Program, information processing information, and information processing device.

Abstract

To properly estimate the state of devices that receive energy from a power source. [Solution] The program according to this embodiment causes a computer to input the first thermal information via an input layer into a thermal reservoir layer, which has multiple virtual nodes that hold a time-series second thermal information observable at each of multiple locations, based on the thermal characteristics of a device including a power source and a device. The program inputs the first thermal information via an input layer, and based on the input first thermal information, outputs an output value via an output layer, which is obtained by linearly combining the time-series second thermal information held by the multiple virtual nodes of the thermal reservoir layer using coupling coefficients between the multiple virtual nodes, and evaluates the state of the device or the state of the device based on the output value.

Description

Technical Field

[0001] The present invention relates to a program, information processing information, and an information processing apparatus.

Background Art

[0002] Conventionally, there has been a device and method for estimating SOH (State Of Health), which is a parameter indicating the capacity degradation of a battery. More specifically, a device and method for estimating the SOH of a battery using SOC (State Of Charge), which is a parameter indicating the remaining capacity of the battery, are known (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Here, with the spread of various IoT (Internet of Things) devices, the applications of batteries are increasing. However, not only is there room for improvement in the method of estimating the remaining capacity of the battery, but there is also room for improvement in the method of evaluating the usage state of the battery.

[0005] The present invention has been made in view of such circumstances, and by using the non-linear output with respect to the temperature of an energy supply source attached to an IoT device or the like, edge computing is performed based on the temperature change of the supply source itself or the heat information input to the supply source, and it is an object of the present invention to provide a program, an information processing method, and an information processing apparatus capable of appropriately estimating the state of a device supplied with energy from the supply source.

Means for Solving the Problems

[0006] The program according to this embodiment causes a computer to perform a process that evaluates the state of the device or the state of the device based on the following characteristics: the heat generated from the device is applied to the supply source, and the voltage of the supply source changes in response to the heat. The program then inputs the time-series first thermal information via an input layer to a thermal reservoir layer which has a plurality of virtual nodes that hold time-series second thermal information observable at each of the plurality of locations, as time-series first thermal information based on the voltage of the supply source changes in response to the heat propagates to a plurality of locations of the device. Based on the input time-series first thermal information, the program outputs an output value via an output layer, which is obtained by linearly combining the time-series second thermal information held by the plurality of virtual nodes of the thermal reservoir layer using coupling coefficients between the plurality of virtual nodes. The program evaluates the state of the device or the state of the device based on the output value. [Effects of the Invention]

[0007] According to the present invention, it becomes possible to appropriately estimate the state of a device that receives energy from a power source. [Brief explanation of the drawing]

[0008] [Figure 1A] This figure shows an example of using TRC in this embodiment. [Figure 1B] This figure shows an example of a battery in which multiple cells are connected together and the change in the temperature characteristics of each cell in this embodiment. [Figure 2] This figure shows an example of the configuration of the information processing device according to this embodiment. [Figure 3] This figure shows an example of the configuration of the thermal reservoir model section. [Figure 4] This figure shows an example of training data used to train the thermal reservoir model. [Figure 5A] This figure shows an example of the learning process for the thermal reservoir model. [Figure 5B] This figure shows an example of the learning process for the thermal reservoir model. [Figure 6A]This figure shows an example of evaluating an evaluation target based on the output of the output layer. [Figure 6B] This figure shows an example of evaluating an evaluation target based on the output of the output layer. [Figure 7] This figure shows an example of learning processing by an information processing device. [Figure 8] This figure shows an example of inference processing by an information processing device. [Modes for carrying out the invention]

[0009] The present invention will be described below based on the drawings illustrating its embodiments. First, an example of implementing Thermal Reservoir Computing (TRC) of the disclosed technology will be described using Figures 1A and 1B.

[0010] Figure 1 shows an example of using TRC in this embodiment. In the example shown in Figure 1A, the device 200 comprises a power source 201 whose supplied energy changes depending on the heat, and a device 202 that operates by receiving the supplied energy. Energy is supplied from the power source 201 to the device 202. The device 202 operates by receiving the supplied energy, but it cannot utilize all of the energy and loses some of it. At this time, the lost energy is converted into heat, and this heat is transmitted to the power source 201 as temperature information. In addition to the lost energy, it is also possible that the device 202 effectively uses the heat, and that heat is transmitted (or conducted) to the power source 201. Thus, the heat generated in the device 200 that is transmitted to the power source 201 may be the heat effectively used by the device 202, or it may be the heat of energy loss. Furthermore, it is also possible that not only the device 202, but also the power source 201 itself generates heat or undergoes temperature changes while in use. In this case, the reason why the power source 201 generates heat is due to the operating state of device 202, and it can be considered that the power source 201 itself also causes thermal fluctuations. Therefore, since the temperature of the power source 201 itself can change, the power source 201 itself can also function as a reservoir element. The power source 201 is affected by this heat, causing a change in voltage and thus a change in the energy it supplies. Therefore, by monitoring the voltage of the power source 201, it is possible to estimate the heat applied to the power source 201, and by performing TRC analysis on the device 200 using this voltage change, it becomes possible to estimate the state of device 202, which receives energy from the power source 201.

