Engine air entrapment electronic control gasoline injection control system and control method
By using the BA-LSTM model and bat algorithm optimization method in the engine clamp gas injection control, the problem of insufficient prediction accuracy of fuel injection volume is solved, and higher prediction accuracy and better engine performance are achieved.
Patent Information
- Application Number
- CN202411907508.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-24
- Publication Date
- 2025-03-28
AI Technical Summary
In the prior art, in the engine clamp gasoline injection control, the prediction accuracy of the injection volume is insufficient, which affects the engine performance and combustion efficiency.
The control method based on the BA-LSTM model is adopted to optimize the prediction parameters of long and short-term memory neural networks through the bat algorithm to improve the prediction accuracy of fuel injection.
It significantly improves the prediction accuracy of fuel injection volume, improves engine performance and combustion efficiency, and reduces the emission of harmful substances in the exhaust gas.
Smart Images
Figure BDA0005204746270000021 
Figure BDA0005204746270000024 
Figure BDA0005204746270000031
Abstract
Description
Technical Field
[0001] The present invention relates to engine control technology, and more specifically, to an engine air-entrained electronically controlled gasoline injection control system and a control method. Background Art
[0002] As people's requirements for automobile performance, fuel economy and environmental protection are increasing, traditional carburetors are gradually being eliminated due to their low control accuracy and high pollution emissions. And with the rapid development of computer technology, sensor technology and control theory, electronic control fuel injection systems have gradually replaced carburetors and become the mainstream fuel supply method.
[0003] On this basis, in order to further optimize the combustion process and improve engine performance, the air-entrained electronically controlled gasoline injection technology was proposed and applied in practice. This technology not only enhances the atomization effect of the fuel and promotes the uniform mixing of fuel and air, but also effectively improves the combustion efficiency, enabling the engine to produce greater power output at lower fuel consumption, while significantly reducing the emission of harmful substances in the exhaust gas. In view of the current air-entrained electronically controlled gasoline injection control technology, the present invention proposes a control technology based on the BA-LSTM model to improve the accuracy of the injection amount. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide an engine air-entrained electronically controlled gasoline injection control system and a control method in view of the deficiencies in the prior art, thereby improving the prediction accuracy of the injection amount.
[0005] The present invention provides an engine air-entrained electronically controlled gasoline injection control method, the method comprising an LSTM network model construction step and an injection quantity prediction step;
[0006] The LSTM network model construction steps include:
[0007] Step 1: Initialize the LSTM network model, randomly generate multiple LSTM parameter combinations, and input all the LSTM parameter combinations into the BA algorithm for optimization to obtain the optimal parameter combination;
[0008] Step 2: collect n sample data including the actual value of the fuel injection amount and the initial predicted value corresponding to the actual value of the fuel injection amount, and input the sample data into the LSTM model model with the optimal parameter combination as the operating parameter for training to obtain the fitness value;
[0009] Step 3: Repeat steps 1 and 2 n-1 times to obtain n fitness values;
[0010] Step 4: Compare and analyze all fitness values, obtain the fitness value with the smallest value, and use the optimal parameter combination corresponding to the fitness value with the smallest value as the target parameter of the LSTM network model;
[0011] The fuel injection amount prediction step comprises:
[0012] Convert the engine data collected at time t into vector x t , and the vector x t Input into the LSTM network model in step 4 to obtain the final injection amount h t .
[0013] Preferably, the mathematical model of the BA algorithm is:
[0014]
[0015] In the formula, f i k is the pulse frequency of the kth iteration; f min Search for the minimum value of the pulse frequency corresponding to the bat individual; f max is the maximum value of the frequency corresponding to the individual bat search pulse; β is a random number in the interval [0,1]; k is the current iteration number; V i k and V i k-1 are the bat rates of the kth and k-1th iterations, respectively, represented by the LSTM parameter combination; and are the bat positions of the kth and k-1th iterations respectively; X * is the current global optimal solution, that is, the optimal parameter combination.
[0016] Preferably, the random number β is determined by the following formula:
[0017]
[0018] In the formula, f i k-1 is the pulse frequency of the k-1th iteration; N max is the maximum number of iterations.
[0019] Preferably, in the BA algorithm, if the current global optimal solution X cannot be calculated within the set number of iterations, * , then the bat speed V i k Calculated by the following formula:
[0020]
[0021] Where V maxis the maximum bat rate; V min is the minimum bat rate; N max is the maximum number of iterations.
