Accurate sensing control method for combustion state of internal combustion engine

Abstract

The invention discloses a precise sensing control method for the combustion state of an internal combustion engine. The method comprises the steps that S1, parameters are collected; s2, establishing a neural network model, training the neural network model, and identifying the running state of the internal combustion engine; s3, a fuzzy language is established, and respective fuzzy adjustment amounts of a fuel injection system, a valve transmission system and an ignition system are obtained; s4, self-adaptive control models of the fuel injection system, the valve transmission system and the ignition system are established respectively, and the self-adaptive adjustment amount of the fuel injection system, the self-adaptive adjustment amount of the valve transmission system and the self-adaptive adjustment amount of the ignition system are obtained; and S5, carrying out weighted fusion on the fuzzy adjustment amount and the adaptive adjustment amount to obtain a final adjustment amount. By collecting multi-dimensional operation parameters in real time, extracting time domain and frequency domain features and combining a deep neural network model, precise recognition and early warning of the operation state of the internal combustion engine are achieved, detection and response can be made in time when the working condition of the internal combustion engine changes slightly, and the problem that a traditional control system is lagged in response is solved.

Description

Technical Field

[0001] The present invention relates to a method for accurately sensing and controlling the combustion state of an internal combustion engine, and belongs to the technical field of internal combustion engines. Background Art

[0002] As the core power unit of modern transportation and industrial production, the performance of the internal combustion engine's control system directly affects power output, fuel economy, and emissions. However, traditional internal combustion engine control systems have many technical bottlenecks: First, their open-loop control based on fixed parameter mapping tables or simple PID closed-loop control strategies do not monitor the engine's operating parameters in real time and accurately, and can often only respond when operating conditions change significantly. This leads to significant lags in the adjustment of key execution actions such as valve opening and closing, fuel injection, and ignition timing. For example, under transient operating conditions such as rapid vehicle acceleration and climbing, the engine cannot adjust to the optimal operating state in a timely manner, resulting in not only unstable power output but also fuel waste and excessive emissions.

[0003] Secondly, existing control methods mainly rely on fixed threshold alarms for detecting abnormal conditions and lack early identification capabilities. When problems such as poor combustion and mechanical failure occur, it is difficult for the system to take effective measures in a timely manner, which can easily lead to chain failures. In addition, fuel injection, valve timing, and ignition systems usually adopt independent control strategies and lack a global collaborative optimization mechanism. The control parameters between subsystems are difficult to coordinate and match, which seriously affects the overall performance of the internal combustion engine. More importantly, although the current advanced adaptive control methods can theoretically adapt to the dynamic characteristics of internal combustion engines, due to their high dependence on accurate physical models, in practical applications, they often cause model mismatch problems due to the strong nonlinearity and time-varying characteristics of internal combustion engines, making it difficult to ensure control accuracy. Summary of the Invention

[0004] Purpose of the invention: In view of the deficiencies in the prior art, the present invention provides a method for accurately sensing and controlling the combustion state of an internal combustion engine. The present invention realizes accurate motion control of the internal combustion engine by combining fuzzy language control and adaptive control.

[0005] Technical solution: A method for accurately sensing and controlling the combustion state of an internal combustion engine, comprising the following steps:

[0006] S1. Parameter acquisition: Collect the parameters of the internal combustion engine under different working conditions, and obtain time domain characteristics, frequency domain characteristics, and working condition characteristics based on the collected parameters;

[0007] S2. Establishing a neural network model based on the time domain features, frequency domain features, and operating condition features acquired in S1, training the neural network model based on the parameters acquired in S1, and identifying the operating state of the internal combustion engine based on the trained neural network model;

[0008] S3, based on the parameters collected by S1 and the operating state of the internal combustion engine obtained by S2, establish a fuzzy language and obtain the fuzzy adjustment values ​​of the fuel injection system, the valve train system, and the ignition system;

[0009] S4. Based on the operating state of the internal combustion engine obtained in S2, adaptive control models of the fuel injection system, the valve drive system, and the ignition system are established respectively, and adaptive adjustment values ​​of the fuel injection system, the valve drive system, and the ignition system are obtained;

[0010] S5. Based on the operating state of the internal combustion engine obtained in S2, the fuzzy adjustment amount and the adaptive adjustment amount obtained in S3 and S4 are weightedly integrated to obtain a final adjustment amount.

[0011] Preferably, the S1 is specifically:

[0012] The parameters collected for the internal combustion engine under different working conditions, i.e., different loads and different speeds, include intake flow, exhaust flow, intake pressure, exhaust pressure, coolant temperature, oil pressure, exhaust emission components, vibration signals, valve train displacement, injection pulse signals, injection amount, and injection timing;

[0013] The acquisition of time domain features is specifically as follows:

[0014] Calculate the effective value of the vibration signal , to reflect the average energy level of the vibration signal: ;

[0015] in, is the number of vibration signal sampling points; is the instantaneous value of the vibration signal at the i-th sampling point;

[0016] Calculating kurtosis , used to evaluate whether there are abnormal impact components in the vibration signal:

[0017] ;

[0018] in, is the mean value of the vibration signal; is the standard deviation of the vibration signal;

[0019] Calculate the pulse factor to help identify the pulse characteristics in the vibration signal: Pulse Factor = ;

[0020] in is the effective value of the vibration signal, is the maximum instantaneous value in the vibration signal;

[0021] The specific steps of obtaining frequency domain features are as follows:

[0022] Perform spectrum analysis on exhaust noise, extract 1 / 3 octave energy spectrum, and analyze the energy distribution of different frequency components;

[0023] Center frequency : ;in, is the frequency band number;

[0024] Frequency band range : ;

[0025] Sound pressure level SPL calculation: SPL=20 ;in, ; is the actual sound pressure;

[0026] Obtain the characteristics of exhaust noise in the frequency domain, thereby reflecting the stability of internal combustion engine combustion processes;

[0027] The operating condition characteristics specifically include speed fluctuation rate, intake pressure gradient, load change rate, intake flow rate, exhaust flow rate, fuel injection amount, exhaust pressure, coolant temperature, oil pressure, exhaust emission composition, injection pulse signal, fuel injection amount, and injection timing; the speed fluctuation rate, intake pressure gradient, and load change rate are obtained by the following methods, and the remaining operating condition characteristics are obtained by direct acquisition:

[0028] Calculate the speed fluctuation rate to measure the change of internal combustion engine speed over time; speed fluctuation rate = ;

[0029] in, is the standard deviation of the rotational speed; is the average value of the rotational speed;

[0030] Calculate the intake pressure gradient to reflect the change trend of intake pressure over time; ;

[0031] in is the intake pressure gradient, For the time point The intake pressure value at the moment, For the time point The intake pressure value at the moment, is the time interval.

[0032] Preferably, the neural network model is established and trained in S2 as follows:

[0033] A key feature set is established based on the time domain features, frequency domain features, and working condition features obtained by S1 as the input of the neural network model input layer;

[0034] Set up at least three hidden layers, select the optimal number of neurons between adjacent hidden layers, and use the ReLU activation function for the neuron output signal of each hidden layer: Perform nonlinear transformation on it, insert Dropout layer between adjacent hidden layers, and set the dropout rate of Dropout layer to 0.3-0.5;

[0035] The number of neurons in the output layer is set to the total number of engine operating state categories, and the Softmax activation function is applied to the output layer. ;in, is the output value of the i-th neuron in the output layer, is the total number of neurons, and the output value of the neuron is mapped to the corresponding probability distribution, which represents the probability of the internal combustion engine being in the corresponding operating state category;

[0036] The time domain features, frequency domain features, and working condition features in S1 are set as training sets for neural network model parameter learning;

[0037] Model initialization: Initialize the parameters of the neural network model;

[0038] Training parameter settings: set the number of training iterations to 50 to 200 rounds, and set the initial learning rate to 0.001-0.0001;

[0039] Training loop: For each training round, traverse all the data in the training set;

[0040] Perform forward propagation: data passes through the input layer, at least one hidden layer, and the output layer in sequence;

[0041] Calculate the loss function: For the running state recognition task, the cross entropy loss function is used:

[0042] Cross-Entropy Loss= ;

[0043] in, is the true value, is the predicted value;

[0044] For the performance index prediction task, the mean square error MSE loss function is used: MSE Loss= ;

[0045] in, is the true value, is the predicted value, is the number of sample sampling points;

[0046] Perform backpropagation: Calculate the gradient of the neural network model parameters:

[0047] Assume that the output layer has neurons, and the corresponding prediction value is , the true value is ;

[0048] For the cross entropy loss function, the gradient of the output layer is calculated as follows: ;

[0049] in, The output layer The linear output of the neuron is the input of the Softmax activation function. The output layer The predicted value of a neuron, The output layer The true value of each neuron;

[0050] For the mean squared error loss function, the gradient of the output layer is calculated as follows: ;

[0051] Starting from the output layer, calculate the gradient of the hidden layer layer by layer; assuming the current layer is The layer has an activation function of ReLU, and the gradient of the layer is calculated as follows: ;

[0052] in: It is Tier The linear output of a neuron, It is Tier The linear output of each neuron; is the output of the ReLU activation function; = ;

[0053] For each layer's weight matrix and the bias vector , its gradient is calculated as follows: ; ;

[0054] in: It is Tier The neuron and Tier The weights between neurons; It is Tier The activation output of each neuron; It is Tier The bias of each neuron;

[0055] Using the calculated gradient, the parameters of the neural network are updated through the Adam optimizer. The parameter update formula is as follows:

[0056] ; ;in, is the learning rate, which controls the step size of parameter updates.

[0057] Preferably, the operating state of the internal combustion engine identified in S2 includes:

[0058] Receive the normalized probability values ​​of six states output by the neural network model, the sum of which is 1, including normal state, overheating state, stall state, poor combustion state, excessive emission state, and mechanical failure state;

[0059] The preset priority order is mechanical failure state > stall state > overheating state > emission exceeding standard state > poor combustion state > normal state. The confidence level is determined based on the probability values ​​of the six states output by the neural network:

[0060] S201. Check whether the confidence level of any operating state reaches a high confidence threshold, that is, the probability of any operating state is ≥ 0.85, and the probability of the operating state is at least twice that of the operating state with the second highest probability. If so, directly determine that the current state is in that operating state;

[0061] S202. If no high confidence state appears, all medium confidence states are screened out, that is, the operating state probabilities are in [0.5, 0.85), and parameter verification is performed on them:

[0062] Overheating: Verify that the coolant temperature is >103°C; Mechanical failure: Verify that the vibration kurtosis is >3.5 or the pulse factor is >5; Poor combustion: Verify that the exhaust CO / HC exceeds the standard by 20%; Excessive emissions: Verify that NOx exceeds the standard by 15% or that particulate matter is abnormal; Stalling: Verify that the speed is <1000 RPM; Normal state: Verify that all parameters are within the safe range; If a state satisfies both the probability condition and the parameter condition, the current operating state is determined to be that state;

[0063] S203. If two or more operating states are high-confidence states, that is, the probability of the operating state is ≥ 0.85, sort them according to the preset priority and check whether the real-time parameters conflict with the operating states. Starting from the highest priority state, select the first operating state whose real-time parameters do not conflict with the operating state as the current operating state;

[0064] S204. If the current operating state cannot be determined in S201-S203, the operating states with operating state probabilities greater than 0.3 are screened, and the state with the highest priority is selected as the current operating state in order of priority;

[0065] S205, dynamic monitoring and control adjustment

[0066] Continuously monitor parameter changes after status determination: When any parameter changes by more than 10%, immediately return to S201 to trigger the determination process.

