffic information respectively. Deep Convolutional Neural Network Models. 5. The method according to claim 4, characterized in that: each network layer of the neural network is to join the neural network at the end of the depth belief network formed by the superposition of a plurality of Gauss-Bernoulli restricted Boltzmann machines ; The training process mainly consists of two parts, one is to train multiple Gauss-Bernoulli restricted Boltzmann machines layer by layer; the other part is parameter fine-tuning under supervised training. 6. The method according to claim 5, characterized in that: in the Gauss-Bernoulli Restricted Boltzmann Machine, each Gauss-Bernoulli Restricted Boltzmann Machine model utilizes energy-based Joint probability expression; the parameter update formula in the fine-tuning process is as follows: Where J(W, b; x, y) is the loss function of the model, which is determined by the parameters W, b, input x and output y, δ is the residual, I represents the number of layers, w ^(l) is the weight parameter, b ^(l) is the bias parameter, a ^(l) is the activation value, m is the number of sample values, a is the learning rate, w ^(l) is the regularization term, and λ is the regularization term coefficient. 7. The method according to claim 6, characterized in that: the process of realizing the real-time prediction of the future short-term operating conditions of the vehicle based on the deep convolutional neural network model can be represented by the following mapping relationship and forward propagation formula: f(traffic information, driving style information; W, b)=v z ^(l+1) = W ^(l) a ^(l) + b ^(l) a ^(l+1) = f(Z ^(l+1) ) where v is the vehicle speed, l represents the number of layers, w ^(l) is the weight parameter, b ^(l) is the bias parameter, a ^(l) is the activation value, z ^{(l + 1)} is the unit input weighted sum. 8. The method according to claim 1, characterized in that: the life model of the power battery in the step 4 adopts the cycle life empirical model of lithium ion power battery: Among them, Q _loss is the loss capacity of the power battery, B _exp is the prefactor, which is inversely proportional to the C _rate , R is the gas constant, T _batt is the average absolute temperature of the battery, and _Ah is the accumulated charging and discharging ampere hours of the power battery. 9. The method according to claim 1, characterized in that: said step 4 further comprises establishing a cost function related to power battery life and driving energy consumption.