WO2022257377A1 - Automatic gear shifting control method based on fuzzy neural network - Google Patents

Automatic gear shifting control method based on fuzzy neural network Download PDF

Info

Publication number
WO2022257377A1
WO2022257377A1 PCT/CN2021/134389 CN2021134389W WO2022257377A1 WO 2022257377 A1 WO2022257377 A1 WO 2022257377A1 CN 2021134389 W CN2021134389 W CN 2021134389W WO 2022257377 A1 WO2022257377 A1 WO 2022257377A1
Authority
WO
WIPO (PCT)
Prior art keywords
layer
fuzzy
neural network
fuzzy neural
input
Prior art date
Application number
PCT/CN2021/134389
Other languages
French (fr)
Chinese (zh)
Inventor
季昌健
张志强
张鹏飞
陈国强
叶伟凡
崔恂
王枭鹏
尹兵
王晨宇
王可
Original Assignee
一汽奔腾轿车有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 一汽奔腾轿车有限公司 filed Critical 一汽奔腾轿车有限公司
Publication of WO2022257377A1 publication Critical patent/WO2022257377A1/en

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F16—ENGINEERING ELEMENTS AND UNITS; GENERAL MEASURES FOR PRODUCING AND MAINTAINING EFFECTIVE FUNCTIONING OF MACHINES OR INSTALLATIONS; THERMAL INSULATION IN GENERAL
    • F16H—GEARING
    • F16H61/00—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing
    • F16H61/02—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing characterised by the signals used
    • F16H61/0202—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing characterised by the signals used the signals being electric
    • F16H61/0204—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing characterised by the signals used the signals being electric for gearshift control, e.g. control functions for performing shifting or generation of shift signal
    • F16H61/0213—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing characterised by the signals used the signals being electric for gearshift control, e.g. control functions for performing shifting or generation of shift signal characterised by the method for generating shift signals
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F16—ENGINEERING ELEMENTS AND UNITS; GENERAL MEASURES FOR PRODUCING AND MAINTAINING EFFECTIVE FUNCTIONING OF MACHINES OR INSTALLATIONS; THERMAL INSULATION IN GENERAL
    • F16H—GEARING
    • F16H61/00—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing
    • F16H2061/0075—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing characterised by a particular control method
    • F16H2061/0081—Fuzzy logic
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F16—ENGINEERING ELEMENTS AND UNITS; GENERAL MEASURES FOR PRODUCING AND MAINTAINING EFFECTIVE FUNCTIONING OF MACHINES OR INSTALLATIONS; THERMAL INSULATION IN GENERAL
    • F16H—GEARING
    • F16H61/00—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing
    • F16H2061/0075—Control functions within control units of change-speed- or reversing-gearings for conveying rotary motion ; Control of exclusively fluid gearing, friction gearing, gearings with endless flexible members or other particular types of gearing characterised by a particular control method
    • F16H2061/0084—Neural networks

Definitions

  • the invention relates to the technical field of automobile control, in particular to an automatic transmission shift control method, in particular to a fuzzy neural network automatic shift control method based on an excellent driver's driving behavior model.
  • Vehicle power, economy and comfort are important indicators for measuring and evaluating vehicles, which not only depend on the quality of the engine, but also are closely related to the performance of the transmission.
  • the gear shift control of automatic transmission is basically based on the two parameters of accelerator pedal and vehicle speed. When the vehicle speed is greater than or less than the set threshold under a certain accelerator opening, the transmission performs upshift or downshift operation.
  • Two shift point calculation methods are mainly used: the first one is to calculate the optimal shift point according to the driving force curve of the adjacent gear of the vehicle; the second is to formulate the shift curve according to the excellent driver's operating habits.
  • the present invention provides a method for applying the T-S fuzzy neural network algorithm to automatic gear shift control.
  • the fuzzy algorithm and the neural network algorithm are integrated, and the trained model can use a small amount of fuzzy rules to generate complex Non-linear function, parameter identification and setting from the perspective of excellent driver operation, to improve the drivability of the vehicle.
  • a kind of automatic shift control method based on fuzzy neural network comprises the following steps:
  • Step 1 Collect the corresponding gear values of vehicles driven by excellent drivers at different accelerator openings and driving speeds;
  • Step 2 Select the T-S model to design the fuzzy neural network controller, and use the collected vehicle speed, accelerator pedal opening and corresponding gear information as training samples to train the training model;
  • the design of the fuzzy neural network controller includes the following steps:
  • the membership function in the front-end network is used to define the center parameter c ij of the horizontal axis position and the self-tuning algorithm of the parameter ⁇ ij defining the width of the fuzzy set;
  • Step 3 Embed the trained model into the shift controller.
  • the target gear is output through calculation;
  • Step 4 The output gear information is applied to the real vehicle system, so as to realize the control output.
  • step 2.1) front-end network calculation includes:
  • the front-end network is divided into four layers for self-tuning of fuzzy control rules:
  • the first layer is the input layer, and the control variables are accelerator pedal opening and vehicle speed, which are used as input samples for fuzzy neural network training; there are 2 neurons in the first layer, recorded as:
  • x 1 and x 2 respectively represent the actual input values of the accelerator pedal opening and vehicle speed of the current vehicle
  • the second layer is the membership function distribution layer, which divides the accelerator pedal opening and vehicle speed into three linguistic variables, and uses the Gaussian membership function for fuzzy processing; among them, the initial discourse domain of the accelerator pedal opening is set to [0,1 ], the language set is defined as [L (low), M (middle), H (high)]; the initial domain of discourse of the actual vehicle speed is set to [0,250], and the language set is defined as [L (low), M (middle), H (high)];
  • the membership function layer of the front-end network has a total of 6 neurons, denoted as:
  • i represents the serial number of the value generated by the first layer of nodes
  • j represents the number of fuzzy sets
  • the third layer is the fuzzy inference layer.
  • Each node is only associated with one of the fuzzy sets of each node in the previous layer.
  • ⁇ j represents the fitness of the jth layer
  • i 1 and i 2 respectively represent the numbers under the membership distribution standard of the second layer
  • the fourth layer implements normalization processing, and the calculation formula is:
  • the 2.2) back-end network calculation includes:
  • the back-end network is divided into three layers for parameter normalization processing, and its structure is consistent with the BP neural network:
  • the first layer is the input layer.
  • the second layer is the hidden layer, and the corresponding nodes correspond to the fuzzy inference rules.
  • the third layer is the output layer, and the output value calculated by the fuzzy neural network is only one, that is, the target gear; the controller uses weighted summation to calculate the final output variable:
  • said 2.3) optimizing for fuzzy neural network self-tuning parameters includes:
  • t represents the expected output of the system
  • y represents the actual output of the system
  • the self-tuning algorithm of the front-end network c ij and ⁇ ij is:
  • the first layer receiving system input:
  • the second layer completes the fuzzy processing of variables and uses the Gaussian function:
  • the third layer realizes the fuzzy reasoning process:
  • the fifth layer working together with the back-end network:
  • the fifth layer weight coefficient correction formula is:
  • ⁇ is the learning rate.
  • the feedback error signal of the fifth layer is:
  • the fifth layer satisfies the weight coefficient of the previous layer:
  • the gradient of the membership layer parameters can be calculated, as follows:
  • the learning algorithm of self-tuning parameters c ij and ⁇ ij can be expressed as:
  • the self-tuning algorithm of the backend network q ij includes:
  • the T-S fuzzy neural network selects a Gaussian membership function for fuzzification. After processing, the data distributed at both ends get a smaller degree of membership and are eliminated.
  • Fig. 1 is the automatic shifting principle of the fuzzy neural network of the present invention
  • Fig. 2 is learning and training algorithm principle of the present invention
  • Fig. 3 is the schematic diagram of T-S fuzzy neural network self-learning method of the present invention
  • Fig. 4 is different number of iterations and training accuracy error figure of the present invention
  • the present invention proposes a shift control strategy for an automatic transmission based on a fuzzy neural network method. Taking the driving habits of an excellent driver as a training sample, the training of the shift map is completed and used in the shift controller, as shown in Figure 1, specifically Include the following:
  • Step 1 Collect the corresponding gear values of vehicles driven by excellent drivers under different accelerator openings and driving speeds.
  • Step 2 Use the collected vehicle speed, accelerator pedal opening and corresponding gear information as training samples to train the training model, as shown in Figure 2.
  • the T-S model is selected to design the fuzzy neural network controller.
  • the controller is mainly divided into two parts. The first part is the back-end network, which is used for standard output; the second part is the front-end network, which is used for fuzzy rule matching. As shown in Figure 3, specifically include:
  • the front-end network is divided into four layers for self-tuning of fuzzy control rules:
  • the first layer is the input layer. It is used to input accurate control variables in the system to the next layer.
  • the control variables of the present invention are: accelerator pedal opening and vehicle speed, which are used as input samples for fuzzy neural network training.
  • x 1 and x 2 respectively represent the actual input values of the accelerator pedal opening and vehicle speed of the current vehicle.
  • the second layer is the distribution layer of membership function, which is used to transform the precise variables input by the first layer into fuzzy control variables.
  • the accelerator pedal opening and vehicle speed are divided into three language variables, and the Gaussian membership function is used for fuzzy processing.
  • the initial domain of discourse of the accelerator pedal opening is set to [0,1]
  • the language set is defined as [L (low), M (medium), H (high)]
  • the initial domain of discourse of the actual vehicle speed is set to [0,250]
  • the language The set is defined as [L (low), M (medium), H (high)].
  • the membership function layer of the front-end network has a total of 6 neurons, denoted as:
  • i represents the serial number of the value generated by the first layer of nodes
  • j represents the number of fuzzy sets.
  • the third layer is the fuzzy inference layer.
  • Each node is associated with one of the fuzzy sets only with each node in the previous layer.
  • calculate the fitness value of the above nine nodes which is recorded as:
  • ⁇ j represents the fitness degree of the jth layer
  • i 1 and i 2 respectively represent the numbers under the membership degree allocation standard of the second layer
  • j 9.
  • the fourth layer implements normalization processing.
  • the calculation formula is:
  • the back-end network is divided into three layers for parameter normalization processing, and its structure is consistent with the BP neural network:
  • the first layer is the input layer.
  • the second layer is the hidden layer, and the corresponding nodes correspond to the fuzzy inference rules, and also have 9 nodes, which are used to calculate the back-end value of the conditional rules. Referred to as:
  • the third layer is the output layer.
  • the output value calculated by the fuzzy neural network is only one, that is, the target gear.
  • the controller uses a weighted sum to calculate the final output variable.
  • the parameters that need to be self-tuned when the fuzzy neural network is controlled are: the membership function is used to define the center parameter c ij of the horizontal axis position and the parameter ⁇ ij that defines the width of the fuzzy set is used for the connection of the back-end BP structure neural network Weight q ij .
  • t represents the expected output of the system
  • y represents the actual output of the system
  • the first layer receiving system input:
  • the second layer completes the fuzzy processing of variables and uses the Gaussian function:
  • the third layer realizes the fuzzy reasoning process:
  • the fifth layer working together with the back-end network:
  • the fifth layer weight coefficient correction formula is:
  • ⁇ is the learning rate.
  • the feedback error signal of the fifth layer is:
  • the fifth layer satisfies the weight coefficient of the previous layer:
  • the gradient of the membership layer parameters can be calculated, as follows:
  • the learning algorithm of self-tuning parameters c ij and ⁇ ij can be expressed as:
  • Step 3 Embed the trained model into the shift controller.
  • the target gear is output through calculation.
  • Step 4 The output gear information is applied to the real vehicle system to realize the control output.
  • Figure 4 shows the relationship between the mean square error and the number of iterations after the fuzzy neural network meets the system design requirements. It can be seen from Fig. 4 that the mean square error of the fuzzy neural network designed by the present invention is reduced to less than 0.1 after 1000 iteration calculations; after 2000 iteration calculations, the mean square error is reduced to less than 0.0001. And with the increase of the number of iterations, the final model of the fuzzy neural network after parameter adjustment is in a state of convergence, and the learning ability is strong. Under the condition of the same learning rate, the larger the number of iteration parameters selected by the controller, the smaller the error after the controller learns.
  • the system error is basically stable within the range of ⁇ 0.01, but too large number of iterations will easily lead to system failure.
  • the training time is too long. Then select the driving shift map of an excellent driver, select the data sampling points for model training, and obtain the actual gear after model training and the training error.

