Estimation support device, computer program, estimation support method, and method for generating trained neural network system

Abstract

To provide an estimation support device, a computer program, a method for supporting estimation, and a method for generating a learned neural network system for supporting estimation of the life of a power storage element.SOLUTION: The estimation support device includes: an estimation index data acquisition unit for acquiring estimation index data of a first time point of an estimation index for estimating the life of a power storage element; a load data acquisition unit for acquiring time-series load data from the first time point to a second time point of the power storage element; and a learned neural network system for inputting the estimation index data of the first time point and the time-series load data from the first time point to the second time point and outputting the estimation index data of the second time point.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an estimation support device, a computer program, an estimation support method, and a method for generating a trained neural network system. [Background technology]

[0002] Energy storage devices are widely used in uninterruptible power supplies, DC or AC power supplies included in stabilized power supplies, etc. The use of energy storage devices is also expanding in energy storage systems that store electricity generated by renewable energy or existing power generation systems. It is known that energy storage devices deteriorate due to repeated charging and discharging.

[0003] Patent document 1 discloses a life prediction device that predicts the life of an electricity storage device under any operating condition by deriving a life prediction formula for the arbitrary operating condition based on actual measurement data under multiple operating conditions of the electricity storage device. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2014 / 103705 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in conventional life prediction formulas, it is sometimes difficult to accurately predict the life because it is not possible to fit the temporary increase in capacity and the temporary decrease in DC resistance that occur in the initial stage of use of the storage element with an appropriate approximation function. In addition, there is no established method for predicting the life of a storage element from an already deteriorated state.

[0006] An object of the present invention is to provide an estimation support device, a computer program, an estimation support method, and a method for generating a trained neural network system that support the estimation of the lifetime of a storage element. [Means for solving the problem]

[0007] An estimation support device according to one embodiment of the present invention includes an estimated index data acquisition unit that acquires estimated index data at a first time point of an estimated index that estimates a lifetime of a storage element, a load data acquisition unit that acquires time-series load data of the storage element from the first time point to a second time point, and a trained neural network system that inputs the estimated index data at the first time point and the time-series load data from the first time point to the second time point, and outputs estimated index data at the second time point. Effect of the Invention

[0008] According to the above aspect, it is possible to assist in estimating the life span of an energy storage element. [Brief description of the drawings]

[0009] [Figure 1] FIG. 2 is a block diagram showing a configuration of an estimation support device. [Diagram 2] FIG. 2 is a schematic diagram showing an example of input / output data of a trained neural network system. [Diagram 3] FIG. 13 is a schematic diagram illustrating an example of an estimated index. [Figure 4] FIG. 4 is a schematic diagram showing an example of time-series load data. [Diagram 5] FIG. 1 is a schematic diagram showing an example of the configuration of a trained neural network system. [Figure 6] FIG. 1 is a schematic diagram showing an example of a method for generating a trained neural network system. [Figure 7] FIG. 1 is a schematic diagram showing a configuration of an estimation support system. [Figure 8] 10 is a flowchart showing a procedure for a process of supporting estimation of a lifetime of an energy storage element. [Figure 9]13 is a flowchart showing the steps of a process for generating a trained neural network system. [Figure 10] FIG. 11 is an explanatory diagram showing a first evaluation of an estimated index after degradation of an energy storage element in an initial state. [Figure 11] FIG. 11 is an explanatory diagram showing a second evaluation of the estimated index after degradation of the storage element in the initial state. [Figure 12] FIG. 11 is an explanatory diagram showing a first evaluation of an estimated index after degradation of a degraded energy storage element. [Figure 13] FIG. 11 is an explanatory diagram showing a second evaluation of the estimated index after degradation of a degraded energy storage element. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] The estimation support device includes an estimated index data acquisition unit that acquires estimated index data at a first point in time of an estimated index that estimates a lifespan of a storage element, a load data acquisition unit that acquires time-series load data of the storage element from the first point in time to a second point in time, and a trained neural network system that inputs the estimated index data at the first point in time and the time-series load data from the first point in time to the second point in time, and outputs estimated index data at the second point in time.

[0011] The computer program causes a computer to execute a process of acquiring estimated indicator data at a first time point of an estimated indicator that estimates a life of a storage element, acquiring time-series load data of the storage element from the first time point to a second time point, inputting the estimated indicator data at the first time point and the time-series load data from the first time point to the second time point into a trained neural network system, and outputting the estimated indicator data at the second time point.

[0012] The estimation support method acquires estimated index data at a first time point of an estimated index that estimates a life of a storage element, acquires time-series load data of the storage element from the first time point to a second time point, inputs the estimated index data at the first time point and the time-series load data from the first time point to the second time point to a trained neural network system, and outputs the estimated index data at the second time point.

