JP2009167968A - Air-fuel ratio control device and air-fuel ratio control method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an air-fuel ratio control device and an air-fuel ratio control method capable of reducing exhaust emission by inhibiting slippage between an air-fuel ratio during actual travel and an estimated air-fuel ratio calculated by using a neural network. <P>SOLUTION: When the estimated air-fuel ratio A/F_n is calculated by inputting a plurality of physical quantity data indicating the state of an engine 1 to the neural network 34 and the air-fuel ratio during a cold start is controlled based on the calculated estimated air-fuel ratio A/F_n, the physical quantity data are accumulated if a difference between the estimated air-fuel ratio A/F_n and the actual air-fuel ratio A/F is larger than a preset prescribed value, and the accumulated physical quantity is inputted to the neural network 34 to make the neural network 34 relearn during an engine stop. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an air-fuel ratio control device and an air-fuel ratio control method, and more particularly to an air-fuel ratio control device and an air-fuel ratio control method that control an air-fuel ratio using a neural network at the time of cold start.

  Conventionally, in order to reduce exhaust gas from an automobile, a value detected by an air-fuel ratio sensor such as an O2 sensor or a linear air-fuel ratio sensor (LAFS) immediately after the cold start is fed back to the fuel injection amount and the air-fuel ratio (A / It is desired to be able to control F). However, in order to realize such control, the air-fuel ratio sensor needs to be activated at the time of starting. As a means for activating the air-fuel ratio sensor at an early stage, preheating of the air-fuel ratio sensor is known, but depending on the conditions, activation of the air-fuel ratio sensor by preheating is often not in time for starting.

  Under such circumstances, as the air-fuel ratio control immediately after the cold start, a plurality of physical quantities that can be measured even at low temperatures representing the state of the engine are detected, and the relationship between the physical quantity and the air-fuel ratio is detected using the detected physical quantities as parameters. There has been proposed an air-fuel ratio control apparatus that learns from a neural network (NN) and performs appropriate air-fuel ratio control using the result (see, for example, Patent Document 1).

JP-A-10-176578

  However, even in the air-fuel ratio control using the above-described neural network, for example, the estimated air-fuel ratio calculated using the neural network is changed due to, for example, the relationship between the physical quantity and the air-fuel ratio being changed due to the influence of aging deterioration or the like. If there is a deviation between the estimated air-fuel ratio and the actual air-fuel ratio during actual driving, the possibility of a deviation between There is a possibility that the accuracy of air-fuel ratio control is lowered and the effect of reducing exhaust gas is lowered.

  For this reason, the present invention suppresses the occurrence of deviation between the estimated air-fuel ratio calculated using the neural network and the air-fuel ratio during actual traveling, and can reduce exhaust gas. An object is to provide an apparatus and an air-fuel ratio control method.

  An air-fuel ratio control apparatus according to a first aspect of the present invention for solving the above-mentioned problem is a state detection means for detecting a plurality of physical quantities representing an engine state, and an estimation derived using a neural network by inputting the plurality of physical quantities as parameters. An air-fuel ratio control device comprising an air-fuel ratio adjustment means capable of adjusting the air-fuel ratio by adjusting the fuel injection amount based on the air-fuel ratio; an actual air-fuel ratio detection means for detecting the actual air-fuel ratio of the engine; Data storage means for storing data of the plurality of physical quantities detected at this time when the difference between the actual air-fuel ratio detected by the actual air-fuel ratio detection means and the estimated air-fuel ratio is equal to or greater than a predetermined value set in advance. When the engine is in a stopped state, the physical quantity data stored in the data storage means is input to the neural network to input the new data. Characterized by comprising a neural network learning means to perform re-learning le network.

  An air-fuel ratio control apparatus according to a second invention for solving the above-described problems is the air fuel ratio control apparatus according to the first invention, wherein the state detection means includes at least a sensor for detecting engine speed, a sensor for detecting intake air pressure, and engine cooling. And a sensor for detecting the temperature of the water.

  An air-fuel ratio control apparatus according to a third aspect of the present invention for solving the above-mentioned problems is the fuel injection system according to the first or second aspect, wherein the air-fuel ratio adjusting means is based on the plurality of physical quantities detected by the state detecting means. A basic fuel injection amount calculating means for calculating an amount; and a correction of the fuel injection amount based on a change amount of the estimated air-fuel ratio calculated using the neural network with respect to the estimated air-fuel ratio calculated immediately before using the neural network And a correction amount calculating means for calculating the amount.

  An air-fuel ratio control method according to a fourth aspect of the present invention for solving the above-described problem is to input a plurality of physical quantity data representing an engine state, calculate an estimated air-fuel ratio using a neural network, and calculate the estimated air-fuel ratio. In the air-fuel ratio control method for controlling the air-fuel ratio at the cold start based on the above, the physical quantity data is accumulated when the difference between the estimated air-fuel ratio and the actually detected air-fuel ratio is larger than a predetermined value set in advance. The physical quantity accumulated when the engine is stopped is input to a neural network, and the neural network is relearned.

  According to the air-fuel ratio control apparatus according to the first aspect described above, the air-fuel ratio control apparatus capable of adjusting the air-fuel ratio at the time of cold start using a neural network, and the actual air-fuel ratio detection means detected by the actual air-fuel ratio detection means. Data storage means for storing data of a plurality of physical quantities detected at this time when the difference between the fuel ratio and the estimated air-fuel ratio is greater than or equal to a predetermined value set in advance, and storage in the data storage means when the engine is stopped Since there is a neural network learning means for inputting the physical quantity data into the neural network and causing the neural network to perform relearning, the difference between the actual air-fuel ratio and the estimated air-fuel ratio caused by aged deterioration is increased. Effective re-learning to reduce the exhaust gas by performing air-fuel ratio feedback control with high accuracy immediately after starting. With wear, it is possible to prevent the calculation load in the engine control unit during travel to re-learned during the engine stop is increased, to implement the relearning efficiently neural network.

  According to the air-fuel ratio control apparatus of the second invention, the state detection means includes at least a sensor for detecting the engine speed, a sensor for detecting the intake air pressure, and a sensor for detecting the coolant temperature of the engine. Therefore, it is possible to control the air-fuel ratio with high accuracy by using a neural network by effectively utilizing a sensor that can detect stable information immediately after the cold start, and the exhaust gas at the cold start can be reduced. Can be reduced.

  Further, according to the air-fuel ratio control apparatus of the third invention, the air-fuel ratio adjusting means, the basic fuel injection amount calculating means for calculating the fuel injection quantity based on the plurality of physical quantities detected by the state detecting means, and the neural network A correction amount calculation means for calculating a correction amount of the fuel injection amount based on a change amount of the estimated air-fuel ratio calculated using the network with respect to the estimated air-fuel ratio calculated immediately before using the neural network is efficiently provided. The fuel injection amount can be calculated.

  According to the air-fuel ratio control method of the fourth invention, in the air-fuel ratio control method for controlling the air-fuel ratio at the time of cold start using a neural network, the estimated air-fuel ratio and the actually detected air-fuel ratio are Since physical quantity data is accumulated when the difference is larger than a predetermined value set in advance and the accumulated physical quantity is input to the neural network when the engine is stopped, the neural network is re-learned. It is possible to effectively prevent the difference between the actual air-fuel ratio and the estimated air-fuel ratio caused by, for example, by re-learning, and to reduce the exhaust gas by performing feedback control of the air-fuel ratio with high accuracy immediately after starting. In order to re-learn when the engine is stopped, the calculation load on the engine control unit is prevented from increasing while driving It is possible to carry out the re-learning of the neural network.

  Embodiments of the present invention will be described in detail in the following examples.

  An embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an air-fuel ratio control apparatus according to the present embodiment, FIG. 2 is a schematic configuration diagram showing an air-fuel ratio control apparatus according to the present embodiment, and FIG. 3 shows an air-fuel ratio control apparatus according to the present embodiment. FIG. 4 is an explanatory diagram showing an example of the learning process of the neural network in this embodiment.

  As shown in FIG. 1, the air-fuel ratio control apparatus according to this embodiment mainly includes a state detection unit 2 as a state detection unit that detects the state of the engine 1, and an engine control unit (ECU) 3 as an air-fuel ratio adjustment unit. It consists of and.

  The state detection unit 2 is a part that detects a plurality of physical quantities representing the state of the engine 1, and in this embodiment, as shown in FIG. 2, as a state detection unit that detects a plurality of physical quantities that can be detected even at low temperatures. A crank angle sensor 21, an intake air pressure sensor 22, a cooling water temperature sensor 23, and an air-fuel ratio sensor (LAFS in this embodiment) 24 as an actual air-fuel ratio detection means are configured.

  Here, the crank angle sensor 21 is provided on the crankshaft 11 to detect the engine speed Ne, the intake air pressure sensor 22 is provided on the intake manifold 12 to detect the intake air pressure (intake manifold pressure) Pb, and the cooling water temperature. A sensor 23 is provided in the water jacket 13 to detect the engine water temperature WT, and an air-fuel ratio sensor 24 is provided in the exhaust manifold 14 to detect the air-fuel ratio A / F based on the oxygen concentration in the exhaust.

  The engine control unit 3 also receives a plurality of physical quantities input from the state detector 2, that is, the engine speed Ne, the intake air pressure Pb, the engine water temperature WT, and the air-fuel ratio A / F detected by the various sensors 21 to 24 described above. Is used to calculate an injector pulse width Pw that keeps the air-fuel ratio at a predetermined constant value, and outputs the calculated result to the injector 4. The injector 4 injects fuel into the engine 1 based on the injector pulse width Pw input from the engine control unit 3 (however, the output of the air-fuel ratio sensor 24 is performed after the air-fuel ratio sensor 24 is activated). .

  More specifically, in this embodiment, the engine control unit 3 uses the basic injector driving time calculation unit 31, learning data as means for calculating the injector pulse width Pw that keeps the above-mentioned air-fuel ratio A / F at a predetermined value. An accumulation unit 32, a neural network learning unit 33, a neural network 34, and an air-fuel ratio correction amount calculation unit 35 are provided.

  The basic injector drive time calculation unit 31 is based on the physical quantities detected by the various sensors 21 to 24 in the state detection unit 2 and is feedback controlled by feedforward control using a map and in a temperature range where the air-fuel ratio sensor 24 can be used. Is a part for calculating the basic injector pulse width Pw_b by a known method. In this embodiment, the engine speed Ne detected by the state detector 2, the intake air pressure Pb, the engine water temperature WT, and the air-fuel ratio A / F are input to the basic injector drive time calculator 31. In normal steady state, the value of the air-fuel ratio A / F is maintained at a predetermined constant value only by the basic injector drive time calculation unit 31.

  The learning data storage unit 32 also has a plurality of physical quantities (in this case, the engine speed Ne, the intake air pressure Pb, the engine water temperature WT, the air-fuel ratio A / F, etc.) acquired in advance by performing tests so as to cover various patterns. ), And physical quantities (in this embodiment, the engine speed Ne, the intake air pressure Pb, the engine water temperature WT, the air-fuel ratio A / F) detected by the various sensors 21 to 24 in the state detector 2. ) Data is added and stored.

  In the learning data storage unit 32, the actual air-fuel ratio A / F detected by the air-fuel ratio sensor 24 after the air-fuel ratio sensor 24 is activated, and the estimated air-fuel ratio A / F_n calculated using the neural network 34 are used. When the difference between and becomes larger than a predetermined value set in advance, the physical quantity data detected by the state detection unit 2 is additionally stored.

  The neural network learning unit 33 is a part that executes relearning of the neural network 34 described later using the physical quantity accumulated in the learning data accumulation unit 32 as a parameter, monitors the state of the engine 1, and determines that the engine 1 is in a stopped state. A function to execute the relearning of the neural network 34 after the learning is completed. That is, the neural network learning unit 33 is configured so that the neural network 34 can be re-learned even when the engine 1 is stopped, and the engine 1 is not stopped, such as when the vehicle is running. In this case, the re-learning of the neural network 34 is suppressed.

  As shown in FIG. 4, the neural network 34 receives time-series data of the engine speed Ne at the cold start, the intake air pressure Pb, the injector pulse width Pw, the engine water temperature WT, and the air-fuel ratio A / F. The air-fuel ratio A / F is learned as an output, and at the time of cold start, the engine speed Ne detected by the crank angle sensor 21, the intake air pressure sensor 22, and the cooling water temperature sensor 23 in the state detection unit 2, respectively. The estimated air-fuel ratio A / F_n is calculated based on the intake air pressure Pb, the engine water temperature WT, and the estimated value of the air-fuel ratio (hereinafter referred to as the estimated air-fuel ratio) calculated by the neural network 34 in the previous control cycle. Has been.

  The neural network 34 also stores the engine speed Ne, the intake air pressure Pb, the injector pulse width Pw, the engine water temperature WT, and the air-fuel ratio A / F newly accumulated in the learning data accumulation unit 32 when the vehicle is running. Is updated as appropriate by performing the re-learning. The estimated air-fuel ratio calculated using the neural network 34 and the actual air-fuel ratio A / F detected by the air-fuel ratio sensor 24 are compared while the engine 1 is in the idle region after warming up. .

  The air-fuel ratio correction amount calculation unit 35 calculates a correction amount ΔPw (hereinafter referred to as an injector pulse width correction amount) ΔCw at the time of cold start of the basic injector pulse width Pw_b calculated by the basic injector drive time calculation unit 31. It is a part to do. In the air-fuel ratio correction amount calculation unit 35, immediately after startup, the estimated air-fuel ratio A / F_n derived in the neural network 34, the estimated air-fuel ratio A / F_n calculated by the neural network 34 in the previous control cycle, and the target air-fuel ratio are calculated. The injector pulse width correction amount ΔPw is calculated from the deviation. During normal travel, the injector pulse width correction amount ΔPw is calculated based on the difference between the actual air-fuel ratio A / F detected by the air-fuel ratio sensor 24 and the target air-fuel ratio.

  That is, in this embodiment, the injector pulse width correction amount ΔPw calculated by the air-fuel ratio correction amount calculation unit 35 based on the estimated air-fuel ratio is added to the basic injector pulse width Pw_b calculated by the basic injector drive time calculation unit 31 at the cold start. Is output to the injector 4 to control the air-fuel ratio A / F, and the normal air-fuel ratio Pw_b calculated by the basic injector driving time calculation unit 31 during normal traveling is set to the actual air-fuel ratio. Based on the above, the injector pulse width Pw obtained by adding the injector pulse width correction amount ΔPw calculated by the air-fuel ratio correction amount calculation unit 35 is output to the injector 4 to control the air-fuel ratio A / F.

  Hereinafter, the operation of the air-fuel ratio control apparatus according to this embodiment will be described with reference to FIG. As shown in FIG. 3, in the air-fuel ratio control apparatus according to this embodiment, after the cold start, first, the air-fuel ratio sensor (LAFS) 24 is activated or not, for example, the air-fuel ratio sensor 24 can be used. It is determined whether or not it is in the temperature range (step S1).

  If it is determined in step S1 that the air-fuel ratio sensor 24 is not activated (NO), the neural network is based on the engine speed Ne, the intake air pressure Pb, and the engine water temperature WT detected by the various sensors 21 to 23 in the state detection unit 2. 34 is used to calculate the estimated air-fuel ratio A / F_n, and the injector pulse width correction amount ΔPw so that the actual air-fuel ratio A / F is maintained at the estimated air-fuel ratio A / F_n in the air-fuel ratio correction amount calculator 35. And the result obtained by adding this to the basic injector pulse width Pw_b calculated by the basic injector drive time calculation unit 31 is output to the injector 4 (step S8).

On the other hand, if it is determined that the air-fuel ratio sensor 24 is activated (YES), feedback control is started using the air-fuel ratio A / F detected by the air-fuel ratio sensor 24 (step S2), and the engine 1 is idle. It is determined whether or not the area is present (step S3).
And when it determines with the engine 1 not being in an idle area | region in step S3 (NO), it returns to the process of step S1.

  On the other hand, if it is determined as a result of the determination in step S3 that the engine 1 is in the idle region (YES), the engine speed Ne and the intake air pressure Pb detected by the various sensors 21 to 23 in the state detection unit 2 are subsequently continued. , And the difference between the estimated air-fuel ratio A / F_n calculated using the neural network 34 based on the engine water temperature WT and the actual air-fuel ratio A / F detected by the air-fuel ratio sensor 24 (hereinafter referred to as air-fuel ratio difference) ΔA / It is determined whether F is within a predetermined value set in advance (step S4).

  In step S4, the absolute value of the air-fuel ratio difference ΔA / F is within a preset fixed value (in the case where the absolute value of the air-fuel ratio difference ΔA / F temporarily exceeds a preset fixed value) If it is determined that the duration is shorter than a predetermined time set in advance (YES), it is determined that the estimated air-fuel ratio A / F_n has been accurately calculated, and the process returns to step S1, and the air-fuel ratio A / The F control process is continued.

  On the other hand, in step S4, the absolute value of the air-fuel ratio difference ΔA / F exceeds a preset constant value (in this embodiment, the absolute value of the air-fuel ratio difference ΔA / F exceeds a preset constant value). If the determination is NO (NO), the learning data is accumulated while the absolute value of the air-fuel ratio difference ΔA / F exceeds a preset constant value. Data of physical quantities detected by the various sensors 21 to 24 by the state detection unit 2 is stored by the unit 32 (step S5).

  Subsequently, it is determined whether or not the engine 1 has stopped (step S6). If it is determined that the engine 1 is operating (NO), the process returns to step S1, and the air-fuel ratio A / F control process is performed. continue.

  On the other hand, when it is determined in step S6 that the engine 1 is stopped (YES), the neural network 34 performs relearning using the physical quantity accumulated in the learning data accumulation unit 32 in the neural network learning unit 33 as an input. Processing is performed (step S7).

  According to the air-fuel ratio control apparatus according to this embodiment described above, after the engine 1 is started and the air-fuel ratio sensor 24 is activated, the estimated air-fuel ratio A / F_n derived using the neural network 34 and the The difference between the actual air-fuel ratio A / F detected by the fuel-fuel ratio sensor 24 is monitored, and the estimated air-fuel ratio A / F_n derived using the neural network 34 and the actual air-fuel ratio A / F detected by the air-fuel ratio sensor 24 are monitored. Data obtained after the air-fuel ratio sensor 24 is activated only when the difference from F is equal to or greater than a predetermined value (in this embodiment, engine speed Ne, intake air pressure Pb, injector pulse width Pw, engine water temperature WT , Air-fuel ratio A / F, etc.) are stored in the learning data storage unit 32, and only when new data is added, the Since the relearning of the network 34 is performed, it is possible to perform the air-fuel ratio feedback control with high accuracy immediately after the start while reducing the calculation load in the engine control unit 3, and to suppress the exhaust gas immediately after the start. Is possible.

  Note that the physical quantity detected by the state detection unit 2 is not limited to that described in the above-described embodiments, and in addition to the above-described sensors constituting the state detection unit 2, other sensors are provided as appropriate. It goes without saying that many physical quantities may be detected.

  The present invention is applicable to an air-fuel ratio control apparatus and an air-fuel ratio control method for controlling an air-fuel ratio using a neural network at the time of cold start.

It is a block diagram which shows the structure of the air fuel ratio control apparatus which concerns on the Example of this invention. It is a schematic block diagram which shows the air fuel ratio control apparatus which concerns on the Example of this invention. It is a flowchart showing operation | movement of the air fuel ratio control apparatus which concerns on the Example of this invention. It is explanatory drawing which shows the structural example of the neural network in the Example of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Engine 2 State detection part 3 Engine control unit 4 Injector 21 Crank angle sensor 22 Intake air pressure sensor 23 Cooling water temperature sensor 24 Air-fuel ratio sensor 31 Basic injector drive time calculation part 32 Learning data storage part 33 Neural network learning part 34 Neural network 35 Empty Fuel ratio correction amount calculation unit

Claims (4)

State detecting means for detecting a plurality of physical quantities representing the state of the engine, and air-fuel ratio adjusting means capable of adjusting the fuel injection amount based on the estimated air-fuel ratio derived using a neural network by inputting the plurality of physical quantities as parameters. In an air-fuel ratio control device comprising:
An actual air-fuel ratio detecting means for detecting an actual air-fuel ratio of the engine;
Data storage means for storing data of the plurality of physical quantities detected at this time when the difference between the actual air-fuel ratio detected by the actual air-fuel ratio detection means and the estimated air-fuel ratio is equal to or greater than a predetermined value set in advance. When,
Neural network learning means for inputting the physical quantity data stored in the data storage means to the neural network and causing the neural network to perform relearning when the engine is in a stopped state. An air-fuel ratio control device.   2. The air-fuel ratio control according to claim 1, wherein the state detecting means includes at least a sensor for detecting an engine speed, a sensor for detecting an intake air pressure, and a sensor for detecting a coolant temperature of the engine. apparatus.   The air-fuel ratio adjusting means calculates basic fuel injection amount calculating means for calculating a fuel injection amount based on the plurality of physical quantities detected by the state detecting means, and the estimated air-fuel ratio calculated using the neural network. 3. The air-fuel ratio control according to claim 1, further comprising correction amount calculation means for calculating a fuel injection amount correction amount based on a change amount with respect to the estimated air-fuel ratio calculated immediately before using a neural network. apparatus. Air-fuel ratio control that inputs data of a plurality of physical quantities representing the state of the engine, calculates an estimated air-fuel ratio using a neural network, and controls the air-fuel ratio at the time of cold start based on the calculated estimated air-fuel ratio In the method
When the difference between the estimated air-fuel ratio and the actually detected air-fuel ratio is larger than a predetermined value set in advance, the physical quantity data is accumulated,
An air-fuel ratio control method, wherein the physical quantity accumulated when the engine is in a stopped state is input to a neural network and the neural network is relearned.

Images