JP7579205B2 - Learning processing device and learning processing method - Google Patents

Description

Embodiments of the present invention relate to a learning processing device and a learning processing method.

One example of a crew operation during railway operations is the manual correction of the target temperature of a vehicle air conditioner to adjust the air conditioning of a railway vehicle. Inside the vehicle air conditioner, a compressor compresses the refrigerant, creating a refrigeration cycle, which is a refrigerant circulation mechanism for exchanging heat between the inside and outside of the vehicle. This circulates the air inside the vehicle, raising and lowering the temperature inside the vehicle. The vehicle air conditioner is also controlled by a control device installed in the vehicle, which adjusts the temperature inside the vehicle. The control device sets the target temperature based on measurements such as the temperature inside the vehicle obtained by a temperature sensor, for example.

However, in reality, the crew may not be satisfied with the above control, for example because the temperature sensor measurement value is significantly different from the target temperature. Therefore, the crew may have to manually enter the target temperature correction value from time to time, which is time-consuming. In other words, it is not possible to substitute for the crew's judgment and operation, and manual operation may be time-consuming.

Patent No. 5840309 JP 2019-202727 A

The objective is to provide a learning processing device and a learning processing method that can reduce the effort required for manual operation.

The learning processing device according to this embodiment includes a data acquisition unit, a model construction unit, and a control unit. The data acquisition unit acquires operation history data of the on-board device installed in the vehicle by the driver and element data related to the vehicle in operation used to determine the operation of the on-board device, and stores the data in the data storage unit. The model construction unit constructs a learning model based on the correspondence between the operation of the on-board device to be learned, which is included in the operation history data, and the element data from the past when the operation to be learned was performed. The control unit generates a control command for controlling the on-board device by applying the element data for the current day to the learning model.

FIG. 1 is a block diagram showing an example of an overall configuration of a railway vehicle according to a first embodiment. FIG. 2 is a block diagram showing an example of the configuration of a learning processing device according to the first embodiment. FIG. 4 is a flowchart showing an example of an operation of the learning processing device in the first embodiment. 5 is a diagram showing an example of a data filter by a data reading unit in the first embodiment. FIG. 2 is a diagram showing an example of the configuration of a learning model in the first embodiment. FIG. 11 is a diagram showing an example of adoption of a machine learning model in the first embodiment. FIG. 2 is a block diagram showing an example of the overall configuration of a railway vehicle in a comparative example. FIG. 11 is a block diagram showing an example of the overall configuration of a railway vehicle according to a second embodiment.

Below, an embodiment of the present invention will be described with reference to the drawings. The present invention is not limited to this embodiment. The drawings are schematic or conceptual, and the ratio of each part is not necessarily the same as the actual one. In the specification and drawings, elements similar to those described above with respect to the previous drawings are given the same reference numerals, and detailed explanations are omitted as appropriate.

First Embodiment
Fig. 1 is a block diagram showing an example of the overall configuration of a railway vehicle 1 in the first embodiment. Fig. 1 shows the railway vehicle 1 having one car. However, the first embodiment is not limited to a one-car formation, and can also be applied to a formation including a plurality of railway vehicles 1. In this case, the vehicle information control device 100 in Fig. 1 is provided at least for each formation including a plurality of railway vehicles 1.

The railway vehicle 1 includes a vehicle information control device 100, an air conditioning control device 101, an air conditioner 102, a load-varying detection unit 103, a temperature and humidity sensor 104, and a learning processing device 205. Note that the air conditioning control device 101 and the air conditioner 102 may also be referred to as an air conditioner 105.

The vehicle information control device 100 is a driver's operation terminal located in the driver's cab. The vehicle information control device 100 is connected to other devices (not shown) in addition to the air conditioning device 105, and receives various data from the motor, brake device, etc. The vehicle information control device 100 displays various data and also accepts operations from devices connected to the vehicle information control device 100.