Abstract

Disclosed in the present invention is an automatic gear shifting control method based on a fuzzy neural network. The method comprises the following steps: collecting gear values corresponding to different throttle openings and different traveling speeds of a vehicle driven by an excellent driver; designing a fuzzy neural network controller by using a T-S model, and training a training model by taking the collected vehicle speeds and acceleration pedal openings and corresponding gear information as training samples; embedding the trained model into a gear shifting controller, and when an actual acceleration pedal opening and an actual vehicle speed are input into the gear shifting controller, outputting a target gear by means of calculation; and output gear information acting on a real vehicle system, thereby realizing a control output. By means of the present invention, a fuzzy algorithm and a neural network algorithm are integrated and used, such that a trained model can generate a complex nonlinear function by using a small number of fuzzy rules, and parameter identification and setting are performed from the perspective of an operation of an excellent driver, thereby improving the drivability of a vehicle.

Description

一种基于模糊神经网络的自动换挡控制方法A Control Method of Automatic Shifting Based on Fuzzy Neural Network 技术领域technical field
本发明涉及汽车控制技术领域，尤其涉及一种自动变速器换挡控制方法，具体为一种基于优秀驾驶员驾驶行为模型的模糊神经网络自动换挡控制方法。The invention relates to the technical field of automobile control, in particular to an automatic transmission shift control method, in particular to a fuzzy neural network automatic shift control method based on an excellent driver's driving behavior model.
背景技术Background technique
车辆动力性、经济性以及舒适性是衡量与评价车辆的重要指标，这不仅取决于发动机的优劣，而且与变速器的性能也息息相关。目前自动变速器换挡控制基本都是基于加速踏板和车辆速度两个参数，当某一油门开度下，车速大于或者小于设定的阈值后，变速器实施升档或者降档操作。主要采用两种换挡点计算方法：第一种根据车辆相邻挡位驱动力曲线计算最优换挡点；第二种是根据优秀的驾驶员操作习惯制定换挡曲线。Vehicle power, economy and comfort are important indicators for measuring and evaluating vehicles, which not only depend on the quality of the engine, but also are closely related to the performance of the transmission. At present, the gear shift control of automatic transmission is basically based on the two parameters of accelerator pedal and vehicle speed. When the vehicle speed is greater than or less than the set threshold under a certain accelerator opening, the transmission performs upshift or downshift operation. Two shift point calculation methods are mainly used: the first one is to calculate the optimal shift point according to the driving force curve of the adjacent gear of the vehicle; the second is to formulate the shift curve according to the excellent driver's operating habits.
国内对自动换挡策略的方法主要集中在控制策略上，且常使用神经网络算法和模糊算法。从提高算法收敛速度、提高车辆动力性、增加车辆变速器系统对外部干扰的鲁棒性等方面进行改进，进行算法集成控制策略较少。因此，自动变速器换挡控制方法需要进一步关注。Domestic methods for automatic shifting strategies mainly focus on control strategies, and neural network algorithms and fuzzy algorithms are often used. Improvements are made in terms of increasing the convergence speed of the algorithm, improving vehicle dynamics, and increasing the robustness of the vehicle transmission system to external disturbances, and there are few algorithm integration control strategies. Therefore, the shift control method of automatic transmission needs further attention.
发明内容Contents of the invention
本发明针对目前存在的问题，提供了一种将T-S模糊神经网络算法应用到自动换挡控制上的方法，集成使用模糊算法和神经网络算法，经过训练的模型可使用少量的模糊规则生成复杂的非线性函数，从优秀驾驶员操作角度进行参数辨识与整定，提高整车驾驶性。Aiming at the current problems, the present invention provides a method for applying the T-S fuzzy neural network algorithm to automatic gear shift control. The fuzzy algorithm and the neural network algorithm are integrated, and the trained model can use a small amount of fuzzy rules to generate complex Non-linear function, parameter identification and setting from the perspective of excellent driver operation, to improve the drivability of the vehicle.
本发明的目的是通过以下技术方案实现的：The purpose of the present invention is achieved through the following technical solutions:
一种基于模糊神经网络的自动换挡控制方法，包括以下步骤：A kind of automatic shift control method based on fuzzy neural network, comprises the following steps:
步骤一、采集优秀驾驶员驾驶车辆在不同油门开度以及行驶车速下对应的挡位值；Step 1. Collect the corresponding gear values of vehicles driven by excellent drivers at different accelerator openings and driving speeds;
步骤二、选用T-S模型进行模糊神经网络控制器设计，将采集的车速、加速踏板开度以及对应的挡位信息作为训练样本，对训练模型进行训练； Step 2. Select the T-S model to design the fuzzy neural network controller, and use the collected vehicle speed, accelerator pedal opening and corresponding gear information as training samples to train the training model;
模糊神经网络控制器设计包括以下步骤：The design of the fuzzy neural network controller includes the following steps:
2.1)前端网络计算；2.1) Front-end network computing;
2.2)后端网络计算；2.2) Back-end network computing;
2.3)针对模糊神经网络自整定参数进行优化2.3) Optimizing the self-tuning parameters of the fuzzy neural network
2.4)前端网络中隶属度函数用于定义横轴位置的中心参数c ij和定义模糊集宽度的参数 σ ij的自整定算法； 2.4) The membership function in the front-end network is used to define the center parameter c ij of the horizontal axis position and the self-tuning algorithm of the parameter σ ij defining the width of the fuzzy set;
2.5)后端网络中用于后端BP结构神经网络的连接权值q ij的自整定算法； 2.5) The self-tuning algorithm for the connection weight q ij of the back-end BP structure neural network in the back-end network;
步骤三、将训练好的模型嵌入换挡控制器中，当实际加速踏板开度和车速输入至换挡控制器，经过计算输出目标挡位；Step 3. Embed the trained model into the shift controller. When the actual accelerator pedal opening and vehicle speed are input to the shift controller, the target gear is output through calculation;
步骤四、输出的挡位信息作用到实车系统，从而实现控制输出。Step 4: The output gear information is applied to the real vehicle system, so as to realize the control output.
进一步地，所述步骤2.1)前端网络计算包括：Further, the step 2.1) front-end network calculation includes:
前端网络分为四层，用于模糊控制规则自整定：The front-end network is divided into four layers for self-tuning of fuzzy control rules:
第一层为输入层，控制变量为油门踏板开度、车速，作为模糊神经网络训练的输入样本；第一层存在2个神经元，记作：The first layer is the input layer, and the control variables are accelerator pedal opening and vehicle speed, which are used as input samples for fuzzy neural network training; there are 2 neurons in the first layer, recorded as:
x＝[x 1,x 2] T x＝[x 1 ,x 2 ] T
其中，x 1,x 2分别表示当前车辆的油门踏板开度、车速的真实输入值； Among them, x 1 and x 2 respectively represent the actual input values of the accelerator pedal opening and vehicle speed of the current vehicle;
