Failure prediction system

Abstract

To provide a failure prediction system capable of predicting a failure of a water heater.SOLUTION: The failure prediction system for a hot water supply device is provided with: a data management unit 23 for reading a reference day drive strength and a reference day pump revolution which are normal time drive data of a hot water supply pump on a reference day, reading an intermediate day drive strength and an intermediate day pump revolution which are normal time drive data of the hot water supply pump on an intermediate day after the reference day, and reading a determination day drive strength and a determination day pump revolution which are determination time drive data of the hot water supply pump on a determination day after the intermediate day from a storage unit 22; a model management unit 50 for generating a mapping model 52 using the reference day drive strength, the intermediate day drive strength, and a difference value between the reference day pump revolution and the intermediate day pump revolution as training data; and a failure prediction information generation unit 54 for generating failure prediction information of the hot water supply device based on a reconstruction error calculated by inputting the determination day drive strength and the difference value between the intermediate day pump revolution and the determination day pump revolution to the mapping model.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a failure prediction system for a hot water heater installed in a home. [Background technology]

[0002] Patent Document 1 discloses a failure estimation system for water heaters installed in each home. This system includes a data storage unit that stores the specification information and usage environment of the water heater, a data acquisition unit that acquires the specification information of the water heater and operating information data related to the operating state from each home, a group data generation unit that generates multiple group data based on at least one of the specification information and the usage environment, a diagnosis unit that diagnoses the maintenance of the water heater to be diagnosed by comparing a reference value set in the group data corresponding to the specification information of the water heater to be diagnosed with the operating information data of the water heater to be diagnosed, and a notification unit that notifies the time for maintenance of the water heater based on the diagnosis results of the diagnosis unit. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-145133 Summary of the Invention [Problem to be solved by the invention]

[0004] In the failure estimation system disclosed in Patent Document 1, failure diagnosis is performed using thresholds based on data indicating the operating status of the water heater, such as the accumulated usage time of the water heater, the accumulated amount of hot water used, and time-series data on the rotation speed of the combustion fan and bath pump. In other words, since failure diagnosis is performed based on data indicating deterioration over time, it is difficult to diagnose failures in water heaters that have been in use for a short period of time. Furthermore, because the technical concept is failure diagnosis based on simple threshold judgment, failure signs, such as whether the water heater being diagnosed is malfunctioning or has a tendency to malfunction, are not taken into consideration.

[0005] In view of the above-mentioned circumstances, an object of the present invention is to provide a failure prediction system that can predict failures in water heaters regardless of the period of use or frequency of use. [Means for solving the problem]

[0006] One of the failure prediction systems for a water heater installed in a home according to the present invention is a failure prediction system for a water heater installed in a home, comprising: a drive data history storage unit that stores a drive strength of a water heater pump and a pump rotation speed of the water heater pump obtained over time during normal operation of the water heater as normal operation data, and stores the drive strength and the pump rotation speed obtained when a failure prediction determination for the water heater is made as determination-time drive data; and a drive data history storage unit that reads out the reference day drive strength and the reference day pump rotation speed, which are the normal operation data on a reference day, from the drive data history storage unit, and stores the average day drive strength and the average day pump rotation speed, which are the normal operation data on an average day after the reference day, as determination-time drive data. The system includes a data management unit that reads from the drive data history storage unit the determination day drive strength and determination day pump rotation speed, which are the determination time drive data for a determination date after the intermediate day, from the drive data history storage unit; a model management unit that generates a mapping model using the reference day drive strength, the intermediate day drive strength, and the difference value between the reference day pump rotation speed and the intermediate day pump rotation speed as training data; a reconstruction error calculation unit that calculates a reconstruction error by inputting the determination day drive strength and the difference value between the intermediate day pump rotation speed and the determination day pump rotation speed into the mapping model; and a failure sign information generation unit that generates failure sign information for the water heating device based on the reconstruction error.