The vehicle information control device 100 also sends data for air conditioning control, such as occupancy rate and calendar information, to the air conditioning control device 101. The vehicle information control device 100 also sends a target temperature correction for the air conditioning control device 101. "Target temperature correction" is a control command for correcting the target temperature set by the air conditioning control device 101. If the driver is not satisfied with the target temperature set by the air conditioning control device 101, he or she can manually input a target temperature correction value using the vehicle information control device 100.

The air conditioning control device 101 acquires temperature and humidity information from a temperature sensor and a humidity sensor (temperature and humidity sensor 104 shown in FIG. 1), and issues control commands to the compressor in the air conditioner 102 while also referring to the status information of the hardware air conditioner 102. The air conditioning control device 101 sets a target temperature based on measured values such as the temperature inside the vehicle acquired by a temperature sensor, for example. This target temperature is corrected by the target temperature correction described above. Also, for example, the air conditioning control device 101 performs feedback control using measured values such as the temperature so that the temperature is within a certain range from the target temperature.

The air conditioner 102 conditions the air inside the railway vehicle 1 based on the control commands of the air conditioning control device 101. The air conditioner 102 has, for example, a compressor. Inside the air conditioner 102, the compressor compresses the refrigerant, creating a refrigeration cycle, which is a refrigerant circulation mechanism for exchanging heat between the inside and outside of the vehicle. This circulates the air inside the vehicle, raising and lowering the temperature inside the vehicle.

The load adaptive detection unit 103 detects, for example, the number of passengers on the railway vehicle 1. From this number of passengers, information such as the occupancy rate can be obtained. The load adaptive detection unit 103 sends the multiplication rate number to the vehicle information control device 100.

The temperature and humidity sensor 104 detects the temperature and humidity inside the railway vehicle 1. The temperature and humidity sensor 104 sends the temperature and humidity information to the air conditioning control device 101.

The learning processing device 205 constructs a learning processing model that substitutes for the operation of the crew from on-board data associated with the operation record of the crew using the vehicle information control device 100. The learning processing device 205 receives on-board data from the vehicle information control device 100. The "on-board data" includes various information about the train 1 in operation. The on-board data includes at least operation record data and element data. The "operation record data" is data indicating a record of the crew operating the on-board device via the train information control device 100. The on-board device is, for example, the air conditioning device 105. The operation record data is, for example, target temperature correction. The "element data" is data related to the train 1 in operation that is used to determine the operation of the on-board device. When the target of the crew's operation is the air conditioning device 105 (air conditioning control device 101), the element data includes, for example, the time, the occupancy rate, and temperature and humidity information inside the train.

The learning processing device 205 also issues control instructions to the vehicle information control device 100 based on the constructed learning model and on-board data obtained during operation. This control instruction is a control command for target temperature correction generated by the learning processing device 205 to replace the driver's operation.

The vehicle information control device 100, which receives a control instruction from the learning processing device 205, sets a target temperature correction in the air conditioning control device 101 based on the control instruction. This makes it possible to automatically control the air conditioning device 105 (air conditioning control device 101) instead of the driver's operation.

The processing of the vehicle information control device 100, the air conditioning control device 101, the learning processing device 205, etc. described above is performed by hardware including circuits such as a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or an FPGA (Field-Programmable Gate Array).

Figure 2 is a block diagram showing an example of the configuration of the learning processing device 205 in the first embodiment.

The learning processing device 205 includes a data acquisition unit 301, a data storage DB (database) 302, a data reading unit 303, a data processing unit 304, a model construction unit 305, and a control instruction unit 306.

The data acquisition unit 301 sequentially acquires on-board data (operation history data and element data) from the vehicle information control device 100 and sequentially accumulates (stores) the data in the data accumulation DB 302.

The crew can switch between a manual operation mode and an automatic operation mode using the learning processing device 205. Operation performance data is accumulated in the manual operation mode. Element data is accumulated in both modes.

More specifically, the data acquisition unit 301 associates the operation history data with the element data by time and accumulates them in the data accumulation DB 302. The data acquisition unit 301 may also acquire operation status information of the railway vehicle 1 in operation and accumulate it in the data accumulation DB 302. "Operation status information" is information included in the on-board data that indicates the operation status of the railway vehicle 1. The operation status information includes, for example, information on at least one of whether or not the train is being operated by a single driver and whether or not the train is being operated as a non-commercial train, such as a deadhead. The operation status information may also include information on the opening and closing of the doors of the railway vehicle 1.