[0013] A method for generating a trained neural network system includes obtaining estimated index data at a first point in time of an estimated index that estimates a lifespan of a storage element, obtaining time-series load data from the first point in time to a second point in time of the storage element, obtaining estimated index data at the second point in time of the storage element, inputting the estimated index data at the first point in time and the time-series load data from the first point in time to the second point in time into a neural network system, and training the neural network system so that the estimated index data output by the neural network system approaches the estimated index data at the second point in time.

[0014] The estimated index data acquisition unit acquires estimated index data at a first time point of an estimated index for estimating the life of the energy storage element. The estimated index may be any index capable of estimating the life of the energy storage element, and includes a characteristic representing the state of the energy storage element. For example, the estimated index at a certain time point makes it possible to estimate the life of the energy storage element at that time point. The estimated index data represents the actual numerical value of the estimated index. The first time point can be any time point, and may be, for example, when the energy storage element starts to be used (when new), or may be a time point after it has been used for a certain period of time (after it has deteriorated).

[0015] The load data acquisition unit acquires time-series load data of the storage element from a first time point to a second time point. The time-series load data is time-series data that is a load on the storage element during use of the storage element, and may be time-series data that is a factor in degradation of the storage element. The second time point may be a time point at which the life of the storage element is estimated (life estimation time point). For example, when the first time point is the current time point and it is desired to estimate the life of the storage element six months from now, the second time point may be six months from the first time point. The time difference (period) between the second time point and the first time point may be appropriately determined according to the time point at which the life is desired to be estimated, and may be, for example, three months, six months, one year, two years, etc.

[0016] The trained neural network system can output estimated index data at a second time point when it receives estimated index data at a first time point and time-series load data from the first time point to a second time point. That is, the trained neural network system is generated by inputting estimated index data at a first time point and time-series load data from the first time point to a neural network system, and learning the estimated index data output by the neural network system to approach the estimated index data at the second time point.

[0017] With the above-mentioned configuration, for example, estimated index data at a desired life estimation time point can be obtained by inputting estimated index data at the current time point (first time point) and time-series load data from the current time point to a desired life estimation time point (second time point) into the trained neural network system. This makes it possible to estimate (predict) the life with high accuracy, and also to estimate the life of the storage element from an already deteriorated state, thereby supporting the estimation of the life of the storage element.

[0018] In the estimation support device, the trained neural network system includes a first neural network layer that receives estimated index data at the first time point and outputs a feature amount of the storage element at the first time point, a second neural network layer that receives the feature amount at the first time point and time-series load data from the first time point to a second time point output by the first neural network layer and outputs the feature amount of the storage element at the second time point, and a third neural network layer that receives the feature amount at the second time point output by the second neural network layer and outputs estimated index data at the second time point.

[0019] The trained neural network system may include a first neural network layer, a second neural network layer, and a third neural network layer. The first neural network layer, upon receiving estimated index data at a first time point, outputs a feature value of the energy storage element at the first time point. The second neural network layer, upon receiving the feature value at the first time point output by the first neural network layer and time-series load data from the first time point to the second time point, outputs a feature value of the energy storage element at the second time point. The third neural network layer, upon receiving the feature value at the second time point output by the second neural network layer, outputs estimated index data at the second time point.

[0020] The first neural network layer can be configured, for example, by a convolutional layer or an encoder including a convolutional layer, and can extract two-dimensional spatially meaningful features contained in the estimated index data based on two-dimensional data (which can be considered an “image”) representing the estimated index data.

[0021] The second neural network layer can be configured with a recurrent neural network (RNN), and more specifically, with a Long Short Term Memory (LSTM) network. The initial state of the LSTM network is set as the feature at the first time point output by the first neural network layer. Time-series load data from the first time point to the second time point is input to each LSTM of the LSTM network.

[0022] The third neural network layer can be configured, for example, by a deconvolution layer or a decoder including a deconvolution layer, and outputs two-dimensional data representing estimated index data at the second time point based on the feature at the second time point output by the second neural network layer.

[0023] With the above-described configuration, the estimated index data at a first time point is regarded as two-dimensional data (image), features that are meaningful in two-dimensional space contained in the estimated index data are extracted, the extracted features are updated using time-series load data to obtain features at a second time point, and the estimated index data is obtained from the features at the second time point, thereby making it possible to obtain the change in the estimated index that estimates the life of the storage element between the first and second times.

[0024] In the estimation support device, the estimation index includes a discharge curve that indicates a change in voltage relative to a change in a discharge-related quantity of the storage element.

[0025] The estimation index may include a discharge curve showing the transition of voltage with respect to the change in the discharge-related quantity of the storage element. The discharge-related quantity may be SOC or discharge capacity. The discharge curve is an index that can be used to estimate the life of the storage element, and for example, represents the transition of voltage with the decrease in SOC (or discharge capacity) during discharge as a curve in a two-dimensional space with the horizontal axis being SOC (or discharge capacity) and the vertical axis being voltage. This makes it possible to obtain the discharge curve at the time of life estimation (second time point).