[0007] According to this configuration, first, normal operation drive data of the water heating apparatus, including the reference day drive strength, the intermediate day drive strength, and the difference between the reference day pump rotation speed and the intermediate day pump rotation speed, are input as training data, and reproduction data that reproduces the input data based on features that express the relationship between the input data is output. A mapping model (autocorrelation model) trained to reduce a reconstruction error, which indicates the difference between the reproduction data and the input data, is generated. A failure sign of the water heating apparatus is determined using the trained mapping model. First, the determination day drive strength and the determination day pump rotation speed, which are determination-time drive data acquired to determine a failure sign of the water heating apparatus, are acquired. The determination day drive strength and the difference between the intermediate day pump rotation speed and the determination day pump rotation speed are input into the trained mapping model, and the reconstruction error is calculated using the output reproduction data. The larger the reconstruction error, the higher the possibility of failure. Therefore, the failure sign information generator can generate failure sign information of the water heating apparatus based on the reconstruction error. This failure prediction system generates the possibility of failure, i.e., failure prediction information, from the judgment operation data on the judgment date and the normal operation operation data on the intermediate date before the judgment date, so it can determine the possibility of failure for a specific usage period. For example, if the intermediate date is set to several weeks before the judgment date, it is possible to predict the possibility that a water heater that was operating normally until several weeks ago will suddenly malfunction. In particular, if the intermediate date is set to the previous judgment date, which is one day before the current judgment date, it is possible to predict the possibility that a water heater that was operating normally until the previous judgment date will suddenly malfunction.

[0008] Of course, the present invention makes it possible to predict the possibility of failure taking into account the entire period of use of the water heater, in other words, to predict the possibility of failure when aging is one of the failure factors. Such a failure prediction system for a water heater according to the present invention includes a drive data history storage unit that stores the drive strength and pump rotation speed of the water heater pump acquired over time during normal operation of the water heater as normal operation data, and stores the drive strength and pump rotation speed acquired when a failure prediction is determined for the water heater as determination operation data, and reads out the reference day drive strength and reference day pump rotation speed, which are the normal operation data on a reference date, from the drive data history storage unit, and reads out the intermediate day drive strength and intermediate day pump rotation speed, which are the normal operation data on an intermediate day after the reference date, from the drive data history storage unit. the determination date drive strength and the determination date pump rotation speed, which are the determination-time drive data for a determination date after the intermediate date, from the drive data history storage unit; a model management unit that generates a mapping model using the reference day drive strength, the intermediate day drive strength, and the ratio between the reference day pump rotation speed and the intermediate day pump rotation speed as training data; a reconstruction error calculation unit that calculates a reconstruction error by inputting the determination day drive strength and the ratio between the reference day pump rotation speed and the determination day pump rotation speed into the mapping model; and a failure sign information generation unit that generates failure sign information for the water heating device based on the reconstruction error.

[0009] This configuration also uses a mapping model generated in a manner similar to the above-described failure prediction system. In this failure prediction system, the reconstruction error is calculated using reproduction data output by inputting the drive strength on the assessment date and the ratio between the pump rotation speed on the reference date and the pump rotation speed on the assessment date. This failure prediction system generates the possibility of failure, i.e., failure prediction information, from the assessment-time drive data on the assessment date and the normal operation drive data on the reference date (e.g., the day the water heater started operating), so it can determine the possibility of failure taking into account the entire usage period. In particular, if the reference date is set to the date on which the normal operation data was first measured and the intermediate date is set to the previous assessment date immediately before the assessment date, it becomes possible to predict the possibility of water heater malfunction, including failures due to aging since the water heater was first installed.

[0010] Furthermore, a failure prediction system that combines the features of the above two failure prediction systems is also within the scope of the present invention. Such a failure prediction system for a water heating apparatus according to the present invention includes a drive data history storage unit that stores the drive strength and pump rotation speed of the hot water pump acquired over time during normal operation of the water heating apparatus as normal operation data, and stores the drive strength and pump rotation speed acquired when a failure prediction determination for the water heating apparatus is made as determination-time drive data, and reads the reference day drive strength and reference day pump rotation speed, which are the normal operation data on a reference date, from the drive data history storage unit, reads the intermediate day drive strength and intermediate day pump rotation speed, which are the normal operation data on an intermediate day after the reference date, from the drive data history storage unit, and stores the determination day drive strength and determination day pump rotation speed, which are the determination-time drive data on a determination date after the intermediate day, in the drive data history storage unit. the data management unit reads the data from the storage unit; a model management unit generates a mapping model using the reference day drive strength, the intermediate day drive strength, and the ratio and difference value between the reference day pump rotation speed and the intermediate day pump rotation speed as training data; a first reconstruction error calculation unit calculates a first reconstruction error as a reconstruction error by inputting the determination day drive strength and the difference value between the intermediate day pump rotation speed and the determination day pump rotation speed into the mapping model; a second reconstruction error calculation unit calculates a second reconstruction error as the reconstruction error by inputting the determination day drive strength and the ratio between the reference day pump rotation speed and the determination day pump rotation speed into the mapping model; and a failure sign information generation unit generates failure sign information of the water heating device based on the first reconstruction error and the second reconstruction error.