[0011] In other words, by performing TRC based on the thermal characteristics of the apparatus 200, which includes the power source 201 and the device 202, it becomes possible to estimate the state of the device 202. As a result, for example, it is possible to appropriately detect an abnormality in the device 202 and stop the energy supply from the power source 201. Furthermore, it is possible to reduce costs by not necessarily requiring a temperature sensor in the power source 201.

[0012] Figure 1B shows an example of a battery in which a plurality of cells in the present embodiment are connected and the change in the temperature characteristics of each cell. In the example shown in Figure 1B, the supply source 201 includes a battery. Also, the supply source 201 may be a primary battery, a secondary battery, or a battery cell formed by stacking these in series or parallel. Further, the supply source 201 is not limited to power supply, and can be applied to the present embodiment as long as heat is applied by a motor or various sensors, etc., and the supply changes. Figure 1B shows the relationship between the temperature T of each cell and time t.

[0013] For example, by continuously applying heat to the battery, the state of the device 202 can be estimated based on the temperature changes at multiple points of the device 200, and it becomes possible to predict a failure or the like of the device 202 based on this estimation.

[0014] Figure 2 is a diagram showing an example of the configuration of the information processing apparatus 50 of the present embodiment. The information processing apparatus 50 includes a control unit 51 that controls the entire apparatus, an input / output unit 52, a memory 53, a learning processing unit 54, an evaluation unit 55, a storage unit 56, and a heat reservoir model unit 60. The information processing apparatus 50 may be configured by a plurality of devices with the processing functions distributed.

[0015] The storage unit 56 can be configured by, for example, a hard disk or a semiconductor memory, etc., and stores a computer program 57 (program product) and required information. The computer program 57 may be downloaded from an external device and stored in the storage unit 56. Also, the computer program 57 recorded on a recording medium (for example, an optical readable disc storage medium such as a CD-ROM) may be read by a recording medium reading unit and stored in the storage unit 56.

[0016] The control unit 51 is configured by incorporating the required number of CPUs (Central Processing Units), MPUs (Micro-Processing Units), GPUs (Graphics Processing Units), etc. The control unit 51 can execute the processing defined by the computer program 57. That is, the processing by the control unit 51 is also the processing by the computer program 57. By executing the computer program 57, the control unit 51 can execute the functions of the learning processing unit 54 and the evaluation unit 55.

[0017] The input / output unit 52 acquires the data necessary for the information processing apparatus 50 to process, and outputs the data obtained as a result of the processing by the information processing apparatus 50. Further, the input / output unit 52 may include a user interface such as a keyboard, a mouse, a display panel, a touch panel, etc. The input / output unit 52 can acquire the first heat-related information in time series input to the device 200 as an object (including the energy supply source 201 and the device 202).

[0018] The memory 53 can be composed of semiconductor memories such as SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), flash memory, etc. The computer program 57 is expanded in the memory 53 so that the control unit 51 can execute the computer program 57.

[0019] The learning processing unit 54 performs the learning processing of the heat reservoir model unit 60. The details of the learning processing will be described later.

[0020] The thermal reservoir model unit 60 is a reservoir computing model and comprises an input layer 61, a thermal reservoir layer 62, a coupling layer 63, and an output layer 64. The thermal reservoir model unit 60 can be implemented using hardware such as an FPGA (Field Programmable Gate Array). The input layer 61 receives time-series first thermal-related information (collectively referred to as "input data") input to the device 200, acquired via the input / output unit 52 under the control of the control unit 51. The location in the device 200 to which the first thermal-related information is input (also referred to as "input point") is not particularly limited, but may be a part of the supply source 201. The first thermal-related information is nonlinearly transformed by the thermal reservoir layer 62 and mapped to a high-dimensional feature space. The mapping to the feature space is represented by the states of multiple virtual nodes in the thermal reservoir layer 62. The coupling layer 63 linearly combines the state values ​​of the multiple virtual nodes in the thermal reservoir layer 62 using coupling coefficients (weight coefficients). The output layer 64 outputs linearly combined output values. The output values ​​output by the thermal reservoir model unit 60 can be used to evaluate the temperature-related state of the device 200. Details of the thermal reservoir model unit 60 will be described later.

[0021] The evaluation unit 55 evaluates the state of the apparatus 200 or device 202 based on the output value output by the thermal reservoir model unit 60. Details of the evaluation unit 55 will be described later.