[0026] In the estimation support device, the estimation index further includes a capacity with respect to a change in a discharge-related quantity of the storage element.

[0027] The estimation index may further include a capacity with respect to a change in a discharge-related quantity of the storage element. The capacity is an index that can be used to estimate the life of the storage element, and is expressed, for example, as a curve showing the change in capacity with a decrease in SOC (or discharge capacity) during discharge in a two-dimensional space with the horizontal axis being SOC (or discharge capacity) and the vertical axis being capacity. Note that since the capacity does not change with a decrease in SOC (or discharge capacity) during one discharge, the change in capacity is expressed as a constant straight line. This makes it possible to obtain the capacity at the time of life estimation (second time point).

[0028] In the estimation support device, the estimation index further includes a direct current resistance to a change in a discharge-related quantity of the storage element.

[0029] The estimation index may further include DC resistance with respect to a change in a discharge-related quantity of the storage element. The DC resistance is an index that can be used to estimate the life of the storage element, and is expressed, for example, as a curve showing the change in resistance with a decrease in SOC (or discharge capacity) during discharge in a two-dimensional space with the horizontal axis being SOC (or discharge capacity) and the vertical axis being resistance. Note that since the resistance does not change with a decrease in SOC (or discharge capacity) during one discharge, the change in DC resistance is expressed as a constant straight line. This makes it possible to obtain the DC resistance at the time of life estimation (second time point).

[0030] In the estimation support device, the time-series load data includes time-series voltage data and time-series temperature data of the storage element.

[0031] The time-series load data may include time-series voltage data and time-series temperature data of the energy storage element, which allows an estimated index at the time of life estimation (second time point) to be calculated, reflecting the progress of deterioration of the energy storage element according to the voltage transition and temperature transition during charging and discharging of the energy storage element during use.

[0032] Hereinafter, embodiments of an estimation support device, a computer program, an estimation support method, and a method for generating a trained neural network system will be described with reference to the drawings.

[0033] 1 is a block diagram showing the configuration of an estimation support device 50. The estimation support device 50 includes a control unit 51 that controls the entire device, a communication unit 52, a storage unit 53, an operation unit 54, a display panel 55, an estimation unit 56, and a learning processing unit 57. The estimation unit 56 includes a trained neural network system 561. The learning processing unit 57 includes a neural network system 571 in which learning has not been completed.

[0034] The control unit 51 is composed of a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The storage unit 53 is composed of a hard disk or a semiconductor memory, etc., and stores required data.

[0035] The communication unit 52 has a wired communication function or a wireless communication function, and can transmit and receive data to and from an external terminal device. Specifically, the communication unit 52 has a function as an estimated index data acquisition unit, and acquires estimated index data at a first time point of an estimated index for estimating the life of the storage element. The storage element means, for example, a storage cell or a storage module. The estimated index may be an index capable of estimating the life of the storage element, and includes a characteristic representing the state of the storage element. For example, the estimated index at a certain time point makes it possible to estimate the life of the storage element at that time point. The estimated index data represents the actual numerical value of the estimated index. The first time point can be any time point, and may be, for example, the time point when the storage element starts to be used (when new), or may be a time point after it has been used for a certain period of time (after deterioration).

[0036] The communication unit 52 also has a function as a load data acquisition unit, and acquires time-series load data from the first time point to the second time point of the storage element for which the estimated index data has been acquired. The time-series load data is time-series data that is a load on the storage element during use of the storage element, and may be time-series data that is a factor in the deterioration of the storage element. The second time point may be a time point at which the life of the storage element is estimated (life estimation time point). For example, when the first time point is the current time point and it is desired to estimate the life of the storage element six months from now, the second time point may be six months from the first time point. The time difference (period) between the second time point and the first time point may be appropriately determined according to the time point at which the life is desired to be estimated, and may be, for example, three months, six months, one year, two years, etc.

[0037] The estimated indicator data and the estimated indicator data acquired via the communication unit 52 are stored in the storage unit 53 in association with an identifier that identifies the energy storage element.

[0038] The display panel 55 is composed of a liquid crystal panel, an organic EL (Electro Luminescence) panel, or the like.

[0039] The operation unit 54 is configured with a keyboard, a pointing device, etc. The operation unit 54 may be a touch panel of the display panel 55.

[0040] The estimation unit 56 (trained neural network system 561) receives estimated index data at a first time point and time-series load data from the first time point to a second time point, and outputs estimated index data at the second time point.

[0041] With the above-mentioned configuration, for example, estimated index data at a desired life estimation time point can be obtained by inputting estimated index data at the current time point (first time point) and time-series load data from the current time point to a desired life estimation time point (second time point) into the trained neural network system. This makes it possible to estimate (predict) the life with high accuracy, and also to estimate the life of the storage element from an already deteriorated state, thereby supporting the estimation of the life of the storage element.