[0022] Preferably, in the BA algorithm, at the bat position X i k Introduce randomness, specifically:
[0023]
[0024] Where ε is a random number in the interval [0,1]; δ is a random vector.
[0025] Preferably, in the BA algorithm, when performing local search, if the current global optimal solution X has been obtained * , then the new optimal solution is obtained by the following calculation:
[0026] X new =X * +αA t ;
[0027] Where α is a random number in the interval [-1,1]; A t is the average loudness.
[0028] Preferably, in the fuel injection amount prediction step, the final fuel injection amount h is obtained. t The mathematical model of the LSTM network model is:
[0029] h t =o t etanh(C t ),
[0030] o t =σ(W xo ·x t +W ho ·h t-1 +b o ),
[0031] C t =f t ·C t-1 +i t tanh(W xc ·x t +W hc ·h t-1 +b c ),
[0032] i t =σ(W xi ·x t +W hi ·h t-1+b i );
[0033] In the formula, o t is the output of the output gate of the LSTM network model; e is the element-by-element multiplication; σ(·) is the sigmoid function; W xo and W ho are the weight matrices for forward propagation and backward propagation in the output gate, respectively; b o is the output gate bias; tanh(·) is the hyperbolic tangent function; C t is the predicted injection amount at the current moment; f t is the output of the forget gate of the LSTM network model; C t-1 is the predicted injection amount at the previous moment; W xc and W hc are the weight matrices for forward propagation and backward propagation in the forget gate respectively; h t-1 is the final injection amount at the previous moment; b c is the forget gate bias; i t is the output of the input gate of the LSTM network model; W xi and W hi are the weight matrices for forward propagation and backward propagation in the input gate, respectively; b i Bias for the input gate.
[0034] An engine gas-injection electronically controlled gasoline injection control system comprises an injector and an ECU; the ECU uses the control method to control the injector to spray a final injection amount h at the current moment t to the cylinders of the engine.
[0035] Beneficial Effects
[0036] The advantages of the present invention are: the BA-LSTM model method used in the present invention determines the prediction parameters of the long short-term memory neural network through the bat algorithm, and while the bat algorithm has good convergence, it retains the advantages of the long short-term memory neural network in classification and prediction, thereby greatly improving the prediction accuracy of the injection amount. DETAILED DESCRIPTION
[0037] The present invention will be further described below in conjunction with the embodiments, but it does not constitute any limitation to the present invention. Any limited number of modifications made by anyone within the scope of the claims of the present invention are still within the scope of the claims of the present invention.
[0038] The present invention discloses an engine air-entrained electronically controlled gasoline injection control method, which comprises an LSTM network model construction step and an injection quantity prediction step.
[0039] Among them, the LSTM network model construction step consists of the following four steps:
[0040] Step 1: Initialize the LSTM network model, randomly generate multiple LSTM parameter combinations, and input all LSTM parameter combinations into the BA algorithm for optimization to obtain the optimal parameter combination. In the LSTM network model, each LSTM parameter combination includes the time window step (look_back, lb), the number of hidden layer units (num_units, nu), the learning rate (learningrate, lr), and the number of training epochs (ep).
[0041] Step 2: collect n sample data including the actual value of the injection amount and the initial predicted value corresponding to the actual value of the injection amount, and perform normalized data processing on these sample data. Input the sample data into the LSTM model model with the optimal parameter combination as the operating parameter for training to obtain the fitness value.
[0042] Step 3: Repeat steps 1 and 2 n-1 times to obtain n fitness values.
[0043] Step 4: Compare and analyze all fitness values, obtain the fitness value with the smallest value, and use the optimal parameter combination corresponding to the fitness value with the smallest value as the target parameter of the LSTM network model.
[0044] In the above step 1, the mathematical model of the BA algorithm is:
[0045]
[0046] In this model, f i k is the pulse frequency of the kth iteration; f min Search for the minimum value of the pulse frequency corresponding to the bat individual; f max is the maximum value of the frequency corresponding to the individual bat search pulse; β is a random number in the interval [0,1]; k is the current iteration number; V i k and V i k-1 are the bat rates of the kth and k-1th iterations, respectively, represented by the LSTM parameter combination; and are the bat positions of the kth and k-1th iterations respectively; X * is the current global optimal solution, that is, the optimal parameter combination.