[0022] The information processing device 50 includes thermal reservoir computing, which realizes reservoir computing by utilizing the device 200 and materials as thermal reservoirs. In this specification, a thermal reservoir refers to a solid, liquid, or gas that can conduct heat. A thermal reservoir will be described below using a solid as an example. The information processing device 50 and the device 200 may be configured as a single unit.

[0023] The apparatus 200 according to this embodiment is configured, for example, as a power source 201 whose supplied energy changes with heat, and a device 202 that operates upon receiving the supplied energy. In this configuration, the apparatus 200 functions as a thermal reservoir, and a nonlinear mapping between parameters related to heat applied to the apparatus 200 (e.g., the temperature of the power source 201) and the temperature response is realized by reservoir computing. Here, a thermal reservoir refers to the entire object to which heat can propagate, and has the function of accumulating and propagating time-series heat-related information input based on the physical characteristics of the object, such as its thermal conductivity, thermal diffusivity, and shape. That is, the apparatus 200 maintains the time-series change in the heat distribution generated inside the apparatus as its internal state. At this time, temperature information observed at multiple locations inside the apparatus or physical quantities that change depending on temperature changes function as a group of virtual nodes, forming the internal state space of the reservoir computing.

[0024] As described above, the entire apparatus 200, including the power source 201 and device 202, is treated as a thermal reservoir. These components each have different material properties, and the temperature distribution generated in the apparatus 200 due to energy loss, etc., responds nonlinearly. This temperature response functions as the internal state of the reservoir in thermal reservoir computing.

[0025] The device 200 includes at least one temperature sensor. The temperature sensor detects the temperature at a predetermined location in the device 200, and the detected value is associated as a virtual node value. That is, each temperature sensor corresponds to a virtual node and functions as an interface for externally observing the internal temperature state of the device 200. Preferably, temperature sensors are provided at multiple locations in the device 200 so that temperatures at multiple locations can be detected. Note that the power source 201 does not necessarily have to be equipped with a temperature sensor. By detecting the voltage of the power source 201 with a voltage detection unit, changes in voltage can be considered as temperature changes based on heat applied to the power source 201.

[0026] In this embodiment, the evaluation unit 55 evaluates the temperature-related state inside the device 200 based on the output value obtained by thermal reservoir computing. The evaluation unit 55 analyzes the temperature distribution and time-series changes of the thermal reservoir (i.e., the entire device 200) and can detect, for example, a steepening of the temperature gradient, a localized temperature rise, or thermal non-uniformity that affects the quality of the device 200 or device 202. When the evaluation target exceeds a predetermined threshold, it is determined to be abnormal, and the state of the device 200 or device 202 is quantitatively evaluated based on the error between the output value and the correct value.

[0027] The device 200 can operate the device 202 in an optimized manner for each environment, even when the environmental conditions (such as ambient temperature and humidity) in which it is installed differ. In this embodiment, multiple models may be generated for each environmental information in order to cope with variations in thermal response due to differences in environmental conditions. That is, multiple thermal reservoir computing models may be switched according to environmental information such as region and climate. As a result, the device is less susceptible to the effects of environmental changes and can predict the temperature distribution in the near future, thereby proactively predicting malfunctions that may occur in the device 200 or device 202.

[0028] Figure 3 shows an example of the configuration of the thermal reservoir model section 60. When the input layer 61 receives the first thermal information in a time series, it inputs the first thermal information in a time series to the thermal reservoir layer 62.

[0029] The thermal reservoir layer 62 comprises multiple virtual nodes. The thermal reservoir layer 62 is realized as a physical phenomenon of heat propagation in a thermal reservoir. That is, the thermal reservoir layer 62 comprises multiple virtual nodes that hold time-series second thermal information (for example, including temperature data or physical quantities that change depending on temperature changes) that can be observed at each of multiple locations (observation points) of the thermal reservoir, based on the thermal conductivity or thermal diffusivity (thermal-related properties) of the thermal reservoir that can propagate heat, or the shape (properties) of the thermal reservoir. The construction and design of the thermal reservoir layer 62 can be performed, for example, by determining physical parameters (such as the dimensions, shape, thermal conductivity (or thermal diffusivity), and material of the thermal reservoir) according to the learning target and learning task, based on simulation experiments.

[0030] The coupling layer 63 linearly combines the time-series second thermal-related information held by multiple virtual nodes of the thermal reservoir layer 62 using coupling coefficients between the multiple virtual nodes. The coupling layer 63 and the output layer 64 are collectively referred to as the readout layer.

[0031] The output layer 64 outputs the output values ​​that are linearly combined in the coupling layer 63. The number of dimensions of the output layer 64 can be set arbitrarily.

[0032] Next, we will explain the thermal reservoir computing method, that is, the learning method for the thermal reservoir model unit 60.