第二层为隶属度函数分配层，将油门踏板开度、车速划分为三个语言变量，选用高斯隶属度函数进行模糊化处理；其中，油门踏板开度的初始论域设置为[0,1]，语言合集定义为[L(低),M(中),H(高)]；实际车速初始论域设置为[0,250]，语言合集定义为[L(低),M(中),H(高)]；The second layer is the membership function distribution layer, which divides the accelerator pedal opening and vehicle speed into three linguistic variables, and uses the Gaussian membership function for fuzzy processing; among them, the initial discourse domain of the accelerator pedal opening is set to [0,1 ], the language set is defined as [L (low), M (middle), H (high)]; the initial domain of discourse of the actual vehicle speed is set to [0,250], and the language set is defined as [L (low), M (middle), H (high)];
前端网络的隶属度函数层共计6个神经元，记作：The membership function layer of the front-end network has a total of 6 neurons, denoted as:
Figure PCTCN2021134389-appb-000001
Figure PCTCN2021134389-appb-000001
其中，i表示第一层节点产生的值的序号，j表示模糊集合个数；Among them, i represents the serial number of the value generated by the first layer of nodes, and j represents the number of fuzzy sets;
第三层为模糊推理层，每个节点仅与上一层每个节点的其中一个模糊集合进行关联，共计存在9个节点，即神经网络控制器同时存在9条模糊推理规则；依据匹配规则，计算上述9个节点的适应度值，记作：The third layer is the fuzzy inference layer. Each node is only associated with one of the fuzzy sets of each node in the previous layer. There are 9 nodes in total, that is, the neural network controller has 9 fuzzy inference rules at the same time; according to the matching rules, Calculate the fitness value of the above 9 nodes, denoted as:
Figure PCTCN2021134389-appb-000002
Figure PCTCN2021134389-appb-000002
其中，α j表示第j层的适应度，i 1，i 2分别表示第二层隶属度分配标准下个数，j＝9，依据隶属度函数定义可知，当训练数据或给定输入在高斯函数赋值附近，对应的语言变量达到较大的值，因此模糊推理方法能够将满足给定输入下的数据计算及筛选。 Among them, α j represents the fitness of the jth layer, i 1 and i 2 respectively represent the numbers under the membership distribution standard of the second layer, j=9, according to the definition of the membership function, when the training data or the given input is in the Gaussian Near the function assignment, the corresponding language variable reaches a larger value, so the fuzzy reasoning method can calculate and filter the data satisfying the given input.
第四层实现归一化处理，计算公式为：The fourth layer implements normalization processing, and the calculation formula is:
Figure PCTCN2021134389-appb-000003
Figure PCTCN2021134389-appb-000003
进一步地，所述2.2)后端网络计算包括：Further, the 2.2) back-end network calculation includes:
后端网络分为三层，用于参数归一化处理，其结构与BP神经网络一致：The back-end network is divided into three layers for parameter normalization processing, and its structure is consistent with the BP neural network:
第一层为输入层，该层共设计3个处理节点节点，其中包含系统输入量以及用于产生常数项的输入恒值；The first layer is the input layer. There are three processing node nodes designed in this layer, which include the system input and the input constant value used to generate constant items;
第二层为隐藏层，相应的节点与模糊推理规则对应，同样具有9个节点，用于计算条件规则的后端值，记作：The second layer is the hidden layer, and the corresponding nodes correspond to the fuzzy inference rules. There are also 9 nodes, which are used to calculate the back-end value of the conditional rules, which are recorded as:
y j＝q j0+q j1x 1+q j2x 2 y j ＝q j0 +q j1 x 1 +q j2 x 2
第三层为输出层，经过模糊神经网络计算后的输出值只有1个，即目标挡位；控制器使用加权求和的方式计算最终输出变量：The third layer is the output layer, and the output value calculated by the fuzzy neural network is only one, that is, the target gear; the controller uses weighted summation to calculate the final output variable:
Figure PCTCN2021134389-appb-000004
Figure PCTCN2021134389-appb-000004
进一步地，所述2.3)针对模糊神经网络自整定参数进行优化包括：Further, said 2.3) optimizing for fuzzy neural network self-tuning parameters includes:
令误差函数为：Let the error function be:
Figure PCTCN2021134389-appb-000005
Figure PCTCN2021134389-appb-000005
其中，t表示系统期望输出，y表示系统实际输出。Among them, t represents the expected output of the system, and y represents the actual output of the system.
进一步地，所述2.4)中，前端网络c ij和σ ij的自整定算法为： Further, in the above 2.4), the self-tuning algorithm of the front-end network c ij and σ ij is:
(a)正向传递信息(a) forward information
第一层，接收系统输入：The first layer, receiving system input:
Figure PCTCN2021134389-appb-000006
Figure PCTCN2021134389-appb-000006
第二层，完成变量模糊化处理，选用高斯函数：The second layer completes the fuzzy processing of variables and uses the Gaussian function:
Figure PCTCN2021134389-appb-000007
Figure PCTCN2021134389-appb-000007
Figure PCTCN2021134389-appb-000008
Figure PCTCN2021134389-appb-000008
其中，i＝1,2,3…n；j＝1,2,3…m分别表示第i个输入节点经过模糊化处理生成j个模糊合集，c ij,σ ij分别表示生成高斯隶属度函数的中心值和宽度值。 Among them, i=1,2,3...n; j=1,2,3...m respectively represent that the i-th input node is fuzzified to generate j fuzzy sets, c ij , σ ij respectively represent the generation of Gaussian membership function The center value and width value of .
第三层，实现模糊推理过程：The third layer realizes the fuzzy reasoning process:
Figure PCTCN2021134389-appb-000009
Figure PCTCN2021134389-appb-000009
第四层，解模糊归一化过程：The fourth layer, defuzzification normalization process:
Figure PCTCN2021134389-appb-000010
Figure PCTCN2021134389-appb-000010
第五层，与后端网络共同作用：The fifth layer, working together with the back-end network:
Figure PCTCN2021134389-appb-000011
Figure PCTCN2021134389-appb-000011
(b)误差反向传播(b) Error Back Propagation
第五层权值系数修正公式为：The fifth layer weight coefficient correction formula is:
Figure PCTCN2021134389-appb-000012
Figure PCTCN2021134389-appb-000012
其中，β为学习率。Among them, β is the learning rate.
则，第五层反馈误差信号为：Then, the feedback error signal of the fifth layer is:
Figure PCTCN2021134389-appb-000013
Figure PCTCN2021134389-appb-000013
此时第五层对前一层权值系数满足：At this time, the fifth layer satisfies the weight coefficient of the previous layer:
Figure PCTCN2021134389-appb-000014
Figure PCTCN2021134389-appb-000014
按照相同构造方式依次向前传递：Pass forward sequentially according to the same construction method:
Figure PCTCN2021134389-appb-000015
Figure PCTCN2021134389-appb-000015
Figure PCTCN2021134389-appb-000016
Figure PCTCN2021134389-appb-000016
Figure PCTCN2021134389-appb-000017
Figure PCTCN2021134389-appb-000017
当f 3计算使用乘法计算中间偏差偏导时，有： When the f 3 calculation uses multiplication to calculate the partial derivative of the intermediate deviation, there is:
Figure PCTCN2021134389-appb-000018
Figure PCTCN2021134389-appb-000018
经由上式，可以计算出隶属层参数的梯度，有：Through the above formula, the gradient of the membership layer parameters can be calculated, as follows:
Figure PCTCN2021134389-appb-000019
Figure PCTCN2021134389-appb-000019
Figure PCTCN2021134389-appb-000020
Figure PCTCN2021134389-appb-000020
考虑学习率，将学习率代入上式，则有：Considering the learning rate and substituting the learning rate into the above formula, there are:
Figure PCTCN2021134389-appb-000021
Figure PCTCN2021134389-appb-000021
Figure PCTCN2021134389-appb-000022
Figure PCTCN2021134389-appb-000022
因此，自整定的参数c ij、σ ij的学习算法可表达为： Therefore, the learning algorithm of self-tuning parameters c ij and σ ij can be expressed as:
Figure PCTCN2021134389-appb-000023