[0011] This failure prediction system also uses a mapping model generated in a manner similar to the two failure prediction systems described above. In this failure prediction system, the reconstruction error is divided into a first reconstruction error and a second reconstruction error. The first reconstruction error is calculated based on the drive strength on the determination day and the difference between the pump rotation speed on the intermediate day and the pump rotation speed on the determination day, and the second reconstruction error is calculated based on the drive strength on the determination day and the ratio between the pump rotation speed on the reference day and the pump rotation speed on the determination day. Therefore, this failure prediction system has the respective characteristics of the two failure prediction systems described above, namely, (1) It is possible to determine the possibility of a malfunction for a specific period of use. For example, if the mid-point date is set to a few weeks before the judgment date, it is possible to predict the possibility that a water heater that was operating normally until a few weeks before will suddenly malfunction. (2) It is possible to determine the possibility of failure taking into account the long period of use, and it is possible to predict malfunctions of water heaters, including failures due to long-term deterioration.

[0012] Furthermore, if the reference date is set to the date on which the normal operation data was first measured and the intermediate date is set to the previous evaluation date immediately preceding the evaluation date, it is possible to determine the possibility of a malfunction taking into account the entire period of use, and it becomes possible to predict, for example, the possibility of a malfunction of the water heater, including malfunctions due to deterioration over the entire period of use, or the possibility of a sudden malfunction of a water heater that was operating normally until recently.

[0013] In the present invention, it is preferable that the mapping model is configured by an autoencoder. An autoencoder is an algorithm for dimensionality reduction using a neural network, and can be preferably used as the mapping model used in the present invention. In addition, since autoencoder algorithms are widely used, it is possible to construct a mapping model relatively inexpensively.

[0014] The most important hot water pump used in a hot water supply system is the hot water filling pump, and its sudden failure poses a major problem. For this reason, it is preferable to target the hot water filling pump as the target of the failure prediction system of the present invention.

[0015] When manually determining a final failure symptom based on the generated failure symptom information, it is preferable to present the failure symptom information in a form that is easy for humans to understand, such as a table, graph, or illustration. For this reason, the present invention proposes providing a display control unit that generates display data for displaying the failure symptom information on a display device. Of course, even when determining a final failure symptom mechanically, it is preferable to provide a display control unit that enables the display of such failure symptom information.

[0016] To mechanically determine the final failure sign, it is necessary to determine in advance a judgment condition that determines the level of the failure sign from the reconstruction error. Setting such a judgment condition enables mechanical judgment of the failure sign of the water heating device without requiring manual intervention. Therefore, the present invention also proposes providing a failure sign judgment unit that judges the failure sign of the water heating device based on the judgment condition set for the reconstruction error. [Brief explanation of the drawings]

[0017] [Figure 1] FIG. 10 is a diagram showing the connection between the water heaters installed in each household and a management computer. [Figure 2] 1 is a functional block diagram showing the schematic configuration of a water heater failure prediction system. [Figure 3] FIG. 2 is a functional block diagram of a failure sign calculation unit. [Figure 4] 10 is a graph showing experimental results for explaining the relationship between the reconstruction error (first reconstruction error) and the detection of a fault sign. [Figure 5] 10 is a graph showing experimental results for explaining the relationship between the reconstruction error (second reconstruction error) and the detection of a fault sign. DETAILED DESCRIPTION OF THE INVENTION

[0018] The failure prediction system predicts failures in water heating apparatuses 1 installed in homes around the country. The failure prediction system is implemented in a management computer 2 installed in, for example, a gas service center. As shown in FIG. 1, a water heating apparatus 1 is installed in each detached house and apartment building. Note that FIG. 1 does not show the water heating apparatus 1 in the apartment building, but in reality, a water heating apparatus 1 is installed in each household in the apartment building. The control units of the water heating apparatuses 1 installed in each household are connected to the management computer 2 via a data communication network DN such as the Internet or a public line. This allows various information to be sent and received between the control units of the water heating apparatuses 1 and the management computer 2.