[0042] The estimation unit 56 and the learning processing unit 57 can be configured by combining hardware such as a CPU (e.g., a multiprocessor having multiple processor cores), GPUs (Graphics Processing Units), DSPs (Digital Signal Processors), FPGAs (Field-Programmable Gate Arrays), etc. Instead of the trained neural network system 561, other machine learning models may be used.

[0043] The following describes details of the trained neural network system 561. The learning processing unit 57 will be described in detail later.

[0044] FIG. 2 is a schematic diagram showing an example of input / output data of the trained neural network system 561. As shown in FIG. 2, the trained neural network system 561 receives, as input data, estimated index data at a first time point (for example, the present time point) and time-series load data from the first time point to a second time point. The estimated index data may include data on the discharge curve, capacity, DC resistance, and dQ / dV curve of the storage element. The dQ / dV curve is a plot of a value obtained by differentiating the remaining capacity Q at the time of discharge by the voltage V. The time-series load data may include voltage data and temperature data. The trained neural network system 561 outputs, as output data, estimated index data at the second time point. The estimated index data at the second time point may include data on the discharge curve, capacity, and DC resistance of the storage element. The estimated index data does not need to include the dQ / dV curve.

[0045] FIG. 3 is a schematic diagram showing an example of an estimation index. In FIG. 3, the horizontal axis indicates SOC (state of charge, remaining capacity), and the vertical axis indicates voltage, capacity, resistance, and dQ / dV. A discharge curve is a curve that represents the transition of voltage with respect to changes in SOC when a storage element is discharged. A discharge curve is an index that can be used to estimate the life of a storage element, and represents the transition of voltage accompanying a decrease in SOC when discharging in a two-dimensional space with SOC on the horizontal axis and voltage on the vertical axis. This makes it possible to determine the discharge curve at the time of life estimation (second time point).

[0046] Capacity is an index that can be used to estimate the lifespan of a storage element, and is expressed, for example, as a curve showing the change in capacity with the decrease in SOC during discharge in a two-dimensional space with the horizontal axis being SOC and the vertical axis being capacity. Note that since capacity does not change with the decrease in SOC during one discharge, the change in capacity is shown as a constant straight line, as shown in Figure 3. This makes it possible to determine the capacity at the time of life estimation (second time point).

[0047] DC resistance is an index that can be used to estimate the lifespan of a storage element, and is expressed, for example, as a curve showing the change in resistance with the decrease in SOC during discharge in a two-dimensional space with the horizontal axis representing SOC and the vertical axis representing resistance. Note that since the resistance does not change with the decrease in SOC during one discharge, the change in DC resistance is expressed as a constant straight line, as shown in Figure 3. This makes it possible to determine the DC resistance at the time of life estimation (second time point).

[0048] The dQ / dV curve is a plot of the value obtained by differentiating the remaining capacity Q during discharge with respect to the voltage V. The dQ / dV curve can represent in two-dimensional space the characteristics of the shoulder peak region near SOC 10%, the low SOC region below SOC 20%, the peak region from SOC 20 to 40%, the transition region between the plateau and peak region from SOC 60 to 80%, and the high SOC region from SOC 60 to 100%.

[0049] In the example of Fig. 3, the horizontal axis is SOC as a discharge-related quantity, but instead of SOC, the discharge capacity, which is the discharge amount (current x time) during discharge, may be used. Note that the discharge curve, capacity, and DC resistance shown in Fig. 3 are examples, and are represented by curves and straight lines whose transitions and values ​​differ depending on the deterioration state and type of the storage element.

[0050] FIG. 4 is a schematic diagram showing an example of time-series load data. In FIG. 4, the horizontal axis indicates time, and the vertical axis indicates voltage or temperature. In FIG. 4, time-series data of voltage (voltage data) and time-series data of temperature (temperature data) from a first time point (t0) to a second time point (tn) are shown. Note that the transitions of the voltage data and temperature data are only an example, and are not limited to the example of FIG. 4. The period from the first time point to the second time point is divided into n equal parts, and the time points are t1, t2, ..., t(k-1), tk, ..., tn. The voltage data can be expressed as {V(t1), V(t2), ..., V(tk), ..., V(tn)}, and the temperature data can be expressed as {T(t1), T(t2), ..., T(tk), ..., T(tn)}.

[0051] As described above, the time-series load data may include time-series voltage data and time-series temperature data of the energy storage element. This makes it possible to obtain an estimated index at the time of life estimation (second time point) that reflects the progress of deterioration of the energy storage element according to the voltage transition and temperature transition during charging and discharging of the energy storage element during use.

[0052] Here, V(tk) may be a sampling value of the voltage at a predetermined time between time t(k-1) and time tk, or may be the average value of the voltage between time t(k-1) and time tk. The interval between time tk and time t(k-1) may be set appropriately to 10 minutes, 30 minutes, 1 hour, 6 hours, 12 hours, 24 hours, or the like.

