The present disclosure relates to the technical field of objective optimization, and in particular, to a multi-objective optimization method and system assisted by a gradient boosted neural network, and a device.
Hybrid electric vehicles derive power from both an internal combustion engine and an electric motor. In the design of hybrid electric vehicles, the energy control manager plays a crucial role in ensuring hybrid performance.
Typically, the energy control management system of a hybrid electric vehicle, as shown in
Considering that vehicle speed and torque are controlled by the driver or external devices, it is challenging to quantify the vehicle speed and torque with specific parameters and models. In the design of the energy control management system, it is necessary to consider different objective optimization problems. These problems are black-box optimization problems. Existing modal analysis methods, such as finite element analysis and experimental modal analysis, can approximate real physical models well, but the established models are generally complex, and model evaluation is time-consuming.
To address this issue, the industry has started using machine learning techniques to fit complex physical models. For example, proxy models like binomial approximation and support vector machine models are used to fit complex functions. While these methods partially address the drawbacks of the modal analysis methods, they fail to meet the requirements of fitting complex multi-objective optimization functions, and have limitations in fitting complex physical models.
An objective of the present disclosure is to provide a multi-objective optimization method and system assisted by a gradient boosted neural network, and a device, to solve issues such as time-consuming function evaluation and inability to obtain an optimal solution of the problem.
To achieve the above objective, the present disclosure provides the following technical solutions:
According to a first aspect, the present disclosure provides a multi-objective optimization method assisted by a gradient boosted neural network, including:
Optionally, the sample data includes at least a maximum threshold range of a battery state of charge when an internal combustion engine is turned off, a minimum threshold range of the battery state of charge when the internal combustion engine is turned on, a minimum speed range for operating the internal combustion engine, a maximum speed range for operating the internal combustion engine, a speed range of the internal combustion engine, a torque value range of the internal combustion engine, and a speed range at which the internal combustion engine is turned off.
Optionally, the optimal solution comprises at least optimal values of a maximum threshold of the battery state of charge when the internal combustion engine is turned off, a minimum threshold of the battery state of charge when the internal combustion engine is turned on, a minimum speed for operating the internal combustion engine, a maximum speed for operating the internal combustion engine, a speed of the internal combustion engine, a torque value of the internal combustion engine, and a speed at which the internal combustion engine is turned off
Optionally, objectives in the multi-objective optimization problem are: battery stress, operation changes, emission, noise, and battery state of charge.
Optionally, said determining the approximate function for the multi-objective optimization problem of the energy control management system based on the gradient boosted neural network algorithm specifically includes:
solving the approximate function by using a Non-dominated Sorting Genetic Algorithm II (NSGA-II), to obtain the optimal solution.
According to a second aspect, the present disclosure provides a multi-objective optimization system assisted by a gradient boosted neural network, including:
According to a third aspect, the present disclosure provides an electronic device including a memory and a processor, where the memory is configured to store a computer program, and the processor runs the computer program to enable the electronic device to perform the multi-objective optimization method assisted by a gradient boosted neural network as described in the first aspect.
According to specific embodiments provided in the present disclosure, the present disclosure has the following technical effects:
Based on the gradient boosted neural network algorithm, the present disclosure transforms a multi-objective optimization problem of an energy control management system into an approximate function and solves the approximate function based on a multi-objective optimization algorithm, thereby solving issues such as time-consuming function evaluation and inability to obtain an optimal solution of the problem.
To describe embodiments of the present disclosure or technical solutions in the prior art more clearly, the accompanying drawings required in the embodiments are briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure. Those of ordinary skill in the art can also obtain other accompanying drawings according to these accompanying drawings without creative efforts.
The technical solutions of the embodiments of the present disclosure are clearly and completely described below with reference to the drawings in the embodiments of the present disclosure. Apparently, the described embodiments are merely a part rather than all of the embodiments of the present disclosure. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.
To make the above objectives, features, and advantages of the present disclosure clearer and more comprehensible, the present disclosure will be further described in detail below with reference to the accompanying drawings and the specific embodiments.
An energy control management system of a hybrid electric vehicle faces complex and expensive multi-objective optimization problems. As shown in
The present disclosure focuses solely on optimizing the control management of the internal combustion engine 102.
The specific multi-objective optimization problem to be solved by the present disclosure is shown in Table 1 and Table 2.
In the above Table 1, the minimum speed (V1) and the maximum (V2) for operating the internal combustion engine 102 and the speed (Voff) at which the internal combustion engine 102 is turned off represent the vehicle speed, while the speed (Rev1, Rev2, Rev3) of the internal combustion engine 102 represent the revolutions per minute (RPM) of the engine internal combustion engine 102.