[0047] In the BA algorithm, the pulse frequency f i and bat rate V i It has a relatively large impact on the performance of the BA algorithm. An adaptive strategy is used to dynamically adjust the pulse frequency f i and bat rate V i, improve the performance of BA algorithm.
[0048] For pulse frequency f i The adaptive adjustment idea is to adjust adaptively based on the change of the bat's objective function value, and adopt the strategy of increasing the pulse frequency to expand the bat's search range. The specific expression is:
[0049]
[0050] That is, the random number β is determined by the following formula:
[0051]
[0052] In the formula, f i k-1 is the pulse frequency of the k-1th iteration; N max is the maximum number of iterations.
[0053] Bat speed V i The adaptive adjustment idea is to adjust the movement rate based on the historical performance of the bat. If the current global optimal solution X cannot be calculated within the set number of iterations, * , then the method of reducing the movement rate is adopted to increase the accuracy of the search. Its mathematical expression is:
[0054]
[0055] Where V max is the maximum bat rate; V min is the minimum bat rate; N max is the maximum number of iterations.
[0056] In order to prevent the BA algorithm from falling into the local optimum, randomness is introduced into the traditional BA algorithm to improve it and form an improved BA algorithm. Specifically:
[0057]
[0058] Where ε is a random number in the interval [0,1]; δ is a random vector.
[0059] In the BA algorithm, if the current global optimal solution X is obtained during local search * , then the new optimal solution is obtained by the following calculation:
[0060] X new =X * +αA t .
[0061] Where α is a random number in the interval [-1,1]; A t is the average loudness.
[0062] After obtaining the target LSTM network model through the above four steps, the fuel injection amount prediction step will obtain the final fuel injection amount through the LSTM network model. Specifically, the engine data collected at time t is converted into a vector x t , and the vector x t Input into the LSTM network model in step 4 to obtain the final injection amount h t .
[0063] LSTM is an improved recurrent neural network, which consists of multiple units. Each unit contains three key gate structures, namely input gate, forget gate and output gate. The specific structure of each gate is as follows:
[0064] The input gate consists of a sigmoid activation function and a point-by-point multiplication operation, and its mathematical expression is:
[0065] i t =σ(W xi ·x t +W hi ·h t-1 +b i ).
[0066] In the formula, i t is the output of the input gate; σ(·) is the sigmoid function; W xi and W hi are the weight matrices for forward propagation and backward propagation in the input gate respectively; h t-1 is the hidden state of the previous time step, i.e. h t-1 is the final injection amount at the previous moment; b i Bias for the input gate.
[0067] The forget gate is controlled by a sigmoid activation function, which produces values between 0 and 1, and its mathematical expression is:
[0068] f t =σ(W xf ·x t +W hf ·h t-1 +b f ).
[0069] In the formula, f t is the output of the forget gate, W xf and W hf are the weight matrices for forward propagation and backward propagation in the forget gate respectively; b f For bias.
[0070] The output f of the forget gate t With the previous cell state C t-1 The element-wise multiplication is then used to update the cell state together with the output of the input gate, which determines the new information to be stored. The mathematical expression for the cell state update is:
[0071] C t =f t ·C t-1 +i t tanh(W xc ·x t +W hc ·h t-1 +b c ).
[0072] In the formula, C t is the unit state at the current time step, that is, the predicted injection amount at the current moment; C t-1 is the unit state of the previous time step, that is, the predicted injection amount at the previous moment; tanh(·) is the hyperbolic tangent function; W xc and W hc are the weight matrices for forward propagation and backward propagation in the forget gate respectively; b c For bias.
[0073] The output gate is usually controlled by a sigmoid activation function, which produces values between 0 and 1. The mathematical expression of the output gate is:
[0074] o t =σ(W xo ·x t +W ho ·h t-1 +b o ).
[0075] In the formula, o t is the output of the output gate; W xo and W ho are the weight matrices for the forward and backward propagation of the output gate, respectively; b o For bias.
[0076] The output o of the output gate t Determines the information to be passed to the output layer at time step t to generate the hidden state h of the current time step t And the final output prediction value. The output of the output gate is related to the cell state C t Multiply element-wise to determine the hidden state h to be passed to the next time step t , its mathematical expression is:
[0077] h t =ot etanh(C t ).
[0078] Where e is element-by-element multiplication.
[0079] The above is only a preferred embodiment of the present invention. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the structure of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent.