[0019] As shown in FIG. 2, the management computer 2 includes a data acquisition unit 21, a drive data history storage unit 22, a data management unit 23, a failure sign calculation unit 5, and a notification unit 24.

[0020] Each water heating device 1 sends a customer ID and drive data to the data acquisition unit 21 via the data communication network DN. The drive data includes the water heating pump drive strength and the water heating pump rotation speed. In this embodiment, a hot water filling pump is used as the hot water pump, so the water heating pump drive strength is referred to as the hot water filling pump drive strength, and the hot water pump rotation speed is referred to as the hot water filling pump rotation speed. The pump drive strength is determined by the duty cycle of the hot water filling pump. The management computer 2 can grasp the specifications, installation location, installation date and time, etc. of each water heating device 1 using the customer ID.

[0021] When each water heating device 1 sends customer IDs, operation data, etc. to management computer 2, a character string indicating other data (differences in rotation speed due to differences in piping length and pressure loss of each water heating device 1) is added and hashed, which allows management computer 2 that receives the data to include differences in rotation speed due to differences in piping length and pressure loss of the customer's water heating device 1 in the operation data. For example, management computer 2 can incorporate information extracted from the customer ID, such as differences in rotation speed due to differences in piping length and pressure loss of each home's water heating device 1, when vectorizing the operation data.

[0022] The frequency of data transmission from water heater 1 to data acquisition unit 21 may be once per hour or once per day. The operation data acquired by data acquisition unit 21 is associated by data management unit 23 with the customer ID and date and time, and stored in operation data history storage unit 22. Data management unit 23 also has the function of assigning identifiers to the stored operation data to distinguish between normal operation data and judgment operation data. Normal operation data includes the operation strength and pump rotation speed of the hot water filling pump that are considered to have been acquired over time when water heater 1 is operating normally. Judgment operation data includes the operation strength and pump rotation speed of the hot water filling pump that are acquired when water heater 1 makes a warning judgment.

[0023] In the failure sign calculation unit 5, an unsupervised learning neural network model (algorithm) configured with an autoencoder is constructed as an example of the mapping model 52. The failure sign calculation unit 5 outputs failure sign information regarding a failure sign based on a reconstruction error obtained by inputting the judgment-time drive data into the mapping model 52, which was generated (the model parameters have been determined) using the normal-time drive data as training data. In one way of using the failure sign information, the failure sign information is provided to the notification unit 24, and the display control unit 24A converts the failure sign information into display data, which is displayed on the display 25, which is an example of a display device. In another way of using the failure sign information, a final failure sign judgment is made from the failure sign information. If some kind of failure sign countermeasure is required, the relevant household is contacted or an inspector is dispatched from the gas service center.

[0024] Next, the configuration of the failure sign computation unit 5 will be described with reference to Fig. 3. In this embodiment, a model management unit 50, a failure sign model 51, a failure sign information generation unit 54, and a failure sign determination unit 55 are configured as hardware and software as functional units for fulfilling the functions of the failure sign computation unit 5. In this embodiment, for ease of explanation, the failure sign model 51 is configured by a mapping model 52 configured as an autoencoder and a reconstruction error calculation unit 53 that calculates a reconstruction error in cooperation with the mapping model 52, but these may also be configured as an integrated unit.

[0025] Model management unit 50 has learning unit 50A and practice unit 50B. Learning unit 50A provides learning commands and training data to mapping model 52, causing mapping model 52 to learn. Practice unit 50B provides practice commands and actual data (practical data), and uses trained mapping model 52 and reconstruction error calculation unit 53 to output a reconstruction error used to predict a failure in water heating apparatus 1. The training data and practice data are provided from data management unit 23.

[0026] During learning, normal operation data, such as the reference day drive strength, the mean day drive strength, and the ratio and difference between the reference day pump rotation speed and the mean day pump rotation speed, are used as training data for the mapping model 52. Because the mapping model 52 operates as an autoencoder, when normal operation data is input, unsupervised learning is performed by updating model parameters such as weights and biases so that the normal operation data is correctly reconstructed, that is, so that the reconstruction error is reduced.