Figure PCTCN2021134389-appb-000023
Figure PCTCN2021134389-appb-000024
Figure PCTCN2021134389-appb-000024
进一步地，所述2.5)中，后端网络q ij的自整定算法包括： Further, in said 2.5), the self-tuning algorithm of the backend network q ij includes:
Figure PCTCN2021134389-appb-000025
Figure PCTCN2021134389-appb-000025
其中：i＝1,2,3…n；j＝1,2,3…m，则有Among them: i=1,2,3...n; j=1,2,3...m, then have
Figure PCTCN2021134389-appb-000026
Figure PCTCN2021134389-appb-000026
进一步地，所述步骤二中，T-S模糊神经网络选用高斯隶属度函数进行模糊化处理。使得处理后分布在两端的数据获得较小的隶属度被剔除。Further, in the second step, the T-S fuzzy neural network selects a Gaussian membership function for fuzzification. After processing, the data distributed at both ends get a smaller degree of membership and are eliminated.
附图说明Description of drawings
图1为本发明模糊神经网络自动换挡原理Fig. 1 is the automatic shifting principle of the fuzzy neural network of the present invention
图2为本发明学习与训练算法原理Fig. 2 is learning and training algorithm principle of the present invention
图3为本发明T-S模糊神经网络自学习方法示意图Fig. 3 is the schematic diagram of T-S fuzzy neural network self-learning method of the present invention
图4为本发明不同迭代次数与训练精度误差图Fig. 4 is different number of iterations and training accuracy error figure of the present invention
具体实施方式Detailed ways
下面结合附图对本发明所提出的技术方案做出进一步阐述和说明。The technical solutions proposed by the present invention will be further elaborated and described below in conjunction with the accompanying drawings.
本发明提出一种基于模糊神经网络方法的自动变速器换挡控制策略，以优秀驾驶员驾驶习惯为训练样本，完成换挡图谱的训练并使用至换挡控制器中，如图1所示，具体包括以下内容：The present invention proposes a shift control strategy for an automatic transmission based on a fuzzy neural network method. Taking the driving habits of an excellent driver as a training sample, the training of the shift map is completed and used in the shift controller, as shown in Figure 1, specifically Include the following:
步骤一：采集优秀驾驶员驾驶车辆在不同油门开度以及行驶车速下对应的挡位值。Step 1: Collect the corresponding gear values of vehicles driven by excellent drivers under different accelerator openings and driving speeds.
步骤二：将采集的车速、加速踏板开度以及对应的挡位信息作为训练样本，对训练模型进行训练，如图2所示。选用T-S模型进行模糊神经网络控制器设计，控制器主要分为两部 分，第一部分是后端网络，目的是用于规范输出；第二部分是前端网络，目的是用于模糊规则匹配。如图3所示，具体包括：Step 2: Use the collected vehicle speed, accelerator pedal opening and corresponding gear information as training samples to train the training model, as shown in Figure 2. The T-S model is selected to design the fuzzy neural network controller. The controller is mainly divided into two parts. The first part is the back-end network, which is used for standard output; the second part is the front-end network, which is used for fuzzy rule matching. As shown in Figure 3, specifically include:
(1)前端网络计算(1) Front-end network computing
前端网络分为四层，用于模糊控制规则自整定：The front-end network is divided into four layers for self-tuning of fuzzy control rules:
第一层为输入层。用于将系统中准确的控制变量输入至下一层，本发明控制变量为：油门踏板开度、车速，作为模糊神经网络训练的输入样本。第一层存在2个神经元，记作：The first layer is the input layer. It is used to input accurate control variables in the system to the next layer. The control variables of the present invention are: accelerator pedal opening and vehicle speed, which are used as input samples for fuzzy neural network training. There are 2 neurons in the first layer, denoted as:
x＝[x 1,x 2] T x＝[x 1 ,x 2 ] T
其中，x 1,x 2分别表示当前车辆的油门踏板开度、车速的真实输入值。 Among them, x 1 and x 2 respectively represent the actual input values of the accelerator pedal opening and vehicle speed of the current vehicle.
第二层为隶属度函数分配层，用于将第一层输入的精确变量转化成模糊控制变量。将油门踏板开度、车速划分为三个语言变量，选用高斯隶属度函数进行模糊化处理。其中油门踏板开度的初始论域设置为[0,1]，语言合集定义为[L(低),M(中),H(高)]；实际车速初始论域设置为[0,250]，语言合集定义为[L(低),M(中),H(高)]。The second layer is the distribution layer of membership function, which is used to transform the precise variables input by the first layer into fuzzy control variables. The accelerator pedal opening and vehicle speed are divided into three language variables, and the Gaussian membership function is used for fuzzy processing. The initial domain of discourse of the accelerator pedal opening is set to [0,1], the language set is defined as [L (low), M (medium), H (high)]; the initial domain of discourse of the actual vehicle speed is set to [0,250], the language The set is defined as [L (low), M (medium), H (high)].
前端网络的隶属度函数层共计6个神经元，记作：The membership function layer of the front-end network has a total of 6 neurons, denoted as:
Figure PCTCN2021134389-appb-000027
Figure PCTCN2021134389-appb-000027
其中，i表示第一层节点产生的值的序号，j表示模糊集合个数。Among them, i represents the serial number of the value generated by the first layer of nodes, and j represents the number of fuzzy sets.
第三层为模糊推理层，每个节点与仅与上一层每个节点的其中一个模糊集合进行关联，共计存在9个节点，说明神经网络控制器同时存在9条模糊推理规则。依据匹配规则，计算上述9个节点的适应度值，记作：The third layer is the fuzzy inference layer. Each node is associated with one of the fuzzy sets only with each node in the previous layer. There are a total of 9 nodes, indicating that there are 9 fuzzy inference rules in the neural network controller. According to the matching rules, calculate the fitness value of the above nine nodes, which is recorded as:
Figure PCTCN2021134389-appb-000028
Figure PCTCN2021134389-appb-000028
其中，α j表示第j层的适应度，i 1，i 2分别表示第二层隶属度分配标准下个数，j＝9。依据隶属度函数定义可知，当训练数据或给定输入在高斯函数赋值附近，对应的语言变量达到较大的值，因此模糊推理方法能够将满足给定输入下的数据计算及筛选。 Among them, α j represents the fitness degree of the jth layer, i 1 and i 2 respectively represent the numbers under the membership degree allocation standard of the second layer, and j=9. According to the definition of the membership function, when the training data or the given input is near the Gaussian function assignment, the corresponding linguistic variable reaches a larger value, so the fuzzy reasoning method can calculate and filter the data satisfying the given input.
第四层实现归一化处理。为保证控输出变量一致性，计算公式为：The fourth layer implements normalization processing. In order to ensure the consistency of the control output variables, the calculation formula is:
Figure PCTCN2021134389-appb-000029
Figure PCTCN2021134389-appb-000029
(2)后端网络计算(2) Back-end network computing
后端网络分为三层，用于参数归一化处理，其结构与BP神经网络一致：The back-end network is divided into three layers for parameter normalization processing, and its structure is consistent with the BP neural network:
第一层为输入层。该层共设计3个处理节点节点，其中包含系统输入量以及用于产生常数项的输入恒值。The first layer is the input layer. There are 3 processing node nodes designed in this layer, which contain the input quantity of the system and the input constant value used to generate the constant item.
第二层为隐藏层，相应的节点与模糊推理规则对应，同样具有9个节点，用于计算条件规则的后端值。记作：The second layer is the hidden layer, and the corresponding nodes correspond to the fuzzy inference rules, and also have 9 nodes, which are used to calculate the back-end value of the conditional rules. Referred to as:
y j＝q j0+q j1x 1+q j2x 2 y j ＝q j0 +q j1 x 1 +q j2 x 2
第三层为输出层。经过模糊神经网络计算后的输出值只有1个，即目标挡位。控制器使用加权求和的方式计算最终输出变量。The third layer is the output layer. The output value calculated by the fuzzy neural network is only one, that is, the target gear. The controller uses a weighted sum to calculate the final output variable.
Figure PCTCN2021134389-appb-000030
Figure PCTCN2021134389-appb-000030
模糊神经网络再进行控制时需要进行自整定的参数分别为：隶属度函数用于定义横轴位置的中心参数c ij和定义模糊集宽度的参数σ ij，用于后端BP结构神经网络的连接权值q ij。 The parameters that need to be self-tuned when the fuzzy neural network is controlled are: the membership function is used to define the center parameter c ij of the horizontal axis position and the parameter σ ij that defines the width of the fuzzy set is used for the connection of the back-end BP structure neural network Weight q ij .
进一步针对模糊神经网络自整定参数进行优化：Further optimize the self-tuning parameters of the fuzzy neural network:
令误差函数为：Let the error function be:
Figure PCTCN2021134389-appb-000031
Figure PCTCN2021134389-appb-000031
其中，t表示系统期望输出，y表示系统实际输出。Among them, t represents the expected output of the system, and y represents the actual output of the system.
(3)前端网络c ij和σ ij的自整定算法 (3) Self-tuning algorithm of front-end network c ij and σ ij
(a)正向传递信息(a) forward information
第一层，接收系统输入：The first layer, receiving system input:
Figure PCTCN2021134389-appb-000032
Figure PCTCN2021134389-appb-000032
第二层，完成变量模糊化处理，选用高斯函数：The second layer completes the fuzzy processing of variables and uses the Gaussian function:
Figure PCTCN2021134389-appb-000033
Figure PCTCN2021134389-appb-000033
Figure PCTCN2021134389-appb-000034
Figure PCTCN2021134389-appb-000034
其中，i＝1,2,3…n；j＝1,2,3…m分别表示第i个输入节点经过模糊化处理生成j个模糊合集，c ij,σ ij分别表示生成高斯隶属度函数的中心值和宽度值。 Among them, i=1,2,3...n; j=1,2,3...m respectively represent that the i-th input node is fuzzified to generate j fuzzy sets, c ij , σ ij respectively represent the generation of Gaussian membership function The center value and width value of .
第三层，实现模糊推理过程：The third layer realizes the fuzzy reasoning process:
Figure PCTCN2021134389-appb-000035
Figure PCTCN2021134389-appb-000035
第四层，解模糊归一化过程：The fourth layer, defuzzification normalization process:
Figure PCTCN2021134389-appb-000036
Figure PCTCN2021134389-appb-000036
第五层，与后端网络共同作用：The fifth layer, working together with the back-end network:
Figure PCTCN2021134389-appb-000037
Figure PCTCN2021134389-appb-000037
(b)误差反向传播(b) Error Back Propagation
第五层权值系数修正公式为：The fifth layer weight coefficient correction formula is:
Figure PCTCN2021134389-appb-000038
Figure PCTCN2021134389-appb-000038
其中，β为学习率。Among them, β is the learning rate.
则，第五层反馈误差信号为：Then, the feedback error signal of the fifth layer is:
Figure PCTCN2021134389-appb-000039
Figure PCTCN2021134389-appb-000039
此时第五层对前一层权值系数满足：At this time, the fifth layer satisfies the weight coefficient of the previous layer:
Figure PCTCN2021134389-appb-000040
Figure PCTCN2021134389-appb-000040
按照相同构造方式依次向前传递：Pass forward sequentially according to the same construction method:
Figure PCTCN2021134389-appb-000041
Figure PCTCN2021134389-appb-000041
Figure PCTCN2021134389-appb-000042
Figure PCTCN2021134389-appb-000042
Figure PCTCN2021134389-appb-000043
Figure PCTCN2021134389-appb-000043
当f 3计算使用乘法计算中间偏差偏导时，有： When the f 3 calculation uses multiplication to calculate the partial derivative of the intermediate deviation, there is:
Figure PCTCN2021134389-appb-000044
Figure PCTCN2021134389-appb-000044
经由上式，可以计算出隶属层参数的梯度，有：Through the above formula, the gradient of the membership layer parameters can be calculated, as follows:
Figure PCTCN2021134389-appb-000045
Figure PCTCN2021134389-appb-000045
Figure PCTCN2021134389-appb-000046
Figure PCTCN2021134389-appb-000046
考虑学习率，将学习率代入上式，则有：Considering the learning rate and substituting the learning rate into the above formula, there are:
Figure PCTCN2021134389-appb-000047
Figure PCTCN2021134389-appb-000047
Figure PCTCN2021134389-appb-000048
Figure PCTCN2021134389-appb-000048
因此，自整定的参数c ij、σ ij的学习算法可表达为： Therefore, the learning algorithm of self-tuning parameters c ij and σ ij can be expressed as:
Figure PCTCN2021134389-appb-000049
Figure PCTCN2021134389-appb-000049
Figure PCTCN2021134389-appb-000050
Figure PCTCN2021134389-appb-000050
(4)后端网络q ij的自整定算法 (4) Self-tuning algorithm of backend network q ij
Figure PCTCN2021134389-appb-000051
Figure PCTCN2021134389-appb-000051
其中：i＝1,2,3…n；j＝1,2,3…m，则有Among them: i=1,2,3...n; j=1,2,3...m, then have
Figure PCTCN2021134389-appb-000052
Figure PCTCN2021134389-appb-000052
步骤三：将训练好的模型嵌入换挡控制器中，当实际加速踏板开度和车速输入至换挡控制器，经过计算输出目标挡位。Step 3: Embed the trained model into the shift controller. When the actual accelerator pedal opening and vehicle speed are input to the shift controller, the target gear is output through calculation.
步骤四：输出的挡位信息作用到实车系统，从而实现控制输出。Step 4: The output gear information is applied to the real vehicle system to realize the control output.
下面给出本发明所提供技术方案的仿真实验数据。The simulation experiment data of the technical solution provided by the present invention is given below.
首先针对不同迭代次数和训练精度下模糊神经网络方法的误差。如图4所示为模糊神经网络达到系统设计要求后的均方误差与迭代次数之间的关系。由图4可知本发明设计的模糊神经网络经过1000次迭代计算后，均方误差降低至0.1一下；经过2000次迭代计算后，均方误差减小至0.0001以下。并且随着迭代次数的增加，模糊神经网络经过参数调整的最终模型处于收敛状态，学习能力较强。相同学习率条件下，控制器选择的迭代次数参数越大，控制器学习后误差越小，当迭代次数达到2000左右时，系统误差基本稳定在±0.01范围内，但迭代次数过大容易导致系统训练时间过长。随后选取优秀驾驶员的驾驶换挡图谱，选取其中的数据采样点进行模型训练，可得到模型训练后的实际挡位以及训练误差。Firstly, the error of the fuzzy neural network method under different iterations and training accuracy is aimed at. Figure 4 shows the relationship between the mean square error and the number of iterations after the fuzzy neural network meets the system design requirements. It can be seen from Fig. 4 that the mean square error of the fuzzy neural network designed by the present invention is reduced to less than 0.1 after 1000 iteration calculations; after 2000 iteration calculations, the mean square error is reduced to less than 0.0001. And with the increase of the number of iterations, the final model of the fuzzy neural network after parameter adjustment is in a state of convergence, and the learning ability is strong. Under the condition of the same learning rate, the larger the number of iteration parameters selected by the controller, the smaller the error after the controller learns. When the number of iterations reaches about 2000, the system error is basically stable within the range of ±0.01, but too large number of iterations will easily lead to system failure. The training time is too long. Then select the driving shift map of an excellent driver, select the data sampling points for model training, and obtain the actual gear after model training and the training error.

