Plug-in hybrid electric vehicle energy management method based on multi-element information fusion

Abstract

The present application relates to a kind of based on multi-element information fusion plug-in hybrid electric vehicle energy management method, comprising: obtaining adaptive fuzzy reasoning system model, vehicle speed sequence prediction model and SOC reference trajectory estimator;Real-time calculation current time transmission ratio, future finite time domain within the vehicle speed prediction sequence and future finite time domain within the SOC final value, based on Pontryagin minimum principle within the finite time domain rolling optimization, obtain the reference covariant variable within the time domain;If there is solution in rolling optimization, then use reference covariant variable in subsequent control domain to carry out open-loop Pontryagin minimum optimization, obtain corresponding control set, if there is no solution in rolling optimization, then execute SOC follow strategy.Compared with prior art, the present application will transmission ratio selection and torque distribution be jointly included in the category of energy management, and multi-element driving information is fused, and double-layer strategy of rolling optimization and SOC follow is proposed, which effectively improves the effect of plug-in hybrid electric vehicle energy management strategy.

Description

Technical Field

[0001] This invention relates to the field of hybrid power energy management, and in particular to a method for energy management of plug-in hybrid electric vehicles based on multi-source information fusion. Background Technology

[0002] Energy shortages and environmental pollution have always been two major challenges facing the automotive industry. New energy vehicles are an important focus for my country to achieve energy transition and environmental improvement. Hybrid vehicles, as a transitional product in the automotive energy transition, combine the advantages of traditional internal combustion engine vehicles and pure electric vehicles. They not only improve the economy and emissions of vehicles to a certain extent, but also solve the problem of short driving range of pure electric vehicles, and have a very broad application market.

[0003] With the rapid development of 5G technology, the construction of intelligent transportation systems has become an important task of my country's "new infrastructure" initiative. Vehicle-road cooperation is an inevitable technological approach for intelligent transportation. It can build a three-layer "end-network-cloud" architecture through environmental perception, data fusion computing, and decision control, strengthening the connection between vehicles, roads, and drivers to provide safe, efficient, and convenient intelligent transportation services. In an intelligent transportation environment, vehicles can wirelessly communicate or exchange information with roads, people, networks, the environment, and infrastructure according to certain communication protocols and standards, thereby improving the driver's experience and increasing traffic efficiency. At the same time, the Internet of Vehicles (IoV) also increases the effective information available to vehicles, bringing new possibilities for vehicle-side decision control.

[0004] Energy management strategy is a key research area for hybrid electric vehicles (HEVs), as it is a crucial factor determining the vehicle's fuel economy, emissions, and driving comfort. Energy management strategy involves the rational distribution of torque demand under the combined action of multiple power sources to fully utilize their respective operating characteristics and achieve complementary advantages. For traditional hybrid electric vehicles (HEVs), energy management strategy requires energy distribution between the engine and electric motor to fully leverage the battery's rapid dynamic response, while simultaneously improving vehicle economy and emissions and extending battery life. Furthermore, the transmission ratio determines the operating range of the engine and electric motor, thus affecting the vehicle's economy, power, and driving smoothness. Therefore, for traditional HEVs, especially those with continuously variable transmissions (CVTs), incorporating transmission ratio selection into the energy management framework is essential.

[0005] Existing energy management strategies for hybrid electric vehicles can be mainly categorized into three types: rule-based, optimization-based, and learning-based. Each strategy has its own advantages and disadvantages. For example, rule-based strategies are robust and simple to design, but their control performance is relatively poor. Optimization-based strategies are often constrained by known operating conditions and computational complexity. A reasonable energy management strategy should combine the advantages of different methods, improving control performance while minimizing resource consumption, to meet the requirements of low energy consumption and low latency.

[0006] Model predictive control (MDC) is one of the most mainstream optimization-based energy management strategies, offering advantages such as online application and good optimization performance. However, the optimization effect of MDC is primarily limited by the accuracy of model predictions. Past research mainly relied on historical information for random predictions, resulting in poor robustness and an inability to adapt to rapidly changing traffic scenarios. With the development of vehicle-to-everything (V2X) technology, intelligent transportation systems are gradually transitioning from concept to reality. V2X communication enables vehicles to monitor surrounding road conditions in real time, providing the possibility of establishing high-precision vehicle speed prediction models based on multi-source information fusion. Therefore, designing energy management strategies in a V2X environment has become a hot research topic. Furthermore, when considering both gear shifting and torque distribution simultaneously, the complexity of the energy management problem increases, making it difficult to guarantee the real-time performance of rolling optimization algorithms within a finite time domain, thus limiting the performance of MDC. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a plug-in hybrid electric vehicle energy management method based on multi-source information fusion, which makes full use of the rich and diverse information available on the vehicle in the vehicle network environment in the design of hybrid electric vehicle energy management strategy.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A method for energy management of plug-in hybrid electric vehicles based on multi-source information fusion includes the following steps:

[0010] S1. Obtain the adaptive fuzzy inference system model, vehicle speed sequence prediction model, and SOC reference trajectory estimator;

[0011] The inputs to the adaptive fuzzy inference system model are the gearbox ratio at the previous moment, the vehicle speed at the current moment, and the normalized vehicle power demand, and the output is the gearbox ratio at the current moment.