Hybrid electric vehicles need to be operated according to specific rules, as outlined below:
SOCmax is not the maximum value of the battery state of charge, but merely represents the level of the battery state of charge, and its state value indicates that the battery 103 is no longer charging.
Operations 1, 2, and 3 may correspond to the torque values and speeds of the internal combustion engine 102 defined in Table 1.
For example, in operation 1, the internal combustion engine 102 operates in the condition of Rev1 and Torque1; in operation 2, the internal combustion engine 102 operates in the condition of Rev2 and Torque2; and in operation 3, the internal combustion engine 102 operates in the condition of Rev3 and Torque3.
To address the complex and expensive multi-objective optimization problems in the energy control management system (also known as the hybrid power system), this present disclosure proposes using a neural network model to fit the function of the multi-objective black-box optimization problem. Due to the difficulty in obtaining gradient information for the problem, the present disclosure utilizes a Kriging model to fit the problem function before using the neural network model, thereby obtaining the gradient information for the problem.
Kriging is a method for predicting a function value at a given point by calculating a weighted average of function values of sample points in the vicinity of that point. This method is closely related to regression analysis in mathematics. Kriging interpolation is a linear unbiased optimal estimation interpolation method that not only fully considers a position relationship between an observation point and an estimation point but also takes into account a spatial relationship between the observation point and the estimation point.
The Kriging model can be expressed in the following form:
R(c, xi, xj) is a correlation coefficient function between any two points with parameter c, representing the correlation between any two points xi and xj. The correlation function relationship is as follows:
In the process of implementing a gradient boosting machine, it is assumed that the prediction function after m iterations is Fm(x) and the corresponding loss function is L(y, Fm(x)). To achieve the fastest reduction of the loss function, a sub-model function of the (m+1)-th iteration is constructed in the direction of gradient descent of the loss function, where the gradient descent direction at this time is:
Since the problem involved in the present disclosure is a multi-objective black-box optimization problem where gradient information of the problem is not available, to address this issue, the present disclosure trains a Kriging model to establish an approximate function for the black-box optimization problem, and then the gradient of the problem is calculated using the above formula.
The loss function adopts the common L2 approach, which calculates the square of a difference between a predicted value and a true value. The specific calculation formula is as follows:
A decision tree is used to construct the sub-model function −gm(x) of the (m+1)-th iteration as a target variable to be predicted (in the case of the L2 loss function, a new target variable is a residual between the latest prediction function and the original target variable). Once a new decision tree model is obtained, predicted values for all data are assumed to be h(x). Therefore, after the (m+1)-th iteration, the overall prediction function should be:
Based on the principle of the gradient boosted neural network algorithm, the implementation process can be broken down into the following steps:
In each iteration, a negative gradient of the current model across all samples is calculated first. Then, a new weak classifier for fitting is trained with the negative gradient as the objective, and the weight of the weak classifier is calculated, to ultimately update the model.
Based on the above content, this embodiment provides a multi-objective optimization method assisted by a gradient boosted neural network. As shown in
Design of Experiments (DOE) is a statistical-based experimental design method, for example a method for uniform sampling, such as random sampling, Latin hypercube sampling, low-discrepancy sequence sampling, etc. Specifically, in this embodiment, Latin hypercube sampling is used as the uniform sampling method. Alternatively, any other DOE can also be used as sampling methods.
In a preferred implementation, step 100 is specifically as follows:
The framework of the method provided in this embodiment, as shown in
As a preferred implementation, step 200 specifically includes: training the gradient boosted neural network based on the sample data and a Kriging model to obtain the approximate function for the multi-objective optimization problem.
To obtain gradient information for the problem, the Kriging model is trained to establish an approximate analytical expression for the multi-objective optimization problem, which is then used for calculating the gradient in the gradient boosted neural network in S103. S102 is merely a specific example of the overall algorithm process and is not limited to the Kriging model; other methods can also be used to obtain the gradient information. In the present disclosure, the Kriging model is first used to approximate the function to obtain gradient information, which is then incorporated into the gradient neural network for the final function fitting.
A GENN model is trained based on the sample data from S101 and the Kriging model from S102, and an approximate function of the multi-objective optimization problem is obtained, for solution set evaluation in S104.
As a preferred implementation, step 300 specifically includes: solving the approximate function by using a multi-objective optimization algorithm NSGA-II, to obtain the optimal solution.