[0067] Perform control adjustments based on the current state:

[0068] Overheating: Reduce fuel injection by 0.2mg and increase cooling fan speed by 20%. Stalling: Advance ignition timing by 3° and increase intake air volume. Mechanical failure: Reduce load to 50% and trigger an audible and visual alarm. Compare actual and expected parameter deviations every 200 milliseconds and fine-tune control commands.

[0069] S306, dynamic update of threshold: the system automatically updates the judgment threshold every 5 seconds:

[0070] Based on operational stability: when the speed fluctuation rate is less than 0.3r / min and the load change rate is less than 8N·m / s, the normal state threshold is increased to 0.75; based on abnormal signals: when the vibration kurtosis is greater than 4.0, the mechanical failure threshold is lowered to 0.6; based on operating condition adaptation: during long-term high-load operation, the poor combustion threshold is lowered to 0.65.

[0071] Preferably, the S3 is specifically:

[0072] Convert the collected load parameters into fuzzy linguistic variables:

[0073] The load parameter is divided into three fuzzy linguistic variables: low load, medium load and high load using triangular membership function, and the definition domains are: low load: definition domain , the membership function is: ; Medium load: Domain range , the membership function is: ; High load: domain range , the membership function is: ;

[0074] in, is the actual measured load value, is the minimum value in the low load range, It is the maximum value of the low load range and also the transition point from low load to medium load. is the center value of the medium load range, which is used to define the membership function of the medium load. It is the maximum value of the medium load range and also the transition point from medium load to high load. This is the maximum value in the high load range.

[0075] Convert the collected speed parameters into fuzzy linguistic variables:

[0076] The speed parameter is divided into three fuzzy linguistic variables: low speed, medium speed and high speed by using trapezoidal membership function: The following definition domain range is determined according to the actual speed range;

[0077] Low speed: Use Z-type membership function, the domain range is [0, ], the membership function is: ;in, and The purpose is to adapt the low-speed fuzzy mathematical model to the real scene, and to determine the adjustable morphological control parameters according to the actual speed range;

[0078] Medium speed: using triangular membership function, the domain range is [ , ], the membership function is:

[0079] ;

[0080] High speed: S-type membership function is used, and the domain range is [ , ∞ ), the membership function is:

[0081] ;

[0082] in, and It is an adjustable shape control parameter determined according to the actual speed range. : The domain boundary of the low speed Z-type membership function and the medium speed triangle membership function, 、 :The medium speed is the endpoint of the domain interval of the triangle membership function itself, which determines the range of the base of the triangle. , ]arrive[ , ] is the interval of medium speed membership change; :The domain boundary of medium speed and high speed is the S-type membership function, [ ,+∞) is the judgment range of high speed; is the actual speed collected;

[0083] The coolant temperature is converted into fuzzy linguistic variables:

[0084] Low temperature: domain range , the membership function is: ;

[0085] Moderate Temperature: Domain Range , the membership function is:

[0086] ;

[0087] High temperature: domain range , the membership function is:

[0088] ;

[0089] in is the collected coolant temperature, is the starting temperature of the low temperature range, is the maximum temperature in the low temperature range, the starting temperature in the medium temperature range, The end temperature of the medium temperature range and the start temperature of the high temperature range, End temperature of the high temperature range. : The center temperature value of low temperature, indicating the center point of the low temperature range, : The center temperature value of the medium temperature, indicating the center point of the medium temperature range, : The center temperature value of high temperature, indicating the center point of the high temperature range, is the standard deviation of the low temperature range, indicating the degree of diffusion in the low temperature range. is the standard deviation of the medium temperature range, indicating the degree of diffusion in the medium temperature range. is the standard deviation of the high temperature range, indicating the degree of diffusion in the high temperature range.

[0090] The oil pressure is converted into fuzzy linguistic variables:

[0091] Low pressure: domain range , the membership function is:

[0092] ;

[0093] Moderate Pressure: Domain Range , the membership function is:

[0094] ;

[0095] High pressure: domain definition , the membership function is:

[0096] ;

[0097] in is the collected oil pressure parameter, It is the minimum value of the oil pressure, usually the lowest safety pressure of the system.

[0098] is the starting pressure of the low pressure range, It is the maximum pressure of the low pressure range and the starting pressure of the medium pressure range. The center pressure of the medium pressure range, It is the maximum pressure of the medium pressure range and the starting pressure of the high pressure range. This is the maximum pressure in the high pressure range.

[0099] The exhaust emission components are converted into fuzzy linguistic variables:

[0100] Low emissions: defining the domain , corresponding to the case where the concentration of pollutants in the exhaust gas is low, the membership function is:

[0101] ;

[0102] Medium emissions: Defining the domain , corresponding to the situation where the concentration of pollutants in the exhaust gas is at a medium level, the membership function is:

[0103] ;

[0104] High emissions: defining the domain , corresponding to the situation where the concentration of pollutants in the exhaust gas is too high, the membership function is:

[0105] ;

[0106] in is the actual measured exhaust emission value, It is the starting emission value of the low emission range, usually the minimum limit of the emission standard. It is the starting point for the transition from low emissions to medium emissions. is the end point of the transition from low emissions to medium emissions. : Starting emission value of the medium emission range, is the central emission value of the medium emission range, It is the starting point for the transition from medium emissions to high emissions. is the starting emission value of the high emission range, It is the maximum emission value in the high emission range.

[0107] Convert vibration signals into fuzzy linguistic variables

[0108] Low Vibration: Domain Definition , corresponding to the case where the vibration amplitude of the internal combustion engine is small, the membership function is:

[0109] ;

[0110] Moderate Vibration: Domain Range , corresponding to the case of medium vibration amplitude of the internal combustion engine, the membership function is:

[0111] ;

[0112] High Vibration: Domain Range , corresponding to the case where the vibration amplitude of the internal combustion engine is large, the membership function is:

[0113] ;

[0114] in is the actual measured vibration signal value. This is the minimum value of the vibration signal, usually the background vibration level during normal operation. is the starting vibration value of the low vibration range. It is the starting point of the transition from low vibration to medium vibration. This is the starting vibration value of the medium vibration range. It is the center vibration value of the medium vibration range. It is the starting point of the transition from medium vibration to high vibration. is the starting vibration value of the high vibration range. It is the maximum vibration value in the high vibration range. After exceeding this vibration value, the membership remains at 1.

[0115] The displacement of the valve train is converted into fuzzy linguistic variables:

[0116] The triangular membership function is used to divide the valve train system displacement into three fuzzy linguistic variables: low displacement, medium displacement, and high displacement:

[0117] The definition domain range is determined according to the displacement range of the actual valve train system;

[0118] Low displacement: domain range , the membership function is:

[0119] ;

[0120] Moderate displacement: Domain range , the membership function is:

[0121] ;

[0122] High displacement: domain range , the membership function is:

[0123] ;

[0124] in, is the lower limit of the low displacement range, less than When , the displacement is completely low; It is the dividing point between low displacement and medium displacement. When the range is within the range, the low displacement membership gradually decreases, and the medium displacement membership gradually increases; It is the dividing point between medium displacement and high displacement. When , it starts to be high displacement; Is the upper limit of the high displacement interval, the displacement is greater than or equal to When , it is high displacement;

[0125] Based on all the aforementioned fuzzy languages, all the operating states of the internal combustion engine, and the corresponding parameters in S1, a fuzzy rule base is constructed. Each rule consists of an IF condition and a THEN conclusion, where the condition is defined by a combination of fuzzy languages. The fuzzy language, the operating state of the internal combustion engine, and the corresponding parameters in S1 obtained in real time are matched against the rule base one by one, and the excitation strength of each rule is calculated.

[0126] Perform membership aggregation operation on the logical operators in the condition part: if it contains AND operator, take the minimum value of the condition part membership as the excitation strength of the rule; if it contains OR operator, take the maximum value of the condition part membership as the excitation strength of the rule;

[0127] According to the excitation intensity of each rule and the corresponding conclusion part, the corresponding membership function is truncated according to the excitation intensity to obtain the truncated output fuzzy set;

[0128] The output fuzzy sets of all rules are combined, that is, the membership function after truncation of any rule and the truncated membership function of the previous rule are maximized to generate the final output fuzzy set;

[0129] Defuzzify the final output fuzzy set:

[0130] Discretize the final output fuzzy set domain into i points , for each discrete point , calculate its corresponding membership value , the center of gravity method is used to convert the final output fuzzy set into an accurate control signal, and the calculation formula is: ;

[0131] in, is the discrete point of the output universe, is its membership value, is the parameter adjustment amount; f1 is the scaling factor, the domain of which is [0 1], and is obtained by querying the mapping function or mapping table based on the throttle and speed.

[0132] The parameter adjustment amount The specific change values ​​of the internal combustion engine control parameters are clarified, including the fuzzy adjustment amount of the fuel injection system, the fuzzy adjustment amount of the valve transmission system, and the fuzzy adjustment amount of the ignition system.

[0133] Preferably, the adaptive adjustment amount of the fuel injection system obtained in S4 is specifically:

[0134] The desired fuel injection system has the characteristics of fast response and small overshoot. The reference model is: ;

[0135] in, is the transfer function of the fuel injection system, is the injection amount of the reference model at time t, The injection pulse signal of the reference model at time t is: For the fuel injection system gain, is the time constant of the fuel injection system, s is the complex frequency domain variable of Laplace transform;

[0136] Substitute the fuel injection system data collected by S1, namely the injection pulse signal, injection amount, and injection time, into the reference model and perform iterative calculation to determine the fuel injection system gain in the model. and the fuel injection system time constant ;

[0137] Determine the error signal, which is defined as the difference between the reference model output and the actual output of the fuel injection system:

[0138] ;in, is the fuel injection system error signal, is the amount of fuel injected by the fuel injection system within time t;

[0139] Set performance indicators: ;in, It is a fuel injection system performance indicator, representing the sum of squares of error signals, and is used to measure the performance of the system.

[0140] To minimize performance , use the gradient descent method to update the control parameters : ;in: is the control parameter at the current moment, is the learning rate, which controls the step size of parameter update; For a certain moment; is the difference between this moment and the previous moment; is the control parameter for the next moment; is the gradient of the performance index with respect to the control parameter, expressed as: ; is the gradient of the error signal with respect to the control parameter;

[0141] Adjustable fuel gain according to the updated and fuel error signal , generating a control signal To adjust the fuel injection system's injection quantity: ;in, is the updated injection pulse signal, is the injection pulse signal before updating; the injection pulse signal after updating Adaptive adjustment for the fuel injection system.

[0142] Preferably, the self-adaptive adjustment amount of the valve transmission system is obtained as follows:

[0143] The desired valvetrain system has stable dynamic response characteristics, and the reference model is the expected trajectory of the valve displacement:

[0144] ;in, Valve displacement The second derivative with respect to time t represents the valve acceleration; is the current input to the solenoid valve at time t, The proportionality constant between the electromagnetic force and the square of the input current; is the actual displacement of the valve at time t; is the spring stiffness coefficient, which represents the restoring force of the spring on the valve displacement; The quality of the valve and its transmission system;

[0145] The error signal is defined as the difference between the reference displacement and the actual displacement: ;in, is the error signal of the valve train system; is the reference displacement of the valve train; is the actual displacement of the valve train;

[0146] Set performance indicators: ;in, It is a performance indicator of the valvetrain system, representing the sum of squares of error signals, and is used to measure the performance of the system.

[0147] To minimize performance , use the gradient descent method to update the control parameters : ;in, is the control parameter at the current moment, is the learning rate, which controls the step size of parameter updates, is the gradient of the performance index with respect to the control parameter: ; is the gradient of the error signal with respect to the control parameter;

[0148] Parameter update: The updated control parameters are used to generate new electromagnetic force signals , thereby adjusting the opening and closing of the valve:

[0149] ;

[0150] in, For in time The valve control parameters at this time are used to adjust the changes of the electromagnetic force signal; In time The new electromagnetic force signal value at time ;

[0151] Updated electromagnetic force It is the adaptive adjustment amount of the valve transmission system.