[0053] Similarly, T(tk) may be a sampled value of the temperature at a predetermined time between time t(k-1) and time tk, or may be the average value of the temperature between time t(k-1) and time tk. Since temperature data often shows more gradual changes than voltage data, the interval between time tk and time t(k-1) may be appropriately determined to be 1 hour, 6 hours, 24 hours, etc.

[0054] Fig. 5 is a schematic diagram showing an example of the configuration of a trained neural network system 561. As shown in Fig. 5, the trained neural network system 561 may include, for example, a first neural network layer 5611, a second neural network layer 5612, and a third neural network layer 5613.

[0055] The first neural network layer 5611, when receiving the estimated index data at the first time point, outputs a feature amount C0 of the storage element at the first time point. The second neural network layer 5612, when receiving the feature amount C0 at the first time point and the time-series load data from the first time point to the second time point output by the first neural network layer 5611, outputs a feature amount Cn of the storage element at the second time point. Here, the time-series load data from the first time point to the second time point includes voltage data V(t1), V(t2), ..., V(tn) and temperature data T(t1), T(t2), ..., T(tn), as illustrated in FIG. 4. The third neural network layer 5613, when receiving the feature amount Cn at the second time point output by the second neural network layer 5612, outputs estimated index data at the second time point.

[0056] The first neural network layer 5611 can be configured, for example, by a convolution layer or an encoder including a convolution layer, and can extract a feature quantity C0 that is meaningful in a two-dimensional space contained in the estimated index data based on two-dimensional data (which can be considered as an "image") representing the estimated index data. The input layer of the first neural network layer 5611 can receive estimated index data of a two-dimensional array of, for example, 100×4. As shown in FIG. 3, the two-dimensional data representing the estimated index data not only includes the transitions of the discharge curve, capacity, DC resistance, and dQ / dV curve in the two-dimensional space, but also includes the (relative) positional relationships between the discharge curve, capacity, DC resistance, and dQ / dV curve in the two-dimensional space.

[0057] The second neural network layer 5612 may be configured with a recurrent neural network (RNN), and more specifically, may be configured with a Long Short Term Memory (LSTM) network. In FIG. 5, the LSTM network is represented by n LSTMs. The input layer of the second neural network layer 5612 may receive, for example, time-series load data in a two-dimensional array of n×2. The initial state of the LSTM network is set as a feature C0 at a first time point output by the first neural network layer 5611. Voltage data V(t1) and temperature data T(t1), voltage data V(t2) and temperature data T(t2), ..., voltage data V(tn) and temperature data T(tn) are input to the n LSTMs of the LSTM network, respectively. The first LSTM updates the feature C0 to the feature C1. The second LSTM updates the feature C1 to the feature C2. Thereafter, the n-th LSTM outputs the feature Cn in the same manner. The number n of LSTMs can be set appropriately according to the time length of the time-series load data, and variable-length time-series load data can be handled. The number n of LSTMs may be, for example, 95.

[0058] The third neural network layer 5613 can be configured, for example, by a deconvolution layer or a decoder including a deconvolution layer, and outputs two-dimensional data representing estimated index data at the second time point based on the feature amount Cn at the second time point output by the second neural network layer 5612.

[0059] With the above-described configuration, the estimated index data at the first time point is regarded as two-dimensional data (image), a feature C0 that is meaningful in two-dimensional space contained in the estimated index data is extracted, the extracted feature is updated by the time-series load data to obtain a feature Cn at the second time point, and the estimated index data is obtained from the feature at the second time point, thereby making it possible to obtain the change in the estimated index that estimates the life of the storage element between the first time point and the second time point.

[0060] An estimated index at a third time point can be obtained by again inputting the estimated index at the second time point output by the trained neural network system 561 and the time-series load data from the second time point to the third time point into the trained neural network system 561. In this manner, the estimated index output by the trained neural network system 561 is again input into the trained neural network system 561, and the process of inputting the corresponding time-series load data is repeated, thereby making it possible to estimate the long-term life of the energy storage element.

[0061] Next, a description will be given of the learning processing unit 57. The learning processing unit 57 trains the neural network system 571 to generate a trained neural network system 561. Note that when the estimation support device 50 does not need to generate the trained neural network system 561, the learning processing unit 57 may be omitted.

[0062] FIG. 6 is a schematic diagram showing an example of a method for generating a trained neural network system 561. As shown in FIG. 6, estimated index data (each data of a discharge curve, capacity, DC resistance, and dQ / dV curve) of a storage element at a first time point and time-series load data (voltage data, temperature data) of the storage element from the first time point to the second time point are collected as learning input data. In addition, estimated index data of the storage element at a second time point is collected as teacher data. The trained neural network system 561 is generated by inputting the collected estimated index data at the first time point and time-series load data from the first time point to the second time point into a neural network system 571, and training the neural network system 571 so that the estimated index data output by the neural network system 571 approaches the estimated index data at the second time point (teacher data). Specifically, the parameters of the neural network system 571 are adjusted so that the value of the loss function between the output data of the neural network system 571 and the teacher data is minimized. When the value of the loss function falls within an acceptable range, the learning can be terminated and a trained neural network system 561 can be generated.