[0012] The input to the vehicle speed sequence prediction model is multivariate driving information, and the output is a vehicle speed prediction sequence within a finite future time domain. The multivariate driving information includes vehicle speed sequence, speed of the vehicle in front, distance to the vehicle, traffic light status, distance, and remaining time.

[0013] The input to the SOC reference trajectory estimator is the demand power sequence in the future finite time domain, the SOC value at the current moment, and the expected SOC value at the end of the cycle. The output is the final SOC value in the future finite time domain.

[0014] S2. Based on the adaptive fuzzy inference system model, vehicle speed sequence prediction model and SOC reference trajectory estimator, calculate the current gearbox transmission ratio, the vehicle speed prediction sequence in the future finite time domain and the final value of SOC in the future finite time domain in real time. Perform rolling optimization in the finite time domain based on the Pontryagin minimum principle to obtain the reference costate variable in this time domain.

[0015] S3. If the rolling optimization has a solution, then the open-loop Pontryagin minimum optimization is performed in the subsequent control domain using the reference costate variables to obtain the corresponding control set, and S2 is repeated in the next control time domain. If the rolling optimization has no solution, then the SOC following strategy is executed based on the final value of SOC in the future finite time domain, and S2 is repeated in the next control time domain.

[0016] Preferably, in step S1, obtaining the adaptive fuzzy inference system model specifically involves:

[0017] Based on historical driving data under different operating conditions, different initial and final SOC conditions are set. Two-dimensional dynamic programming is used to optimize the selection of gearbox transmission ratio and torque distribution under different operating conditions and different initial and final SOC conditions, and the transmission ratio and SOC change trajectory are obtained as expert experience.

[0018] An adaptive fuzzy inference system model is constructed. The transmission ratio at the previous moment, the vehicle speed at the current moment, and the normalized vehicle power demand are taken as inputs, and the transmission ratio at the current moment is taken as output. The adaptive fuzzy inference system model is trained using expert experience on transmission ratio, and the trained adaptive fuzzy inference system model is obtained.

[0019] Preferably, in step S1, obtaining the vehicle speed sequence prediction model specifically involves:

[0020] Acquire historical multi-dimensional driving information and divide it into multiple training sets according to different geographical regions;

[0021] A first neural network model is constructed, and the first neural network model is trained using training sets corresponding to different geographical regions to obtain vehicle speed sequence prediction models corresponding to different geographical regions.

[0022] Preferably, in step S2, when the vehicle starts, the corresponding vehicle speed sequence prediction model is obtained according to the start and end points of the trip. During the driving process, multi-dimensional driving information is obtained. The vehicle speed prediction sequence in the future finite time domain is calculated in real time according to the vehicle speed sequence prediction model. After the trip ends, the multi-dimensional driving information and actual vehicle speed information obtained in this trip are saved for updating the training vehicle speed sequence prediction model.

[0023] Preferably, in step S1, obtaining the SOC reference trajectory estimator specifically involves:

[0024] Based on historical driving data under different operating conditions, different initial and final SOC conditions are set. Two-dimensional dynamic programming is used to optimize the selection of gearbox transmission ratio and torque distribution under different operating conditions and different initial and final SOC conditions, and the transmission ratio and SOC change trajectory are obtained as expert experience.

[0025] Obtain historical driving data;

[0026] A second neural network model is constructed, taking the demand power sequence in the future finite time domain, the SOC value at the current moment, and the expected SOC value at the end of the cycle as inputs, and the final SOC value in the future finite time domain as outputs. The second neural network model is trained using expert experience on the SOC change trajectory to obtain the SOC reference trajectory estimator.

[0027] Preferably, the cost function calculation formula for two-dimensional dynamic programming is as follows:

[0028]

[0029] Where Cost(t) represents the equivalent cost of using the corresponding control law at time t, and b e For engine specific fuel consumption, t delta For the time spent walking, P e (t) represents the engine output power at time t, Ratio(t) and Ratio(t-1) represent the gearbox transmission ratios at times t and t-1, and α is the transmission ratio change penalty factor.