Claims (7)

  1. 一种基于模糊神经网络的自动换挡控制方法，其特征在于，包括以下步骤：A kind of automatic shift control method based on fuzzy neural network, it is characterized in that, comprises the following steps:
    步骤一、采集优秀驾驶员驾驶车辆在不同油门开度以及行驶车速下对应的挡位值；Step 1. Collect the corresponding gear values of vehicles driven by excellent drivers at different accelerator openings and driving speeds;
    步骤二、选用T-S模型进行模糊神经网络控制器设计，将采集的车速、加速踏板开度以及对应的挡位信息作为训练样本，对训练模型进行训练；Step 2. Select the T-S model to design the fuzzy neural network controller, and use the collected vehicle speed, accelerator pedal opening and corresponding gear information as training samples to train the training model;
    模糊神经网络控制器设计包括以下步骤：The design of the fuzzy neural network controller includes the following steps:
    2.1)前端网络计算；2.1) Front-end network computing;
    2.2)后端网络计算；2.2) Back-end network computing;
    2.3)针对模糊神经网络自整定参数进行优化；2.3) Optimizing the self-tuning parameters of the fuzzy neural network;
    2.4)前端网络中隶属度函数用于定义横轴位置的中心参数c ij和定义模糊集宽度的参数σ ij的自整定算法； 2.4) The membership function in the front-end network is used to define the center parameter c ij of the horizontal axis position and the self-tuning algorithm of the parameter σ ij defining the width of the fuzzy set;
    2.5)后端网络中用于后端BP结构神经网络的连接权值q ij的自整定算法； 2.5) The self-tuning algorithm for the connection weight q ij of the back-end BP structure neural network in the back-end network;
    步骤三、将训练好的模型嵌入换挡控制器中，当实际加速踏板开度和车速输入至换挡控制器，经过计算输出目标挡位；Step 3. Embed the trained model into the shift controller. When the actual accelerator pedal opening and vehicle speed are input to the shift controller, the target gear is output through calculation;
    步骤四、输出的挡位信息作用到实车系统，从而实现控制输出。Step 4: The output gear information is applied to the real vehicle system, so as to realize the control output.
  2. 如权利要求1所述的一种基于模糊神经网络的自动换挡控制方法，其特征在于，所述步骤2.1)前端网络计算包括：A kind of automatic shift control method based on fuzzy neural network as claimed in claim 1, is characterized in that, described step 2.1) front-end network calculation comprises:
    前端网络分为四层，用于模糊控制规则自整定：The front-end network is divided into four layers for self-tuning of fuzzy control rules:
    第一层为输入层，控制变量为油门踏板开度、车速，作为模糊神经网络训练的输入样本；第一层存在2个神经元，记作：The first layer is the input layer, and the control variables are accelerator pedal opening and vehicle speed, which are used as input samples for fuzzy neural network training; there are 2 neurons in the first layer, recorded as:
    x＝[x 1,x 2] T x＝[x 1 ,x 2 ] T
    其中，x 1,x 2分别表示当前车辆的油门踏板开度、车速的真实输入值； Among them, x 1 and x 2 represent the actual input values of the accelerator pedal opening and vehicle speed of the current vehicle respectively;
    第二层为隶属度函数分配层，将油门踏板开度、车速划分为三个语言变量，选用高斯隶属度函数进行模糊化处理；其中，油门踏板开度的初始论域设置为[0,1]，语言合集定义为[L(低),M(中),H(高)]；实际车速初始论域设置为[0,250]，语言合集定义为[L(低),M(中),H(高)]；The second layer is the membership function distribution layer, which divides the accelerator pedal opening and vehicle speed into three linguistic variables, and uses the Gaussian membership function for fuzzy processing; among them, the initial discourse domain of the accelerator pedal opening is set to [0,1 ], the language set is defined as [L (low), M (middle), H (high)]; the initial domain of discourse of the actual vehicle speed is set to [0,250], and the language set is defined as [L (low), M (middle), H (high)];
    前端网络的隶属度函数层共计6个神经元，记作：The membership function layer of the front-end network has a total of 6 neurons, denoted as:
    Figure PCTCN2021134389-appb-100001
    Figure PCTCN2021134389-appb-100001
    其中，i表示第一层节点产生的值的序号，j表示模糊集合个数；Among them, i represents the serial number of the value generated by the first layer of nodes, and j represents the number of fuzzy sets;
    第三层为模糊推理层，每个节点仅与上一层每个节点的其中一个模糊集合进行关联，共计存在9个节点，即神经网络控制器同时存在9条模糊推理规则；依据匹配规则，计算上述9个节点的适应度值，记作：The third layer is the fuzzy inference layer. Each node is only associated with one of the fuzzy sets of each node in the previous layer. There are 9 nodes in total, that is, the neural network controller has 9 fuzzy inference rules at the same time; according to the matching rules, Calculate the fitness value of the above 9 nodes, denoted as:
    Figure PCTCN2021134389-appb-100002
    Figure PCTCN2021134389-appb-100002
    其中，α j表示第j层的适应度，i 1，i 2分别表示第二层隶属度分配标准下个数，j＝9，依据隶属度函数定义可知，当训练数据或给定输入在高斯函数赋值附近，对应的语言变量达到较大的值； Among them, α j represents the fitness of the jth layer, i 1 and i 2 respectively represent the numbers under the membership distribution standard of the second layer, j=9, according to the definition of the membership function, when the training data or the given input is in the Gaussian Near the function assignment, the corresponding language variable reaches a larger value;
    第四层实现归一化处理，计算公式为：The fourth layer implements normalization processing, and the calculation formula is:
    Figure PCTCN2021134389-appb-100003
    Figure PCTCN2021134389-appb-100003
  3. 如权利要求1所述的一种基于模糊神经网络的自动换挡控制方法，其特征在于，所述2.2)后端网络计算包括：A kind of automatic shift control method based on fuzzy neural network as claimed in claim 1, is characterized in that, described 2.2) back-end network calculation comprises:
    后端网络分为三层，用于参数归一化处理，其结构与BP神经网络一致：The back-end network is divided into three layers for parameter normalization processing, and its structure is consistent with the BP neural network:
    第一层为输入层，该层共设计3个处理节点节点，其中包含系统输入量以及用于产生常数项的输入恒值；The first layer is the input layer. There are three processing node nodes designed in this layer, which include the system input and the input constant value used to generate constant items;
    第二层为隐藏层，相应的节点与模糊推理规则对应，同样具有9个节点，用于计算条件规则的后端值，记作：The second layer is the hidden layer, and the corresponding nodes correspond to the fuzzy inference rules. There are also 9 nodes, which are used to calculate the back-end value of the conditional rules, which are recorded as:
    y j＝q j0+q j1x 1+q j2x 2 y j ＝q j0 +q j1 x 1 +q j2 x 2
    第三层为输出层，经过模糊神经网络计算后的输出值只有1个，即目标挡位；控制器使用加权求和的方式计算最终输出变量：The third layer is the output layer, and the output value calculated by the fuzzy neural network is only one, that is, the target gear; the controller uses weighted summation to calculate the final output variable:
    Figure PCTCN2021134389-appb-100004
    Figure PCTCN2021134389-appb-100004
  4. 如权利要求1所述的一种基于模糊神经网络的自动换挡控制方法，其特征在于，所述2.3)针对模糊神经网络自整定参数进行优化包括：A kind of automatic shifting control method based on fuzzy neural network as claimed in claim 1, is characterized in that, described 2.3) optimizing for fuzzy neural network self-tuning parameters comprises:
    令误差函数为：Let the error function be:
    Figure PCTCN2021134389-appb-100005
    Figure PCTCN2021134389-appb-100005
    其中，t表示系统期望输出，y表示系统实际输出。Among them, t represents the expected output of the system, and y represents the actual output of the system.
  5. 如权利要求1所述的一种基于模糊神经网络的自动换挡控制方法，其特征在于，所述2.4)中，前端网络c ij和σ ij的自整定算法为： A kind of automatic shift control method based on fuzzy neural network as claimed in claim 1, is characterized in that, in described 2.4), the self-tuning algorithm of front-end network c ij and σ ij is:
    (a)正向传递信息(a) forward information
    第一层，接收系统输入：The first layer, receiving system input:
    Figure PCTCN2021134389-appb-100006
    Figure PCTCN2021134389-appb-100006
    第二层，完成变量模糊化处理，选用高斯函数：The second layer completes the fuzzy processing of variables and uses the Gaussian function:
    Figure PCTCN2021134389-appb-100007
    Figure PCTCN2021134389-appb-100007
    Figure PCTCN2021134389-appb-100008
    Figure PCTCN2021134389-appb-100008
    其中，i＝1,2,3…n；j＝1,2,3…m分别表示第i个输入节点经过模糊化处理生成j个模糊合集，c ij,σ ij分别表示生成高斯隶属度函数的中心值和宽度值； Among them, i=1,2,3...n; j=1,2,3...m respectively represent that the i-th input node is fuzzified to generate j fuzzy sets, c ij , σ ij respectively represent the generation of Gaussian membership function The center value and width value of ;
    第三层，实现模糊推理过程：The third layer realizes the fuzzy reasoning process:
    Figure PCTCN2021134389-appb-100009
    Figure PCTCN2021134389-appb-100009
    第四层，解模糊归一化过程：The fourth layer, defuzzification normalization process:
    Figure PCTCN2021134389-appb-100010
    Figure PCTCN2021134389-appb-100010
    第五层，与后端网络共同作用：The fifth layer, working together with the back-end network:
    Figure PCTCN2021134389-appb-100011
    Figure PCTCN2021134389-appb-100011
    (b)误差反向传播(b) Error Back Propagation
    第五层权值系数修正公式为：The fifth layer weight coefficient correction formula is:
    Figure PCTCN2021134389-appb-100012
    Figure PCTCN2021134389-appb-100012
    其中，β为学习率；Among them, β is the learning rate;
    则，第五层反馈误差信号为：Then, the feedback error signal of the fifth layer is:
    Figure PCTCN2021134389-appb-100013
    Figure PCTCN2021134389-appb-100013
    此时第五层对前一层权值系数满足：At this time, the fifth layer satisfies the weight coefficient of the previous layer:
    Figure PCTCN2021134389-appb-100014
    Figure PCTCN2021134389-appb-100014
    按照相同构造方式依次向前传递：Pass forward sequentially according to the same construction method:
    Figure PCTCN2021134389-appb-100015
    Figure PCTCN2021134389-appb-100015
    Figure PCTCN2021134389-appb-100016
    Figure PCTCN2021134389-appb-100016
    Figure PCTCN2021134389-appb-100017
    Figure PCTCN2021134389-appb-100017
    当f 3计算使用乘法计算中间偏差偏导时，有： When the f 3 calculation uses multiplication to calculate the partial derivative of the intermediate deviation, there is:
    Figure PCTCN2021134389-appb-100018
    Figure PCTCN2021134389-appb-100018
    经由上式，可以计算出隶属层参数的梯度，有：Through the above formula, the gradient of the membership layer parameters can be calculated, as follows:
    Figure PCTCN2021134389-appb-100019
    Figure PCTCN2021134389-appb-100019
    Figure PCTCN2021134389-appb-100020
    Figure PCTCN2021134389-appb-100020
    考虑学习率，将学习率代入上式，则有：Considering the learning rate and substituting the learning rate into the above formula, there are:
    Figure PCTCN2021134389-appb-100021
    Figure PCTCN2021134389-appb-100021
    Figure PCTCN2021134389-appb-100022
    Figure PCTCN2021134389-appb-100022
    因此，自整定的参数c ij、σ ij的学习算法表达为： Therefore, the learning algorithm of self-tuning parameters c ij and σ ij is expressed as:
    Figure PCTCN2021134389-appb-100023
    Figure PCTCN2021134389-appb-100023
    Figure PCTCN2021134389-appb-100024
    Figure PCTCN2021134389-appb-100024
  6. 如权利要求1所述的一种基于模糊神经网络的自动换挡控制方法，其特征在于，所述2.5)中，后端网络q ij的自整定算法包括： A kind of automatic shift control method based on fuzzy neural network as claimed in claim 1, is characterized in that, in described 2.5), the self-tuning algorithm of backend network q ij comprises:
    Figure PCTCN2021134389-appb-100025
    Figure PCTCN2021134389-appb-100025
    其中：i＝1,2,3…n；j＝1,2,3…m，则有Among them: i=1,2,3...n; j=1,2,3...m, then have
    Figure PCTCN2021134389-appb-100026
    Figure PCTCN2021134389-appb-100026
  7. 如权利要求1所述的一种基于模糊神经网络的自动换挡控制方法，其特征在于，所述 步骤二中，T-S模糊神经网络选用高斯隶属度函数进行模糊化处理。A kind of automatic shift control method based on fuzzy neural network as claimed in claim 1, is characterized in that, in described step 2, T-S fuzzy neural network selects Gaussian membership function to carry out fuzzification processing.
PCT/CN2021/134389 2021-06-10 2021-11-30 Automatic gear shifting control method based on fuzzy neural network WO2022257377A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110646436.8 2021-06-10
CN202110646436.8A CN113324026A (en) 2021-06-10 2021-06-10 Automatic gear shifting control method based on fuzzy neural network