[0063] In the above example, a configuration was described in which the estimated indices include all of the discharge curve, capacity, DC resistance, and dQ / dV curve, but the estimated indices may be any one or any two of the discharge curve, capacity, DC resistance, and dQ / dV curve.

[0064] FIG. 7 is a schematic diagram showing a configuration of an estimation support system. As shown in FIG. 7, an estimation support device 50 is connected to external terminal devices 10, 20, and 30 via a communication network 1. The terminal devices 10, 20, and 30 can be configured with a personal computer, a tablet terminal, a smartphone, or the like. The terminal device 10 is a terminal device used by a power storage system operation business operator, the terminal device 20 is a terminal device used by a reuse business operator, and the terminal device 30 is a terminal device used by a refurbishment business operator. Note that examples of businesses are not limited to the example in FIG. 7.

[0065] The energy storage system operator uses the terminal device 10 to transmit to the estimation support device 50 the estimated index of the energy storage element (for example, new or deteriorated) at the current time (first time point) and time-series load data (predicted load state data) of the energy storage element from the current time to the life estimation time point (second time point). Here, the predicted load state data can be predicted from the operation plan of the energy storage element and past operation results. The estimation support device 50 obtains an estimated index at the life estimation time point based on the estimated index at the current time point and the time-series load data from the current time point to the life estimation time point, and transmits the obtained estimated index at the life estimation time point to the terminal device 10. This allows the energy storage system operator to estimate the life of the energy storage element at a required time point and to plan or review an operation plan for the energy storage element.

[0066] The reuse business operator uses the terminal device 20 to transmit the estimated index of the current energy storage element (degraded state) and time-series load data of the energy storage element from the current time to the estimated life time to the estimation support device 50. Here, the time-series load data may be predicted from the general usage of the used energy storage element, etc. The estimation support device 50 obtains an estimated index at the estimated life time based on the estimated index at the current time and the time-series load data from the current time to the estimated life time, and transmits the obtained estimated index at the estimated life time to the terminal device 10. This enables the reuse business operator to estimate the remaining life of the used energy storage element and appropriately determine the selling price in the secondhand market and the reuse destination of the energy storage element, etc.

[0067] The refurbishment business operator uses the terminal device 30 to transmit to the estimation support device 50 the estimated index of the current energy storage element (degraded state) and time-series load data of the energy storage element from the current time to the estimated lifespan point. Here, the time-series load data may be predicted from the general usage of the used energy storage element. The estimation support device 50 obtains the estimated index at the estimated lifespan point based on the estimated index at the current time and the time-series load data from the current time to the estimated lifespan point, and transmits the obtained estimated index at the estimated lifespan point to the terminal device 10. This enables the refurbishment business operator to stratify reusable energy storage elements and assemble used energy storage elements.

[0068] 8 is a flowchart showing the procedure of the process for supporting estimation of the lifetime of an energy storage element. For convenience, the process will be described below as being performed by control unit 51. Control unit 51 acquires estimated index data of the energy storage element at a first time point (S11), and acquires time-series load data of the energy storage element from the first time point to a second time point (S12).

[0069] The control unit 51 inputs the acquired estimated index data and time-series load data to the trained neural network system 561 (S13), and outputs the estimated index data at the second time point from the trained neural network system 561 (S14).

[0070] The control unit 51 determines whether or not to perform a long-term estimation (S15), and if an estimation is to be performed (YES in S15), the output estimated index data at the second time point is set as estimated index data at the first time point (S16), and the process continues from step S12 onwards. If a long-term estimation is not to be performed (NO in S15), the control unit 51 ends the process.

[0071] 9 is a flowchart showing the procedure of the generation process of the trained neural network system 561. The learning processing unit 57 reads out the neural network system 571 from the learning processing unit 57 (S21), and sets the parameters of the neural network system 571 to initial values ​​(S22). The parameters of the neural network include, for example, weights, biases, parameters of the filter used in convolution, the number of blocks of LSTM, and the like.

[0072] The learning processing unit 57 acquires estimated index data at a first time point as learning input data and estimated index data at a second time point as teacher data (S23), and acquires time-series load data from the first time point to the second time point as learning input data (S24). Note that each piece of acquired data is classified as training data from data previously collected for learning.

[0073] The learning processing unit 57 inputs the acquired estimated index data at the first time point and the time-series load data from the first time point to the second time point to a neural network system 571 (S25), and adjusts parameters so that the value of a loss function based on the estimated index data output by the neural network system 571 and the estimated index data at the second time point is minimized (S26).

[0074] The learning processing unit 57 determines whether the value of the loss function is within an allowable range (S27), and if it is not within the allowable range (NO in S27), repeats the processes from step S23 onwards. If the value of the loss function is within the allowable range (YES in S27), the learning processing unit 57 determines whether there is data for other first and second time points (S28).