[0033] Figure 4 shows an example of training data used for learning the thermal reservoir model unit 60. First thermal-related information in time series for various states corresponding to the device 200 shown in Figure 4 is input. In the example shown in Figure 4, for convenience, the states are denoted as 1, 2, ..., m. For example, state 1 corresponds to time point 1 (time t1=t0), state 2 corresponds to time point 2 (time t2=t0+dt), ... state m corresponds to time point m (time t mThis corresponds to (=t0+m·dt). The first thermal information may include, for example, environmental information. The first thermal information may also include temperature data or thermal data. The training input data collected corresponding to states 1, 2, ..., m of the device 200 are represented by vectors G1, G2, and Gm. Also, when the state of the device 200 is state 1, 2, ..., m, the correct values ​​(true values) of the output values ​​representing the state of the device 200 are represented by vectors D1, D2, ..., Dm. Vectors D1, D2, ..., Dm can serve as training data.

[0034] Furthermore, when the learning input data G1 is input to the thermal reservoir layer 62, the values ​​indicating the state of multiple virtual nodes in the thermal reservoir layer 62 are represented by vector S1. Vector S1 corresponds to the time-series second thermal-related information held by the multiple virtual nodes, and includes, for example, temperature data or physical quantities that change depending on temperature changes related to the apparatus 200 or device 202. When vector S1 is input to the coupling layer 63, the vector of output values ​​output by the coupling layer 63 is represented by Y1. Also, when the learning input data G2 is input to the thermal reservoir layer 62, the values ​​indicating the state of multiple virtual nodes in the thermal reservoir layer 62 are represented by vector S2, and when vector S2 is input to the coupling layer 63, the vector of output values ​​output by the coupling layer 63 is represented by Y2. Similarly, when the training input data Gm is input to the thermal reservoir layer 62, the values ​​representing the states of multiple virtual nodes in the thermal reservoir layer 62 are represented by vector Sm, and when vector Sm is input to the coupled layer 63, the vector of output values ​​output by the coupled layer 63 is represented by Ym.

[0035] In this embodiment, the correct value is defined based on the temperature (or voltage) at each location when the device 202 is operating stably. For example, when the device 200 is operating properly, if the temperature information observed at multiple locations corresponding to the internal structure does not change during the next operation, then the temperature distribution at that time corresponds to the correct value.

[0036] More specifically, consider the case where device 202 is operating correctly at time t1, and is operating at the next time t2. In this case, if temperatures T1 and T2 at each observation point are equal, that is, if the temperature change dT = T2 - T1 = 0, then it is determined that there is no temperature disturbance. In such a state, it is possible to determine that the operation of device 202 is stable, and this temperature distribution represents the ideal correct state.

[0037] The actual temperature information consists of multiple temperature values ​​observed by multiple temperature sensors installed inside or on the surface of the device 200, and these are treated as vector quantities. Therefore, the correct value can be expressed as any linear combination of temperature vectors at each observation point. In other words, not only when the temperature is perfectly constant at all observation points, but also when a linear combination of multiple temperature distributions is used, if the device 202 is operating normally, the temperature distribution is also treated as the correct value. Thus, the correct value in this embodiment represents a thermal equilibrium state corresponding to the stability of the device 202's operation, and is used as a criterion for minimizing the error with the output value vector during the learning phase.

[0038] Figure 5 shows an example of the learning process of the thermal reservoir model unit 60. Conventional recurrent neural networks are complex, nonlinear, and high-dimensional models, requiring large amounts of data and processing time for learning. In thermal reservoir computing (TRC), by implementing the thermal reservoir layer 62 with the entire device, i.e., a thermal reservoir capable of heat propagation (see Figure 3), it becomes possible to realize a high-dimensional, nonlinear model of the nonlinear temperature response caused by the heat input to (applied to) the device, independently of the neural network. In the thermal reservoir layer 62, the state of multiple virtual nodes is randomly set according to the properties of the thermal reservoir, including its thermal-related properties (thermal conductivity, thermal diffusivity, etc.), so there is no need to adjust the parameters within the thermal reservoir layer 62 through learning. That is, the output of the output layer 64 is given by the coupling layer 63 as a linear combination of the states of the thermal reservoir layer 62 (time-series temperature changes at observation points inside the device). The coupling coefficients (weight coefficients) of the coupling layer 63 can be trained using a simple algorithm such as linear regression so that the output of the output layer 64 is the same as or close to the correct value.

[0039] As shown in Figure 5A, the virtual nodes of the thermal reservoir layer 62 are denoted as s1, s2, s3, ..., sj. Here, q may be the number of observation points, N may be the number of discrete values, and j = q·N. The virtual node s corresponds to the time-series second thermal information (temperature data, etc.) observed at multiple locations (observation points) of the device 200. The output values ​​(output nodes) of the output layer 64 are denoted as y1, y2, ..., yr. The coupling coefficients from virtual node s1 to output values ​​y1, y2, ..., yr are denoted as w11, w12, ..., w1r. Similarly, the coupling coefficients from virtual node sj to output values ​​y1, y2, ..., yr are denoted as wj1, wj2, ..., wjr.

[0040] As shown in Figure 5B, if we represent the output value vector as Y, the virtual node value vector corresponding to the temperature sensed at each point of the display or projector as S, and the coupling coefficient matrix as W, then in the equation Y = S·W, we can calculate the coupling coefficients of the coupling coefficient matrix W so that the output value vector Y approaches the correct value vector D.

[0041] As described above, the learning processing unit 54 functions as a decision unit and acquires training data including time-series first thermal information (thermal data or temperature data: including cases where voltage data is considered thermal data / temperature data) and the correct value of the output, which are brought about by the heat applied to the device 200. Based on the acquired training data, when the time-series first thermal information is input to the thermal reservoir layer 62, the thermal reservoir model unit 60 is trained by determining the coupling coefficients so that the output value, which is output by linearly combining the time-series second thermal information held by multiple virtual nodes of the thermal reservoir layer 62 using coupling coefficients between multiple virtual nodes, approaches the correct value. In this case, the correct value is expressed as a linear combination of temperature vectors at multiple observation points and represents a thermal equilibrium state in which no malfunction occurs in the operation of the device 202.

[0042] As described above, since learning can be performed using a simple configuration such as the coupled layer 63 and a simple algorithm called linear regression, a large amount of data is not required, the processing time is relatively short, and it is possible to operate with low power consumption, as well as real-time learning. As a result, the information processing device 50 of this embodiment can be applied to edge computing such as IoT devices that predict temperature changes.

[0043] The output value y and the output value vector Y are not physical quantities with units, but rather numerical values ​​resulting from calculations. It is necessary to determine how these calculation results relate to the state of the thermal reservoir, particularly the temperature-related state, i.e., the thermal non-uniformity related to the stability of the device 200.

[0044] Figure 6 shows an example of evaluating an object to be evaluated based on the output of the output layer 64. The evaluation unit 55 evaluates the state related to the device 202 included in the apparatus 200, which is the object to be evaluated, based on the output value (including a prediction of the temperature distribution in the near future) output by the output layer 64. This state of the object to be evaluated includes steepening of the temperature gradient, localized temperature rise, or the degree of thermal non-uniformity affecting the device 202. Figure 6A shows the case where the output layer 64 outputs multiple output values. Multiple evaluations 1, 2, ..., r are set up corresponding to each output value. If the value of the output value is above a predetermined threshold, the evaluation corresponding to that output value can be determined as the evaluation result of the apparatus 200 (or device 202) as the object to be evaluated. The state that evaluations 1, 2, ..., r represent can be determined as appropriate. Evaluations 1, 2, ..., r may each be evaluations of different states (for example, the temperature of device 202, the quality of products manufactured by device 202, etc.), or they may be temporal changes in a specific state (for example, a prediction result of the degree of degradation of device 202, etc.).

[0045] Figure 6B shows the case where the output layer 64 outputs a single output value y. For example, if the output value y is above a threshold, it can be determined that the device 202 of the apparatus 200 is normal (in thermal equilibrium), and if the output value y is below the threshold, it can be determined that the object being evaluated is abnormal (there is a possibility of temperature disturbance). Based on this evaluation result, the evaluation unit 55 can reduce the probability of device 202 malfunction by stopping the energy supply to device 202 in advance according to the predicted change in temperature distribution. Note that the evaluation method is not limited to the example in Figure 6. Now, an embodiment of this design will be described.

[0046] <Examples> Based on the characteristics of the apparatus 200, which includes a power source 201 that changes the amount of energy supplied depending on the heat and a device 202 that operates by receiving the energy, the control unit 51 inputs the time-series first thermal information via the input layer 61 to a thermal reservoir layer 62 which has multiple virtual nodes that hold time-series second thermal information observable at each of the multiple locations, by propagating time-series first thermal information based on the voltage of the power source 201 that changes depending on the heat generated from the apparatus 200 (including the power source 201 and / or device 202) to the power source 201.

[0047] For example, the control unit 51 may use the voltage detected by the voltage detection unit, which detects the voltage when energy is supplied from the power source 201, to obtain time-series data of the said voltage as first thermal-related information, input it to the input layer 61, and output it to the thermal reservoir layer 62. Alternatively, the control unit 51 may use thermal data or time-series data of temperature data sensed by a temperature sensor provided in the power source 201 as second thermal-related information, input it to the input layer 61, and output it to the thermal reservoir layer 62.

[0048] Based on the input time-series first thermal information, the control unit 51 linearly combines the time-series second thermal information held by multiple virtual nodes of the thermal reservoir layer 62 in the coupling layer 63 using coupling coefficients between the multiple virtual nodes, and outputs an output value via the output layer 64.

[0049] The evaluation unit 55 evaluates the state of the apparatus 200 or the state of the device 202 based on the output value output from the output layer 64. For example, by pre-training data for normal and abnormal output values ​​using a learning model, it is possible to set thresholds for normal or abnormal.

[0050] The above process makes it possible to appropriately estimate the state of the apparatus 200 or device 202. As a result, malfunctions in the operation of device 202 can be reduced, and failures of device 202 can be detected in advance through simulation or other means.

[0051] The learning processing unit 54, based on training data including a time-series first thermal information based on a voltage that changes in response to the heat applied to the power source 201 and the correct value of the output, determines the coupling coefficients so that when the time-series first thermal information is input to the thermal reservoir layer, the output value, which is produced by linearly combining the time-series second thermal information held by multiple virtual nodes using coupling coefficients between multiple virtual nodes, approaches the correct value.

[0052] For example, the learning processing unit 54 uses first thermal information from when the device 202 begins to malfunction and first thermal information from when it is operating stably as training data. Furthermore, by performing thermal reservoir computing using this training data, the learning processing unit 54 can appropriately set the coupling coefficients of the coupling layer 63.

[0053] Through the above process, a thermal reservoir model can be trained using training data, and the output values ​​from this thermal reservoir model can be used to more accurately estimate the state of the apparatus 200 or device 202.

[0054] The power source 201 may include a battery, and the device 202 may operate by receiving power from the battery. Furthermore, as described above, the battery may be a primary battery, a secondary battery, or a battery cell consisting of these stacked in series and parallel. Moreover, the power source 201 is not limited to power supply; any device that changes the supplied material (energy, power, motive force, information, processing capacity, etc.) when heat is applied, such as a motor or various sensors, can be applied to this embodiment.

[0055] Device 202 can be any device that operates by receiving energy from power source 201. For example, device 202 may be computer-related devices (laptops, desktop PCs, tablet devices, smartphones, smartwatches, e-readers, game consoles, smart glasses, VR headsets, external hard drives, USB memory sticks, SSDs / HDDs), input devices (keyboards, mice, touchpads, stylus pens, joysticks, game controllers, microphones, scanners, cameras / webcams), output devices (monitors, printers, speakers, headphones / earphones, projectors, VR displays), communication / network devices (routers, modems, Wi-Fi access points, switching hubs, Bluetooth® devices, IoT game consoles). Examples of applicable products include, but are not limited to, sensors and IoT devices (temperature sensors, acceleration sensors, proximity sensors, GPS modules, smart home devices, smart lights, smart locks, smart speakers, smart thermostats), home appliances and lifestyle devices (refrigerators, washing machines, microwave ovens, robotic vacuums, air conditioners, televisions, induction cooktops, electric toothbrushes), automotive and mobility-related devices (vehicles, car navigation systems, dashcams, electric vehicle control devices, sensors related to autonomous driving), medical and wearable devices (heart rate monitors, blood pressure monitors, thermometers, smart scales, electroencephalographs, blood glucose monitors), or industrial and special-purpose devices (robot arms with sensors, 3D printers, drones, rangefinders, vibration meters).

[0056] The evaluation unit 55 may also include evaluating whether the device 202 is functioning normally. If the evaluation unit 55 compares the output value with a threshold and determines that the device 202 is abnormal, the control unit 51 may stop supplying energy from the power source 201. This makes it possible to prevent a malfunction from occurring by stopping the energy supply in advance if a malfunction is predicted to occur in the device 202 while evaluating its condition.

[0057] As described above, the control unit 51 may input information regarding the voltage that changes in response to the heat applied to the power source 201 as first thermal-related information to the input layer 61, instead of the data measured by the temperature sensor. This makes it possible to reduce the number of temperature sensors.

[0058] <Learning Process> Figure 7 shows an example of a learning process performed by the information processing device 50. For convenience, the main unit of the process will be described below as the control unit 51. The control unit 51, which detects that heat is applied to the device 200 from, for example, the supply source 201 and / or device 202, acquires training data including input data (time-series first thermal-related information) and correct output values ​​of the thermal reservoir resulting from the heat applied to the device 200 (S11). The time-series first thermal-related information may include, for example, thermal data and temperature data. The control unit 51 (or learning processing unit 54) then inputs input data (thermal data for learning the near-future temperature response) to the thermal reservoir layer 62 via the input layer 61 based on the acquired training data (S12).

[0059] The control unit 51 (or learning processing unit 54) adjusts the coupling coefficient (weight coefficient) of the coupling layer 63 so that the output value output by the output layer 64 approaches the correct value (S13). Here, the correct value is the value corresponding to the temperature at which the device 202 operates stably, for example, the thermal equilibrium state. In thermal reservoir computing, learning is performed using simple algorithms such as linear regression, resulting in low learning costs and the ability to perform learning in real time. The control unit 51 determines whether the difference between the output value and the correct value (the error in the predicted temperature disturbance) is within an acceptable range (S14), and if it is not within an acceptable range (NO in S14), it continues the processing from step S12 onward.

[0060] If the difference between the output value and the correct value is within an acceptable range (YES in S14), the control unit 51 (or learning processing unit 54) stores the adjusted coupling coefficients (model weights for predicting near-future temperatures) in the storage unit 56 (S15) and terminates the process. These coupling coefficients are used in the subsequent inference process (Figure 8) to predict the near-future temperature distribution from the input thermal / temperature data and to determine the appropriate timing for maintenance of the device 200 or replacement of the device 202.

[0061] <Inference Processing> Figure 8 shows an example of inference processing by the information processing device 50. The control unit 51 acquires first thermal-related information in a time series based on a second time point that is different from the first time point (S21). The first thermal-related information at the second time point is, for example, the temperature applied to the supply source 201 and the voltage affecting the energy supply. The control unit 51 then inputs the first thermal-related information in a time series related to the second time point, acquired via the input layer 61, to the thermal reservoir layer 62 (S22).

[0062] The control unit 51 (or evaluation unit 55) acquires the output value (including a prediction of the temperature distribution in the near future) output by the output layer 64 when the first thermal information at the second time point is input to the device 200 (S23), and evaluates the temperature-related state of the device 200 to be evaluated (such as steepening of the temperature gradient or thermal non-uniformity) based on the acquired output value (S24).

[0063] As described above, the control unit 51 inputs the time-series first thermal information to a thermal reservoir layer, which has multiple virtual nodes that hold time-series second thermal information observable at each of multiple locations as the time-series first thermal information input to the thermal reservoir propagates to multiple locations of the thermal reservoir, based on the thermal-related characteristics of the thermal reservoir (the entire device) that is capable of heat propagation. The control unit 51 can read out an output value (predicted thermal state) which is output by linearly combining the time-series second thermal information held by the multiple virtual nodes using coupling coefficients between the multiple virtual nodes. This makes it possible to appropriately estimate the state of the device 200 or device 202.

[0064] (Note 1) On the computer, Based on the thermal characteristics of a device including a power source whose supplied energy changes in response to heat and a device that operates by receiving said energy, the heat generated from the device is applied to the power source, and a time-series first thermal information based on the voltage of the power source which changes in response to said heat is propagated to multiple locations of the device, thereby inputting the time-series first thermal information via an input layer to a thermal reservoir layer having multiple virtual nodes that hold time-series second thermal information observable at each of the multiple locations. Based on the input time-series first thermal information, the output value is output via the output layer by linearly combining the time-series second thermal information held by the multiple virtual nodes of the thermal reservoir layer using coupling coefficients between the multiple virtual nodes, and outputting the output value. Based on the output value, evaluate the state of the apparatus or the state of the device. A program that executes a process. (Note 2) Based on training data including a time-series first thermal information based on a voltage that changes in response to the heat applied to the supply source and the correct value of the output, when the time-series first thermal information is input to the thermal reservoir layer, the coupling coefficients are determined such that the output value, which is produced by linearly combining the time-series second thermal information held by the plurality of virtual nodes using coupling coefficients between the plurality of virtual nodes, approaches the correct value. The program described in Appendix 1 that causes the computer to execute the processing. (Note 3) The power source includes a battery, and the device operates by receiving power from the battery, according to the program described in Appendix 1. (Note 4) The evaluation described above includes evaluating whether the device is functioning correctly or not. If the device is evaluated as abnormal, the energy supply from the power source is stopped. A program described in any one of the appendices 1 to 3 that causes the computer to perform the processing. (Note 5) The program according to any one of Appendix 1 to 4, wherein the input includes inputting, instead of data measured by a temperature sensor, information regarding a voltage that changes in response to heat applied to the supply source as the first thermal-related information. (Note 6) Computers Based on the thermal characteristics of a device including a power source whose supplied energy changes in response to heat and a device that operates by receiving said energy, the heat generated from the device is applied to the power source, and a time-series first thermal information based on the voltage of the power source which changes in response to said heat is propagated to multiple locations of the device, thereby inputting the time-series first thermal information via an input layer to a thermal reservoir layer having multiple virtual nodes that hold time-series second thermal information observable at each of the multiple locations. Based on the input time-series first thermal information, the output value is output via the output layer by linearly combining the time-series second thermal information held by the multiple virtual nodes of the thermal reservoir layer using coupling coefficients between the multiple virtual nodes, and outputting the output value. Based on the output value, evaluate the state of the apparatus or the state of the device. An information processing method that performs a process. (Note 7) An input layer that inputs first thermal information based on a time series, which is based on the voltage of the supply source that changes in response to the heat, of an apparatus including a supply source whose supplied energy changes in response to the heat and a device that operates by receiving the said energy, to which the heat generated from the apparatus is applied to the supply source, and said heat changes in response to the heat. A thermal reservoir layer to which time-series first thermal information is input from the input layer, comprising a plurality of virtual nodes that hold time-series second thermal information observable at each of the plurality of locations as the first thermal information propagates to a plurality of locations of the apparatus based on characteristics including the thermal characteristics of the apparatus, A coupling layer that linearly combines the time-series second heat-related information held by the plurality of virtual nodes of the heat reservoir layer, based on the first heat-related information in time series input to the heat reservoir layer, using coupling coefficients between the plurality of virtual nodes, An output layer that outputs the second thermal information, which is linearly coupled in the coupling layer, as an output value, An information processing apparatus comprising: an evaluation unit that evaluates the state of the apparatus or the state of the device based on the output value.

[0065] The matters described in each embodiment can be combined with each other. Furthermore, the independent and dependent claims described in the claims can be combined with each other in any combination, regardless of the form of reference. In addition, the claims use a form in which claims referencing two or more other claims (multi-claim form), but are not limited to this. A form in which multi-claims referencing at least one multi-claim (multi-multi-claim) may also be used. [Explanation of Symbols]

[0066] 50...Information processing unit, 51...Control unit, 52...Input / output unit, 53...Memory, 54...Learning processing unit, 55...Evaluation unit, 56...Storage unit, 57...Computer program, 60...Thermal reservoir model unit, 61...Input layer, 62...Thermal reservoir layer, 63...Coupling layer, 64...Output layer, 200...Device, 201...Supply unit, 202...Device

Claims

1. On the computer, Based on the thermal characteristics of a device including a power source whose supplied energy changes in response to heat and a device that operates upon receiving said energy, heat generated from the device is applied to the power source, and a time-series first thermal information based on the voltage of the power source which changes in response to said heat is propagated to multiple locations in the device. The time-series first thermal information is input via an input layer to a thermal reservoir layer having multiple virtual nodes that hold time-series second thermal information observable at each of the multiple locations. Based on the input time-series first thermal information, the output value is output via the output layer by linearly combining the time-series second thermal information held by the multiple virtual nodes of the thermal reservoir layer using coupling coefficients between the multiple virtual nodes, and outputting the output value. Based on the output value, evaluate the state of the apparatus or the state of the device. A program that executes a process.

2. Based on training data including a time-series first thermal information based on a voltage that changes in response to the heat applied to the supply source and the correct value of the output, when the time-series first thermal information is input to the thermal reservoir layer, the coupling coefficients are determined such that the output value, which is produced by linearly combining the time-series second thermal information held by the plurality of virtual nodes using coupling coefficients between the plurality of virtual nodes, approaches the correct value. The program according to claim 1, which causes the computer to perform the processing.

3. The program according to claim 1, wherein the power source includes a battery, and the device operates by receiving power from the battery.

4. The evaluation described above includes evaluating whether the device is functioning correctly or not. If the device is evaluated as abnormal, the energy supply from the power source is stopped. A program according to any one of claims 1 to 3 that causes the computer to perform processing.

5. Computers Based on the thermal characteristics of a device including a power source whose supplied energy changes in response to heat and a device that operates upon receiving said energy, heat generated from the device is applied to the power source, and a time-series first thermal information based on the voltage of the power source which changes in response to said heat is propagated to multiple locations in the device. The time-series first thermal information is input via an input layer to a thermal reservoir layer having multiple virtual nodes that hold time-series second thermal information observable at each of the multiple locations. Based on the input time-series first thermal information, the output value is output via the output layer by linearly combining the time-series second thermal information held by the multiple virtual nodes of the thermal reservoir layer using coupling coefficients between the multiple virtual nodes, and outputting the output value. Based on the output value, evaluate the state of the apparatus or the state of the device. An information processing method that performs a process.

6. An input layer receives heat generated from a device including a power source that changes the amount of energy supplied in response to heat and a device that operates by receiving said energy, and inputs first heat-related information based on a time series of the voltage of the power source that changes in response to said heat, A thermal reservoir layer to which time-series first thermal information is input from the input layer, comprising a plurality of virtual nodes that hold time-series second thermal information observable at each of the plurality of locations as the first thermal information propagates to a plurality of locations of the apparatus based on characteristics including the thermal characteristics of the apparatus, A coupling layer that linearly combines the time-series second heat-related information held by the plurality of virtual nodes of the heat reservoir layer, based on the first heat-related information in time series input to the heat reservoir layer, using coupling coefficients between the plurality of virtual nodes, An output layer that outputs the second thermal information, which is linearly coupled in the coupling layer, as an output value, An information processing apparatus comprising: an evaluation unit that evaluates the state of the apparatus or the state of the device based on the output value.

Images