[0030] Preferably, in step S2, during driving, multi-dimensional driving information is acquired, and the vehicle speed prediction sequence within the future finite time domain is calculated in real time according to the vehicle speed sequence prediction model; the demand power sequence within the future finite time domain is calculated according to the vehicle speed prediction sequence within the future finite time domain, and the transmission ratio of the gearbox at the current moment is obtained according to the adaptive fuzzy inference system model; the final value of SOC within the future finite time domain is obtained according to the SOC reference trajectory estimator.

[0031] Preferably, in step S2, only the SOC value needs to be considered as a state variable during the rolling optimization process. The Hamiltonian function used in the rolling optimization method is specifically as follows:

[0032]

[0033] Where SOC(t) is the SOC value at time t, P eng P(t) represents the engine output power at time t, λ(t) represents the costate variable at time t, and P req (t) represents the power demand at time t. Let ω(SOC) be the engine fuel consumption rate, and ω(SOC) be the SOC penalty factor. The rate of change of SOC;

[0034] The costate variables that satisfy the boundary conditions are saved as reference costate variables for subsequent open-loop optimization in the control domain, and as initial costate variables for the next rolling optimization.

[0035] Preferably, the costate variable is calculated as follows:

[0036] λ1=λ0,i=1

[0037] λ2=λ0+δ,i=2

[0038]

[0039] Where λ0 is the initial reference costate variable, and δ is the step size of the costate variable change. Let SOC be the final state value of the solution domain at the previous time i-1. Let SOC be the final state SOC value of the solution domain at time i-2. target The final value of SOC in the future finite time domain obtained in step (2) is given.

[0040] Preferably, in step S3, a SOC following strategy is implemented, which uses the final SOC value within a finite future time domain as the reference final state SOC value. Based on the reference final state SOC value, the expected SOC value at the next moment is obtained. From this, the output power of the power battery can be calculated, and the engine output power can be obtained by combining it with the transmission ratio at the current moment. The expected SOC value at the next moment can be obtained by the following formula:

[0041]

[0042] Among them, SOC ref The obtained reference final state SOC value is given by H, where H is the prediction time window length.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) The two-dimensional dynamic programming algorithm is used to optimize the working conditions in the cloud database offline, which solves the problem of shrinking feasible solution set when considering both gear shifting and torque distribution, and improves the optimization potential of energy management strategy.

[0045] (2) By training in the cloud, the vehicle-side application fully leverages the advantages of distributed computing, reduces the computing load on the vehicle side, and transmits the driving data of different vehicles back to the cloud to achieve rapid accumulation of training datasets and accelerate model training.

[0046] (3) By using deep learning algorithms to incorporate multi-dimensional information such as the vehicle's historical speed sequence, the speed of the vehicle in front, the distance between vehicles, the status of traffic lights, the distance and the remaining time into the prediction of future speed sequences, the accuracy and robustness of the prediction model are significantly improved, which provides a guarantee for the control effect of the rolling optimization algorithm.

[0047] (4) The gearbox transmission ratio at the corresponding time is obtained by using an adaptive fuzzy inference system. Online torque allocation is performed using a dual strategy of rolling optimization and SOC following, which decouples the transmission ratio selection and torque allocation problem, ensuring the reliability and real-time performance of the algorithm. Attached Figure Description

[0048] Figure 1 This is a flowchart of the present invention;

[0049] Figure 2 This is a schematic diagram of a two-dimensional dynamic programming algorithm for offline optimization.

[0050] Figure 3 This is a structural diagram of an adaptive fuzzy inference system model.

[0051] Figure 4 Input the membership function graph into the adaptive fuzzy inference system model;

[0052] Figure 5 This is a flowchart of the vehicle-side energy management strategy. Detailed Implementation

[0053] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operating procedures. It should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, and not for limiting the scope of the claims. The description in this section pertains only to a few typical embodiments, but the scope of protection of the present invention is not limited to the following embodiments. Substitution of identical or similar prior art with some technical features in the embodiments is also within the scope of the description and protection of the present invention.

[0054] As used herein, "an embodiment" or "embodiment" refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. In the description of the invention, it should be understood that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0055] Example 1:

[0056] This invention aims to solve two major problems in the operation of hybrid electric vehicles: transmission ratio selection and torque distribution. It achieves a "vehicle-cloud interconnected" energy management method through a closed-loop approach combining cloud data collection, model training, and on-vehicle online optimization and data feedback. Specifically, a two-dimensional dynamic programming algorithm is used in the cloud to globally optimize for different road conditions, obtaining the optimal transmission ratio selection law and SOC change trajectory. The optimized data is then used to train an adaptive fuzzy inference system for transmission ratio selection and a shallow neural network for generating the reference SOC trajectory. Simultaneously, a deep learning model is trained using multi-dimensional information collected from the actual vehicle, including historical vehicle speed, preceding vehicle speed, distance, traffic light status, distance, and remaining time, to predict future speed sequences. On the vehicle side, by downloading the pre-trained transmission ratio selection, speed prediction, and reference SOC estimation models from the cloud, a dual strategy of rolling optimization within a finite time domain and SOC following is applied to ensure the real-time performance and reliability of energy management.

[0057] Compared to existing energy management strategies, this application considers both gear shifting and torque distribution, fully utilizes the big data advantages brought by the Internet of Vehicles environment to make the vehicle speed prediction model more accurate, and effectively improves the optimization potential of the energy management strategy by using two-dimensional dynamic programming for offline optimization. The robustness of the method is improved by using online rolling optimization and SOC following dual strategies.

[0058] Specifically, a plug-in hybrid electric vehicle energy management method based on multi-source information fusion, such as... Figure 1 As shown, it includes the following steps:

[0059] (1) Obtain the adaptive fuzzy inference system model, vehicle speed sequence prediction model and SOC reference trajectory estimator;

[0060] The adaptive fuzzy inference system model, also known as the gearbox ratio selection model or shift strategy set, takes the gearbox ratio of the previous moment, the vehicle speed at the current moment, and the normalized vehicle power demand as inputs, and outputs the gearbox ratio at the current moment.

[0061] The vehicle speed sequence prediction model, also known simply as the vehicle speed prediction model, takes multivariate driving information as input and outputs a predicted vehicle speed sequence within a finite future time domain. The multivariate driving information includes the vehicle speed sequence, the speed of the vehicle in front, the distance between vehicles, the status of traffic lights, the distance, and the remaining time.

[0062] The SOC reference trajectory estimator, also known simply as the SOC estimator, takes the demand power sequence over a finite future time domain, the current SOC value, and the expected SOC value at the end of the cycle as inputs, and outputs the final SOC value over a finite future time domain.

[0063] Specifically, it includes the following steps:

[0064] (1.1) Based on historical driving data under different working conditions, different initial and final SOC conditions are set. Two-dimensional dynamic programming is used to optimize the selection of gearbox transmission ratio and torque distribution under different working conditions and different initial and final SOC conditions. The transmission ratio and SOC change trajectory are obtained as expert experience.

[0065] In this example, the computational principle of two-dimensional dynamic programming is as follows: Figure 2 As shown, for each time step, the state space is discretized into a matrix, and each reachable point (t) on the matrix is ​​represented by a matrix. n SOC n Ratio n All possible solutions can be transferred to the state space at the next time step based on the constraints, and the solution that minimizes the value function is stored at the currently reachable point. The specific formula for calculating the cost function of two-dimensional dynamic programming is as follows:

[0066]

[0067] Where Cost(t) represents the equivalent cost of using the corresponding control law at time t, and b e For engine specific fuel consumption, t delta For the time spent walking, P e (t) represents the engine output power at time t, Ratio(t) and Ratio(t-1) represent the gearbox transmission ratios at times t and t-1, and α is the transmission ratio change penalty factor.

[0068] This application uses two-dimensional dynamic programming to optimize the selection of transmission ratio and torque distribution under different working conditions. Compared with the traditional one-dimensional dynamic programming, this method adds the gearbox transmission ratio as a second state variable on the basis of the original single state variable SOC value. This can avoid the failure of the dynamic programming no-aftereffect condition and the shrinking of the feasible solution set caused by the influence of the transmission ratio change penalty factor during the calculation process, improve the upper limit of offline optimization, and make the obtained expert experience more accurate and reliable.

[0069] (1.2) Construct an adaptive fuzzy inference system model. Take the gearbox ratio at the previous moment, the vehicle speed at the current moment and the normalized vehicle power demand as inputs, and the gearbox ratio at the current moment as output. Use the expert experience of the gearbox ratio to train the adaptive fuzzy inference system model to obtain the trained adaptive fuzzy inference system model.

[0070] Specifically, in this example, the structure of the adaptive fuzzy inference system is as follows: Figure 3 As shown, the model includes 3 input variables and 27 fuzzy rules. The model inputs are the normalized power demand at the current moment, the vehicle speed, and the gearbox ratio at the previous moment. The model output is the optimal gearbox ratio at the current moment. The input membership function obtained after training is as follows: Figure 4 As shown, all of them adopt the form of Gaussian membership function.

[0071] This application utilizes expert experience data to train an adaptive fuzzy inference system to obtain an online transmission ratio selection strategy. By decoupling the transmission ratio selection and torque distribution problems, the complexity of energy management problems can be effectively reduced and the timeliness of the strategy can be improved.

[0072] (1.3) Obtain historical multi-dimensional driving information and divide it into multiple training sets according to different geographical regions; construct the first neural network model and train the first neural network model using the training sets corresponding to different geographical regions to obtain the vehicle speed sequence prediction model corresponding to different geographical regions.

[0073] Multi-dimensional driving information, such as the vehicle's historical speed, the speed of the vehicle in front, the distance between vehicles, traffic light status, distance, and remaining time, can be obtained through real-vehicle data collection and simulation. Considering that road conditions and speed limits vary across different regions, a speed sequence prediction model is trained based on the historical multi-dimensional driving information for different regions and roads.

[0074] To ensure the accuracy and effectiveness of the vehicle speed sequence prediction model, it needs to be updated regularly. During subsequent driving, multi-dimensional information can be obtained through vehicle sensors, V2X communication, and other means. After the vehicle trip is completed, the corresponding multi-dimensional information data and actual vehicle speed information are uploaded to the corresponding cloud database for model training.

[0075] After training is completed, multiple vehicle speed sequence prediction models corresponding to different regions and roads are stored in the cloud. When the vehicle starts, the vehicle speed prediction model corresponding to the region and road is downloaded from the cloud according to the start and end points of the trip. Multi-dimensional information is obtained by means of vehicle sensors, V2X communication, etc., and the vehicle speed prediction model is applied to predict the future vehicle speed sequence. If there is no corresponding road in the cloud database for the trip range, the most similar model can be matched and downloaded from the cloud database based on the real-time traffic information provided by the navigation system.

[0076] Specifically, in this example, the first neural network model uses an N-BEATS deep learning network as the vehicle speed prediction model. The model's input includes the vehicle's historical speed sequence, the speed of the vehicle in front, the distance between vehicles, the traffic light status, the distance, and the remaining time. The output is the predicted vehicle speed within a finite future time domain. The main parameters of the prediction model in this example are shown in Table 1, where H represents the prediction time domain length, and RMSE is the root mean square error.

[0077] Table 1. Main training parameters of the vehicle speed prediction model

[0078]

[0079]

[0080] For model-based predictive energy management strategies, the higher the model prediction accuracy, the better the optimization effect. However, the longer the prediction time, the closer the final result is to the global optimum. Therefore, considering both prediction accuracy and prediction time domain, this example selects the N-BEATS model with a prediction time domain of 10s as the vehicle speed sequence prediction model.

[0081] (1.4) Obtain historical driving data; construct a second neural network model, with the demand power sequence in the future finite time domain, the SOC value at the current moment, and the expected SOC value at the end of the cycle as input, and the final SOC value in the future finite time domain as output. Train the second neural network model using SOC change trajectory expert experience to obtain the SOC reference trajectory estimator.

[0082] Specifically, in this example, a shallow neural network is used to obtain the final state reference SOC value in the finite time domain. The network input includes the normalized demand power prediction value for the next 10 seconds calculated by the vehicle speed prediction model in S3, the SOC value at the current time, and the expected SOC value at the end of the loop. The number of hidden layers is 1, the number of neurons is 25, the training method is Levenberg-Marquardt, and the loss function is mean squared error (MSE).

[0083] In order to enable the obtained SOC reference trajectory estimator to incorporate operating condition information within a finite future time domain, this application introduces the future demand power sequence as an input parameter. At the same time, in order to make the SOC value at the end of the trip close to the expected value, the expected SOC value at the end of the trip is also used as an input parameter of the model. For plug-in hybrid electric vehicles, the expected SOC value at the end of the trip is generally the lower limit of the battery operating range.

[0084] It is understandable that the training of the aforementioned adaptive fuzzy inference system model, vehicle speed sequence prediction model, and SOC reference trajectory estimator can be carried out offline by acquiring relevant data, or the relevant data can be uploaded to the cloud for training and updating in the cloud to obtain an adaptive fuzzy inference system model, vehicle speed sequence prediction model, and SOC reference trajectory estimator that can be applied online. The trained model is then periodically written to the vehicle controller, and the vehicle controller can apply the trained adaptive fuzzy inference system model, vehicle speed sequence prediction model, and SOC reference trajectory estimator when the vehicle is in motion.

[0085] (2) Based on the adaptive fuzzy inference system model, vehicle speed sequence prediction model and SOC reference trajectory estimator, calculate the gearbox transmission ratio at the current moment, the vehicle speed prediction sequence in the future finite time domain and the final value of SOC in the future finite time domain in real time. Based on the Pontryagin minimum principle, perform rolling optimization in the finite time domain to obtain the reference costate variable in the time domain.

[0086] For details, see Figure 1 ,as follows:

[0087] (2.1) Determine whether the current time is within the effective control time domain of the previous prediction. If yes, proceed to step (3); otherwise, proceed to the following steps:

[0088] (2.2) During the driving process, acquire multi-dimensional driving information such as vehicle speed, speed of the vehicle in front, distance, traffic light status, distance and remaining time, and calculate the vehicle speed prediction sequence in the future finite time domain in real time based on the vehicle speed sequence prediction model;

[0089] (2.3) The power demand sequence in the future finite time domain is calculated based on the vehicle speed prediction sequence in the future finite time domain, and the gearbox transmission ratio at the current moment is obtained based on the adaptive fuzzy inference system model;

[0090] (2.4) Obtain the final value of SOC in the future finite time domain, i.e. the reference final state SOC value, based on the SOC reference trajectory estimator.

[0091] (2.5) Using the information obtained in steps (2.2)-(2.4), a rolling optimization in the finite time domain is performed using a method based on the Pontryagin minimum principle. In this rolling optimization process, only the SOC value needs to be considered as a state variable, and the gearbox transmission ratio selection has been given in step (2.3). To improve the calculation speed of the rolling optimization, the engine operating range in the two-dimensional dynamic programming in step S1 is used as the discrete range of the engine output power. When the required power is negative, the engine is assumed not to operate, and the power battery performs braking energy recovery within a preset range. The Hamiltonian function used in this rolling optimization method is specifically as follows:

[0092]

[0093] Where SOC(t) is the SOC value at time t, P eng P(t) represents the engine output power at time t, λ(t) represents the costate variable at time t, and P req (t) represents the power demand at time t. Let ω(SOC) be the engine fuel consumption rate, and ω(SOC) be the SOC penalty factor. The rate of change of SOC;

[0094] The calculation method for costate variables is as follows:

[0095] λ1=λ0,i=1

[0096] λ2=λ0+δ,i=2

[0097]

[0098] Where λ0 is the initial reference costate variable, and δ is the step size of the costate variable change. Let SOC be the final state value of the solution domain at the previous time i-1. Let SOC be the final state SOC value of the solution domain at time i-2. target The reference final state SOC value obtained in step (2.4);

[0099] (2.6) Save the costate variables that satisfy the boundary conditions as reference costate variables for subsequent open-loop optimization of the control domain, and as initial costate variables for the next rolling optimization.

[0100] (3) If the rolling optimization has a solution, the open-loop Pontryagin minimum optimization is performed in the subsequent control domain using the reference costate variable to obtain the corresponding control set, and step (2) is repeated in the next control time domain. If the rolling optimization has no solution, the SOC following strategy is executed according to the final value of SOC in the future finite time domain, and step (2) is repeated in the next control time domain.

[0101] Specifically, in this example, the SOC following strategy mainly includes the following steps:

[0102] (3.1) Based on the final SOC value obtained in step (2) within the future finite time domain, use it as the reference final state SOC value, and obtain the expected SOC value at the next moment based on the reference final state SOC value;

[0103] The expected SOC value at the next moment can be obtained by the following formula:

[0104]

[0105] Among them, SOC ref The obtained reference final state SOC value, where H is the prediction time window length;

[0106] (3.2) The output power of the power battery is calculated based on the expected SOC value at the next moment;

[0107] (3.3) Combine the gearbox transmission ratio obtained in step (2.3) at the current moment and calculate the engine output power through the required power.

[0108] During the operation of a hybrid electric vehicle, this application can be used to obtain a near-globally optimal online energy management strategy under different driving routes. When the operation of the hybrid electric vehicle ends, this application can be used to complete the aggregation of historical driving data and the training of cloud models. It can effectively combine the advantages of large cloud data storage and strong computing power with the real-time collection of vehicle data and timely feedback of results, so as to realize the integration of vehicle-road cooperation in an intelligent connected environment.

[0109] The main application process of the energy management strategy proposed in this application at the vehicle end is as follows: Figure 5 As shown, the system obtains the future demand power sequence within a finite time domain based on multi-dimensional information such as historical vehicle speed sequences, preceding vehicle speed, distance, traffic light status, distance, and remaining time. It then uses an adaptive fuzzy inference system model and a SOC reference trajectory estimator to obtain the gearbox transmission ratio and the expected final state SOC value within the finite time domain, respectively. The Pontryagin minimum algorithm is used for rolling solutions within the finite time domain. If the algorithm has a solution within a given computation time, the optimal control sequence is applied to each power source, and the current costate variable is used as the initial costate variable for the next control time domain. If there is no solution, an SOC following strategy is used for energy allocation, and this process is repeated at the next time step.

[0110] This application divides the energy management strategy for plug-in hybrid electric vehicles into two parts: cloud and vehicle. The cloud is responsible for training and updating the vehicle speed prediction model, SOC reference trajectory estimator, and transmission ratio selection model related to the driving route. The vehicle is responsible for the online application of the three models, real-time energy management allocation, and the transmission of driving data. The energy management strategy design considers vehicle-side information such as vehicle speed, battery SOC value, and transmission ratio, as well as roadside information such as the speed of the vehicle ahead, distance, traffic light status, distance, and remaining time. This completes the development of a hybrid electric vehicle energy management strategy based on multi-source information fusion under the background of intelligent connected vehicles, effectively improving the economic efficiency of the energy management strategy.

[0111] Hybrid vehicle energy management was performed using a rule-based energy management strategy, a one-dimensional dynamic programming strategy, a two-dimensional dynamic programming strategy, and a multi-dimensional information-based strategy proposed in this application. The results are shown in Table 2, where the corresponding fuel consumption per 100 kilometers is the equivalent result after unifying the terminal SOC value.

[0112] Table 2 Equivalent fuel consumption per 100 kilometers and terminal SOC value for four energy management strategies

[0113]

[0114] It can be seen that the offline optimization algorithm proposed in this application, two-dimensional dynamic programming, can achieve the best economic effect. Compared with the traditional one-dimensional dynamic programming, the economic efficiency is improved by 2.68%. The online energy management strategy based on multi-source information fusion proposed in this application improves the economic efficiency by 17.73% compared with the rule-based strategy, and the terminal SOC value can be closer to the expected SOC value.

[0115] It should be noted that this application can be implemented in software and / or a combination of software and hardware, for example, using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In one embodiment, the software program of this application can be executed by a processor to implement the steps or functions described above. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices. Furthermore, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0116] Furthermore, a portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions invoking the methods of this application may be stored in a fixed or removable recording medium, and / or transmitted via data streams in broadcast or other signal carrying media, and / or stored in the working memory of a computer device operating according to the program instructions. Here, one embodiment of this application includes an apparatus comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein, when the computer program instructions are executed by the processor, the apparatus is triggered to operate the methods and / or technical solutions based on the foregoing embodiments of this application.

[0117] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for energy management of plug-in hybrid electric vehicles based on multi-source information fusion, characterized in that, Includes the following steps: S1. Obtain the adaptive fuzzy inference system model, vehicle speed sequence prediction model, and SOC reference trajectory estimator; The inputs to the adaptive fuzzy inference system model are the gearbox ratio at the previous moment, the vehicle speed at the current moment, and the normalized vehicle power demand, and the output is the gearbox ratio at the current moment. The input to the vehicle speed sequence prediction model is multivariate driving information, and the output is a vehicle speed prediction sequence within a finite future time domain. The multivariate driving information includes vehicle speed sequence, speed of the vehicle in front, distance to the vehicle, traffic light status, distance, and remaining time. The input to the SOC reference trajectory estimator is the demand power sequence in the future finite time domain, the SOC value at the current moment, and the expected SOC value at the end of the cycle. The output is the final SOC value in the future finite time domain. S2. Based on the adaptive fuzzy inference system model, vehicle speed sequence prediction model and SOC reference trajectory estimator, calculate the current gearbox transmission ratio, the vehicle speed prediction sequence in the future finite time domain and the final value of SOC in the future finite time domain in real time. Perform rolling optimization in the finite time domain based on the Pontryagin minimum principle to obtain the reference costate variable in this time domain. S3. If the rolling optimization has a solution, then the open-loop Pontryagin minimum optimization is performed in the subsequent control domain using the reference costate variable to obtain the corresponding control set, and S2 is repeated in the next control time domain. If the rolling optimization has no solution, then the SOC following strategy is executed according to the final value of SOC in the future finite time domain, and S2 is repeated in the next control time domain. In step S1, obtaining the adaptive fuzzy inference system model specifically involves: Based on historical driving data under different operating conditions, different initial and final SOC conditions are set. Two-dimensional dynamic programming is used to optimize the selection of transmission ratio and torque distribution under different operating conditions and different initial and final SOC conditions. The transmission ratio and SOC change trajectory are obtained as expert experience. The expert experience data is used to train an adaptive fuzzy inference system to obtain an online transmission ratio selection strategy, thus decoupling the transmission ratio selection and torque distribution problems. An adaptive fuzzy inference system model is constructed. The transmission ratio of the previous time step, the vehicle speed at the current time step, and the normalized vehicle power demand are taken as inputs, and the transmission ratio of the current time step is taken as output. The adaptive fuzzy inference system model is trained using expert experience on transmission ratio of transmission ratio, and the trained adaptive fuzzy inference system model is obtained. In step S1, obtaining the SOC reference trajectory estimator specifically involves: Based on historical driving data under different operating conditions, different initial and final SOC conditions are set. Two-dimensional dynamic programming is used to optimize the selection of gearbox transmission ratio and torque distribution under different operating conditions and different initial and final SOC conditions, and the transmission ratio and SOC change trajectory are obtained as expert experience. Obtain historical driving data; A second neural network model is constructed, with the demand power sequence in the future finite time domain, the SOC value at the current moment, and the expected SOC value at the end of the cycle as inputs, and the final SOC value in the future finite time domain as output. The second neural network model is trained using SOC change trajectory expert experience to obtain the SOC reference trajectory estimator. The specific formula for calculating the cost function in two-dimensional dynamic programming is as follows: in, express The equivalent cost of using the corresponding control law at all times. For engine fuel consumption, Because the time is long compared to a walk, for Engine output power at all times. and for Time and The gearbox ratio at any given time, This is the penalty factor for changes in transmission ratio.

2. The energy management method for plug-in hybrid electric vehicles based on multi-source information fusion according to claim 1, characterized in that, In step S1, obtaining the vehicle speed sequence prediction model specifically involves: Acquire historical multi-dimensional driving information and divide it into multiple training sets according to different geographical regions; A first neural network model is constructed, and the first neural network model is trained using training sets corresponding to different geographical regions to obtain vehicle speed sequence prediction models corresponding to different geographical regions.

3. The energy management method for plug-in hybrid electric vehicles based on multi-source information fusion according to claim 2, characterized in that, In step S2, when the vehicle starts, the corresponding vehicle speed sequence prediction model is obtained according to the start and end points of the trip. During the driving process, multi-dimensional driving information is obtained. The vehicle speed prediction sequence in the future finite time domain is calculated in real time according to the vehicle speed sequence prediction model. After the trip ends, the multi-dimensional driving information and actual vehicle speed information obtained in this trip are saved for updating the training vehicle speed sequence prediction model.

4. The energy management method for plug-in hybrid electric vehicles based on multi-source information fusion according to claim 1, characterized in that, In step S2, multi-dimensional driving information is acquired during driving, and the vehicle speed prediction sequence within the future finite time domain is calculated in real time based on the vehicle speed sequence prediction model; the demand power sequence within the future finite time domain is calculated based on the vehicle speed prediction sequence within the future finite time domain; the transmission ratio of the gearbox at the current moment is obtained based on the adaptive fuzzy inference system model; and the final value of SOC within the future finite time domain is obtained based on the SOC reference trajectory estimator.

5. The energy management method for plug-in hybrid electric vehicles based on multi-source information fusion according to claim 1, characterized in that, In step S2, only the SOC value needs to be considered as a state variable during the rolling optimization process. The Hamiltonian function used in the rolling optimization method is specifically as follows: in, for SOC value at time, for Engine output power at all times. for Costate variables at time intervals, for Power demand at all times Engine fuel consumption rate. As the SOC penalty factor, The rate of change of SOC; The costate variables that satisfy the boundary conditions are saved as reference costate variables for subsequent open-loop optimization in the control domain, and as initial costate variables for the next rolling optimization.

6. The energy management method for plug-in hybrid electric vehicles based on multi-source information fusion according to claim 5, characterized in that, The calculation method for costate variables is as follows: in, As the initial reference costate variable, The step size of the costate variable change. For the previous moment The final state SOC value in the solution domain. For the best moment The final state SOC value in the solution domain. The final value of SOC in the future finite time domain obtained in step (2) is given.

7. The energy management method for plug-in hybrid electric vehicles based on multi-source information fusion according to claim 1, characterized in that, In step S3, the SOC following strategy is executed, which uses the final SOC value within a finite future time domain as the reference final state SOC value. Based on the reference final state SOC value, the expected SOC value at the next moment is obtained. From this, the output power of the power battery can be calculated. Combined with the transmission ratio at the current moment, the engine output power is obtained. The expected SOC value at the next moment can be obtained by the following formula: in, The obtained reference final state SOC value, To predict the length of the time window.

Images