[0075] If there is data for other first and second time points (YES in S28), the learning processing unit 57 repeats the processes from step S23 onwards. If there is no data for other first and second time points (NO in S28), the learning processing unit 57 stores the generated trained neural network system 561 in the estimation unit 56 (S29) and ends the process. In the example of FIG. 9, the loss function is evaluated for each piece of training data (online learning), but alternatively, a configuration in which multiple training data are trained together and then the value of the loss function is evaluated (mini-batch learning, batch learning) may be used. By performing (mini) batch learning, it is possible to prevent learning that is overly fitted to outlier data.

[0076] The estimation support device 50 can also be realized by using a general-purpose computer equipped with a CPU (processor), a GPU, a RAM (memory), etc. That is, a computer program that defines the procedures of each process, such as that shown in Fig. 8 and Fig. 9, is loaded into a RAM (memory) of the computer, and the computer program is executed by a CPU (processor), thereby realizing the estimation support device 50 on the computer. The computer program may be recorded on a recording medium and distributed.

[0077] Next, the estimation accuracy of the estimation support device 50 will be described.

[0078] FIG. 10 is an explanatory diagram showing a first evaluation of an estimation index after deterioration of an initial state storage element. FIG. 10A shows a capacity transition after deterioration of an initial state storage element. In FIG. 10A, the horizontal axis indicates the number of charge / discharge cycles, and the vertical axis indicates the capacity. The 0th cycle indicates that the storage element is in an unused initial state. The charge / discharge cycles were performed between SOC 25% and 100%, and the temperature was 35° C. As the number of charge / discharge cycles increases, the deterioration of the storage element progresses. As shown in FIG. 10A, the estimation result of the capacity transition by the estimation support device 50 changes in line with the transition of the actual measurement value, and a highly accurate estimation result was obtained.

[0079] FIG. 10B shows the transition of DC resistance after deterioration of the storage element in the initial state. In FIG. 10B, the horizontal axis shows the number of charge / discharge cycles, and the vertical axis shows the resistance value. The 0th cycle shows that the storage element is in an unused initial state. The charge / discharge cycles were performed between SOC 25% and 100%, and the temperature was 35°C. As the number of charge / discharge cycles increases, the deterioration of the storage element progresses. As shown in FIG. 10B, the estimation result of the estimation support device 50 for the transition of DC resistance changes in line with the transition of the actual measured value, and a highly accurate estimation result was obtained.

[0080] FIG. 11 is an explanatory diagram showing a second evaluation of the estimation index after deterioration of an initial state storage element. FIG. 11A shows a discharge curve when an initial state storage element is deteriorated by performing 1200 cycles of charge and discharge, and FIG. 11B shows a discharge curve when an initial state storage element is deteriorated by performing 2750 cycles of charge and discharge. In FIG. 11, the horizontal axis shows the discharge capacity, and the vertical axis shows the voltage. Note that the discharge capacity indicates the capacity used for discharging, and the SOC is obtained by subtracting the discharge capacity from the capacity at the start of discharging. As shown in FIG. 11A and FIG. 11B, the difference between the estimated result of the discharge curve by the estimation support device 50 and the actual measured value is small, and an accurate estimation result was obtained.

[0081] FIG. 12 is an explanatory diagram showing a first evaluation of the estimation index after degradation of a degraded storage element. FIG. 12A shows the capacity transition after degradation of a degraded storage element. In FIG. 12A, the horizontal axis shows the number of charge / discharge cycles, and the vertical axis shows the capacity. In FIG. 12, the 1500th cycle is used as the initial input value. This shows that the storage element is already in a degraded state. The charge / discharge cycle was performed between SOC 25% and 100%, and the temperature was 35° C. As the number of charge / discharge cycles increases, the degradation of the storage element progresses. As shown in FIG. 12A, the estimation result of the capacity transition by the estimation support device 50 changes along with the transition of the actual measurement value, and a highly accurate estimation result was obtained.

[0082] FIG. 12B shows the transition of DC resistance after degradation of a degraded energy storage element. In FIG. 12B, the horizontal axis shows the number of charge / discharge cycles, and the vertical axis shows the resistance value. The 1500th cycle shows that the energy storage element is in a degraded state. The charge / discharge cycles were performed between SOC 25% and 100%, and the temperature was 35°C. As the number of charge / discharge cycles increases, the degradation of the energy storage element progresses. As shown in FIG. 12B, the estimation result of the estimation support device 50 for the transition of DC resistance changes in line with the transition of the actual measured value, and a highly accurate estimation result was obtained.

