JP2010095067A - Hybrid car, computer device, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a hybrid car optimally controlling torque distribution between an engine and an electric motor according to various driver's habits and use environments and requiring no high-load computation. <P>SOLUTION: A hybrid controller 12 includes a torque distribution control unit 30 outputting an indication value of a torque to an engine 10 and an electric motor 11 to be decided based on a required torque of a driver within a predetermined time, a rotation speed of the engine 10, a regenerative current to a battery 17 supplying a power source to the electric motor 11, and a value indicating a charge state of the battery 17. The control unit 30 is learned so that a torque distribution value is output in association with an input value, and pre-simulation using a dynamic programming method is adopted to the learning. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a hybrid vehicle, a computer apparatus, and a program.

  In order to achieve low fuel consumption, hybrid vehicles equipped with both an engine powered by gasoline or light oil and an electric motor driven by electricity are becoming popular. Generally, a hybrid vehicle is driven by an electric motor in a situation where a large engine torque is temporarily required, such as at the time of departure. As a result, the load on the engine can be reduced. On the other hand, it is driven by the engine in a situation where a low torque continues during high speed running. Thereby, the battery for electric motors is charged (for example, refer patent document 1).

Such a hybrid vehicle is equipped with an assist map based on results obtained by an automobile manufacturer conducting an experiment in advance. In order to optimally control the engine and the electric motor, an optimum assist map is selected from a plurality of assist maps based on the driver's required torque and the driving state of the vehicle at that time. In addition, in order to ensure running stability in an electric vehicle provided with a motor corresponding to each drive wheel, a technique of incorporating a control neural network and an updating neural network that have been learned in advance is also known (see Patent Document 2). ).
JP-A-11-32442 Japanese Patent Laid-Open No. 10-285707

  As for passenger cars, automatic cars are widely used, and there are relatively few cases where driver's driving habits are reflected in the running state. Therefore, it is sufficient that there are several tens of types (at most hundreds of types) of assist maps. On the other hand, in large vehicles such as trucks and buses, manual transmissions are the mainstream, so there are many situations where driver's driving habits are reflected in the running state. For example, some drivers frequently make sudden starts and accelerations, while others do not perform such actions.

  Also, large vehicles such as trucks and buses have their usage environment and conditions of use, such as long-distance truck flights or high-speed buses that travel on expressways for long periods of time and courier services or route buses that frequently start and stop. Compared to passenger cars, it is very different.

  Therefore, when performing hybrid control using an assist map prepared in advance by the manufacturer, if there are few variations of the assist map, it may not be possible to cope with various driver's habits and usage environments. Therefore, the number of types of assist maps ranges from thousands to tens of thousands. However, a large memory capacity is required to accommodate an enormous variety of assist maps in a vehicle-mounted memory. Furthermore, the process of selecting an appropriate assist map from the assist map also requires enormous processing capacity and requires high load calculation. Therefore, it is not preferable to adopt a control for selecting a plurality of assist maps similar to a passenger car as it is for a large vehicle such as a truck or a bus.

  On the other hand, in order to avoid this problem, the learned control neural network and the updating neural network as described in Patent Document 2 can be used when creating the assist map, but the following problems occur. In other words, unlike an electric vehicle described in Patent Document 2, a hybrid vehicle that requires an assist map has a delicate relationship between an engine and an electric motor, and a control neural network is required to perform accurate control at the start of vehicle operation. The learning time for the network becomes enormous and the price increases. In addition, a high load calculation is required to construct an updating neural network.

  The present invention has been made under such a background, and can optimally control the torque distribution between the engine and the electric motor in accordance with various driver's habits and usage environments, and It is an object of the present invention to provide a hybrid vehicle, a computer apparatus, and a program that do not require high load calculation.

  The hybrid vehicle according to the present invention is a hybrid vehicle including an engine, an electric motor, and a calculation unit that calculates torque to be distributed to the engine and the electric motor based on a driver's requested torque and a running state of the vehicle. The engine and the motor are determined in accordance with the driver's required torque, the engine rotation speed, the regenerative current to the battery that supplies power to the motor, and the value indicating the state of charge of the battery within a predetermined time. Torque distribution means for outputting the indicated value of torque is provided, and this torque distribution means is trained to output the torque distribution value in correspondence with the input value, and this learning uses a dynamic programming method. The previous simulation was adopted.

  For example, as the data used for the prior simulation, time-series data based on the engine rotation speed of a vehicle having the same power performance and the driver's required torque, or data after a test run using the hybrid vehicle is used.

  Further, the torque distribution means can comprise a neural network that has been trained to output a torque instruction value associated with the input value.

  At this time, the neural network can calculate and generate an optimization map for calculating the indicated value of the torque by interpolation or extrapolation when an approximate value approximate to the input value is input.

  In addition, the computer device of the present invention targets the hybrid vehicle of the present invention and performs a simulation for calculating an optimum torque instruction value corresponding to an input value using a dynamic programming technique.

  Further, the program of the present invention is installed in the information processing apparatus to cause the information processing apparatus to realize the functions of the calculation means and the torque distribution means in the hybrid vehicle of the present invention.

  According to the present invention, it is possible to provide a hybrid vehicle that can optimally control the torque distribution between the engine and the electric motor in accordance with various driver's habits and usage environments and can be reduced in price.

(About the configuration of the hybrid vehicle 1)
A configuration of a hybrid vehicle 1 according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a main part configuration diagram of a hybrid vehicle 1. As shown in FIG. 1, the hybrid vehicle 1 is a calculation unit that calculates the torque distributed to the engine 10 and the electric motor 11 based on the engine 10, the electric motor 11, the driver's requested torque and the traveling state of the vehicle. And a hybrid control unit 12.

  Other configurations of the hybrid vehicle 1 include an accelerator sensor 15 that detects driver torque information in accordance with the driver's accelerator operation amount, an engine control unit 16 that controls the engine 10, and a battery that supplies power to the motor 11. 17, a battery control unit 18 that controls the battery 17, an inverter 19 that adjusts the amount of power supplied to the electric motor 11 by the battery 17, and a CAN (Control Area Network) 20 that transmits information on each part of the vehicle to the hybrid control unit 12. Prepare. The hybrid control unit 12 includes a torque distribution control unit 30. The torque distribution control unit 30 performs main processing in hybrid control.

(Basic concept of information processing performed by hybrid vehicle 1 according to the embodiment of the present invention)
A basic concept of information processing performed by the hybrid vehicle 1 according to the embodiment of the present invention will be described with reference to FIG. As shown in FIG. 2, the information processing performed by the hybrid vehicle 1 according to the embodiment of the present invention is based on the information acquired at time T1, and the torque distribution control unit 30 generates an optimization map. The two-dimensional lookup table generated based on this is used in the traveling at time T2. Further, based on the information acquired at time T2, the torque distribution control unit 30 generates a two-dimensional lookup table generated based on the optimization map, and uses this two-dimensional lookup table in traveling at time T3.

  That is, when N is an integer equal to or greater than 2, the torque distribution control unit 30 generates an assist map (that is, a two-dimensional lookup table) based on the information acquired at time T (N−1), and this assist map is converted to time. Used for TN travel. Thus, since it is possible to travel while sequentially generating the assist map, it is possible to optimally control the torque distribution between the engine and the electric motor in accordance with various driver's habits and usage environments. Note that the optimal control is a control in which the fuel consumption is improved as compared with the same type vehicle which is not a conventional hybrid vehicle, and is high in terms of the SOC (State of Charge) which is the remaining amount of the battery. The control keeps the value constant. An assist map is generated so that both of these controls are optimized simultaneously.

  In addition, the longer the predetermined time corresponding to the time Ti (i is an integer from 1 to N), the larger the amount of information to be acquired and the generation of an optimization map with a wider numerical range. can do. However, since the time for shifting from time Ti to time T (i + 1) also becomes longer, the real-time property of the control is deteriorated. Therefore, it is preferable to set the length of the predetermined time corresponding to the time Ti as appropriate according to the use state of the vehicle and the environment of the road. For example, if the vehicle mainly travels on a highway, the change in the road environment is small, and the predetermined time may be longer. On the other hand, in the case of a vehicle that mainly travels in an urban area, the change in the road environment is large, so the predetermined time is preferably short. The predetermined time due to the difference between the vehicles can be determined in advance according to the difference between the vehicles such as a truck, a bus, and a passenger car. Further, the actual running state, for example, the use frequency of the brake may be detected and appropriately changed according to the detected value. Note that the torque distribution control unit 30 shown in FIG. 1 can receive input of predetermined time setting information. For example, the predetermined time ranges from 1 second to several minutes.

(Overall processing flow)
Next, an overall processing flow of information processing performed by the hybrid vehicle 1 according to the embodiment of the present invention will be described with reference to FIG. In the embodiment of the present invention, as shown in FIG. 3, optimization maps # 1 to #N corresponding to simulations # 1 to #N are offline with respect to hybrid control unit 12 shown in FIG. Is generated. That is, simulations # 1 to #N are performed by a dynamic programming method for a predetermined vehicle type (step S1), and an optimization map # 1 corresponding to a plurality of simulations # 1 to #N is obtained based on information obtained as a result of the simulation. ~ # N are generated (step S2). These simulations are prior simulations.

  When optimization maps # 1 to #N corresponding to simulations # 1 to #N are generated (step S2), they are not mounted on the torque distribution control unit 30 of the hybrid control unit 12 as they are. The neural network 50 is made to learn the correspondence between #N and the optimization maps # 1 to #N (step S3). Then, the learned neural network 50 is mounted on the torque distribution control unit 30 (step S4). As a result, it is possible to generate an optimization map corresponding to the case where approximate values are input instead of the simulation input values themselves.

  That is, when the neural network 50 of the torque distribution control unit 30 obtains an input value that approximates simulation #i (i is an integer from 1 to N) in actual running, for example, interpolation or extrapolation is performed. Thus, an optimization map #i corresponding to the simulation #i can be generated and output.

  Here, a difference in memory capacity required for mounting between the case where the optimization maps # 1 to #N are directly mounted on the torque distribution control unit 30 and the case where the neural network 50 is mounted on the torque distribution control unit 30 will be described. .

  FIG. 4 is a diagram showing an example of a two-dimensional lookup table for calculating the assist torque from the “driver's required output” and the “engine speed” determined by the optimization map. As shown in FIG. 4, a two-dimensional lookup table serving as an assist map is created and stored in a real-time torque distribution unit 70, which will be described later, and "driver's request output" (lateral direction) and "engine rotation speed". ”(Longitudinal direction) is a map in which assist torque corresponding to“ is recorded. The unit of each numerical value of the assist torque, that is, the torque of the electric motor 11 is “Nm”. In the two-dimensional lookup table shown in FIG. 4, the assist torque values are “driver's required output average value”, “driver's required output dispersion value”, “engine rotational speed average value”, and “engine rotation”. It varies depending on input values such as “speed dispersion value” and “battery chargeable electric energy”. Details of these values will be described later. As the input value changes variously, there are countless optimization maps. Therefore, if the optimization map itself is mounted on the torque distribution control unit 30, an enormous number of optimization maps corresponding to the number of combinations of input values is required. Moreover, the higher the resolution, the greater the number of combinations of input values. Therefore, the total number is from thousands to tens of thousands.

  On the other hand, the neural network 50 includes neurons, “weights” and threshold values θ given to the neurons. Therefore, even in the case of the neural network 50 subjected to a huge amount of learning in step S3, the memory of the torque distribution control unit 30 holds the neuron, its “weight”, and the threshold θ, that is, the result of learning. Only information on hundreds of optimization maps. That is, in the neural network 50, even when an approximate value is input instead of a learned input value, several hundreds of optimization maps calculate an optimization map corresponding to the approximate value in an interpolation or extrapolation manner. is there. Therefore, it can be seen that the storage capacity can be extremely small as compared with the case where the torque distribution control unit 30 holds an enormous optimization map corresponding to various input values.

(Configuration of neural network 50)
Here, the configuration of the neural network 50 will be described. FIG. 5 shows an example of neurons constituting the neural network 50. The neurons constituting the neural network 50 are as shown in FIG. FIG. 5 schematically shows a neuron. FIG. 5 shows an example in which “driver's requested output average value” W _{Drv_req} is an input signal. In the neuron, based on the input signals W _{Drv_req1 to} W _{Drv_reqN} and the weights w1 to wN,
net = W _{Drv_req1} × w1 + W _{Drv_req2} × w2 + ... + W _{Drv_reqN} × wN
out = f (net−θ)
The output signal out is output by performing the calculation.

That is, when the net (membrane potential) exceeds the threshold θ, the neuron outputs “1”, and otherwise outputs “0”. As described above, in the neuron, by setting the weights w1 to wN and the threshold value θ for the input signals W _{Drv_req1 to} W _{Drv_reqN, the} conditions under which the output signal out becomes “1” can be variously changed. A neural network 50 is configured using such neurons as shown in FIG. In the example of FIG. 6, five feature quantity vectors “driver's required output average value”, “driver's required output dispersion value”, “engine rotational speed average value”, and “engine rotational speed dispersion value” are used. In this example, “battery chargeable electric energy” is an input signal, and five types of optimization maps # 1 to # 5 are output signals. In the example of FIG. 6, the optimization map # 3 is output. In FIG. 6, five types of output signals are illustrated, but in actuality, it may be several tens to several hundreds or more. In this way, an optimization map for determining the values on the horizontal axis and the vertical axis of the two-dimensional lookup table shown in FIG. 4 is output. The neural network 50 is a well-known technique and may have various known configurations other than the configuration shown in FIG.

  FIG. 7 is a diagram showing a table in which “weights” w11 to w5n4 and threshold values θ set for the neurons N11 to N5n shown in FIG. 6 are recorded. As shown in FIG. 7, only “weights” w11 to w5n4 and threshold value θ are recorded in the table. Therefore, the torque distribution control unit 30 does not need to hold a plurality of optimization maps # 1 to #N themselves, and may hold a table in which “weights” w11 to w5n4 and a threshold θ are recorded as shown in FIG. . Thereby, the amount of information held by the torque distribution control unit 30 can be reduced.

(Generation of optimization map by computer device 40)
Next, generation of optimization maps # 1 to #N corresponding to simulations # 1 to #N by computer apparatus 40 in step S1 of FIG. 3 will be described. In the embodiment of the present invention, a technique called a dynamic programming technique is used to generate optimization maps # 1 to #N. The dynamic programming method is a well-known technique. Therefore, a detailed description of the dynamic programming method itself is omitted, and the relationship between the dynamic programming method and the hybrid control of the hybrid vehicle 1 will be mainly described below.

  The dynamic programming technique is an effective technique for solving the resource allocation problem in a short time. In other words, the purpose is to distribute certain limited resources to the investees and maximize (or minimize) the profits obtained. Consider a case where such a dynamic programming method is applied to the hybrid control of the hybrid vehicle 1.

If the driving operation is the same when traveling in a certain traveling pattern, the amount of power regeneration by the electric motor 11 is determined. Therefore, when this regenerative electric power (corresponding to the above-mentioned resource) is input (distributed) to the driver's required torque at any scene, the fuel consumption (corresponding to the above-mentioned profit) is maximized (minimum in terms of fuel consumption) Can be reduced to the problem of resource allocation.
When formulating this resource allocation problem,
f _N (k _N ) =
min _{x1 + x2 + ... + xN} [g ₁ (x ₁ , ω _e1 ) + g ₂ (x ₂ , ω _e2 ) + ... + g _N (x _N , ω _eN )]
... (1)
_{However, k N = x 1 + x} 2 + ... + x N
Is a constraint condition. Each variable is as follows.
f _N (k _N ): Fuel consumption (cc) when power is applied for k _N
k _N : Total energy input (kWh)
x _N : Energy input per sampling (kW)
g _N (x _N , ω _eN ): Fuel injection amount (cc / sec) when x _N energy is input
ω _eN
: Engine speed (RPM)
In order to solve Equation (1) numerically, it is expressed as a function recurrence equation based on the principle of optimality.
f _N (k _N ) = min _{0 ≦} x _{N ≦} k _N [g _n (x _n ) + f _n−1 (k _n −x _n )] (n = 2, 3,..., N)
f ₁ (k ₁ ) = min _{0 ≦ xN = k1 ≦ kN} [g ₁ (x ₁ )]
Thus, it becomes possible to obtain a solution to the dynamic programming problem. Here, 0 ≦ x _N ≦ W _{m_max} (kW) due to hardware restrictions
0 ≦ x _N ≦ W _{Bat_max} (kW)

_{0 ≦ x N ≦ T m_max ×} (ω e / 60 × 2π) / 1000 (kW)
0 ≦ x _N ≦ W _{Drv_req} −T _{e_min} × ω _e / 60 × 2π / 1000 (kW)
0 ≦ k _N ≦ Σ _{k = 1 to N} × _k
Shall be satisfied. However,
T _{m_max} : Maximum motor torque (Nm) (hardware limit of the motor 11)
_{Te_min} : engine minimum torque (Nm) (minimum output torque of engine 10)
W _{Drv_req} : Driver's request output (kW) (Driver's request output)
W _{m_max} : Maximum motor output (kW) (maximum output of electric motor 11)
W _{Bat_max} : Maximum battery output (kW) (battery 17 output limit)
It is.

  The neural network 50 learns a plurality of sets (several tens to several hundreds) of the relationship of the optimization map with respect to the five feature quantity vectors. As a result, an approximate optimization map is calculated by interpolation for the approximate feature vector. Further, an optimization map is calculated by extrapolation using the other learning data for the feature amount vector in the unlearned range. As described above, the neuron takes the product of the input signal and the “weight”, and the output signal out becomes “1” when the sum of the products exceeds the threshold θ. Therefore, even if the input value is a value that approximates one predetermined value, if the sum of the product of the value and the “weight” exceeds the threshold θ, it is the output signal out corresponding to the predetermined value. “1” is obtained. Thus, it is an advantage to use the neural network 50 that an approximate value can be used. Note that the data to be learned by the neural network 50 is various time series data such as the engine rotation speed and the driver's required torque of a vehicle having the same power performance, and already acquired data. However, data of a test run after completion of the vehicle may be used.

(Regarding the configuration of the torque distribution control unit 30)
Next, the configuration of the torque distribution control unit 30 will be described with reference to FIG. As shown in FIG. 8, the torque distribution control unit 30 includes a data analysis unit 60 that generates data to be input to the neural network 50, and a real-time torque distribution unit 70 that performs torque distribution based on the optimization map generated by the neural network 50. It consists of.

(About operation | movement of the torque distribution control part 30)
The torque distribution control unit 30 includes five vectors that become feature quantity vectors at predetermined time intervals, such as “driver output information”, “engine speed (rotation speed) information”, “battery regenerative current information”, “Battery SOC information” and “vehicle speed information” are updated and input to the data analysis unit 60. This predetermined time is, for example, T1 (T2,..., TN) seconds described in FIG. The data analysis unit 60 analyzes the information to obtain “driver's required output average value”, “driver's required output variance value”, “engine rotational speed average value”, “engine rotational speed variance value”. And “battery chargeable electric energy” are generated and output to the neural network 50.

  That is, the data analysis unit 60 generates “driver's required output average value and variance value” based on “driver's required output information”, “engine speed information” and “vehicle speed information”, and “engine speed information”. ”Generate“ engine speed average value and variance value ”based on“ driver's requested output information ”and“ vehicle speed information ”, and“ battery input ”based on“ battery regenerative current information ”and“ battery SOC information ” “Amount of possible power” is generated. Note that the idling data is irrelevant to the assist control and is therefore excluded in software.

  FIG. 9 shows the relationship between the SOC and the amount of power that can be input. The data analysis unit 60 inputs the battery chargeable power amount to the neural network 50 based on the relationship between the SOC and the chargeable power amount shown in FIG.

  When the neural network 50 obtains the “driver's requested output average value and variance value”, “engine speed average value and variance value”, and “battery chargeable electric energy” that are inputs from the data analysis unit 60. Then, an optimization map corresponding to the input value is generated by interpolation or extrapolation from simulation #i obtained from input values approximate to these input values or from an optimization map corresponding to a plurality of simulations. The calculation of the optimization map is performed every predetermined time (T1, T2,..., TN).

  The real-time torque distribution unit 70 generates a two-dimensional lookup table that becomes an assist map in which the assist torque shown in FIG. 4 is described from the optimization map calculated every predetermined time. As shown in FIG. 4, this two-dimensional lookup table is a map that covers assist torques corresponding to “driver's required output” and “engine speed” in a wide numerical range. However, the current two-dimensional lookup table is generated a predetermined time after the previous two-dimensional lookup table is generated. Therefore, only the portion of the two-dimensional lookup table corresponding to the input value range within the predetermined time becomes the assist torque based on the actually measured value, and the other portion becomes the initial value or the assist torque generated before. In general, since there are few situations in which the driving state of the vehicle or the environment of the road changes dramatically, there are few practical problems even if only the currently required numerical range is the assist torque based on the actually measured value.

  Next, the operation of the torque distribution control unit 30 will be described with reference to the flowchart of FIG. When the key of the hybrid vehicle 1 is turned on (step S10), a two-dimensional lookup table (assist map) calculated from an optimization map corresponding to the initially set input value is set (step S11). That is, since the hybrid vehicle 1 has not yet traveled, the measured value of the input value cannot be obtained, and the data analysis unit 60 outputs the input value that is initially set to the neural network 50.

  When the hybrid vehicle 1 starts traveling (Yes in step S12), the torque distribution control unit 30 updates the input value input to the data analysis unit 60 with the actual measurement value (step S13). This update is performed every second in this embodiment.

  When the accumulated travel time has passed a predetermined time (for example, each t seconds of T1,..., TN in FIG. 2) (Yes in step S14), the past feature quantity vector at the timing when the two-dimensional lookup table is generated and the current The Euclidean distances of the feature vector are compared every second. If the Euclidean distance exceeds the specified value (Yes in step S15), the neural network 50 calculates and updates an optimal two-dimensional lookup table using the current feature vector at that moment (step S16). ). At this time, the accumulated travel time is cleared. If the specified value is not exceeded, the use of the same two-dimensional lookup table is continued until the specified value is exceeded. Note that the cumulative travel time is a travel time that does not include the stopped time. Steps S12 to S16 are repeated during traveling (No in Step S17), and when the key is turned off (Yes in Step S17), the process ends.

  As a result, the optimization map selected by the neural network 50 is input to the real-time torque distribution unit 70 every time the accumulated traveling time passes a predetermined time. The real-time torque distribution unit 70 creates a two-dimensional lookup table based on the selected optimization map, and calculates assist torque from “driver's required output information” and “engine speed information”. Based on the calculated value, the real-time torque distribution unit 70 outputs an engine torque instruction and an electric motor torque instruction to the engine control unit 16 and the inverter 19, respectively.

(About the program)
Each unit of the hybrid control unit 12 is a general-purpose information processing device (CPU (Central Processing Unit), DSP (Digital Signal) that operates by predetermined software (a program referred to in claims).
Processor), a microprocessor (microcomputer), or the like. For example, a general-purpose information processing apparatus has a memory, a CPU, an input / output port, and the like. The CPU of the general-purpose information processing apparatus reads and executes a control program as a predetermined program from a memory or the like. Thereby, the function of each part of the hybrid control part 12 is implement | achieved in a general purpose information processing apparatus.

  Even if the control program executed by the general-purpose information processing apparatus is stored in the memory or the like of the general-purpose information processing apparatus before the hybrid control unit 12 is shipped, It may be stored in a memory of the information processing apparatus. Further, a part of the control program may be stored in a memory of a general-purpose information processing device after the hybrid control unit 12 is shipped. The control program stored in the memory or the like of a general-purpose information processing apparatus after shipment of the hybrid control unit 12 is an installation of what is stored in a computer-readable recording medium such as a CD-ROM. Alternatively, it may be an installed version downloaded via a transmission medium such as the Internet.

  The control program includes not only a program that can be directly executed by a general-purpose information processing apparatus, but also a program that can be executed by being installed on a hard disk or the like. Also included are those that are compressed or encrypted.

(Regarding the effect of the embodiment of the present invention)
As described above, the hybrid vehicle 1 according to the embodiment of the present invention uses the neural network 50 that has learned the correspondence between the simulations # 1 to #N and the corresponding optimization maps # 1 to #N. Thereby, it is possible to perform the same control as storing all the optimization maps without storing all the optimization maps corresponding to the input values. Therefore, it is possible to optimally control the torque distribution between the engine and the electric motor in accordance with various driver's habits and usage environments.

  In addition, by using a dynamic programming method in simulations # 1 to #N, an optimization map can be generated by calculation in a short time without collecting a large amount of experimental data.

  In Patent Documents 1 and 2, a neural network is used for vehicle control. However, this neural network does not generate an optimization map. In Patent Document 2, learning is performed in advance on a neural network. However, there is no disclosure or suggestion that a dynamic programming technique is used for the learning.

  In addition, in order to verify the effect of the adaptive assist control using the optimization map according to the embodiment of the present invention, a running simulation is performed on various evaluation patterns assuming suburban roads, urban roads, mountain roads, and the like. It was. Here, there are few types of optimization map selection control (referred to as conventional HV control) employed in conventional passenger cars, and adaptive assist control (adaptive) described in the above-described embodiment employed in the hybrid vehicle 1. Type HV control). The results are shown in FIGS.

  FIG. 11 shows the fuel consumption reduction amount (fuel consumption reduction rate) between the conventional HV control and the adaptive HV control. As shown in FIG. 11, it can be seen that the adaptive HV control reduces the fuel consumption for any travel pattern as compared to the conventional HV control. In particular, on suburban and urban roads, the fuel consumption reduction rate is close to 2%. Moreover, the engine load area | region during assistance in the running pattern of the mountain road shown in FIG. 11 is shown in FIG. 12, FIG. FIG. 12 is a diagram showing the relationship between engine torque and engine speed in conventional HV control. FIG. 13 is a diagram showing the relationship between engine torque and engine speed in adaptive HV control. In both FIG. 12 and FIG. 13, the horizontal axis represents the engine rotation speed, and the vertical axis represents the engine torque. The hatched portion in the figure indicates the engine load region. As can be seen from FIGS. 12 and 13, both engine load regions are different. This is a result of using the engine mounted by adaptive control more efficiently.

(Modification)
The embodiment of the present invention can be variously modified without departing from the gist thereof. For example, instead of the neural network 50, a small computer device such as a CPU or a DSP may be provided.

  The hybrid vehicle 1 is assumed to be a vehicle including a manually operated clutch 13 and a transmission 14. However, a transmission as an automatic vehicle may be provided in place of the clutch 13 and the transmission 14.

  Further, the embodiment of the present invention has been described as being applied to a large vehicle such as a truck or a bus. However, when applied to a passenger car, the hybrid control of a passenger car can achieve more accurate hybrid control than before. It is useful in.

It is a principal part block diagram of the hybrid vehicle which concerns on embodiment of this invention. It is a figure for demonstrating the basic concept of the information processing which the hybrid vehicle shown in FIG. 1 performs. It is a figure explaining the learning process of the torque distribution control part of the hybrid vehicle shown in FIG. It is a figure which shows an example of the two-dimensional lookup table produced | generated based on the optimization map shown in FIG. It is a figure which shows the neuron which comprises the neural network shown in FIG. It is a block diagram of the neural network shown in FIG. It is a figure which shows the recording table and threshold value of the weight and threshold value of the neural network which are shown in FIG. It is a block block diagram of the torque distribution control part of the hybrid vehicle shown in FIG. FIG. 9 is a diagram for explaining the amount of electric power that can be input to the battery shown in FIG. 8 input to the neural network of the hybrid vehicle shown in FIG. 1. It is a flowchart which shows operation | movement of the torque distribution control part shown in FIG. The fuel consumption reduction amount (fuel consumption reduction rate) as a result of comparing the conventional HV control adopted by the conventional hybrid vehicle with the applied HV control adopted by the hybrid vehicle according to the embodiment of the present invention is shown. FIG. It is a figure which shows the relationship between the engine rotational speed and engine torque of the conventional HV control in the traveling pattern of the mountain road in FIG. It is a figure which shows the relationship between the engine rotational speed and engine torque of applied type HV control in the traveling pattern of the mountain road in FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Hybrid vehicle, 10 ... Engine, 11 ... Electric motor, 12 ... Hybrid control part (calculation means), 13 ... Clutch, 14 ... Transmission, 15 ... Acceleration sensor, 16 ... Engine control part, 17 ... Battery, 18 ... Battery control , 19 ... inverter, 20 ... CAN, 30 ... torque distribution control unit (torque distribution means), 40 ... computer device, 50 ... neural network (torque distribution means), 60 ... data analysis unit (torque distribution means), 70 ... Torque distribution unit (torque distribution means)

Claims (6)

In a hybrid vehicle comprising an engine, an electric motor, and calculation means for calculating torque distributed to the engine and the electric motor based on a driver's requested torque and a running state of the vehicle,
The calculation means corresponds to a driver's required torque, a rotation speed of the engine, a regenerative current to a battery that supplies power to the electric motor, and a value indicating a charge state of the battery within a predetermined time. Torque distribution means for outputting an instruction value of torque for the engine and the electric motor determined by
The torque distribution means is trained to output the torque distribution value corresponding to the input value,
For this learning, we used pre-simulation using dynamic programming techniques.
A hybrid vehicle characterized by that. The hybrid vehicle according to claim 1,
The data used for the preliminary simulation uses time-series data based on the engine rotation speed of a vehicle with the same power performance and the driver's required torque, or data after a test run using the hybrid vehicle,
A hybrid vehicle characterized by that. The hybrid vehicle according to claim 1 or 2,
The hybrid vehicle comprising a neural network that is trained to output an instruction value of the torque associated with an input value. The hybrid vehicle according to claim 3,
The neural network calculates and generates an optimization map for calculating the indicated value of the torque by interpolation or extrapolation when an approximate value approximate to the input value is input.
A hybrid vehicle characterized by that. Targeting the hybrid vehicle according to any one of claims 1 to 4,
A computer apparatus for performing a simulation for calculating an optimum value of the torque corresponding to the input value by using a dynamic programming technique.   5. A program that, when installed in an information processing apparatus, causes the information processing apparatus to realize the functions of the calculation means and the torque distribution means in a hybrid vehicle according to claim 1.

Images