[0027] In this embodiment, the reconstruction error calculation unit 53 is made up of a first reconstruction error calculation unit 53A that calculates a first reconstruction error, and a second reconstruction error calculation unit 53B that calculates a second reconstruction error.

[0028] When the first reconstruction error is calculated by first reconstruction error calculation unit 53A, the determination day drive strength (determination drive data on the day of determination) and the difference value between the intermediate day pump rotation speed (normal drive data) and the determination day pump rotation speed (determination drive data on the day of determination) are input to the mapping model. The first reconstruction error is considered to indicate a difference between the correlation characteristics of the determination day drive strength and the difference value between the intermediate day pump rotation speed and the determination day pump rotation speed, and the correlation characteristics during normal operation. Therefore, if the intermediate day is set to be several weeks before the determination day (for example, the previous determination day or the measurement day), it is possible to predict the possibility that a water heater that had been operating normally until several weeks ago will suddenly malfunction, based on the first reconstruction error.

[0029] When the second reconstruction error is calculated by second reconstruction error calculation unit 53B, the determination day drive strength (determination time drive data on the day of determination) and the ratio between the reference day pump rotation speed (normal drive data) and the determination day pump rotation speed (determination time drive data on the day of determination) are input to the mapping model. The second reconstruction error is considered to indicate a difference between the correlation characteristics of the determination day drive strength and the ratio between the reference day pump rotation speed and the determination day pump rotation speed, and the correlation characteristics during normal operation, so it is also possible to predict from the second reconstruction error the possibility of water heater malfunction, including failure due to aging deterioration, or a sudden malfunction of a water heater that was operating normally until recently.

[0030] The autoencoder is generated by using only normal operation drive data as training data (learning data), and the reconstruction error is small when drive data during normal operation is input, whereas the reconstruction error is large when drive data in an operating state in which a failure is predicted is input. Utilizing this, failure sign information generator 54 generates failure sign information for water heating apparatus 1 based on the first reconstruction error and the second reconstruction error.

[0031] As described above, this failure prediction system can detect signs of failure that could not be detected using conventional logic by using the difference in revolutions between the bath pump revolutions measured last time and the revolutions measured on the evaluation date, and the ratio between the bath pump revolutions measured the first time (for example, when the hot water supply device 1 was installed) and the revolutions measured on the evaluation date.

[0032] FIG. 4 is a graph showing experimental results for explaining the relationship between the reconstruction error (first reconstruction error) and the detection of a fault sign in the above-described embodiment of the fault prediction system. The vertical axis of this graph represents the reconstruction error (first reconstruction error), and the horizontal axis represents the difference (rotation speed difference value) between the intermediate day pump speed (the previously measured pump speed, which is the measurement value immediately before the judgment day) and the judgment day pump speed (the pump speed measured on the current day). In this graph, a group of points where the reconstruction error is distributed horizontally near zero (group A surrounded by a horizontally long ellipse in FIG. 4) represents a normal operating state, i.e., a group of data from which no fault signs can be detected. In contrast, a group of points where the rotation speed difference value is distributed vertically near 1500 rpm (group B surrounded by a vertically long ellipse in FIG. 4) represents an abnormal data group created as dummy data, and the reconstruction error is large. Therefore, when the reconstruction error falls within a predetermined range (the range of group B: one of the judgment conditions), a fault sign can be detected.

[0033] FIG. 5 is a graph showing experimental results for explaining the relationship between the reconstruction error (second reconstruction error) and the detection of a fault sign in the above-described embodiment of the fault prediction system. The vertical axis of this graph represents the reconstruction error (second reconstruction error), and the horizontal axis represents the ratio (%) of the reference day pump rotation speed (initial measurement pump rotation speed) to the evaluation day pump rotation speed (current day measurement pump rotation speed). In this graph, when the reconstruction error is close to zero and the ratio of the pump rotation speed between the initial and current day is close to 100% (a state in which there is almost no decrease from the initial measurement), a group of points (surrounded by a horizontally long ellipse in FIG. 5: group C) that are distributed together indicate a normal operating state, i.e., a group of data from which no fault signs can be detected. In contrast, when the ratio of the rotation speed between the initial and current day is between 90% and 70%, the reconstruction error (second reconstruction error) increases and the group of points (surrounded by a vertically long ellipse in FIG. 5: group D) that are distributed vertically represent an abnormal data group created as dummy data, and have a larger reconstruction error than group C. From this, when the reconstruction error falls within a predetermined range (range of group D: one type of judgment condition), it becomes possible to estimate a sign of a failure.

[0034] [Another embodiment] (1) In the above-described embodiment, the reconstruction error calculation unit 53, which constitutes the failure prediction model 51 together with the mapping model 52, which is an autoencoder, is composed of the first reconstruction error calculation unit 53A and the second reconstruction error calculation unit 53B. However, instead, the reconstruction error calculation unit 53 may be composed of only the first reconstruction error calculation unit 53A or only the second reconstruction error calculation unit 53B.

[0035] (2) In the above-described embodiment, the data handled by the failure prediction model 51 were the reference day drive strength, the reference day pump rotation speed, the intermediate day drive strength, the intermediate day pump rotation speed, the judgment day drive strength, and the judgment day pump rotation speed. In addition to this, other data may also be handled, such as the cumulative usage time of the water heating device 1, the cumulative hot water usage amount of the water heating device 1, the rotation speed of the combustion fan of the water heating device 1, and the temperature of the thermistor provided in the water heating device 1.

[0036] (3) In the above-described embodiment, the data stored in the driving data history storage unit 22 is acquired online via the data communication network DN, but at least a portion of the data may be acquired offline.

[0037] (4) The drive strength included in the drive data can include values ​​such as the length of the piping and pressure loss of each water heater 1 in addition to the value defined by the duty cycle of the hot water pump. In this case, the drive data is converted into multidimensional vector data and provided to the failure prediction model 51.

[0038] (5) The reference date and the intermediate date can be selected arbitrarily as long as they are measurement dates that are earlier than the assessment date and the reference date is a date that is earlier than the intermediate date.

[0039] (6) In the above-described embodiment, the ratio between the reference day pump rotation speed and the intermediate day pump rotation speed, and the difference between the intermediate day pump rotation speed and the evaluation day pump rotation speed were distinguished. However, these ratios and difference values ​​are difference values ​​that indicate the difference between two different pump rotation speeds. Therefore, in another embodiment of the present invention, the ratio and difference value may be interchanged, or the use may be limited to either the ratio or the difference value. In other words, in the present invention, the ratio and difference value can be replaced with the difference value described above.

[0040] (7) The functional units shown in Figures 2 and 3 are created for explanatory purposes, and each functional unit may be further divided or integrated with other functional units.

[0041] The configurations disclosed in the above embodiments (including other embodiments, the same applies below) can be applied in combination with configurations disclosed in other embodiments, as long as no contradiction arises. Furthermore, the embodiments disclosed in this specification are examples, and the embodiments of the present invention are not limited to these, and can be modified as appropriate within the scope that does not deviate from the purpose of the present invention. [Industrial Applicability]

[0042] The present invention can be applied to a system that predicts a failure in a water heater. [Explanation of symbols]

[0043] 1: Hot water supply equipment 2: Management computer 5: Failure prediction calculation unit 21: Data acquisition section 22: Driving data history storage section 23: Data Management Department 24: Information Department 24A: Display control unit 25: Display 50: Model Management Department 50A: Learning Department 50B: Practical Department 51: Failure prediction model 52: Mapping model 53: Reconstruction error calculation unit 53A: First reconstruction error calculation unit 53B: Second reconstruction error calculation unit 54: Failure sign information generation unit 55:Failure sign determination unit DN: Data communication network

Claims

1. A failure prediction system for a hot water heater installed in a home, a drive data history storage unit that stores the drive strength and the pump rotation speed of the hot water pump acquired over time during normal operation of the hot water supply device as normal drive data, and stores the drive strength and the pump rotation speed acquired when a warning sign is determined for the hot water supply device as determination drive data; and a data management unit that reads out the reference day drive intensity and the reference day pump rotation speed, which are the normal operation data on a reference day, from the drive data history storage unit, reads out the intermediate day drive intensity and the intermediate day pump rotation speed, which are the normal operation data on an intermediate day after the reference day, from the drive data history storage unit, and reads out the judgment day drive intensity and the judgment day pump rotation speed, which are the judgment operation data on a judgment day after the intermediate day, from the drive data history storage unit; a model management unit that generates a mapping model using the reference day driving strength, the intermediate day driving strength, and a difference value between the reference day pump rotation speed and the intermediate day pump rotation speed as training data; a reconstruction error calculation unit that calculates a reconstruction error by inputting the determination day drive strength and a difference value between the intermediate day pump rotation speed and the determination day pump rotation speed into the mapping model; a failure sign information generation unit that generates failure sign information of the water heating apparatus based on the reconstruction error; A failure prediction system equipped with

2. The failure prediction system according to claim 1 , wherein the intermediate date is a previous evaluation date immediately preceding the evaluation date.

3. A failure prediction system for a hot water heater installed in a home, a drive data history storage unit that stores the drive strength and the pump rotation speed of the hot water pump acquired over time during normal operation of the hot water supply device as normal drive data, and stores the drive strength and the pump rotation speed acquired when a warning sign is determined for the hot water supply device as determination drive data; and a data management unit that reads out the reference day drive intensity and the reference day pump rotation speed, which are the normal operation data on a reference day, from the drive data history storage unit, reads out the intermediate day drive intensity and the intermediate day pump rotation speed, which are the normal operation data on an intermediate day after the reference day, from the drive data history storage unit, and reads out the judgment day drive intensity and the judgment day pump rotation speed, which are the judgment operation data on a judgment day after the intermediate day, from the drive data history storage unit; a model management unit that generates a mapping model using the reference day driving strength, the intermediate day driving strength, and the ratio between the reference day pump rotation speed and the intermediate day pump rotation speed as training data; a reconstruction error calculation unit that calculates a reconstruction error by inputting the determination day drive strength and the ratio between the reference day pump rotation speed and the determination day pump rotation speed into the mapping model; a failure sign information generation unit that generates failure sign information of the water heating apparatus based on the reconstruction error; A failure prediction system equipped with

4. The failure prediction system according to claim 3 , wherein the reference date is the date on which the normal operation data was first measured.

5. A failure prediction system for a hot water heater installed in a home, a drive data history storage unit that stores the drive strength and the pump rotation speed of the hot water pump acquired over time during normal operation of the hot water supply device as normal drive data, and stores the drive strength and the pump rotation speed acquired when a warning sign is determined for the hot water supply device as determination drive data; and a data management unit that reads out the reference day drive intensity and the reference day pump rotation speed, which are the normal operation data on a reference day, from the drive data history storage unit, reads out the intermediate day drive intensity and the intermediate day pump rotation speed, which are the normal operation data on an intermediate day after the reference day, from the drive data history storage unit, and reads out the judgment day drive intensity and the judgment day pump rotation speed, which are the judgment operation data on a judgment day after the intermediate day, from the drive data history storage unit; a model management unit that generates a mapping model using the reference day driving strength, the intermediate day driving strength, and the ratio and difference between the reference day pump rotation speed and the intermediate day pump rotation speed as training data; a first reconstruction error calculation unit that calculates a first reconstruction error as a reconstruction error by inputting the determination day drive strength and a difference value between the intermediate day pump rotation speed and the determination day pump rotation speed into the mapping model; a second reconstruction error calculation unit that calculates a second reconstruction error as the reconstruction error by inputting the determination day drive strength and the ratio between the reference day pump rotation speed and the determination day pump rotation speed into the mapping model; a failure sign information generating unit that generates failure sign information of the water heating apparatus based on the first reconstruction error and the second reconstruction error; A failure prediction system equipped with

6. 6. The failure prediction system according to claim 5, wherein the reference date is the date on which the normal operation data was first measured, and the intermediate date is the previous evaluation date immediately preceding the evaluation date.

7. The failure prediction system according to claim 5 , wherein the mapping model is configured by an autoencoder.

8. The failure prediction system according to claim 5, wherein the hot water supply pump is a hot water filling pump.

9. The failure sign system according to claim 1 , further comprising a display control unit that generates display data for displaying the failure sign information on a display device.

10. The failure prediction system according to claim 1 , further comprising a failure sign determination unit that determines a failure sign of the water heating apparatus based on a determination condition set for the reconstruction error.

Images