[0083] FIG. 13 is an explanatory diagram showing a second evaluation of the estimation index after degradation of a degraded storage element. FIG. 13A shows a discharge curve when the storage element is degraded by charging and discharging for 2250 cycles (corresponding to 750 cycles when the 1500th cycle is the initial value of the degraded state), and FIG. 13B shows a discharge curve when the storage element is degraded by charging and discharging for 2750 cycles (corresponding to 1250 cycles when the 1500th cycle is the initial value of the degraded state). In FIG. 13, the horizontal axis shows the discharge capacity, and the vertical axis shows the voltage. The discharge capacity indicates the capacity used for discharging, and the SOC is obtained by subtracting the discharge capacity from the capacity at the start of discharging. As shown in FIG. 13A and FIG. 13B, the difference between the estimated result of the discharge curve by the estimation support device 50 and the actual measurement value is small, and a highly accurate estimation result was obtained.

[0084] The present invention is intended to be illustrative and not restrictive in all respects, and the scope of the present invention is defined by the claims, and includes all modifications within the scope and meaning equivalent to the claims. [Explanation of symbols]

[0085] 1. Communication Network 10, 20, 30 Terminal Equipment 50 Estimation support device 51 Control section 52 Communications Department 53 Storage section 54 Operation section 55 Display Panel 56 Estimation part 561 Trained Neural Network System 5611 First Neural Network Layer 5612 Second Neural Network Layer 5613 Third Neural Network Layer 57 Learning processing unit 571 Neural Network System

Claims

1. an estimated indicator data acquisition unit that acquires estimated indicator data at a first time point of an estimated indicator that estimates a lifetime of the energy storage element; a load data acquisition unit that acquires time-series load data of the energy storage element from the first time point to the second time point; a trained neural network system that receives the estimated index data at the first time point and time-series load data from the first time point to the second time point and outputs the estimated index data at the second time point; Equipped with The trained neural network system comprises: a first neural network layer that receives the estimated index data at the first time point and outputs a feature amount of the energy storage element at the first time point; a second neural network layer that receives the feature amount at a first time point output by the first neural network layer and time-series load data from the first time point to a second time point, and outputs the feature amount at the second time point of the energy storage element; a third neural network layer that receives the feature amount at the second time point output by the second neural network layer and outputs estimated indicator data at the second time point; Equipped with Estimation support device.

2. The estimated indicator is A discharge curve showing a change in voltage with respect to a change in SOC or discharge capacity of the storage element is included. The estimation support device according to claim 1 .

3. The estimated indicator is Further, the capacity with respect to the change in SOC or discharge capacity of the storage element is included. The estimation support device according to claim 1 or 2.

4. The estimated indicator is Further, the DC resistance to the change in SOC or discharge capacity of the storage element is included. The estimation support device according to any one of claims 1 to 3.

5. The time series load data is The storage element includes time-series voltage data and time-series temperature data, The estimation support device according to any one of claims 1 to 4.

6. On the computer, Obtaining estimated indicator data at a first time point for an estimated indicator for estimating a lifetime of the energy storage element; acquiring time-series load data of the energy storage element from the first time point to the second time point; The estimated index data at the first time point and the time-series load data from the first time point to the second time point, a first neural network layer that receives the estimated index data at the first time point and outputs a feature amount of the energy storage element at the first time point; a second neural network layer that receives the feature amount at a first time point output by the first neural network layer and time-series load data from the first time point to a second time point, and outputs the feature amount at the second time point of the energy storage element; a third neural network layer that receives the feature amount at the second time point output by the second neural network layer and outputs estimated indicator data at the second time point; and outputting the estimated indicator data at the second time point. A computer program that executes a process.

7. Obtaining estimated indicator data at a first time point for an estimated indicator for estimating a lifetime of the energy storage element; acquiring time-series load data of the energy storage element from the first time point to the second time point; The estimated index data at the first time point and the time-series load data from the first time point to the second time point, a first neural network layer that receives the estimated index data at the first time point and outputs a feature amount of the energy storage element at the first time point; a second neural network layer that receives the feature amount at a first time point output by the first neural network layer and time-series load data from the first time point to a second time point, and outputs the feature amount at the second time point of the energy storage element; a third neural network layer that receives the feature amount at the second time point output by the second neural network layer and outputs estimated indicator data at the second time point; and outputting the estimated indicator data at the second time point. Estimation support method.

8. Obtaining estimated indicator data at a first time point for an estimated indicator for estimating a lifetime of the energy storage element; acquiring time-series load data of the energy storage element from the first time point to the second time point; acquiring estimated indicator data of the energy storage element at the second time point; The estimated index data at the first time point and the time-series load data from the first time point to the second time point, a first neural network layer that receives the estimated index data at the first time point and outputs a feature amount of the energy storage element at the first time point; a second neural network layer that receives the feature amount at a first time point output by the first neural network layer and time-series load data from the first time point to a second time point, and outputs the feature amount at the second time point of the energy storage element; a third neural network layer that receives the feature amount at the second time point output by the second neural network layer and outputs estimated indicator data at the second time point; and training the neural network system so that the estimated indicator data output by the neural network system approaches the estimated indicator data at the second time point. How to generate a trained neural network system.

Images