The data storage DB (data storage unit) 302 stores on-board data. The data storage DB 302 is realized, for example, by a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a hard disk, etc.

The data reading unit 303 reads out the learning target data from the data accumulation DB 302. The learning target data includes operation history data and element data to be considered for modeling. The data reading unit 303 also filters the data, such as excluding one-man operation and non-commercial running from the above data. That is, the data reading unit 303 reads out the operation history data and element data used to build the learning model from the data accumulation DB 302 based on the operation status information. More specifically, the data reading unit 303 reads out the operation history data and element data used to build the learning model from the data accumulation DB 302, excluding at least one of one-man operation and non-commercial running. During one-man operation, the conductor does not board the train, and the driver also serves as the conductor. Therefore, during one-man operation, the driver may not have time to operate the air conditioning. During non-commercial running, there are almost no passengers on the railway car 1. By excluding data from one-man operation and non-commercial driving, it is possible to select data that is more meaningful as the results of air conditioning operation and reduce the data load. Details of the data filter will be explained later with reference to Figure 4.

The data acquisition unit 301 may be configured not to store data that is excluded by the data filter of the data reading unit 303 in the data accumulation DB 302.

The data processing unit 304 divides the data received from the data reading unit 303 into a data set for learning and an evaluation set for model construction. The data processing unit 304 also functions as a data division unit.

The model construction unit 305 constructs (generates) a learning model based on the correspondence between the operation of the air conditioner 105 to be learned, which is included in the operation history data, and element data in the past when the operation to be learned was performed. The "operation to be learned" is, for example, target temperature correction. More specifically, the model construction unit 305 constructs a learning model based on the correspondence between the operation to be learned and element data associated with the operation to be learned by time.

The model construction unit 305 also tries multiple machine learning methods and adopts the most appropriate method as the learning processing model. That is, the model construction unit 305 evaluates the multiple constructed learning models and sets the learning model with the highest evaluation as the learning model to be applied by the control instruction unit 306.

The control instruction unit (control unit) 306 issues control instructions based on the learning model. The control instruction unit 306 receives operation substitution data from the data acquisition unit 301 and applies it to the learning model. The operation substitution data is element data acquired on the day on which the operation substitution of the driver is performed. The control instruction unit 306 applies the element data on the day to the learning model to generate a control command for controlling the air conditioning device 105. The control instruction unit 306 outputs a control command to the air conditioning control device 101 (air conditioning device 105) via, for example, the vehicle information control device 100.

Next, we will explain the processing flow of the learning processing device 205.

Figure 3 is a flow diagram showing an example of the operation of the learning processing device 205 in the first embodiment.

First, the data acquisition unit 301 acquires on-board data from the vehicle information control device 100 (S201). The data acquisition unit 301 also stores the acquired on-board data in the data accumulation DB 302. Here, the vehicle information control device 100 manages time series data at sampling intervals of less than one second, but the data acquisition unit 301 can further downsample and store the data at any specified interval, such as every 15 seconds. In other words, the data acquisition unit 301 may thin out the acquired on-board data (operation history data and element data) and accumulate it in the data accumulation DB 302.

Next, the data reading unit 303 reads out the target temperature correction value (objective variable Y of the classification model) to be operated (S202). In the specific example of the first embodiment, for example, the in-vehicle temperature, the outside air temperature, the occupancy rate, the time period, and the humidity are set as the judgment factor data group Xi for the target temperature correction value to be operated, which is Y of the model formula of the formula (1).

Next, the data reading unit 303 reads out the judgment factor data (the explanatory variable group Xi of the classification model) that is the judgment criterion for the target temperature correction (S203). Any data may be selected as the judgment factor data (factor data), or data whose correlation coefficient with respect to the target temperature correction value is equal to or greater than a certain threshold value may be selected.

Next, the data reading unit 303 extracts data from the data accumulation DB 302 by performing data filtering (S204). For example, when the data reading unit 303 reads data, it can narrow down the data to be learned and reduce the processing load by not reading data from one-man operation (when the conductor is not on board and the driver also serves as the conductor), not reading data from non-commercial runs (when the train number is not set), and reading data from situations where the space has changed significantly after the train stops at a station and the doors are closed.

Figure 4 is a diagram showing an example of a data filter by the data reading unit 303 in the first embodiment. The time series data shown in Figure 4 includes, for example, the date and time, the outside temperature, the occupancy rate, and the presence or absence of one-man operation.

As shown in FIG. 4, data extraction during one-man operation can be performed by filtering data using equipment signals (operation status information) such as a one-man switch signal. A switch for switching on and off one-man operation is located in the vehicle information control device 100, for example, so that the driver can operate it. In this case, the data reading unit 303 filters and reads out time-series data for cases when one-man operation is not performed. This makes it possible to build a model, as will be explained later in step S209, using data on conditions under which the driver and conductor operate under relatively detailed conditions.

The data reading unit 303 may also read out operation record data and element data used to construct a learning model from the data accumulation DB 302 based on at least one of the value of the element data and a specified operation of the railway vehicle 1. That is, the data reading unit 303 may, for example, filter by the value of the element data. For example, the data reading unit 303 applies a filter that reads out data on a situation in which the temperature, etc. of the interior space of the vehicle changes significantly after the railway vehicle 1 stops at a station and the doors are closed. While the doors are open, air is likely to be exchanged between the inside and outside of the railway vehicle 1, making the interior temperature, etc. unstable. By excluding cases in which the interior temperature, etc. is likely to become unstable, it is possible to select data that is more meaningful as the history of air conditioning operation and reduce the data load.

Next, the data processing unit 304 obtains both short-term trend data for the most recent few days and data that takes into account periodicity for several months during the same season last year from the data output by the data reading means, and processes them into data for model construction (S205, S206). That is, the data processing unit 304 extracts operation performance data and element data used to construct a learning model so that the model construction unit 305 constructs a learning model that takes into account at least one of trend fluctuation and seasonal fluctuation. Trend fluctuation indicates, for example, fluctuations in the value of time-series data from the current day to a specified number of days in the past. The specified number of days is, for example, 2 to 3 days. For example, if the current day is summer, the data processing unit 304 extracts data from the summer of one year ago and two years ago as seasonal fluctuations. Either the short-term trend data or the data that takes into account periodicity may be selected, and an arbitrary period may be specified.

Next, the data processing unit 304 divides the data, for example, into the most recent three days' worth of data for model evaluation and data other than the most recent three days' worth of data for model learning (S207). That is, the data processing unit 304 divides the operation history data and element data used to build the learning model into learning data for building the learning model and evaluation data for evaluating the built learning model.

Next, the model construction unit 305 executes steps S209 to S212 for the number of machine learning methods to be tried (S208). The model construction unit 305 constructs classification models using multiple supervised machine learning methods, and adopts the best model from the construction results of the classification model to be operated on as the learning processing model (S209 to S212). General-purpose machine learning methods include decision trees, k-nearest neighbors, random forests, support vector machines, and deep learning, and generally there is no method that is always optimal in every situation. Therefore, the model construction unit 305 adopts the method with the best score among predetermined indices for the learning data as the learning processing model.

Figure 5 is a diagram showing an example of the configuration of a learning model in the first embodiment. Figure 5 shows a decision tree model.

As shown in Figure 5, by branching out the decision-making element data such as outside temperature and occupancy rate for each condition from the top, an appropriate target temperature correction value can be obtained. In the example shown in Figure 5, either 0°C correction or -2.0°C correction is selected based on the outside temperature (°C), inside temperature (°C), occupancy rate (percentage (full rate)), and time (hour) in the element data for that day. Note that 0°C correction indicates that no temperature correction is performed.

Figure 6 is a diagram showing an example of the adoption of a machine learning model in the first embodiment. Figure 6 shows an example in which three machine learning techniques, a decision tree, a k-nearest neighbor method, and a random forest, are modeled using the same training data and the accuracy rates are compared when collated with evaluation data.

In the example shown in FIG. 6, the accuracy rates of the learning models constructed using the decision tree, k-nearest neighbor method, and random forest are 75%, 81%, and 79%, respectively. Therefore, the model obtained using the k-nearest neighbor method has the highest accuracy rate for the evaluation data. The model construction unit 305 adopts the model obtained using the k-nearest neighbor method as the learning processing model. Note that the above three methods are only examples, and any machine learning method may be added or removed from the methods. Regarding indices for evaluating the classification model, there are accuracy rate, precision rate, recall rate, F-value, etc., and any indices can be selected.

Next, the control instruction unit 306 instructs the vehicle information control device 100 to set a target temperature correction value based on the learning processing model obtained by the model construction unit 305 (S213). At this time, if the decision tree shown in FIG. 5 is constantly applied, for example, the correction value 0°C and the correction value -2.0°C may be frequently alternately specified, resulting in hunting. Hunting may lead to equipment failure. Therefore, the control instruction unit 306 issues a control instruction at a predetermined fixed period (for example, 15 minutes) or with data at a timing when the interior space and passengers have settled down 15 seconds after the door is closed. That is, the control instruction unit 306 outputs a control command to the air conditioner 105 at each first predetermined period. The control instruction unit 306 also stops outputting a control command to the air conditioner 105 while the door of the railway vehicle 1 is open and for a second predetermined period after the door is closed. This makes it possible to reduce the number of times the learning processing device 205 executes automatic control when the vehicle interior space is unstable, and to prevent the target temperature change from becoming unstable.

The processing flow in Figure 3 may be applied to each vehicle individually, or the data from each vehicle may be combined and applied to the entire train.

As described above, according to the first embodiment, the data acquisition unit 301 acquires operation history data and element data, and accumulates them in the data accumulation DB 302. The model construction unit 305 constructs a learning model based on the correspondence between the operation to be learned and element data from the past when the operation to be learned was executed. The control instruction unit 306 generates a control command for controlling the air conditioning device 105 (air conditioning control device 101) by applying the learning model to the element data for the day. This allows the learning model to replace the judgment and operation of the driver regarding the operation of the air conditioning device 105, such as target temperature correction.

Figure 7 is a block diagram showing an example of the overall configuration of a railway vehicle 1 in a comparative example. In the comparative example shown in Figure 7, a learning processing device 205 is not provided.

In the example shown in FIG. 7, the air conditioning control device 101 performs feedback control using measured values such as temperature so that the temperature is within a certain range from the target temperature. In addition, the system may also have a function to automatically correct the target temperature when the occupancy rate or humidity obtained from the load compensation detection unit that detects the passenger load is equal to or exceeds a certain threshold. In reality, however, if the crew is not satisfied with the automatic correction function, they may take the time to manually provide a target temperature correction value at any time. As with the automatic correction function, crew members are likely to be aware of factors such as the temperature, humidity, and occupancy rate inside the car, and also to manually operate the system while taking into account operational information such as rush hour times. In addition, crew members often manually correct the target temperature according to the operation of each route or each train formation.

In contrast to this, in the first embodiment, by accumulating decision-making factor data such as outside temperature, in-car temperature, and occupancy rate along with the driver's manual correction operation data, it is possible to accumulate data including the circumstances in which the human system will operate the air conditioning. In addition, by using machine learning with the manual correction operation data as training data, it is possible to build a learning model that can replace the driver's operation. As a result, it is possible to reduce the effort required for the driver to perform manual operations.

Recently, an environment is being established where large amounts of on-board data managed by railway vehicle information control devices can be stored on ground servers and analyzed. This makes it possible to analyze data downloaded to the ground, apply machine learning, and formulate, i.e., model, relationships between various pieces of performance data. Therefore, for example, the data storage DB 302 shown in FIG. 2 may be provided on the ground side, not limited to on-board. If the above-mentioned ground server cannot be secured, as shown in FIG. 1, the learning processing device 205 may store and model data on the train. In this case, it is preferable that the data acquisition unit 301 thins out the data stored in the data storage DB 302 as described above.

The railway vehicle 1 may further include a detection unit that detects information such as door opening/closing information. In this case, the detection results of the detection unit are included in the on-board data and stored in the data storage DB 302.

Furthermore, crew operation of the air conditioning control device 101 is not limited to target temperature correction. For example, it can also be applied to other crew operations related to the air conditioner 102, such as line flow fans (circulators for use inside railway vehicles), ventilation fans, or fan strength adjustments. Furthermore, the object of crew operation is not limited to the air conditioning device 105, and may be other on-board devices.

As described with reference to FIG. 1, the first embodiment is applicable to a formation consisting of multiple railway cars 1. The learning processing device 205 in the first embodiment is, for example, attached to the vehicle information control device 100. Therefore, similar to the vehicle information control device 100, the learning processing device 205 is provided at least for each formation including multiple railway cars 1. In this case, the learning processing device 205 shown in FIG. 1 acquires main line data including individual on-board data of the multiple railway cars 1 from the vehicle information control device 100. This makes it possible to substitute for the operation of the air conditioning devices 105 of the multiple railway cars 1 with a small number of devices.

The learning processing device 205 may also be included inside the vehicle information control device 100.

Second Embodiment
8 is a block diagram showing an example of the overall configuration of a railway vehicle 1 according to the second embodiment. The second embodiment differs from the first embodiment in that a learning processing device 205 transmits and receives data and signals to and from an air conditioning control device 101 instead of the vehicle information control device 100.

The vehicle information control device 100 sends the on-board data to the air conditioning control device 101.

The air conditioning control device 101 sends the on-board data to the learning processing device 205.

The learning processing device 205 is attached to, for example, the air conditioning device 105 (air conditioning control device 101). Therefore, the learning processing device 205 is provided for each air conditioning device 105. Note that the learning processing device 205 may be provided for each railway vehicle 1. The learning processing device 205 receives on-board data from the air conditioning control device 101 and sends target temperature corrections to the air conditioning control device 101.

More specifically, the data acquisition unit 301 acquires operation record data and element data from the air conditioning device 105 (air conditioning control device 101) without going through the vehicle information control device 100, which receives operations related to the operation of the railway vehicle 1 from the crew. The control instruction unit 306 outputs a control command to the air conditioning device 105 (air conditioning control device 101) without going through the vehicle information control device 100. This makes it possible to build a learning model and perform operation substitution individually for multiple railway vehicles 1. For example, if the manufacturers of the air conditioning device 105 or the railway vehicle 1 are different, the operating tendency may differ for each manufacturer. Also, for example, a weakly air-conditioned car may have a different operating tendency from a normal car. Also, for example, when multiple formations are connected to form one formation, interface problems may occur between the new formation and the old formation. Also, for example, if the manufacturers of the vehicle information control device 100 and the learning processing device 205 are different, interface problems may occur. In such cases, it is more preferable to build a learning model for each vehicle or each air conditioning device 105.

Other configurations of the learning processing device 205 in the second embodiment are similar to the corresponding configurations of the learning processing device 205 in the first embodiment, so detailed descriptions thereof will be omitted. The learning processing device 205 in the second embodiment can obtain the same effects as the first embodiment.

The learning processing device 205 may be included inside the air conditioning control device 101.

At least a part of the learning processing device 205 and the learning processing method according to this embodiment may be configured as hardware or software. When configured as software, a program that realizes at least a part of the functions of the learning processing device 205 and the learning processing method may be stored in a recording medium such as a flexible disk or CD-ROM, and may be read and executed by a computer. The recording medium is not limited to removable ones such as magnetic disks and optical disks, but may be a fixed recording medium such as a hard disk device or memory. In addition, a program that realizes at least a part of the functions of the learning processing device 205 and the learning processing method may be distributed via a communication line (including wireless communication) such as the Internet. Furthermore, the program may be encrypted, modulated, or compressed and distributed via a wired line or wireless line such as the Internet, or stored in a recording medium.

Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are within the scope of the invention and its equivalents as set forth in the claims, as well as the scope and gist of the invention.

1 Railway vehicle, 100 Vehicle information control device, 101 Air conditioning control device, 102 Air conditioner, 103 Load response detection unit, 104 Temperature and humidity sensor, 205 Learning processing device, 301 Data acquisition unit, 302 Data storage DB, 303 Data reading unit, 304 Data processing unit, 305 Model construction unit, 306 Control instruction unit

Claims (14)

a data acquisition unit that acquires operation record data by a driver of an on-board device mounted on a vehicle and element data related to the vehicle in operation that is used to determine the operation of the on-board device, and stores the data in a data storage unit;
a model construction unit that constructs a learning model based on a correspondence relationship between an operation of the on-board device that is a learning target included in the operation record data and the element data in the past when the operation of the learning target was executed;
A learning processing device comprising: a control unit that generates a control command for controlling the on-board device by applying the element data for the current day to the learning model. The data acquisition unit causes the operation result data and the element data to be associated with each other by time and stored in the data storage unit;
The learning processing device according to claim 1 , wherein the model construction unit constructs the learning model based on a correspondence relationship between the operation of the learning target and the element data associated with the operation of the learning target by time. The learning processing device according to claim 1 or 2, wherein the data acquisition unit thins out the acquired operation history data and element data and accumulates them in the data accumulation unit. The data acquisition unit acquires operation status information of the vehicle during operation and accumulates the information in the data accumulation unit;
The learning processing device according to claim 1 , further comprising a data reading unit that reads out, from the data storage unit, the operation history data and the element data used for constructing the learning model, based on the operation status information. The operation status information includes information regarding at least one of one-man operation and non-commercial running,
5. The learning processing device according to claim 4, wherein the data reading unit reads out the operation record data and the element data used to construct the learning model from the data storage unit, excluding at least one of one-man operation and non-commercial driving. The learning processing device according to any one of claims 1 to 5, further comprising a data reading unit that reads out the operation record data and the element data used to construct the learning model from the data storage unit based on at least one of the values of the element data and a predetermined operation of the vehicle. The learning processing device according to any one of claims 1 to 6, further comprising a data processing unit that extracts the operation record data and the element data used to construct the learning model so that the model construction unit constructs the learning model taking into account at least one of trend fluctuation and seasonal fluctuation. The model construction unit includes:
Constructing a plurality of said learning models using a plurality of machine learning techniques;
Evaluating the constructed learning models;
The learning processing device according to claim 1 , wherein the learning model with the highest evaluation is set as the learning model to be applied by the control unit. The learning processing device according to any one of claims 1 to 8, wherein the control unit outputs the control command to the on-board device at each first predetermined period that is set in advance. The learning processing device according to any one of claims 1 to 9, wherein the control unit stops outputting the control command to the on-board device while the door of the vehicle is open and for a second predetermined period after the door is closed. The learning processing device is provided at least for each formation including a plurality of the vehicles,
The data acquisition unit acquires the operation record data and the element data from a vehicle information control device that accepts operations related to the operation of the vehicle from the driver,
The learning processing device according to claim 1 , wherein the control unit outputs the control command to the on-board device via the vehicle information control device. the learning processing device is provided for each of the on-board devices or for each of the vehicles;
the data acquisition unit acquires the operation record data and the element data from the on-board device without going through a vehicle information control device that accepts operations related to the operation of the vehicle from the driver;
The learning processing device according to claim 1 , wherein the control unit outputs the control command to the on-board device without using the vehicle information control device. The learning processing device according to any one of claims 1 to 12, wherein the on-board device is an air conditioning device. data on operation history of an on-board device by a driver mounted on the vehicle and element data on the vehicle during operation used to determine the operation of the on-board device are acquired and stored in a data storage unit;
constructing a learning model based on a correspondence relationship between an operation of the on-board device to be learned, which is included in the operation record data, and the element data in the past when the operation of the learning target was executed;
generating a control command for controlling the on-board device by applying the element data for the current day to the learning model.

Images