[0152] Preferably, the ignition system adaptive adjustment amount is obtained as follows:

[0153] The desired ignition system has precise ignition advance angle control characteristics, and the reference model is:

[0154] ;

[0155] in, is the reference ignition advance angle; is the internal combustion engine speed; is the compression pressure; is the excess air coefficient of the mixture; 、 、 、 are the reference model coefficients, which are estimated based on the parameters collected by S1 using the least squares method;

[0156] The error signal is defined as the difference between the reference ignition advance angle and the actual ignition advance angle: ;

[0157] in, is the error signal of the ignition system, is the reference ignition advance angle, is the actual ignition advance angle;

[0158] Set performance indicators: ;

[0159] in: It is an ignition system performance indicator, representing the sum of squares of error signals, and is used to measure the performance of the system.

[0160] To minimize performance , use the gradient descent method to update the control parameters :

[0161] ;

[0162] in: Control parameters at the next moment; is the control parameter at the current moment; is the learning rate, which controls the step size of parameter update; is the gradient of the performance index with respect to the control parameter, expressed as: ;

[0163] is the gradient of the error signal with respect to the control parameter;

[0164] The updated control parameters are used to generate new ignition advance angle adjustment , thereby adjusting the ignition timing:

[0165] ;

[0166] Updated ignition advance angle It is the adaptive adjustment amount of the ignition system.

[0167] Preferably, the S5 is specifically:

[0168] S501, determine the weight coefficient of the fusion of the fuzzy adjustment amount and the adaptive adjustment amount obtained in S3 and S4:

[0169] Monitor the operating data of the internal combustion engine in S1 at real-time speed and load in real time, and calculate the operating stability index, including speed stability index and load stability index, specifically:

[0170] Calculate the speed stability index, that is, calculate the standard deviation of the speed fluctuation rate in a 3-second sliding time window:

[0171] The speed of the internal combustion engine is collected in real time. The collection frequency is set between 10 and 100 times per second. A sliding time window of 3 seconds is set to store the continuously collected speed data. The average speed value is calculated within the sliding time window. , the formula is as follows: ;in, is the number of data points in the window, It is The speed value at the data point,

[0172] Calculate the standard deviation of the speed fluctuation rate within the window : ;

[0173] This standard deviation reflects the degree of speed fluctuation within the window. As time goes by, the sliding time window continues to move forward, adding the latest speed data point each time and discarding the first data point in the time period, keeping the window length at 3 seconds.

[0174] Repeat the above calculation process to obtain the real-time speed fluctuation standard deviation. If the standard deviation is less than 0.5 r / min, the speed is determined to be in a stable state; if it is greater than or equal to 0.5 r / min, the speed is determined to be in an unstable state.

[0175] Calculate the load stability index, that is, calculate the absolute value of the load change rate:

[0176] Collect the load data of the internal combustion engine in real time. The collection frequency is consistent with the speed data collection frequency to ensure the time synchronization and consistency of the data. Calculate the load change at two adjacent collection moments. With time interval The ratio of load change rate is as follows: ;in, is the load value at the current moment, is the load value at the previous moment, is the time interval between two acquisition moments, usually ranging from 0.01 seconds to 1 second. Since we are concerned about the severity of the load change, regardless of whether the load increases or decreases, the absolute value of the load change rate is taken: ;

[0177] The larger the absolute value of the load change rate, the more drastic the load change. If the absolute value of the load change rate is less than 10 N·m / s, the load is determined to be in a stable state; if it is greater than or equal to 10 N·m / s, the load is determined to be in an unstable state. The error between the actual speed output and the expected speed output of the internal combustion engine is calculated, and the speed stability index and the load stability index are combined to set the weight coefficient values ​​under different speed fluctuation standard deviations and load change rate absolute values, as well as error states and operating states: If the speed fluctuation standard deviation is less than 0.5 r / min, the absolute value of the load change rate is less than 10 N·m / s, the speed error is less than 50 r / min, and the operating state is normal, the fuzzy control weight coefficient is set. , adaptive control weight coefficient If the speed fluctuation standard deviation is less than 0.5 r / min, the absolute value of the load change rate is less than 10 N·m / s, the speed error is greater than or equal to 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient. , adaptive control weight coefficient If the speed fluctuation standard deviation is greater than or equal to 0.5 r / min, the absolute value of the load change rate is greater than or equal to 10 N·m / s, the speed error is less than 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient. , adaptive control weight coefficient If the speed fluctuation standard deviation is greater than or equal to 0.5 r / min, the absolute value of the load change rate is greater than or equal to 10 N·m / s, the speed error is greater than or equal to 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient , adaptive control weight coefficient ; If the operating state is overheating, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the running state is stall state, then the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is a poor combustion state, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is the emission exceeding the standard state, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is a mechanical failure state, the fuzzy control weight coefficient is set to 0.7, the adaptive control weight coefficient is 0.3;

[0178] S502: Perform weighted fusion on the fuzzy adjustment amount and the adaptive adjustment amount according to the weight coefficient fused in S501 to obtain the final adjustment amount:

[0179] According to the determined weight coefficients, the fuzzy adjustment values ​​and adaptive adjustment values ​​of the fuel injection system, valve train system, and ignition system are substituted into the following formulas for weighted fusion: ;

[0180] in, is the final adjustment amount, and are the weight coefficients of fuzzy control and adaptive control respectively, Parameter adjustment for the fuel injection system, valve train system, and ignition system , namely the fuzzy adjustment amount of the fuel injection system, the fuzzy adjustment amount of the valve system, and the fuzzy adjustment amount of the ignition system; The adaptive adjustment amount of the fuel injection system, valve train system, and ignition system, including the adaptive adjustment amount of the fuel injection system, i.e. the updated injection pulse signal , the adaptive adjustment amount of the valve transmission system is the updated electromagnetic force , ignition system adaptive adjustment amount, that is, the updated ignition advance angle .

[0181] Beneficial Effects: This invention achieves precise identification and early warning of the internal combustion engine's operating status by collecting multidimensional operating parameters in real time and extracting time-domain and frequency-domain features, combined with a deep neural network model. This system can promptly detect and respond to even the slightest changes in the engine's operating conditions, addressing the response lag of traditional control systems. During transient conditions such as sudden acceleration or hill climbing, the system can complete state identification and control parameter adjustments within milliseconds, ensuring the engine maintains optimal operating conditions. This ensures smooth power delivery and significantly improves fuel economy. It employs a dynamic fusion strategy of fuzzy control and adaptive control. Under steady-state conditions, fuzzy control is prioritized to ensure system stability, while under transient or abnormal conditions, adaptive control is automatically weighted to improve response speed. This intelligent hybrid control approach perfectly balances system stability and dynamic performance. Furthermore, a global optimization algorithm based on operating conditions enables coordinated control of the fuel injection system, valvetrain system, and ignition system. By dynamically adjusting the control parameters of each subsystem, the entire powertrain remains in optimal operating condition. DETAILED DESCRIPTION

[0182] A method for accurately sensing and controlling the combustion state of an internal combustion engine comprises the following steps:

[0183] S1. Parameter acquisition: Collect the parameters of the internal combustion engine under different working conditions, and obtain time domain characteristics, frequency domain characteristics, and working condition characteristics based on the collected parameters;

[0184] The S1 is specifically:

[0185] The parameters collected for the internal combustion engine under different working conditions, i.e., different loads and different speeds, include intake flow, exhaust flow, intake pressure, exhaust pressure, coolant temperature, oil pressure, exhaust emission components, vibration signals, valve train displacement, injection pulse signals, injection amount, and injection timing;

[0186] The acquisition of time domain features is specifically as follows:

[0187] Calculate the effective value of the vibration signal , to reflect the average energy level of the vibration signal: ;

[0188] in, is the number of vibration signal sampling points; is the instantaneous value of the vibration signal at the i-th sampling point;

[0189] Calculating kurtosis , used to evaluate whether there are abnormal impact components in the vibration signal: ;

[0190] in, is the mean value of the vibration signal; is the standard deviation of the vibration signal;

[0191] Calculate the pulse factor to help identify the pulse characteristics in the vibration signal: Pulse Factor = ;

[0192] in is the effective value of the vibration signal, is the maximum instantaneous value in the vibration signal;

[0193] The specific steps of obtaining frequency domain features are as follows:

[0194] Perform spectrum analysis on exhaust noise, extract 1 / 3 octave energy spectrum, and analyze the energy distribution of different frequency components;

[0195] Center frequency : ;in, is the frequency band number;

[0196] Frequency band range : ;

[0197] Sound pressure level SPL calculation: SPL=20 ;in, ; is the actual sound pressure;

[0198] Obtain the characteristics of exhaust noise in the frequency domain, thereby reflecting the stability of internal combustion engine combustion processes;

[0199] The operating condition characteristics specifically include speed fluctuation rate, intake pressure gradient, load change rate, intake flow rate, exhaust flow rate, fuel injection amount, exhaust pressure, coolant temperature, oil pressure, exhaust emission composition, injection pulse signal, fuel injection amount, and injection timing; the speed fluctuation rate, intake pressure gradient, and load change rate are obtained by the following methods, and the remaining operating condition characteristics are obtained by direct acquisition:

[0200] Calculate the speed fluctuation rate to measure the change of internal combustion engine speed over time: Speed ​​fluctuation rate = ;

[0201] in, is the standard deviation of the rotational speed; is the average value of the rotational speed;

[0202] Calculate the intake pressure gradient to reflect the change trend of intake pressure over time; ;

[0203] in, is the intake pressure gradient, For the time point The intake pressure value at the moment, For the time point The intake pressure value at the moment, is the time interval.

[0204] S2. Establishing a neural network model based on the time domain features, frequency domain features, and operating condition features acquired in S1, training the neural network model based on the parameters acquired in S1, and identifying the operating state of the internal combustion engine based on the trained neural network model;

[0205] The neural network model is established and trained in S2 specifically as follows:

[0206] A key feature set is established based on the time domain features, frequency domain features, and working condition features obtained by S1 as the input of the neural network model input layer;

[0207] Set at least three hidden layers and select the optimal number of neurons between adjacent hidden layers.

[0208] After optimizing the number of neurons, the number of neurons in the three hidden layers is 10, 20, and 20 respectively. Determine the number of neurons in each layer:

[0209] 1. Based on a single-layer neural network, the number of neurons is n*[1 2 3 4 5 6 7 8 9 10], n=5~10, assuming 8n is the best.

[0210] 2. Based on 8n neurons, optimize the two-layer neural network, 8n*f1*[0.1 0.9; 0.2 0.8; 0.3 0.7; 0.40.6 0.5 0.5;], where f1∈[0.2 0.3] or so, assuming [0.5 0.5] is the best.

[0211] 3. Based on 8n neurons, optimize the three-layer neural network, 8n*f2*[0.33 0.33 0.33; 0.25 0.50.25; 0.2 0.4 0.4; 0.3 0.4 0.3;], where f2∈[0.1 0.2] or so, assuming [0.2 0.4 0.4] is the best.

[0212] 4. And so on, optimize the four- and five-layer neural network.