Publications (1)

Publication Number Publication Date
WO2022257377A1 true WO2022257377A1 (en) 2022-12-15

Family

ID=77420356

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/134389 WO2022257377A1 (en) 2021-06-10 2021-11-30 Automatic gear shifting control method based on fuzzy neural network

Country Status (2)

Country Link
CN (1) CN113324026A (en)
WO (1) WO2022257377A1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116092688A (en) * 2023-04-10 2023-05-09 广东技术师范大学 Fuzzy logic-based blood pressure grading quantization method, system and storage medium
CN116452865A (en) * 2023-04-03 2023-07-18 南通大学 Jumping type attention lung pathological image classification method based on fuzzy logic

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113324026A (en) * 2021-06-10 2021-08-31 一汽奔腾轿车有限公司 Automatic gear shifting control method based on fuzzy neural network

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6078857A (en) * 1995-10-19 2000-06-20 Hyundai Motor Company Apparatus for deciding a shift pattern suitable for a driver's driving habit using neural network operation and fuzzy inference and a control method thereof
DE102006007717A1 (en) * 2006-02-20 2007-08-23 Robert Bosch Gmbh Gear selection predicting method for vehicle driven by e.g. diesel engine, involves determining probability value by artificial intelligence, which is realized by neural network, fuzzy system, fuzzy preceptron or expert system
CN106828500A (en) * 2017-01-19 2017-06-13 西华大学 Electric automobile geared automatic transmission schedule optimization method
CN108284442A (en) * 2017-01-24 2018-07-17 中国北方车辆研究所 A kind of mechanical arm flexible joint control method based on fuzzy neural network
CN111027618A (en) * 2019-12-09 2020-04-17 西华大学 Automobile dynamic property and economic property expectation quantification method
CN113324026A (en) * 2021-06-10 2021-08-31 一汽奔腾轿车有限公司 Automatic gear shifting control method based on fuzzy neural network

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07127718A (en) * 1993-10-29 1995-05-16 Aisin Aw Co Ltd Control device for automatic transmission
DE19527323A1 (en) * 1995-07-26 1997-01-30 Siemens Ag Circuit arrangement for controlling a device in a motor vehicle
CN103605285A (en) * 2013-11-21 2014-02-26 南京理工大学 Fuzzy nerve network control method for automobile driving robot system

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6078857A (en) * 1995-10-19 2000-06-20 Hyundai Motor Company Apparatus for deciding a shift pattern suitable for a driver's driving habit using neural network operation and fuzzy inference and a control method thereof
DE102006007717A1 (en) * 2006-02-20 2007-08-23 Robert Bosch Gmbh Gear selection predicting method for vehicle driven by e.g. diesel engine, involves determining probability value by artificial intelligence, which is realized by neural network, fuzzy system, fuzzy preceptron or expert system
CN106828500A (en) * 2017-01-19 2017-06-13 西华大学 Electric automobile geared automatic transmission schedule optimization method
CN108284442A (en) * 2017-01-24 2018-07-17 中国北方车辆研究所 A kind of mechanical arm flexible joint control method based on fuzzy neural network
CN111027618A (en) * 2019-12-09 2020-04-17 西华大学 Automobile dynamic property and economic property expectation quantification method
CN113324026A (en) * 2021-06-10 2021-08-31 一汽奔腾轿车有限公司 Automatic gear shifting control method based on fuzzy neural network

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116452865A (en) * 2023-04-03 2023-07-18 南通大学 Jumping type attention lung pathological image classification method based on fuzzy logic
CN116452865B (en) * 2023-04-03 2023-11-07 南通大学 Jumping type attention lung pathological image classification method based on fuzzy logic
CN116092688A (en) * 2023-04-10 2023-05-09 广东技术师范大学 Fuzzy logic-based blood pressure grading quantization method, system and storage medium

Also Published As

Publication number Publication date
CN113324026A (en) 2021-08-31

Similar Documents

Publication Publication Date Title
WO2022257377A1 (en) Automatic gear shifting control method based on fuzzy neural network
US11958365B2 (en) Method for dual-motor control on electric vehicle based on adaptive dynamic programming
US6208981B1 (en) Circuit configuration for controlling a running-gear or drive system in a motor vehicle
Khayyam et al. Adaptive cruise control look-ahead system for energy management of vehicles
CN103324085B (en) Based on the method for optimally controlling of supervised intensified learning
CN101630144B (en) Self-learning inverse model control method of electronic throttle
Lopez et al. Game-theoretic lane-changing decision making and payoff learning for autonomous vehicles
JPH04293626A (en) Vehicle integrated control device
CN109910863B (en) Control method based on special transmission for plug-in hybrid electric vehicle
CN112668799A (en) Intelligent energy management method and storage medium for PHEV (Power electric vehicle) based on big driving data
CN112429005B (en) Pure electric vehicle personalized gear shifting rule optimization method considering transmission efficiency and application
Bo et al. A Q-learning fuzzy inference system based online energy management strategy for off-road hybrid electric vehicles
Razi et al. Predictive energy management of hybrid electric vehicles via multi-layer control
CN111027618B (en) Automobile dynamic property and economic expectation quantification method
CN114384916A (en) Adaptive decision-making method and system for off-road vehicle path planning
JP2009129366A (en) Sensibility estimation system of vehicle
CN115492928B (en) Economical efficiency, dynamic performance and safety comprehensive optimal gear shifting rule optimization method
Li et al. The application of an ANFIS and BP neural network method in vehicle shift decision
Liu et al. Driving intention identification based on long short-term memory neural network
Kerbel et al. Residual policy learning for powertrain control
Bovee et al. Model-based torque shaping for smooth acceleration response in hybrid electric vehicles
Xia et al. Shift control of vehicle automatic transmission based on traffic congestion identification
CN116834757A (en) Vehicle load value estimation method based on BP neural network
CN117002472B (en) Energy management optimization method and system for hybrid electric vehicle
Liu et al. The Optimization of Electric Vehicles Gearshift Controller Parameters Using Reinforcement Learning Algorithm

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21944873

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE