JP2001105847A - Air conditioner for vehicle - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an air conditioner for a vehicle capable of a necessary and sufficient oil return control of a compressor adapted to environment conditions and vehicle condition of the operation time. SOLUTION: In an air conditioner for vehicle 1 executing an initial operation of a compressor 2 for oil returning at start up of an internal combustion engine E, a completion timing of the initial operation is decided based on a judgment if a necessary and sufficient oil return is completed or not with using an initial operation duration calculator 8 using a neural network and processing an input signal 9 indicating operational environment conditions of the air conditioner for vehicle.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vehicle air conditioner provided with an external variable displacement compressor.

[0002]

2. Description of the Related Art In a compressor used in a cooling cycle device of a vehicle air conditioner, a mixed fluid of a refrigerant and a lubricating oil is brought into direct contact with a sliding portion from a suction side during a circulation of the refrigerant in the compressor. Let it lubricate.

However, in the above-described cooling cycle device, when the cooling cycle device is started, a required amount of oil may not be returned to the compressor due to a low oil concentration in the oil separator, resulting in a shortage of lubricating oil for the compressor. There was a problem. To solve this problem, JP-A-9-104225 discloses that
Even if the operation stop signal is output by the instructing means after the start of the compressor, by controlling the compressor to be forcibly operated for a predetermined time, it is possible to prevent a shortage of lubricating oil at the time of starting the compressor. Such a configuration is disclosed.

Japanese Patent Application Laid-Open No. 9-329087 discloses a configuration in which the refrigerant is circulated through a compressor for a predetermined time depending on the use environment after the internal combustion engine is started, so that the oilless state is not continued in the compressor for a long time. It has been disclosed.

[0005]

In each of the above-mentioned prior arts, the time for forcibly turning on the compressor is a predetermined fixed time. However, whether the refrigerant circulates sufficiently in the cooling cycle and the required amount of oil returns to the compressor depends on environmental conditions (atmospheric temperature), vehicle conditions (internal combustion engine rotation speed), and the like. The appropriate time for turning on the power is not constant.
Therefore, according to the configuration in which the time for forcibly turning on is fixed as in the above-described prior art, the oil is still insufficient, or the oil is sufficiently returned, but the ON state continues and useless operation is performed. Failure will occur.

[0006] This problem is remarkable in the case of a clutchless compressor. That is, in the clutchless compressor, the compressor is constantly rotating, and the oil circulates inside the compressor body when there is no need to send the refrigerant to the outside. Therefore, there is a possibility that the compressor may rotate at a high speed even though the oil remains in the pipe or the heat exchanger and the inside of the compressor is short of oil, and in such a state, the compressor is likely to fail. .

In the case of a swash plate type clutchless compressor, the capacity is changed by using the pressure in the cycle, so that the time until the compressor is switched from the internal circulation to the external circulation depends on the conditions at the time of starting. It changes from tenths of a second to about 4 or 5 seconds. For this reason, when the time of the forced operation of the compressor is fixed within a predetermined fixed time as seen in the above-described related art, the above-described problem is more likely to occur.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned disadvantages in the prior art, and to provide a necessary and sufficient oil return control of a compressor adapted to environmental conditions and vehicle conditions at that time. It is to provide a device.

[0009]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, according to the present invention, in a vehicle air conditioner equipped with an external variable displacement compressor, the compressor is forcibly driven at the time of starting the vehicle internal combustion engine and an oil return is performed. In addition to performing the initial operation, the compressor is continued to be driven until it can be determined that the oil in the cooling cycle has returned to the compressor to the extent that it does not interfere with the operation of the compressor. The above judgment is made based on the pressure in the cooling cycle, or based on the operating environment conditions such as the temperature of the evaporator or the condenser, the outside air temperature, the engine speed, the driving condition of the air conditioner (A / C), the temperature in the passenger compartment, the amount of solar radiation and the like. Is determined by the neural network. What driving environment condition is selected is arbitrary.

A judgment processing program for this judgment processing is started at regular intervals (for example, every 0.1 seconds).
Once the compressor OFF signal is generated, the OFF state is continued until a reset operation such as an internal combustion engine stop operation or an air conditioner (A / C) stop operation is performed. As described above, since the time of the oil return operation is controlled using the neural network based on the situation of the vehicle,
It is possible to minimize necessary and sufficient oil return operation.

The above determination may be made by setting a minimum time for oil return completion in advance and adding an additional time calculated under various conditions to the minimum time. This additional time is calculated by a neural network constituted by operating environment conditions such as the pressure of the refrigeration cycle, the temperature of the evaporator or the condenser, the outside air temperature, the engine rotation speed, the A / C driving state, the vehicle compartment temperature, and the amount of solar radiation. What driving environment condition is selected is arbitrary.

According to this configuration, the operation time of the oil return operation can be set to a time corresponding to the driving condition of the vehicle at the time, and the oil return operation is performed to a necessary minimum without any excess or shortage, and there is no waste. Further, since the minimum time for oil return completion is set in advance, it is possible to cope with an unexpected situation.

According to the present invention, in order to realize a necessary and sufficient oil return control adapted to environmental conditions and vehicle conditions, in a vehicle air conditioner equipped with an external variable displacement compressor, the vehicle air conditioner is used when starting the vehicle internal combustion engine. There has been proposed a configuration in which the amount of oil in the compressor is estimated, and the driving of the compressor is controlled based on the amount.

The above estimation may be performed using a neural network constituted by the pressure of the refrigeration cycle, the temperature of the evaporator or the condenser, the outside air temperature, the engine speed, the A / C driving state, and the like.

As a result, the operation time for the oil return operation can be set to a time corresponding to the driving condition of the vehicle, and the oil return operation is performed to the minimum necessary without excess or deficiency, so that there is no waste.

In any of the configurations, when operating the compressor for an oil return operation of returning the oil in the cooling cycle to the compressor, it is preferable to set the capacity to a maximum, and to use a clutchless compressor as the compressor. Can be adopted.

According to the first aspect of the present invention, there is provided an air conditioner for a vehicle, comprising an external variable displacement compressor, wherein the compressor performs an initial operation for oil return when the internal combustion engine for the vehicle is started. By using a neural network that inputs and processes a plurality of signals indicating the operating environment conditions of the vehicle air conditioner, it is determined whether or not necessary and sufficient oil return has been performed to determine the end timing of the initial operation. A vehicle air conditioner characterized by the above is proposed.

According to the second aspect of the present invention, the environmental conditions include at least a refrigerant pressure in the cooling cycle, an evaporator temperature, an outside air temperature, an engine speed, an A / C driving state, a vehicle interior temperature, and an amount of solar radiation. A vehicle air conditioner according to claim 1 is proposed.

According to the third aspect of the present invention, the neural network is constituted by a computer program, and the computer program is started and operated at regular time intervals, and is turned off until the compressor is reset once the compressor is turned off. Claim 1 in which
Or the vehicle air conditioner of 2 is proposed.

According to the fourth aspect of the present invention, the teacher signal used for learning the neural network is normalized from 0.02 to 0.98 and input.
Or the air conditioner for vehicles of 3 is proposed.

According to the fifth aspect of the present invention, the sigmoid function used in the neural network is a recent function based on a set of a plurality of linear functions, and the plurality of linear functions are a value of the original sigmoid function and the value of the recent function. The input conversion range of the sigmoid function is divided into a plurality of ranges and the coefficients of the respective linear functions in the respective divided ranges are set so that the error from the approximate value obtained by the linear function constituting the predetermined range is within a predetermined range. Claims 1, 2, and 3
Or a vehicle air conditioner according to 4 is proposed.

According to a sixth aspect of the present invention, there is provided a vehicle air conditioner comprising an external variable displacement compressor, wherein the compressor performs an initial operation for oil return when the vehicle internal combustion engine is started. The first means for setting the minimum time for the initial operation and the neural network for inputting and processing a plurality of signals indicating the operating environment conditions of the vehicle air conditioner are used to perform a necessary and sufficient oil return necessary for the oil return. A second means for calculating an additional operation time to be added to the minimum time, and determining the end timing of the initial operation from the minimum time and the additional time in response to the first and second means. A vehicle air conditioner characterized by the following is proposed.

According to the seventh aspect of the present invention, the environmental conditions include at least a refrigerant pressure in the cooling cycle, an evaporator temperature, an outside air temperature, an engine rotation speed, an A / C driving state, a vehicle interior temperature, and an amount of solar radiation. A vehicle air conditioner according to claim 6 is proposed.

According to an eighth aspect of the present invention, there is provided the vehicle air conditioner according to the sixth or seventh aspect, wherein the teacher signal used for learning the neural network is normalized and input from 0.02 to 0.98. Suggested.

According to the ninth aspect of the present invention, the sigmoid function used in the neural network is a recent function based on a set of a plurality of linear functions.
The input conversion range of the sigmoid function is divided into a plurality of ranges so that the error between the original value of the sigmoid function and the approximate value obtained by the linear function constituting the recent function is within a predetermined range. The air conditioner for a vehicle according to claim 6, 7 or 8, wherein the coefficients of the respective linear functions are set.

According to the tenth aspect of the present invention, there is provided an air conditioner for a vehicle, comprising an external variable displacement compressor, wherein the compressor performs an initial operation for oil return when the internal combustion engine for the vehicle is started. Estimating the oil amount in the compressor at the time of starting the internal combustion engine using a neural network that inputs and processes a plurality of signals indicating the operating environment conditions of the vehicle air conditioner, and terminating the initial operation according to the estimation result The vehicle air conditioner is characterized in that the air conditioner is determined.

According to the eleventh aspect, the environmental conditions include at least a refrigerant pressure in the cooling cycle, an evaporator temperature, an outside air temperature, an engine rotation speed, an A / C driving state, a vehicle interior temperature, and an amount of solar radiation. A vehicle air conditioner according to claim 10 is proposed.

According to the twelfth aspect of the present invention, there is provided an air conditioner for a vehicle according to the tenth or eleventh aspect, wherein a teacher signal used in learning of a neural network is input after being normalized from 0.02 to 0.98. Suggested.

According to the invention of claim 13, the sigmoid function used for the neural network is a recent function based on a set of a plurality of linear functions, and the plurality of linear functions are:
The input conversion range of the sigmoid function is divided into a plurality of ranges so that the error between the original value of the sigmoid function and the approximate value obtained by the linear function constituting the recent function is within a predetermined range. 10. The coefficient of each linear function in the above is set.
A vehicle air conditioner according to 1 or 12 is proposed.

[0030]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a configuration diagram showing an example of an embodiment of a vehicle air conditioner according to the present invention. Vehicle air conditioner 1
The cooling cycle 5 includes an external variable displacement clutchless compressor 2, an evaporator 3, and a condenser 4, which are rotationally driven by an internal combustion engine E which is a driving source of the vehicle. Reference numeral 6 denotes a fan for cooling the condenser. Since the configuration of the cooling cycle 5 itself is publicly known, the details of each part will not be described.

Reference numeral 7 denotes a control unit for controlling the operation of the compressor 2. The control unit 7 forcibly drives the compressor 2 when the internal combustion engine E is started to supply the necessary and sufficient oil. An initial operation time calculator 8 for calculating the operation time of the initial operation necessary for returning a large amount in accordance with the environmental condition each time is connected. The initial operation time calculator 8 receives supply of a sensor output required for calculation as an input signal 9 from the control unit 7, and the control unit 7 outputs operation time data 10 indicating a calculation result. Thus, the control unit 7 is configured to be able to control the start-up operation of the compressor 2 so that the amount of lubricating oil does not become excessive or insufficient when the compressor 2 is started.

The control unit 7 includes a microcomputer system having a known configuration. The control unit 7 includes a PD sensor 1 for detecting a pressure level in the condenser 4.
1, the post-evaporation temperature signal 12S from the post-evaporation temperature sensor 12 for detecting the air temperature after passing through the evaporator 3, and the outside air temperature from the outside air temperature sensor 13 for detecting the vehicle outside temperature. Signal 13S, internal combustion engine E
An engine speed signal 14S from a speed sensor 14 for detecting the speed of the vehicle, an A / C signal 15S indicating ON / OFF of an air conditioner (A / C) switch 15, and a vehicle room temperature sensor 16 for detecting the temperature inside the vehicle. The vehicle room temperature signal 16S and the solar radiation signal 17S from the solar radiation sensor 17 for detecting the amount of solar radiation into the vehicle compartment are input.

Then, each of these signals 11S to 17S
Is also input as an input signal 9 to the initial operation time calculator 8 via the control unit 7, and the operation time data 10 calculated based on the input signal 9 is output from the initial operation time calculator 8.
Is output, and the operation time data 10 is input to the control unit 7.

Next, the initial operation time calculator 8 will be described in detail.

As shown in FIG. 2, the initial operation time calculator 8 according to the embodiment of the present invention uses a so-called neural network, and selects in advance from so-called plural environmental factors used in a vehicle air conditioner. The signals representing the states of the plurality of environmental factors and the air conditioners are used as input signals, and the initial operation time for oil return to the compressor 2 is calculated based on these input signals as described later. Is output.

Here, among the so-called plural environmental factors used in the vehicle air conditioner, the input signal 9 of the initial operation time calculator S in the embodiment of the present invention is P
D signal 11S, post-evaporation temperature signal 12S, outside air temperature signal 13
S, the engine speed signal 14S, the A / C signal 15S, the vehicle room temperature signal 16S, and the solar radiation signal 17S are selected.
These are examples, and the input signal 9 does not necessarily need to be limited to these. Each of these sensor values is to be input in a predetermined signal format.

More specifically, the initial operation time calculator 8 can be realized by, for example, a so-called microcomputer and software. When the initial operation time calculator 8 is realized by a microcomputer and software, when the initial operation time calculator 8 is used in the vehicle air conditioner 1, the initial operation time calculator 8 is used in the control unit 7 to perform operation control and the like in the vehicle air conditioner 1. If a microcomputer to be used is used, it is not necessary to provide a separate microcomputer only for the initial operation time calculator 8, so that an installation space for the microcomputer is not required, thereby simplifying the configuration and reducing costs. Becomes

The neural network itself shown in FIG. 2 that constitutes the initial operation time calculator 8 has a well-known and well-known configuration. FIG. 3 shows an example of the configuration of the neural network according to the embodiment of the present invention. It is shown. Hereinafter, a neural network according to an embodiment of the present invention will be described with reference to FIG.

The neural network has an input layer, a hidden layer, and an output layer, and the input signal 9 described above is applied to the input layer.

In this neural network,
The following sigmoid function is used.

Y _j = 1 / {1 + exp (− | x _j |)} (Expression 1)

Here, x _j is a value obtained by subtracting the bias bj from the value obtained by multiplying each input signal Vi to the layer i by the weight Wij (x _j = Σ (Vi × Wij) −bj), and y _j
Is the output signal from layer i (the input signal to layer j).
Note that this sigmoid function is, as shown in FIG.
The value of y is determined in the range of 0 to 1 with respect to the change of the input variable x of -∞ to + ∞.

The initial operation time calculator 8 in the embodiment of the present invention uses the value of the circulation amount of the refrigerant as the teacher signal at the time of learning the neural network. In a state where the air conditioner is circulating, since the refrigerant and the oil are composed of a mixed fluid, the circulation amount of the refrigerant obtained in an experiment can be measured in advance and used as an alternative value of the oil return amount. Because. The circulation amount of the refrigerant measured in advance based on environmental conditions, operating conditions, and the like is given to the neural network as a teacher signal, and the weight Wij and the bias bj are learned so that the initial operating time set value output at that time becomes 1. Ask by. In this way, in each of the other cases, similarly, the acquired data for each input signal is input to the initial operation time calculator 8, and the initial operation time set value becomes the corresponding initial operation time by learning. To obtain the weight Wij and the bias bj.

Then, the weight Wij and the bias bj obtained in this way are set in the initial operation time calculator 8 and can be actually used as the initial operation time calculator 8.

Here, each input signal to the neural network is actually normalized from the minimum value to the maximum value of the calculated data from “0” to “1”, and the weight Wij of each input signal Can be evaluated using the same criteria.

On the other hand, in consideration of the fact that the output characteristics of the sigmoid function "0" and "1" are the saturation output values, the teacher signal changes the maximum value from the minimum value of the measurement data to "0. 02 "to" 0.98 ". In learning in a neural network using the sigmoid function as an input function of a neuron, in consideration of learning efficiency and safety, this is more effective than normalizing the teacher signal from “0” to “1” as described above. , A slightly narrow range from “0.02” to “0.9”
This is because the normalization to "8" not only improves the learning efficiency and the security, but also increases the convergence speed.

Since the vehicle air conditioner 1 is configured as described above, the initial operation time for oil return is appropriately set in the initial operation time calculator 8 according to the situation of the vehicle. As a result, the initial operation for the oil return of the compressor 2 is performed without excess and deficiency, and the oil return operation without waste can be performed.

In the above embodiment, the initial operation time calculator 8 is configured to use the sigmoid function shown in the equation 1, but the straight line of the sigmoid function shown in the equation 1 is used. It may be configured to use an approximation function.

That is, the sigmoid function used in the initial operation time calculator 8 is an equation including an exponential function as shown in the above-mentioned equation 1, but as shown in FIG. A linear approximation at an appropriate interval and a configuration using a plurality of linear functions may be used. By expressing the sigmoid function with a plurality of linear functions in this way, the memory capacity for storing the sigmoid function is reduced, and the calculation time is shortened.

For example, when the sigmoid function shown in Equation 1 is stored in an 8-bit so-called built-in microcomputer, an exponential function and a so-called floating-point arithmetic library are used in order to obtain a high-precision function value. Known and well-known software is required, which increases the capacity of the storage element (for example, ROM) and the operation time.

On the other hand, since the linear function does not include division, there is no need for a floating-point operation, and the operation accuracy can be maintained even when an integer operation is used.
There is an advantage that the storage capacity of the storage element such as M and the calculation time are reduced.

FIG. 5 shows a specific example of the sigmoid function approximated by a straight line. The so-called approximation function shown in FIG. 5 is divided into 17 ranges of the variable x so that the error based on the sigmoid function of the formula 1 is within ± 0.005, for example. The sigmoid function is represented as a linear function (y = a _i × x + b _i ; i = 1 to 17). In this embodiment, the sigmoid function approximated to the straight line shown in FIG. 5 may be stored in a storage element such as a ROM and used.

FIG. 6 shows a result obtained by calculating an error between the sigmoid function shown in the equation 1 and the sigmoid function approximated by a straight line shown in FIG. According to this, Equation 1
The magnitude of the error between the sigmoid function and the sigmoid function approximated by a straight line is ± 0.0
It can be confirmed that it is within 05.

In the example shown in FIG. 5, the sigmoid function is approximated by 17 straight lines so that the error is within ± 0.005. However, the present invention is not limited to this. From a practical point of view, the error is within 0.03 (± 3%
If there is no error, the number of straight lines used for approximation can be set as appropriate within the range of the error.

The approximation does not necessarily have to be a broken line, and in some cases it may be better to make the approximation discontinuous from the viewpoint of reducing the number of straight lines (the number of divisions of the input range) and increasing the calculation speed.

Further, although the case where a so-called logarithmic sigmoid function is used has been described above, the logarithmic sigmoid function is not necessarily required.
Of course, a sigmoid function of a type may be used, and a sigmoid function of another type may be used. yj = {tanh (xj) +1} / 2 (Equation 2)

Further, the input signal of the initial operation time calculator 8 in the above-described embodiment of the present invention is selected from the viewpoint of having a relatively large influence on the initial operation time calculation value in performing the initial operation time calculation. However, it is needless to say that the input signals need not be limited to these, and the type and number of the input signals are variously set in consideration of the required accuracy of the set values and the like. Good thing.

FIG. 7 is a block diagram showing another embodiment of the present invention.

The vehicle air conditioner 20 shown in FIG. 7 is provided with a minimum time setting unit 21, an additional operation time calculation unit 22, and an addition unit 23 in place of the initial operation time calculation unit 8, so that the oil return can be adjusted without excess or shortage. The only difference from the embodiment shown in FIG. 1 is that the configuration is such that the optimal value of the initial operation time for execution can be set from the operating environment conditions at that time. Therefore, the same reference numerals are given to the portions corresponding to the respective portions in FIG. 1 among the respective portions in FIG. 7, and the description thereof will be omitted.

The minimum time setting section 21 is for setting the minimum time which is the minimum value of the initial operation time for the required oil return to the compressor 2, and is set from the minimum time setting section 21. The minimum time data D1 indicating the minimum time is output.

The additional operation time calculating section 22 is configured as a neural network for input processing of the input signal 9 including a plurality of signals indicating the operating environment conditions of the vehicle air conditioner 1. It is learned that an additional operation time to be added to the minimum time in order to perform an oil return is calculated. The additional operation time calculation unit 22 outputs additional operation time data D2 indicating the calculated additional operation time.

The minimum time data D1 and the additional operation time data D2 are input to the adder 23, where the additional operation time is added to the minimum time to determine the operating environment condition of the vehicle air conditioner 20 at that time. A suitable optimal initial operation time is obtained, and the initial operation time data D3 indicating the obtained initial operation time is output.

The initial operation time data D3 corresponds to the operation time data 10 shown in FIG. 1, and the control unit 7 performs the initial operation of the compressor 2 based on the initial operation time data D3.

Since the vehicle air conditioner 20 is configured as described above, in addition to the effect obtained in the vehicle air conditioner 1, the initial operation is performed only for the minimum time set by the minimum time setting unit 21. In addition, an effect that an unexpected situation can be dealt with can be obtained.

It should be noted that the additional operation time calculation unit 22 may also be configured as a neural network similar to that described with reference to FIGS.

FIG. 8 is a block diagram showing still another embodiment of the present invention.

The vehicle air conditioner 30 shown in FIG. 8 includes an oil amount estimator 31 for estimating the oil amount in the compressor 2 when the internal combustion engine E is started, instead of the initial operation time calculator 8 in FIG. This embodiment differs from the embodiment shown in FIG. Therefore, the same reference numerals are given to the portions corresponding to the respective portions in FIG. 1 among the respective portions in FIG. 8, and the description thereof will be omitted.

The oil amount estimator 31 is used for the vehicle air conditioner 1.
Is configured as a neural network that performs input processing of an input signal 9 including a plurality of signals indicating the operating environment conditions, and estimates an oil amount in the compressor 2 at the time of starting the internal combustion engine E based on the input signal 9. It is something that has been learned. From the oil amount estimator 31, oil amount data D4 indicating the estimated oil amount is output and input to the control unit 7.

The control unit 7 calculates an initial operation time required to perform a sufficient and sufficient oil return to the condenser 2 based on the oil amount indicated by the oil amount data D4, and calculates the calculated initial operation time. The compressor 2 is initially operated for the time.

Since the vehicle air conditioner 30 is configured as described above, the initial operation is performed without waste for an appropriate time corresponding to the oil amount at that time.

The oil amount estimator 31 may of course be configured as a neural network similar to that described with reference to FIGS. 2 to 6 for the initial operation time calculator 8.

[0073]

According to the present invention, the oil return to the compressor is required and sufficiently performed by using the neural network based on a plurality of operating environment conditions used in the vehicle air conditioner. In addition, the operation time of the oil return operation can be set to a time suitable for the current driving condition of the vehicle by utilizing the existing plurality of sensor outputs in the vehicle air conditioner. The oil return operation is performed to the minimum necessary without excess or shortage, and there is no waste. In addition, when a configuration in which the minimum time for oil return completion is set in advance is used, unexpected situations can be dealt with.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of an embodiment of a vehicle air conditioner according to the present invention.

FIG. 2 is a configuration diagram showing a configuration of an initial operation time calculator of FIG. 1;

FIG. 3 is a configuration diagram showing a configuration of a neural network constituting the initial operation time calculator shown in FIG. 2;

FIG. 4 is a characteristic diagram showing a logarithmic sigmoid function used in the neural network shown in FIG. 3;

FIG. 5 is a characteristic diagram showing a sigmoid function approximated by a straight line.

FIG. 6 is a characteristic diagram showing an error between an original sigmoid function and a sigmoid function approximated by a straight line.

FIG. 7 is a block diagram showing another embodiment of the vehicle air conditioner according to the present invention.

FIG. 8 is a block diagram showing still another embodiment of the vehicle air conditioner according to the present invention.

[Explanation of symbols]

 1, 20, 30 Vehicle air conditioner 2 Compressor 3 Evaporator 4 Capacitor 5 Cooling cycle 7 Control unit 8 Initial operation time calculator 9 Input signal 10 Operating time data 11 PD sensor 11S PD signal 12 Temperature sensor after evaporation 12S Temperature signal after evaporation 13 outside air temperature sensor 13S outside air temperature signal 14 rotation sensor 14S engine speed signal 15 air conditioner (A / C) switch 15S A / C signal 16 car room temperature sensor 16S car room temperature signal 17 solar radiation sensor 17S solar radiation signal E internal combustion engine

Claims (13)

[Claims] An external variable displacement compressor is provided,
In a vehicle air conditioner configured to cause the compressor to perform an initial operation for oil return when a vehicle internal combustion engine is started, a neural network that inputs and processes a plurality of signals indicating operating environment conditions of the vehicle air conditioner is provided. The vehicle air conditioner according to claim 1, wherein whether or not a necessary and sufficient oil return has been performed is used to determine the end timing of the initial operation. 2. The vehicle according to claim 1, wherein the environmental conditions include at least a refrigerant pressure in the cooling cycle, an evaporator temperature, an outside air temperature, an engine rotation speed, an air conditioner driving state, a vehicle interior temperature, and an amount of solar radiation. Air conditioner. 3. The neural network is constituted by a computer program. The computer program is started and operated at regular intervals, and once the compressor is turned off, the off state is continued until the compressor is reset. The vehicle air conditioner according to claim 1 or 2. 4. The air conditioner for a vehicle according to claim 1, wherein a teacher signal used for learning of the neural network is normalized from 0.02 to 0.98 and input. 5. A sigmoid function used in the neural network is a recent function based on a set of a plurality of linear functions, and the plurality of linear functions are obtained from a value of an original sigmoid function and a linear function constituting the recent function. The input conversion range of the sigmoid function is divided into a plurality of ranges so that an error from an approximate value to be obtained is within a predetermined range, and a coefficient of each linear function in each of the divided ranges is set. The vehicle air conditioner according to 1, 2, 3, or 4. 6. An external variable displacement type compressor is provided.
A first means for setting a minimum time of the initial operation, wherein the compressor is configured to cause the compressor to perform an initial operation for oil return when the internal combustion engine of the vehicle is started, and an operation of the air conditioner for the vehicle. Second means for calculating an additional operation time to be added to the minimum time necessary for the necessary and sufficient oil return to be performed using a neural network for inputting a plurality of signals indicating environmental conditions, An air conditioner for a vehicle, wherein the end timing of the initial operation is determined from the minimum time and the additional time in response to first and second means. 7. The vehicle according to claim 6, wherein the environmental conditions include at least a refrigerant pressure in the cooling cycle, an evaporator temperature, an outside air temperature, an engine rotation speed, an air conditioner driving state, a vehicle interior temperature, and an amount of solar radiation. Air conditioner. 8. The vehicle air conditioner according to claim 6, wherein a teacher signal used for learning of the neural network is normalized from 0.02 to 0.98 and input. 9. A sigmoid function used in the neural network is a recent function based on a set of a plurality of linear functions, and the plurality of linear functions are determined by a value of an original sigmoid function and a linear function constituting the recent function. The input conversion range of the sigmoid function is divided into a plurality of ranges so that an error from the obtained approximate value is within a predetermined range, and coefficients of respective linear functions in the respective divided ranges are set. The vehicle air conditioner according to 6, 7, or 8. 10. An air conditioner for a vehicle, comprising an external variable displacement compressor, wherein the compressor performs an initial operation for oil return when the internal combustion engine for a vehicle is started. Estimating the oil amount in the compressor at the time of starting the internal combustion engine using a neural network that inputs a plurality of signals indicating environmental conditions, and determining the end timing of the initial operation according to the estimation result. A vehicle air conditioner characterized by the above-mentioned. 11. The vehicle according to claim 10, wherein the environmental conditions include at least a refrigerant pressure, an evaporator temperature, an outside air temperature, an engine rotation speed, an air conditioner driving state, a vehicle interior temperature, and an amount of solar radiation in the cooling cycle. Air conditioner. 12. The air conditioner for a vehicle according to claim 10, wherein a teacher signal used at the time of learning of the neural network is normalized and input from 0.02 to 0.98. 13. A sigmoid function used in a neural network is a recent function based on a set of a plurality of linear functions, and the plurality of linear functions are obtained from a value of an original sigmoid function and a linear function constituting the recent function. 11. An input conversion range of a sigmoid function is divided into a plurality of ranges so that an error from an approximate value obtained falls within a predetermined range, and coefficients of respective linear functions in each of the divided ranges are set. , 11 or 12.