[0213] 5. Compare, 8n, 8n*f1*[0.5 0.5], 8n*f2*[0.2 0.4 0.4], 8n*f3*[0.25 0.25 0.250.25], 8n*f4*[0.2 0.2 0.2 0.2 0.2], and take the best one as the final architecture and parameters;

[0214] The ReLU activation function is used for the output signal of neurons in each hidden layer: Perform nonlinear transformation on it, insert Dropout layer between adjacent hidden layers, and set the dropout rate of Dropout layer to 0.3-0.5;

[0215] The number of neurons in the output layer is set to the total number of engine operating state categories, and the Softmax activation function is applied to the output layer. ;in, is the output value of the i-th neuron in the output layer, is the total number of neurons, and the output value of the neuron is mapped to the corresponding probability distribution, which represents the probability of the internal combustion engine being in the corresponding operating state category;

[0216] The time domain features, frequency domain features, and working condition features in S1 are set as training sets for neural network model parameter learning;

[0217] Model initialization: Initialize the parameters of the neural network model. This embodiment uses the Xavier initialization method, specifically:

[0218] Determine the number of neurons in each layer

[0219] Input layer:

[0220] First hidden layer:

[0221] Second hidden layer:

[0222] The third hidden layer:

[0223] Output layer: (Determined based on the total number of internal combustion engine operating status categories)

[0224] in, is the number of neurons in the input layer, 、 、 are the number of neurons in the first, second and third hidden layers respectively, is the number of neurons in the output layer;

[0225] Initialize the weight matrix

[0226] For each layer, according to the ReLU activation function, use the Xavier initialization formula: ;

[0227] in, is the number of input neurons in the current layer, is the number of output neurons in the current layer. represents the weight matrix, Represents a normal distribution

[0228] Specific initialization steps:

[0229] Assume that the number of features in the input layer , the number of neurons in the output layer .

[0230] The weight matrix input to the first hidden layer :

[0231] ;

[0232] ;

[0233] Standard deviation ;

[0234] Initialize the weight matrix The shape is ;

[0235] The weight matrix from the first hidden layer to the second hidden layer :

[0236] ;

[0237] ;

[0238] Standard deviation ;

[0239] Initialize the weight matrix The shape is ;

[0240] The weight matrix from the second hidden layer to the third hidden layer :

[0241] ;

[0242] ;

[0243] Since the number of neurons must be an integer, assuming ;

[0244] Standard deviation ;

[0245] Initialize the weight matrix The shape is ;

[0246] The weight matrix from the third hidden layer to the output layer :

[0247] ;

[0248] ;

[0249] Standard deviation ;

[0250] Initialize the weight matrix The shape is ;

[0251] Training parameter settings: set the number of training iterations to 50 to 200 rounds, and set the initial learning rate to 0.001-0.0001;

[0252] Training loop: For each training round, traverse all the data in the training set;

[0253] Perform forward propagation: data passes through the input layer, at least one hidden layer, and the output layer in sequence;

[0254] Calculate the loss function: For the running state recognition task, the cross entropy loss function is used: Cross-EntropyLoss= ,

[0255] in, is the true value, is the predicted value;

[0256] For the performance index prediction task, the mean square error MSE loss function is used: MSE Loss= ;

[0257] in, is the true value, is the predicted value, is the sample size;

[0258] Perform backpropagation: Calculate the gradient of the neural network model parameters:

[0259] Assume that the output layer has neurons, and the corresponding prediction value is , the true value is ;

[0260] For the cross entropy loss function, the gradient of the output layer is calculated as follows: ;

[0261] in, The output layer The linear output of the neuron is the input of the Softmax activation function. The output layer The predicted value of a neuron, The output layer The true value of each neuron;

[0262] For the mean squared error loss function, the gradient of the output layer is calculated as follows: ;

[0263] Starting from the output layer, calculate the gradient of the hidden layer layer by layer; assuming the current layer is The layer has an activation function of ReLU, and the gradient of the layer is calculated as follows: ;

[0264] in: It is Tier The linear output of a neuron, It is Tier The linear output of each neuron; is the output of the ReLU activation function; = ;

[0265] For each layer's weight matrix and the bias vector , its gradient is calculated as follows: ; ;

[0266] in: It is Tier The neuron and Tier The weights between neurons; It is Tier The activation output of each neuron; It is Tier The bias of each neuron;

[0267] Using the calculated gradient, the parameters of the neural network are updated through the Adam optimizer. The parameter update formula is as follows: ; ;in, is the learning rate, which controls the step size of parameter updates.

[0268] The operating state of the internal combustion engine identified in S2 includes:

[0269] Receive the normalized probability values ​​of six states output by the neural network model, including normal state, overheating state, stall state, poor combustion state, excessive emission state, and mechanical failure state;

[0270] The preset priority order is mechanical failure state > stall state > overheating state > emission exceeding standard state > poor combustion state > normal state. The confidence level is determined based on the probability values ​​of the six states output by the neural network:

[0271] S201. Check whether the confidence level of any operating state reaches a high confidence threshold, that is, the probability of any operating state is ≥ 0.85, and the probability of the operating state is at least twice that of the operating state with the second highest probability. If so, directly determine that the current state is in that operating state;

[0272] S202. If no high confidence state appears, all medium confidence states are screened out, that is, the operating state probabilities are in [0.5, 0.85), and parameter verification is performed on them:

[0273] Overheating: Verify that the coolant temperature is >103°C; Mechanical failure: Verify that the vibration kurtosis is >3.5 or the pulse factor is >5; Poor combustion: Verify that the exhaust CO / HC exceeds the standard by 20%; Excessive emissions: Verify that NOx exceeds the standard by 15% or that particulate matter is abnormal; Stalling: Verify that the speed is <1000 RPM; Normal state: Verify that all parameters are within the safe range;

[0274] If only one state satisfies both the probability condition and the parameter condition, the current operating state is determined to be that state; if multiple states satisfy both the probability condition and the parameter condition, starting from the highest priority state, the first operating state whose real-time parameters do not conflict with the operating state that appears is selected as the current operating state.

[0275] S203. If two or more operating states are high-confidence states, that is, the probability of the operating state is ≥0.85, they are sorted according to the preset priority and the real-time parameters are checked to see if they conflict with the existing operating states. Starting from the highest priority state, the first operating state whose real-time parameters do not conflict with the existing operating state is selected as the current operating state. For example, if both the overheat and stall states are high-confidence states and the speed is detected to be less than 800 RPM, the stall state is prioritized.

[0276] S204. If the current operating state cannot be determined in S201-S203, the operating states with operating state probabilities greater than 0.3 are screened, and the state with the highest priority is selected as the current operating state in order of priority; for example, when the emission exceeding standard probability is 0.4 and the combustion failure probability is 0.6, the emission exceeding standard is ultimately determined to be the emission exceeding standard because the emission exceeding standard has a higher priority.

[0277] S205, dynamic monitoring and control adjustment

[0278] Continuously monitor parameter changes after status determination: When any parameter changes by more than 10%, immediately return to S201 to trigger the determination process.

[0279] Perform control adjustments based on the current state:

[0280] Overheating: Reduce fuel injection by 0.2mg and increase cooling fan speed by 20%; Stalling: Advance ignition timing by 3° and increase intake air volume; Mechanical failure: Reduce load to 50% and trigger an audible and visual alarm; Compare actual and expected parameter deviations every 200 milliseconds and fine-tune control commands;

[0281] S306, threshold dynamic update:

[0282] The system automatically updates the judgment threshold every 5 seconds:

[0283] Based on operational stability: When the speed fluctuation rate is less than 0.3 r / min and the load change rate is less than 8 N· m / s, the normal state threshold is increased to 0.75;

[0284] Based on abnormal signals: When the vibration kurtosis is greater than 4.0, the mechanical fault threshold is reduced to 0.6;

[0285] Based on operating condition adaptation: During long-term high-load operation, the combustion failure threshold is reduced to 0.65.

[0286] S3, based on the parameters collected by S1 and the operating state of the internal combustion engine obtained by S2, establish a fuzzy language and obtain the fuzzy adjustment values ​​of the fuel injection system, the valve train system, and the ignition system;

[0287] Convert the collected load parameters into fuzzy linguistic variables:

[0288] The load parameter is divided into three fuzzy linguistic variables: low load, medium load and high load using triangular membership function, and the definition domains are: low load: definition domain , the membership function is:

[0289] ;

[0290] Medium load: Domain range , the membership function is: ;

[0291] High load: Domain range , the membership function is:

[0292] ;

[0293] in, is the actual measured load value, is the minimum value in the low load range, It is the maximum value of the low load range and also the transition point from low load to medium load. is the center value of the medium load range, which is used to define the membership function of the medium load. It is the maximum value of the medium load range and also the transition point from medium load to high load. This is the maximum value in the high load range.

[0294] Convert the collected speed parameters into fuzzy linguistic variables:

[0295] The speed parameter is divided into three fuzzy linguistic variables: low speed, medium speed, and high speed using the trapezoidal membership function:

[0296] The following definition domain ranges are determined according to the actual speed range;

[0297] Low speed: Use Z-type membership function, the domain range is [0, ], the membership function is:

[0298] ;

[0299] in, and In order to make the low-speed fuzzy mathematical model adapt to the real scene, the adjustable morphological control parameters are determined according to the actual speed range. The specific acquisition method is:

[0300] ;

[0301] in, is the reference ignition advance angle, is the internal combustion engine speed, is the compression pressure, is the excess air coefficient of the mixture, 、 、 、 is the reference model coefficient;

[0302] Based on the parameters collected by S1, the least squares method is used to estimate the parameters 、 、 、 .

[0303] The goal of the least squares method is to minimize the sum of squared errors:

[0304] ;

[0305] in, is the sum of squared errors, is the total number of collected data, is the i-th data of the internal combustion engine speed, is the i-th data of compression pressure, is the excess air coefficient of the mixture of the i-th data.

[0306] To minimize , we are 、 、 、 Find the partial derivatives separately and set them to zero:

[0307] ;

[0308] ;

[0309] ;

[0310] ;

[0311] Simplify the above equations into matrix form:

[0312] ;

[0313] By solving the above matrix equation, we can get the parameters 、 、 、 estimated value.

[0314] Medium speed: using triangular membership function, the domain range is [ , ], the membership function is: ; High speed: Use S-type membership function, the domain range is [ , ∞ ), the membership function is ;in, and It is an adjustable shape control parameter determined according to the actual speed range. : The domain boundary of the low speed Z-type membership function and the medium speed triangle membership function, 、 :The medium speed is the endpoint of the domain interval of the triangle membership function itself, which determines the range of the base of the triangle. , ]arrive[ , ] is the interval of medium speed membership change; :The domain boundary of medium speed and high speed is the S-type membership function, [ ,+∞) is the judgment range of high speed; is the actual speed collected;

[0315] Convert the collected coolant temperature parameters into fuzzy linguistic variables

[0316] Low temperature: domain range , the membership function is: ;

[0317] Moderate Temperature: Domain Range , the membership function is:

[0318] ;

[0319] High temperature: domain range , the membership function is:

[0320] ;

[0321] in is the collected coolant temperature, is the starting temperature of the low temperature range, is the maximum temperature in the low temperature range, the starting temperature in the medium temperature range, The end temperature of the medium temperature range and the start temperature of the high temperature range, End temperature of the high temperature range. : The center temperature value of low temperature, indicating the center point of the low temperature range, : The center temperature value of the medium temperature, indicating the center point of the medium temperature range, : The center temperature value of high temperature, indicating the center point of the high temperature range, is the standard deviation of the low temperature range, indicating the degree of diffusion in the low temperature range. is the standard deviation of the medium temperature range, indicating the degree of diffusion in the medium temperature range. is the standard deviation of the high temperature range, indicating the degree of diffusion in the high temperature range.