Claims
1. An engine air-injection electronically controlled gasoline injection control method, characterized in that: The method includes an LSTM network model building step and a fuel injection amount prediction step; The LSTM network model construction steps include: Step 1: Initialize the LSTM network model, randomly generate multiple LSTM parameter combinations, and input all the LSTM parameter combinations into the BA algorithm for optimization to obtain the optimal parameter combination; Step 2: collect n sample data including the actual value of the fuel injection amount and the initial predicted value corresponding to the actual value of the fuel injection amount, and input the sample data into the LSTM model model with the optimal parameter combination as the operating parameter for training to obtain the fitness value; Step 3: Repeat steps 1 and 2 n-1 times to obtain n fitness values; Step 4: Compare and analyze all fitness values, obtain the fitness value with the smallest value, and use the optimal parameter combination corresponding to the fitness value with the smallest value as the target parameter of the LSTM network model; The fuel injection amount prediction step comprises: Convert the engine data collected at time t into vector x t , and the vector x t Input into the LSTM network model in step 4 to obtain the final injection amount h t .
2. The engine air-injection electronically controlled gasoline injection control method according to claim 1, characterized in that: The mathematical model of the BA algorithm is: In the formula, f i k is the pulse frequency of the kth iteration; f min Search for the minimum value of the pulse frequency corresponding to the bat individual; f max is the maximum value of the frequency corresponding to the individual bat search pulse; β is a random number in the interval [0,1]; k is the current iteration number; V i k and V i k-1 are the bat rates of the kth and k-1th iterations, respectively, represented by the LSTM parameter combination; and are the bat positions of the kth and k-1th iterations respectively; X * is the current global optimal solution, that is, the optimal parameter combination.
3. The engine air-injection electronically controlled gasoline injection control method according to claim 2, characterized in that: The random number β is determined by the following formula: In the formula, f i k-1 is the pulse frequency of the k-1th iteration; N max is the maximum number of iterations.
4. The engine air-injection electronically controlled gasoline injection control method according to claim 2, characterized in that: In the BA algorithm, if the current global optimal solution X cannot be calculated within the set number of iterations, * , then the bat speed V i k Calculated by the following formula: Where V max is the maximum bat rate; V min is the minimum bat rate; N max is the maximum number of iterations.
5. The engine air-injection electronically controlled gasoline injection control method according to claim 2, characterized in that: In the BA algorithm, at the bat position Introduce randomness, specifically: Where ε is a random number in the interval [0,1]; δ is a random vector.
6. The engine air-injection electronically controlled gasoline injection control method according to claim 2, characterized in that: In the BA algorithm, if the current global optimal solution X is obtained during local search * , then the new optimal solution is obtained by the following calculation: X new =X * +αA t ; Where α is a random number in the interval [-1,1]; A t is the average loudness.
7. The engine air-injection electronically controlled gasoline injection control method according to claim 1, characterized in that: In the fuel injection amount prediction step, the final fuel injection amount h is obtained. t The mathematical model of the LSTM network model is: h t =o t etanh(C t ), o t =σ(W xo ·x t +W ho ·h t-1 +b o ), C t =f t ·C t-1 +i t ·tanh(W xc ·x t +W hc ·h t-1 +b c ), i t =σ(W xi ·x t +W hi ·h t-1 +b i ); In the formula, o t is the output of the output gate of the LSTM network model; e is the element-by-element multiplication; σ(·) is the sigmoid function; W xo and W ho are the weight matrices for forward propagation and backward propagation in the output gate, respectively; b o is the output gate bias; tanh(·) is the hyperbolic tangent function; C t is the predicted injection amount at the current moment; f t is the output of the forget gate of the LSTM network model; C t-1 is the predicted injection amount at the previous moment; W xc and W hc are the weight matrices for forward propagation and backward propagation in the forget gate respectively; h t-1 is the final injection amount at the previous moment; b c is the forget gate bias; i t is the output of the input gate of the LSTM network model; W xi and W hi are the weight matrices for forward propagation and backward propagation in the input gate, respectively; b i Bias for the input gate.
8. An engine air-injection electronically controlled gasoline injection control system, characterized in that: The method comprises a fuel injector and an ECU; the ECU controls the fuel injector to spray a final fuel injection amount h at the current moment by using the control method as described in any one of claims 1 to 7. t to the cylinders of the engine.