The optimal solution is optimal values for the maximum threshold of the battery state of charge when the internal combustion engine 102 is turned off, the minimum threshold of the battery state of charge when the internal combustion engine 102 is turned on, the minimum speed for operating the internal combustion engine 102, the maximum speed for operating the internal combustion engine 102, the speeds of the internal combustion engine 102 in the first operation to the third operation, the torque values of the internal combustion engine 102 in the first operation to third operation, and the speed at which the internal combustion engine 102 is turned off.
In this embodiment, the existing classical and widely used multi-objective optimization algorithm NSGA-II is employed to handle the multi-objective optimization problem fitted in step 300, where other multi-objective optimization algorithms are also suitable to be embedded into the overall algorithm process.
Following the general process of multi-objective optimization algorithms, the implementation process of NSGA-II is illustrated in
In step S400, the working state of internal combustion engine 102 can be controlled by an power control system 101 of the energy control management system according to the optimal solution.
In some embodiments, the power control system 101 can execute the operation of turning off the internal combustion engine 102 according to the optimal solution. For example, if the current vehicle speed is lower than the optimal value of a speed Voff at which the internal combustion engine 102 is turned off, the internal combustion engine 102 is turned off; if the battery state of charge is greater than the optimal value of SOCmax, the internal combustion engine 102 is turned off; if the battery state of charge is less than the optimal value of SOCmin, the internal combustion engine 102 is turned on.
In some embodiments, the power control system 101 may execute the following operations according to the optimal solution: when the internal combustion engine 102 is on and the speed is lower than the optimal value of the minimum speed V1 for operating the internal combustion engine 102, the first operation is executed; when the internal combustion engine 102 is on and the speed is between the optimal value of minimum speed V1 for operating the internal combustion engine 102 and the optimal value of maximum speed V2 for operating the internal combustion engine 102, the second operation is executed; and when the internal combustion engine 102 is on and its speed is greater than the optimal value of maximum speed V2 for operating the internal combustion engine 102, the third operation is executed. During the first operation, the internal combustion engine 102 operates in the condition of optimal values of Rev1 and Torque1. During the second operation, the internal combustion engine 102 operates in the condition of optimal values of Rev2 and Torque2. During the third operation, the internal combustion engine 102 operates in the condition of optimal values of Rev3 and Torque3. Furthermore, the power control system 101 can execute predetermined operations according to the optimal solution, which correspond to the minimum speed, maximum speed, and torque values of the internal combustion engine 102. This enables the internal combustion engine 102 and electric motor 104 to operate in a working state that conforms to each other's physical work characteristics, thereby improves energy efficiency, reduces noise during vehicle operation, lowers emissions, and reduces battery load.
In order to perform the method in Embodiment 1 to achieve corresponding functions and technical effects, a multi-objective optimization system assisted by a gradient boosted neural network is provided below.
This embodiment provides a multi-objective optimization system assisted by a gradient boosted neural network, including:
This embodiment of the present disclosure provides an electronic device, including a memory and a processor, where the memory is configured to store a computer program, and the processor runs the computer program to enable the electronic device to execute the multi-objective optimization method assisted by a gradient boosted neural network in Embodiment 1.
Optionally, the electronic device may be a server.
In addition, an embodiment of the present disclosure further provides a computer readable storage medium storing a computer program. The computer program is executed by a processor to implement the multi-objective optimization method assisted by a gradient boosted neural network in Embodiment 1.
Each embodiment in the description is described in a progressive mode, each embodiment focuses on differences from other embodiments, and references can be made to each other for the same and similar parts between embodiments. Since the system disclosed in an embodiment corresponds to the method disclosed in an embodiment, the description is relatively simple, and for related contents, references can be made to the description of the method.
Particular examples are used herein for illustration of principles and implementation modes of the present disclosure. The descriptions of the above embodiments are merely used for assisting in understanding the method of the present disclosure and its core ideas. In addition, those of ordinary skill in the art can make various modifications in terms of particular implementation modes and the scope of application in accordance with the ideas of the present disclosure. In conclusion, the content of the description shall not be construed as limitations to the present disclosure.
Number | Date | Country | Kind |
---|---|---|---|
202310691691.3 | Jun 2023 | CN | national |
The present disclosure is a continuation-in-part of International Patent Application No. PCT/CN2024/072216, filed on Jan. 15, 2024, which claims priority to the Chinese Patent Application No. 202310691691.3 filed with the China National Intellectual Property Administration on Jun. 12, 2023, and entitled “MULTI-OBJECTIVE OPTIMIZATION METHOD AND SYSTEM ASSISTED BY GRADIENT BOOSTED NEURAL NETWORK, AND DEVICE”, which is incorporated herein by reference in its entirety.
Number | Date | Country | |
---|---|---|---|
Parent | PCT/CN2024/072216 | Jan 2024 | WO |
Child | 18985734 | US |