[0322] The oil pressure is converted into fuzzy linguistic variables:

[0323] Low pressure: domain range , the membership function is:

[0324] ;

[0325] Moderate Pressure: Domain Range , the membership function is:

[0326] ;

[0327] High pressure: domain definition , the membership function is:

[0328] ;

[0329] in is the collected oil pressure parameter, It is the minimum value of the oil pressure, usually the lowest safety pressure of the system.

[0330] is the starting pressure of the low pressure range, It is the maximum pressure of the low pressure range and the starting pressure of the medium pressure range. The center pressure of the medium pressure range, It is the maximum pressure of the medium pressure range and the starting pressure of the high pressure range. This is the maximum pressure in the high pressure range.

[0331] The exhaust emission components are converted into fuzzy linguistic variables:

[0332] Low emissions: defining the domain , corresponding to the case where the concentration of pollutants in the exhaust gas is low, the membership function is:

[0333] ;

[0334] Medium emissions: Defining the domain , corresponding to the situation where the concentration of pollutants in the exhaust gas is at a medium level, the membership function is:

[0335] ;

[0336] High emissions: defining the domain , corresponding to the situation where the concentration of pollutants in the exhaust gas is too high, the membership function is:

[0337] ;

[0338] in is the actual measured exhaust emission value, It is the starting emission value of the low emission range, usually the minimum limit of the emission standard. It is the starting point for the transition from low emissions to medium emissions. is the end point of the transition from low emissions to medium emissions. : Starting emission value of the medium emission range, is the central emission value of the medium emission range, It is the starting point for the transition from medium emissions to high emissions. is the starting emission value of the high emission range, It is the maximum emission value in the high emission range.

[0339] The vibration signal is converted into fuzzy linguistic variables using the Z-type membership function.

[0340] Low Vibration: Domain Definition , corresponding to the case where the vibration amplitude of the internal combustion engine is small, the membership function is:

[0341] ;

[0342] Moderate Vibration: Domain Range , corresponding to the case of medium vibration amplitude of the internal combustion engine, the membership function is:

[0343] ;

[0344] High Vibration: Domain Range , corresponding to the case where the vibration amplitude of the internal combustion engine is large, the membership function is:

[0345] ;

[0346] in is the actual measured vibration signal value. This is the minimum value of the vibration signal, usually the background vibration level during normal operation. is the starting vibration value of the low vibration range. It is the starting point of the transition from low vibration to medium vibration. This is the starting vibration value of the medium vibration range. It is the center vibration value of the medium vibration range. It is the starting point of the transition from medium vibration to high vibration. is the starting vibration value of the high vibration range. It is the maximum vibration value in the high vibration range. After exceeding this vibration value, the membership remains at 1.

[0347] The displacement of the valve train is converted into fuzzy linguistic variables:

[0348] The triangular membership function is used to divide the valve train system displacement into three fuzzy linguistic variables: low displacement, medium displacement, and high displacement:

[0349] The definition domain range is determined according to the displacement range of the actual valve train system;

[0350] Low displacement: domain range , the membership function is:

[0351] ;

[0352] Moderate displacement: Domain range , the membership function is:

[0353] ;

[0354] High displacement: domain range , the membership function is:

[0355] ;

[0356] in, is the lower limit of the low displacement range, less than When , the displacement is completely low; It is the dividing point between low displacement and medium displacement. When the range is within the range, the low displacement membership gradually decreases, and the medium displacement membership gradually increases; It is the dividing point between medium displacement and high displacement. When , it starts to be high displacement; Is the upper limit of the high displacement interval, the displacement is greater than or equal to When the displacement is high, a fuzzy rule base is constructed based on all the above-mentioned fuzzy languages, all the operating states of the internal combustion engine, and the corresponding parameters in S1. The fuzzy rules in this embodiment are shown in Table 1, where Z represents the injection quantity, ignition advance angle, and valve opening, and has five subsets with language values ​​of VS, S, M, B, and VB. E and F represent the load and speed, and have five subsets with language values ​​of NM, NS, ZE, PS, and PM.

[0357] Table 1 Fuzzy rule table:

[0358]

[0359] Rule 1: Basic stable operation under normal operating conditions

[0360] IF (load = low load) AND (speed = low speed) AND (state = low confidence normal / slightly abnormal state), THEN (injection amount = small baseline value), (ignition advance angle = small advance angle baseline value), (valve opening = small opening baseline value).

[0361] Note: When the engine is in a low load, low speed and basically normal state, it maintains stable operation with smaller injection amount, advance angle and valve opening to ensure fuel economy and low emissions.

[0362] Rule 2: Stable operation at medium load

[0363] IF (load = medium load) AND (speed = medium speed) AND (state = low confidence normal / slightly abnormal state), THEN (injection amount = medium baseline value), (ignition advance angle = medium advance angle baseline value), (valve opening = medium opening baseline value).

[0364] Note: Under moderate operating conditions and normal status, use moderate parameter settings to ensure a balance between power output and fuel economy.

[0365] Rule 3: High-load power boost

[0366] IF (load = high load) AND (speed = medium speed) AND (state = low confidence normal / slightly abnormal state), THEN (injection amount = larger value), (ignition advance angle = larger advance angle value), (valve opening = larger opening value).

[0367] Note: Under high load and medium speed, increase the fuel injection amount, advance the ignition and increase the valve opening to provide sufficient power to meet high load requirements.

[0368] Rule 4: Economic adjustment at low load and high speed

[0369] IF (load = low load) AND (speed = high speed) AND (state = low confidence normal / slightly abnormal state), THEN (injection amount = medium to small value), (ignition advance angle = medium to small advance angle value), (valve opening = medium to small opening value).

[0370] Note: At low load and high speed, appropriately reduce the injection amount and ignition advance angle, and at the same time reduce the valve opening to avoid excessive fuel consumption and improve fuel economy.

[0371] Rule 5: Extreme performance adjustment for high load and high speed

[0372] IF (load = high load) AND (speed = high speed) AND (state = low confidence normal / slightly abnormal state), THEN (injection amount = maximum allowable value), (ignition advance angle = maximum safe advance angle value), (valve opening = maximum allowable opening value).

[0373] Note: Under extreme operating conditions of high load and high speed, adjust the injection quantity, ignition advance angle and valve opening to the maximum allowable values ​​to ensure the maximum power output of the engine, but strictly monitor the operating status to prevent overload.

[0374] Rule 6: Overheating warning and prevention

[0375] IF (status = high confidence overheat status) AND (coolant temperature = high), THEN (reduce injection quantity to X% of current), (retard ignition timing by Y crankshaft angle degrees), (increase coolant pump speed to Z rpm).

[0376] Description: By reducing the amount of fuel injection and delaying the ignition time, the combustion temperature is lowered, while the coolant circulation is accelerated to prevent overheating.

[0377] Rule 7: Adjustment when coolant temperature is high but load is low

[0378] IF (status = medium confidence abnormal state) AND (coolant temperature = medium to high) AND (load = low load), THEN (injection amount remains at current value), (ignition timing is advanced by α crankshaft angle), (maintain current coolant pump speed).

[0379] Note: When overheating with medium confidence and low load, the ignition timing is appropriately advanced to optimize combustion while maintaining the coolant pump speed to ensure normal heat dissipation.

[0380] Rule 8: Stall Recovery (High Confidence)

[0381] IF (state = high confidence stall state) AND (speed = low speed), THEN (increase injection amount to current A%), (advance ignition timing by B crankshaft angle), (increase throttle opening to C%).

[0382] Note: When a high-confidence stall occurs and the engine speed is low, increase the fuel injection amount, ignition advance angle, and throttle opening to increase the intake volume and help the engine return to normal speed.

[0383] Rule 9: Stall Recovery (Medium Confidence)

[0384] IF (state = medium confidence abnormal state) AND (speed = low speed) AND (load = low load), THEN (injection amount is increased slightly), (ignition timing is advanced by a small angle), (maintain current throttle opening).

[0385] Note: When stalling with medium confidence and at low load and speed, slightly adjust the injection amount and ignition timing to avoid stalling due to insufficient air intake.

[0386] Rule 10: Adjustment of poor combustion status (high confidence)

[0387] IF (status = high confidence poor combustion status) AND (unburned hydrocarbons in exhaust emissions = high), THEN (check whether the injection system is faulty), (adjust the injection amount to a more accurate Dmg), (optimize the ignition advance angle to E degrees crankshaft angle).

[0388] Note: For high-confidence poor combustion and high levels of unburned hydrocarbons in the exhaust gas, check the injection system and accurately adjust the injection amount and ignition advance angle to improve combustion efficiency.

[0389] Rule 11: Adjustment of the poor combustion state (medium confidence)

[0390] IF (status = medium confidence abnormal state) AND (carbon monoxide in exhaust emissions = medium to high) AND (the mixture is too rich), THEN (reduce the injection amount Fmg), (retard the ignition timing G degrees crankshaft angle), (check whether the air filter is blocked).

[0391] Note: When the combustion is poor with medium confidence and the mixture is too rich and the carbon monoxide is high, reduce the fuel injection amount and delay the ignition timing. At the same time, check the air filter to ensure sufficient intake air and optimize the combustion process.

[0392] Rule 12: Adjustment of Emission Exceedance Status (High Confidence)

[0393] IF (status = high confidence emission exceeding standard status) AND (nitrogen oxides in exhaust gas = high), THEN (retard ignition timing by H crankshaft angle degrees), (increase exhaust gas recirculation rate to 1%), (check whether the three-way catalytic converter is working properly).

[0394] Note: When high-confidence emissions exceed the standard and nitrogen oxides are high, delay the ignition timing to lower the combustion temperature, increase the exhaust gas recirculation rate to reduce NOx generation, and check the three-way catalytic converter to ensure that it is properly purifying the exhaust gas.

[0395] Rule 13: Adjustment of Exceeding Emissions Standards (Medium Confidence)

[0396] IF (state = medium confidence abnormal state) AND (carbon soot in exhaust = medium to high) AND (load = medium load), THEN (optimize injection quantity distribution), (advance ignition timing by J crankshaft angle), (adjust intake air quantity to K%).

[0397] Note: When emissions exceed the standard with medium confidence and the exhaust contains high levels of soot, and the load is medium, optimize the fuel injection distribution, advance the ignition timing, and adjust the intake volume to improve the combustion process and reduce soot emissions.

[0398] Rule 14: Emergency Adjustment for Mechanical Failure (High Confidence)

[0399] IF (status = high confidence mechanical fault status) AND (vibration signal amplitude = high), THEN (reduce engine power output to L%), (switch to backup control mode), (immediately issue a fault alarm signal).

[0400] Note: When a high-confidence mechanical fault occurs and the vibration amplitude is high, reduce the engine power output, switch to the backup control mode to maintain basic operation, and immediately alarm to prompt maintenance.

[0401] Rule 15: Mechanical Fault Status (Medium Confidence) Monitoring and Adjustment

[0402] IF (status = medium confidence abnormal state) AND (vibration signal amplitude = medium to high) AND (running time = long), THEN (gradually reduce the injection amount Mmg), (check whether the oil pressure is normal), (record the fault characteristic information).

[0403] Note: In the case of a medium-confidence mechanical fault with high vibration and long-term operation, gradually reduce the fuel injection volume to reduce the engine load, check the oil pressure to ensure normal lubrication, and record the fault characteristics for subsequent diagnosis.

[0404] Each rule consists of an IF condition and a THEN conclusion, where the condition is defined by a combination of fuzzy language. The fuzzy language obtained in real time by the internal combustion engine, the internal combustion engine operating status, and the corresponding parameters in S1 are matched with the rule base one by one to calculate the excitation strength of each rule.

[0405] Perform membership aggregation operation on the logical operators in the condition part: if it contains AND operator, take the minimum value of the condition part membership as the excitation strength of the rule; if it contains OR operator, take the maximum value of the condition part membership as the excitation strength of the rule;

[0406] According to the excitation intensity of each rule and the corresponding conclusion part, the corresponding membership function is truncated according to the excitation intensity to obtain the truncated output fuzzy set;

[0407] The output fuzzy sets of all rules are combined, that is, the membership function after truncation of any rule and the truncated membership function of the previous rule are maximized to generate the final output fuzzy set;

[0408] Defuzzify the final output fuzzy set:

[0409] Discretize the final output fuzzy set domain into i points , for each discrete point , calculate its corresponding membership value , the center of gravity method is used to convert the final output fuzzy set into an accurate control signal, and the calculation formula is: ;

[0410] in, is the discrete point of the output universe, is its membership value, is the parameter adjustment amount; f1 is the scaling factor, the domain of which is [0 1], and is obtained by querying the mapping function or mapping table based on the throttle and speed.

[0411] The parameter adjustment amount The specific change values ​​of the internal combustion engine control parameters are clarified, including the fuzzy adjustment amount of the fuel injection system, the fuzzy adjustment amount of the valve transmission system, and the fuzzy adjustment amount of the ignition system.

[0412] S4. Based on the operating state of the internal combustion engine obtained in S2, adaptive control models of the fuel injection system, the valve drive system, and the ignition system are established respectively, and adaptive adjustment values ​​of the fuel injection system, the valve drive system, and the ignition system are obtained;

[0413] The desired fuel injection system has the characteristics of fast response and small overshoot. The reference model is:

[0414] ;

[0415] in, is the transfer function of the fuel injection system, is the injection amount of the reference model at time t, The injection pulse signal of the reference model at time t is: For the fuel injection system gain, is the time constant of the fuel injection system, s is the complex frequency domain variable of Laplace transform;

[0416] Substitute the fuel injection system data collected by S1, namely the injection pulse signal, injection amount, and injection time, into the reference model and perform iterative calculation to determine the fuel injection system gain in the model. and the fuel injection system time constant ;

[0417] Determine the error signal, which is defined as the difference between the reference model output and the actual output of the fuel injection system:

[0418] ;

[0419] in, is the fuel injection system error signal, is the amount of fuel injected by the fuel injection system within time t;

[0420] Set performance indicators: ;

[0421] in, It is a fuel injection system performance indicator, representing the sum of squares of error signals, and is used to measure the performance of the system.

[0422] To minimize performance , use the gradient descent method to update the control parameters :

[0423] ;

[0424] in: is the control parameter at the current moment, is the learning rate, which controls the step size of parameter update; For a certain moment; is the difference between this moment and the previous moment; is the control parameter for the next moment; is the gradient of the performance index with respect to the control parameter, expressed as: ; is the gradient of the error signal with respect to the control parameter;

[0425] Adjustable fuel gain according to the updated and fuel error signal , generating a control signal To adjust the fuel injection system's injection quantity:

[0426] ;

[0427] in, is the updated injection pulse signal, is the injection pulse signal before updating; the injection pulse signal after updating Adaptive adjustment for the fuel injection system.

[0428] The desired valvetrain system has stable dynamic response characteristics, and the reference model is the expected trajectory of the valve displacement:

[0429] ;

[0430] in, Valve displacement The second derivative with respect to time t represents the valve acceleration; is the current input to the solenoid valve at time t, The proportionality constant between the electromagnetic force and the square of the input current; is the actual displacement of the valve at time t; is the spring stiffness coefficient, which represents the restoring force of the spring on the valve displacement; The quality of the valve and its transmission system;

[0431] The error signal is defined as the difference between the reference displacement and the actual displacement:

[0432] ;

[0433] in, is the error signal of the valve train system; is the reference displacement of the valve train; is the actual displacement of the valve train;

[0434] Set performance indicators: ;

[0435] in, It is a performance indicator of the valvetrain system, representing the sum of squares of error signals, and is used to measure the performance of the system.

[0436] To minimize performance , use the gradient descent method to update the control parameters :

[0437] ;

[0438] in, is the control parameter at the current moment, is the learning rate, which controls the step size of parameter updates, is the gradient of the performance index with respect to the control parameter: ;

[0439] is the gradient of the error signal with respect to the control parameter;

[0440] Parameter update: The updated control parameters are used to generate new electromagnetic force signals , thereby adjusting the opening and closing of the valve:

[0441] ;

[0442] in, For in time The valve control parameters at this time are used to adjust the changes of the electromagnetic force signal; In time The new electromagnetic force signal value at time ;

[0443] Updated electromagnetic force It is the adaptive adjustment amount of the valve transmission system.

[0444] The desired ignition system has precise ignition advance angle control characteristics, and the reference model is:

[0445] ;

[0446] in, is the reference ignition advance angle; is the internal combustion engine speed; is the compression pressure; is the excess air coefficient of the mixture; 、 、 、 are the reference model coefficients, which are estimated based on the parameters collected by S1 using the least squares method;

[0447] The error signal is defined as the difference between the reference ignition advance angle and the actual ignition advance angle: ;

[0448] in, is the error signal of the ignition system, is the reference ignition advance angle, is the actual ignition advance angle;

[0449] Set performance indicators: ;

[0450] in: It is an ignition system performance indicator, representing the sum of squares of error signals, and is used to measure the performance of the system.

[0451] To minimize performance , use the gradient descent method to update the control parameters :

[0452] ;

[0453] in: Control parameters at the next moment; is the control parameter at the current moment; is the learning rate, which controls the step size of parameter update; is the gradient of the performance index with respect to the control parameter, expressed as:

[0454] is the gradient of the error signal with respect to the control parameter;

[0455] The updated control parameters are used to generate new ignition advance angle adjustment , thereby adjusting the ignition timing:

[0456] ;

[0457] Updated ignition advance angle It is the adaptive adjustment amount of the ignition system.

[0458] S5. Based on the operating state of the internal combustion engine obtained in S2, the fuzzy adjustment amount and the adaptive adjustment amount obtained in S3 and S4 are weighted and fused to obtain the final adjustment amount:

[0459] S501, determine the weight coefficient of the fusion of the fuzzy adjustment amount and the adaptive adjustment amount obtained in S3 and S4:

[0460] Monitor the operating data of the internal combustion engine in S1 at real-time speed and load in real time, and calculate the operating stability index, including speed stability index and load stability index, specifically:

[0461] Calculate the speed stability index, that is, calculate the standard deviation of the speed fluctuation rate in a 3-second sliding time window:

[0462] The speed of the internal combustion engine is collected in real time. The collection frequency is set between 10 and 100 times per second. A sliding time window of 3 seconds is set to store the continuously collected speed data. The average speed value is calculated within the sliding time window. , the formula is as follows: ;

[0463] in, is the number of data points in the window, It is The speed value at the data point,

[0464] Calculate the standard deviation of the speed fluctuation rate within the window : ;

[0465] This standard deviation reflects the degree of speed fluctuation within the window. As time goes by, the sliding time window continues to move forward, adding the latest speed data point each time and discarding the first data point in the time period, keeping the window length at 3 seconds.

[0466] Repeat the above calculation process to obtain the real-time speed fluctuation standard deviation. If the standard deviation is less than 0.5 r / min, the speed is determined to be in a stable state; if it is greater than or equal to 0.5 r / min, the speed is determined to be in an unstable state.

[0467] Calculate the load stability index, that is, calculate the absolute value of the load change rate:

[0468] Collect the load data of the internal combustion engine in real time. The collection frequency is consistent with the speed data collection frequency to ensure the time synchronization and consistency of the data. Calculate the load change at two adjacent collection moments. With time interval The ratio of load change rate is as follows:

[0469] ;

[0470] in, is the load value at the current moment, is the load value at the previous moment, The time interval between two acquisition moments is usually between 0.01 seconds and 1 second;

[0471] Since we are concerned about the severity of the load change, whether the load is increasing or decreasing, we take the absolute value of the load change rate: ;

[0472] The larger the absolute value of the load change rate, the more drastic the load change. If the absolute value of the load change rate is less than 10 N·m / s, the load is determined to be in a stable state; if it is greater than or equal to 10 N·m / s, the load is determined to be in an unstable state.

[0473] Calculate the error between the actual speed output and the expected speed output of the internal combustion engine, and combine the speed stability index and load stability index to set the weight coefficient values ​​under different speed fluctuation standard deviations and load change rate absolute values, as well as error state and operating state combinations:

[0474] If the speed fluctuation standard deviation is less than 0.5 r / min, the absolute value of the load change rate is less than 10 N·m / s, the speed error is less than 50 r / min, and the operating state is normal, then the fuzzy control weight coefficient is set. , adaptive control weight coefficient ;

[0475] If the speed fluctuation standard deviation is less than 0.5 r / min, the absolute value of the load change rate is less than 10 N·m / s, the speed error is greater than or equal to 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient , adaptive control weight coefficient If the speed fluctuation standard deviation is greater than or equal to 0.5 r / min, the absolute value of the load change rate is greater than or equal to 10 N·m / s, the speed error is less than 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient. , adaptive control weight coefficient If the speed fluctuation standard deviation is greater than or equal to 0.5 r / min, the absolute value of the load change rate is greater than or equal to 10 N·m / s, the speed error is greater than or equal to 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient , adaptive control weight coefficient ; If the operating state is overheating, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the running state is stall state, then the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is a poor combustion state, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is the emission exceeding the standard state, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is a mechanical failure state, the fuzzy control weight coefficient is set to 0.7, the adaptive control weight coefficient is 0.3;

[0476] S502. Based on the weight coefficients obtained in S501, the fuzzy adjustment amount and the adaptive adjustment amount are weightedly fused to obtain a final adjustment amount. Based on the determined weight coefficients, the fuzzy adjustment amount and the adaptive adjustment amount of the fuel injection system, the valve train system, and the ignition system are respectively substituted into the following formula for weighted fusion:

[0477]

[0478] in, is the final adjustment amount, and are the weight coefficients of fuzzy control and adaptive control respectively, Parameter adjustment for the fuel injection system, valve train system, and ignition system , namely the fuzzy adjustment amount of the fuel injection system, the fuzzy adjustment amount of the valve system, and the fuzzy adjustment amount of the ignition system; The adaptive adjustment amount of the fuel injection system, valve train system, and ignition system, including the adaptive adjustment amount of the fuel injection system, i.e. the updated injection pulse signal , the adaptive adjustment amount of the valve transmission system is the updated electromagnetic force , ignition system adaptive adjustment amount, that is, the updated ignition advance angle .

Claims

1. A method for accurately sensing and controlling the combustion state of an internal combustion engine, characterized by: The following steps are involved: S1. Parameter acquisition: Collect the parameters of the internal combustion engine under different working conditions, and obtain time domain characteristics, frequency domain characteristics, and working condition characteristics based on the collected parameters; S2. Establishing a neural network model based on the time domain features, frequency domain features, and operating condition features acquired in S1, training the neural network model based on the parameters acquired in S1, and identifying the operating state of the internal combustion engine based on the trained neural network model; S3, based on the parameters collected by S1 and the operating state of the internal combustion engine obtained by S2, establish a fuzzy language and obtain the fuzzy adjustment values ​​of the fuel injection system, the valve train system, and the ignition system; S4. Based on the operating state of the internal combustion engine obtained in S2, adaptive control models of the fuel injection system, the valve drive system, and the ignition system are established respectively, and adaptive adjustment values ​​of the fuel injection system, the valve drive system, and the ignition system are obtained; S5. Based on the operating state of the internal combustion engine obtained in S2, the fuzzy adjustment amount and the adaptive adjustment amount obtained in S3 and S4 are weightedly integrated to obtain a final adjustment amount.

2. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 1, characterized in that: The S1 is specifically: The parameters collected for the internal combustion engine under different working conditions, i.e., different loads and different speeds, include intake flow, exhaust flow, intake pressure, exhaust pressure, coolant temperature, oil pressure, exhaust emission components, vibration signals, valve train displacement, injection pulse signals, injection amount, and injection timing; The acquisition of time domain features is specifically as follows: Calculate the effective value of the vibration signal , to reflect the average energy level of the vibration signal: in, is the number of vibration signal sampling points; is the instantaneous value of the vibration signal at the i-th sampling point; Calculating kurtosis , used to evaluate whether there are abnormal impact components in the vibration signal: in, is the mean value of the vibration signal; is the standard deviation of the vibration signal; Calculate the pulse factor to help identify pulse characteristics in vibration signals: Pulse Factor = in is the effective value of the vibration signal, is the maximum instantaneous value in the vibration signal; The specific steps of obtaining frequency domain features are as follows: Perform spectrum analysis on exhaust noise, extract 1 / 3 octave energy spectrum, and analyze the energy distribution of different frequency components; Center frequency : in, is the frequency band number; Frequency band range : Sound pressure level SPL calculation: SPL=20 in, ; is the actual sound pressure; Obtain the characteristics of exhaust noise in the frequency domain, thereby reflecting the stability of internal combustion engine combustion processes; The operating condition characteristics specifically include speed fluctuation rate, intake pressure gradient, load change rate, intake flow rate, exhaust flow rate, fuel injection amount, exhaust pressure, coolant temperature, oil pressure, exhaust emission composition, injection pulse signal, fuel injection amount, and injection timing; the speed fluctuation rate, intake pressure gradient, and load change rate are obtained by the following methods, and the remaining operating condition characteristics are obtained by direct acquisition: Calculate the speed fluctuation rate to measure the change of internal combustion engine speed over time; Speed ​​fluctuation rate = in, is the standard deviation of the rotational speed; is the average value of the rotational speed; Calculate the intake pressure gradient to reflect the change trend of intake pressure over time: in, is the intake pressure gradient, For the time point The intake pressure value at the moment, For the time point The intake pressure value at the moment, is the time interval.

3. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 2, characterized in that: The neural network model is established and trained in S2 specifically as follows: A key feature set is established based on the time domain features, frequency domain features, and working condition features obtained by S1 as the input of the neural network model input layer; Set up at least three hidden layers, select the optimal number of neurons between adjacent hidden layers, and use the ReLU activation function for the neuron output signal of each hidden layer: Perform nonlinear transformation on it, insert Dropout layer between adjacent hidden layers, and set the dropout rate of Dropout layer to 0.3-0.5; The number of neurons in the output layer is set to the total number of engine operating state categories, and the Softmax activation function is applied to the output layer. ;in, is the output value of the i-th neuron in the output layer, is the total number of neurons, and the output value of the neuron is mapped to the corresponding probability distribution, which represents the probability of the internal combustion engine being in the corresponding operating state category; The time domain features, frequency domain features, and working condition features in S1 are set as training sets for neural network model parameter learning; Model initialization: Initialize the parameters of the neural network model; Training parameter settings: set the number of training iterations to 50 to 200 rounds, and set the initial learning rate to 0.001-0.0001; Training loop: For each training round, traverse all the data in the training set; Perform forward propagation: data passes through the input layer, at least one hidden layer, and the output layer in sequence; Calculate the loss function: For the running state recognition task, the cross entropy loss function is used: Cross-Entropy Loss= , in, is the true value, is the predicted value; For the performance indicator prediction task, the mean square error MSE loss function is used: MSE Loss= ; in, is the true value, is the predicted value, is the number of sample sampling points; Perform backpropagation: Calculate the gradient of the neural network model parameters: Assume that the output layer has neurons, and the corresponding prediction value is , the true value is ; For the cross entropy loss function, the gradient of the output layer is calculated as follows: in, The output layer The linear output of the neuron is the input of the Softmax activation function. The output layer The predicted value of a neuron, The output layer The true value of each neuron; For the mean squared error loss function, the gradient of the output layer is calculated as follows: Starting from the output layer, calculate the gradient of the hidden layer layer by layer; assuming the current layer is The layer has an activation function of ReLU, and the gradient of the layer is calculated as follows: in: It is Tier The linear output of a neuron, It is Tier The linear output of each neuron; is the output of the ReLU activation function; = For each layer's weight matrix and the bias vector , its gradient is calculated as follows: in: It is Tier The neuron and Tier The weights between neurons; It is Tier The activation output of each neuron; It is Tier The bias of each neuron; Using the calculated gradient, the parameters of the neural network are updated through the Adam optimizer. The parameter update formula is as follows: in, is the learning rate, which controls the step size of parameter updates.

4. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 3, characterized in that: The operating state of the internal combustion engine identified in S2 includes: Receive the normalized probability values ​​of six states output by the neural network model, including normal state, overheating state, stall state, poor combustion state, excessive emission state, and mechanical failure state; The preset priority order is mechanical failure state > stall state > overheating state > emission exceeding standard state > poor combustion state > normal state. The confidence level is determined based on the probability values ​​of the six states output by the neural network: S201. Check whether the confidence level of any operating state reaches a high confidence threshold, that is, the probability of any operating state is ≥ 0.85, and the probability of the operating state is at least twice that of the operating state with the second highest probability. If so, directly determine that the current state is in that operating state; S202. If no high confidence state appears, all medium confidence states are screened out, that is, the operating state probabilities are in [0.5, 0.85), and parameter verification is performed on them: Overheating state: verify whether the coolant temperature is greater than 103°C; Mechanical failure: Verify whether the vibration kurtosis is greater than 3.5 or the impulse factor is greater than 5; Poor combustion: Verify whether the exhaust CO / HC exceeds the standard by 20%; Excessive emissions: Verify whether NOx exceeds the standard by 15% or particulate matter is abnormal; Stall state: Verify that the speed is <1000 RPM; Normal status: Verify that all parameters are within the safe range; If only one state satisfies both the probability and parameter conditions, the current operating state is determined to be that state; if multiple states satisfy both the probability and parameter conditions, starting from the highest priority state, the first operating state whose real-time parameters do not conflict with the existing operating state is selected as the current operating state; S203. If two or more operating states are high-confidence states, that is, the probability of the operating state is ≥ 0.85, sort them according to the preset priority and check whether the real-time parameters conflict with the operating states. Starting from the highest priority state, select the first operating state whose real-time parameters do not conflict with the operating state as the current operating state; S204. If the current operating state cannot be determined in S201-S203, the operating states with operating state probabilities greater than 0.3 are screened, and the state with the highest priority is selected as the current operating state in order of priority; S205, dynamic monitoring and control adjustment Continuously monitor parameter changes after status determination: When any parameter changes by more than 10%, immediately return to S201 to trigger the determination process. Perform control adjustments based on the current state: Overheating: Reduce fuel injection by 0.2mg and increase cooling fan speed by 20% Stall state: advance the ignition timing by 3 degrees and increase the intake volume Mechanical failure: reduce the load to 50% and trigger the sound and light alarm Compare actual and expected parameter deviations every 200 milliseconds and fine-tune control instructions; S306: Dynamic update of threshold The system automatically updates the judgment threshold every 5 seconds: Based on operational stability: When the speed fluctuation rate is less than 0.3r / min and the load change rate is less than 8N·m / s, the normal state threshold is increased to 0.75; Based on abnormal signals: When the vibration kurtosis is greater than 4.0, the mechanical fault threshold is reduced to 0.6; Based on operating condition adaptation: During long-term high-load operation, the combustion failure threshold is reduced to 0.

65.

5. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 4, characterized in that: The S3 is specifically: Convert the collected load parameters into fuzzy linguistic variables: The load parameter is divided into three fuzzy linguistic variables: low load, medium load and high load using the triangular membership function. The low load is defined as follows: , the membership function is: Medium load: Domain range , the membership function is: High load: Domain range , the membership function is: in, is the actual measured load value, is the minimum value in the low load range, It is the maximum value of the low load range and also the transition point from low load to medium load. is the center value of the medium load range, which is used to define the membership function of the medium load. It is the maximum value of the medium load range and also the transition point from medium load to high load. It is the maximum value of the high load range; Convert the collected speed parameters into fuzzy linguistic variables: The speed parameter is divided into three fuzzy linguistic variables: low speed, medium speed, and high speed using the trapezoidal membership function: The following definition domain ranges are determined according to the actual speed range; Low speed: Use Z-type membership function, the domain range is [0, ], the membership function is: in, and The purpose is to adapt the low-speed fuzzy mathematical model to the real scene, and to determine the adjustable morphological control parameters according to the actual speed range; Medium speed: using triangular membership function, the domain range is [ , ], the membership function is: High speed: S-type membership function is used, and the domain range is [ , ∞ ), the membership function is: in, and It is an adjustable shape control parameter determined according to the actual speed range. : The domain boundary of the low speed Z-type membership function and the medium speed triangle membership function, 、 :The medium speed is the endpoint of the domain interval of the triangle membership function itself, which determines the range of the base of the triangle. , ]arrive[ , ] is the interval of medium speed membership change; :The domain boundary of medium speed and high speed is the S-type membership function, [ ,+∞) is the judgment range of high speed; is the actual speed collected; Convert the collected coolant temperature parameters into fuzzy linguistic variables: Low temperature: domain range , the membership function is: Moderate Temperature: Domain Range , the membership function is: High temperature: domain range , the membership function is: in is the collected coolant temperature, is the starting temperature of the low temperature range, is the maximum temperature in the low temperature range, the starting temperature in the medium temperature range, The end temperature of the medium temperature range and the start temperature of the high temperature range, End temperature of the high temperature range; : The center temperature value of low temperature, indicating the center point of the low temperature range, : The center temperature value of the medium temperature, indicating the center point of the medium temperature range, : The center temperature value of high temperature, indicating the center point of the high temperature range, is the standard deviation of the low temperature range, indicating the degree of diffusion in the low temperature range. is the standard deviation of the medium temperature range, indicating the degree of diffusion in the medium temperature range. is the standard deviation of the high temperature range, indicating the degree of diffusion in the high temperature range; The oil pressure is converted into fuzzy linguistic variables: Low pressure: domain range , the membership function is: Moderate Pressure: Domain Range , the membership function is: High pressure: domain definition , the membership function is: in is the collected oil pressure parameter, It is the minimum value of the oil pressure, usually the lowest safety pressure of the system; is the starting pressure of the low pressure range, It is the maximum pressure of the low pressure range and the starting pressure of the medium pressure range; The center pressure of the medium pressure range, It is the maximum pressure of the medium pressure range and the starting pressure of the high pressure range; It is the maximum pressure in the high pressure range; The exhaust emission components are converted into fuzzy linguistic variables: Low emissions: defining the domain , corresponding to the case where the concentration of pollutants in the exhaust gas is low, the membership function is: Medium emissions: Defining the domain , corresponding to the situation where the concentration of pollutants in the exhaust gas is at a medium level, the membership function is: High emissions: defining the domain , corresponding to the situation where the concentration of pollutants in the exhaust gas is too high, the membership function is: in is the actual measured exhaust emission value, It is the starting emission value of the low emission range, usually the minimum limit of the emission standard; It is the starting point for the transition from low emissions to medium emissions. is the end point of the transition from low emissions to medium emissions. : Starting emission value of the medium emission range, is the central emission value of the medium emission range, It is the starting point for the transition from medium emissions to high emissions. is the starting emission value of the high emission range, is the maximum emission value in the high emission range; Vibration signals are converted into fuzzy linguistic variables: Low Vibration: Domain Definition , corresponding to the case where the vibration amplitude of the internal combustion engine is small, the membership function is: Moderate Vibration: Domain Range , corresponding to the case of medium vibration amplitude of the internal combustion engine, the membership function is: High Vibration: Domain Range , corresponding to the case where the vibration amplitude of the internal combustion engine is large, the membership function is: in is the actual measured vibration signal value; is the minimum value of the vibration signal, usually the background vibration level during normal operation; is the starting vibration value of the low vibration range; It is the starting point of the transition from low vibration to medium vibration; is the starting vibration value of the medium vibration range; It is the center vibration value of the medium vibration range; It is the starting point of the transition from medium vibration to high vibration; is the starting vibration value of the high vibration range; is the maximum vibration value in the high vibration range, and the membership remains at 1 after exceeding this vibration value; The displacement of the valve train is converted into fuzzy linguistic variables: The triangular membership function is used to divide the valve train system displacement into three fuzzy linguistic variables: low displacement, medium displacement, and high displacement: The definition domain range is determined according to the displacement range of the actual valve train system; Low displacement: domain range , the membership function is: Moderate displacement: Domain range , the membership function is: High displacement: domain range , the membership function is: in, is the lower limit of the low displacement range, less than When , the displacement is completely low; It is the dividing point between low displacement and medium displacement. When the range is within the range, the low displacement membership gradually decreases, and the medium displacement membership gradually increases; It is the dividing point between medium displacement and high displacement. When , it starts to be high displacement; Is the upper limit of the high displacement interval, the displacement is greater than or equal to When , it is high displacement; Based on all the aforementioned fuzzy languages, all the operating states of the internal combustion engine, and the corresponding parameters in S1, a fuzzy rule base is constructed. Each rule consists of an IF condition and a THEN conclusion, where the condition is defined by a combination of fuzzy languages. The fuzzy language, the operating state of the internal combustion engine, and the corresponding parameters in S1 obtained in real time are matched against the rule base one by one, and the excitation strength of each rule is calculated. Perform membership aggregation operation on the logical operators in the condition part: if it contains AND operator, take the minimum value of the condition part membership as the excitation strength of the rule; if it contains OR operator, take the maximum value of the condition part membership as the excitation strength of the rule; According to the excitation intensity of each rule and the corresponding conclusion part, the corresponding membership function is truncated according to the excitation intensity to obtain the truncated output fuzzy set; The output fuzzy sets of all rules are combined, that is, the membership function after truncation of any rule and the truncated membership function of the previous rule are maximized to generate the final output fuzzy set; Defuzzify the final output fuzzy set: Discretize the final output fuzzy set domain into i points , for each discrete point , calculate its corresponding membership value , the center of gravity method is used to convert the final output fuzzy set into an accurate control signal, and the calculation formula is: in, is the discrete point of the output universe, is its membership value, is the parameter adjustment amount; f1 is the scaling factor, whose domain is [0 1], and is obtained by querying the mapping function or mapping table based on the throttle and speed; The parameter adjustment amount The specific change values ​​of the internal combustion engine control parameters are clarified, including the fuzzy adjustment amount of the fuel injection system, the fuzzy adjustment amount of the valve transmission system, and the fuzzy adjustment amount of the ignition system.

6. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 5, characterized in that: The adaptive adjustment amount of the fuel injection system obtained in S4 is specifically: The desired fuel injection system has the characteristics of fast response and small overshoot. The reference model is: in, is the transfer function of the fuel injection system, is the injection amount of the reference model at time t, The injection pulse signal of the reference model at time t is: For the fuel injection system gain, is the time constant of the fuel injection system, s is the complex frequency domain variable of Laplace transform; Substitute the fuel injection system data collected by S1, namely the injection pulse signal, injection amount, and injection time, into the reference model and perform iterative calculation to determine the fuel injection system gain in the model. and the fuel injection system time constant ; Determine the error signal, which is defined as the difference between the reference model output and the actual output of the fuel injection system: in, is the fuel injection system error signal, is the amount of fuel injected by the fuel injection system within time t; Set performance indicators: in, It is a fuel injection system performance indicator, representing the sum of squares of error signals, and is used to measure the performance of the system. To minimize performance , the control parameters are updated using the gradient descent method : in: is the control parameter at the current moment, is the learning rate, which controls the step size of parameter update; For a certain moment; is the difference between this moment and the previous moment; is the control parameter for the next moment; is the gradient of the performance index with respect to the control parameter, expressed as: is the gradient of the error signal with respect to the control parameter Adjustable fuel gain according to the updated and fuel error signal , generating a control signal To adjust the fuel injection system's injection quantity: in, is the updated injection pulse signal, is the injection pulse signal before updating; the injection pulse signal after updating Adaptive adjustment for the fuel injection system.

7. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 6, characterized in that: The self-adaptive adjustment amount of the valve transmission system is specifically obtained as follows: The desired valvetrain system has stable dynamic response characteristics, and the reference model is the expected trajectory of the valve displacement: in, Valve displacement The second derivative with respect to time t represents the valve acceleration; is the current input to the solenoid valve at time t, The proportionality constant between the electromagnetic force and the square of the input current; is the actual displacement of the valve at time t; is the spring stiffness coefficient, which represents the restoring force of the spring on the valve displacement; The quality of the valve and its transmission system; The error signal is defined as the difference between the reference displacement and the actual displacement: in, is the error signal of the valve train system; is the reference displacement of the valve train; is the actual displacement of the valve train; Set performance indicators: in, It is a performance indicator of the valvetrain system, representing the sum of squares of error signals, and is used to measure the performance of the system. To minimize performance , the control parameters are updated using the gradient descent method : in, is the control parameter at the current moment, is the learning rate, which controls the step size of parameter updates, is the gradient of the performance index with respect to the control parameter: is the gradient of the error signal with respect to the control parameter; Parameter update: The updated control parameters are used to generate new electromagnetic force signals , thereby adjusting the opening and closing of the valve: in, For in time The valve control parameters at this time are used to adjust the changes of the electromagnetic force signal; In time The new electromagnetic force signal value at time ; Updated electromagnetic force It is the adaptive adjustment amount of the valve train system.

8. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 7, characterized in that: The ignition system adaptive adjustment amount is specifically obtained as follows: The desired ignition system has precise ignition advance angle control characteristics, and the reference model is: in, is the reference ignition advance angle; is the internal combustion engine speed; is the compression pressure; is the excess air coefficient of the mixture; 、 、 、 are the reference model coefficients, which are estimated based on the parameters collected by S1 using the least squares method; The error signal is defined as the difference between the reference ignition advance angle and the actual ignition advance angle: in, is the error signal of the ignition system, is the reference ignition advance angle, is the actual ignition advance angle; Set performance indicators: in: It is an ignition system performance indicator, representing the sum of squares of error signals, and is used to measure the performance of the system. To minimize performance , the control parameters are updated using the gradient descent method : in: Control parameters at the next moment; is the control parameter at the current moment; is the learning rate, which controls the step size of parameter update; is the gradient of the performance index with respect to the control parameter, expressed as: is the gradient of the error signal with respect to the control parameter The updated control parameters are used to generate new ignition advance angle adjustment , thereby adjusting the ignition timing: Updated ignition advance angle It is the adaptive adjustment amount of the ignition system.

9. The method for accurately sensing and controlling the combustion state of an internal combustion engine according to claim 8, characterized in that: The S5 is specifically: S501, determine the weight coefficient of the fusion of the fuzzy adjustment amount and the adaptive adjustment amount obtained in S3 and S4: Monitor the operating data of the internal combustion engine in S1 at real-time speed and load in real time, and calculate the operating stability index, including speed stability index and load stability index, specifically: Calculate the speed stability index, that is, calculate the standard deviation of the speed fluctuation rate in a 3-second sliding time window: The speed of the internal combustion engine is collected in real time. The collection frequency is set between 10 and 100 times per second. A sliding time window of 3 seconds is set to store the continuously collected speed data. The average speed value is calculated within the sliding time window. , the formula is as follows: in, is the number of data points in the window, It is The speed value at the data point, Calculate the standard deviation of the speed fluctuation rate within the window : This standard deviation reflects the degree of speed fluctuation within the window. As time goes by, the sliding time window continues to move forward, adding the latest speed data point each time and discarding the first data point in the time period, keeping the window length at 3 seconds. Repeat the above calculation process to obtain the real-time speed fluctuation standard deviation. If the standard deviation is less than 0.5 r / min, the speed is determined to be in a stable state; if it is greater than or equal to 0.5 r / min, the speed is determined to be in an unstable state. Calculate the load stability index, that is, calculate the absolute value of the load change rate: Collect the load data of the internal combustion engine in real time. The collection frequency is consistent with the speed data collection frequency to ensure the time synchronization and consistency of the data. Calculate the load change at two adjacent collection moments. With time interval The ratio of load change rate is as follows: in, is the load value at the current moment, is the load value at the previous moment, The time interval between two acquisition moments is usually between 0.01 seconds and 1 second; Since we are concerned about the severity of the load change, whether the load is increasing or decreasing, we take the absolute value of the load change rate: The larger the absolute value of the load change rate, the more drastic the load change. If the absolute value of the load change rate is less than 10 N·m / s, the load is determined to be in a stable state; if it is greater than or equal to 10 N·m / s, the load is determined to be in an unstable state. Calculate the error between the actual speed output and the expected speed output of the internal combustion engine, and combine the speed stability index and load stability index to set the weight coefficient values ​​under different speed fluctuation standard deviations and load change rate absolute values, as well as error state and operating state combinations: If the speed fluctuation standard deviation is less than 0.5 r / min, the absolute value of the load change rate is less than 10 N·m / s, the speed error is less than 50 r / min, and the operating state is normal, then the fuzzy control weight coefficient is set. , adaptive control weight coefficient ; If the speed fluctuation standard deviation is less than 0.5 r / min, the absolute value of the load change rate is less than 10 N·m / s, the speed error is greater than or equal to 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient , adaptive control weight coefficient ; If the speed fluctuation standard deviation is greater than or equal to 0.5 r / min, the absolute value of the load change rate is greater than or equal to 10 N·m / s, the speed error is less than 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient , adaptive control weight coefficient ; If the speed fluctuation standard deviation is greater than or equal to 0.5 r / min, the absolute value of the load change rate is greater than or equal to 10 N·m / s, the speed error is greater than or equal to 50 r / min, and the operating state is normal, then set the fuzzy control weight coefficient , adaptive control weight coefficient ; If the operating state is overheating, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the running state is stall state, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is a poor combustion state, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is the emission exceeding the standard, the fuzzy control weight coefficient is set , adaptive control weight coefficient ; If the operating state is a mechanical failure state, the fuzzy control weight coefficient is set to 0.7, the adaptive control weight coefficient is 0.3; S502: Perform weighted fusion on the fuzzy adjustment amount and the adaptive adjustment amount according to the weight coefficient fused in S501 to obtain the final adjustment amount: According to the determined weight coefficients, the fuzzy adjustment values ​​and adaptive adjustment values ​​of the fuel injection system, valve train system, and ignition system are substituted into the following formulas for weighted fusion: in, is the final adjustment amount, and are the weight coefficients of fuzzy control and adaptive control respectively, Parameter adjustment for the fuel injection system, valve train system, and ignition system , namely the fuzzy adjustment amount of the fuel injection system, the fuzzy adjustment amount of the valve system, and the fuzzy adjustment amount of the ignition system; The adaptive adjustment amount of the fuel injection system, valve train system, and ignition system, including the adaptive adjustment amount of the fuel injection system, i.e. the updated injection pulse signal , the adaptive adjustment amount of the valve transmission system is the updated electromagnetic force , ignition system adaptive adjustment amount, that is, the updated ignition advance angle .

Images