Power consumption estimation apparatus, power consumption estimation method, and power consumption estimation program

Abstract

A power consumption estimation apparatus (100) comprises: a time sampling processing unit (13) that converts a plurality of device data respectively acquired from a plurality of devices (12) into time-series device data; a process mining unit (15) that analyzes a state transition of each of the plurality of devices (12) on the basis of the time-series device data, and outputs state transition time-series data indicating an operating state related to the power consumption of each of the plurality of devices (12); and a power consumption estimation unit (16) that estimates the power consumption of each of the plurality of devices (12) on the basis of total power consumption data of the plurality of devices (12) and the state transition time-series data.

Description

Power consumption estimation apparatus, power consumption estimation method, and power consumption estimation program

[0001] The present disclosure relates to a power consumption estimation apparatus, a power consumption estimation method, and a power consumption estimation program.

[0002] A wide variety of devices are operated in facilities such as factories. It is desirable to individually measure the power consumption of these devices. In such a case, for example, a sensor for measuring power consumption is provided for each device, but installing sensors for all devices in the facility requires a considerable cost.

[0003] As one of the methods for solving the above problem, Patent Document 1 discloses a technique that estimates average power consumption, which is an average value of power consumption of each device, from total power consumption data obtained from a power meter as total power consumption of a plurality of devices and operating state data obtained as data indicating whether each of the plurality of devices is in an operating state from each device, and estimates device power consumption, which is power consumption of each device, from the average power consumption.

[0004] International Publication No. 2023 / 228816

[0005] However, with the technique disclosed in Patent Document 1, it is necessary to manually edit various data (log data, sensor data, etc.) obtained from devices, convert the data into data representing the operating state of the devices, and input the converted data. This work requires labor, and also requires specialized knowledge about the structure and operation of the devices, leading to the problem that power consumption cannot be estimated easily.

[0006] An object of the present disclosure is to provide a power consumption estimation apparatus, a power consumption estimation method, and a power consumption estimation program that allow even a person without specialized knowledge about devices to estimate the power consumption of devices with less labor by automating the work of converting various data obtained from devices into data representing the operating state of the devices.

[0007] The power consumption estimation device of this disclosure comprises: a time sampling processing unit that converts multiple device data acquired from multiple devices into time-series device data; a process mining unit that analyzes the state transitions of each of the multiple devices based on the time-series device data and outputs state transition time-series data indicating the operating state related to the power consumption of each of the multiple devices; and a power consumption estimation unit that estimates the power consumption of each of the multiple devices based on the total power consumption data of the multiple devices and the state transition time-series data.

[0008] The power consumption estimation method of this disclosure is a method performed by a computer and includes the steps of: converting a plurality of device data obtained from a plurality of devices into time-series device data; analyzing the state transitions of each of the plurality of devices based on the time-series device data and outputting state transition time-series data indicating the operating state relating to the power consumption of each of the plurality of devices; and estimating the power consumption of each of the plurality of devices based on the total power consumption data of the plurality of devices and the state transition time-series data.

[0009] The power consumption estimation program of this disclosure causes a computer to perform the following steps: convert multiple device data obtained from multiple devices into time-series device data; analyze the state transitions of each of the multiple devices based on the time-series device data and output state transition time-series data indicating the operating state related to the power consumption of each of the multiple devices; and estimate the power consumption of each of the multiple devices based on the total power consumption data of the multiple devices and the state transition time-series data.

[0010] According to this disclosure, it is possible to provide a power consumption estimation device, a power consumption estimation method, and a power consumption estimation program that enable even a person without specialized knowledge of the equipment to estimate the power consumption of the equipment with minimal effort.

[0011] This is a block diagram showing an example of the configuration of the power consumption estimation device according to Embodiment 1. This is a block diagram showing an example of the hardware configuration of the power consumption estimation device according to Embodiment 1. This is a flowchart showing an example of processing in the time sampling processing unit. This is an example of a functional block diagram of the noise reduction processing unit. This is a flowchart showing an example of processing in the noise reduction processing unit. This is an example of a functional block diagram of the process mining unit. (A) is an explanatory diagram showing an example of state identification of a data group by the state extraction unit, (B) is an explanatory diagram showing an example of state label correction by the transition relationship analysis unit, and (C) is a schematic diagram showing an example of state transition time series data. This is a flowchart showing an example of processing in the process mining unit. This is a flowchart showing an example of processing in the power consumption estimation unit. This is a block diagram showing an example of the configuration of the power consumption estimation device according to Embodiment 2. This is an explanatory diagram showing an example of processing in the data expansion unit of the power consumption estimation device according to Embodiment 2. This is a flowchart showing an example of processing in the data expansion unit. This is a block diagram showing an example of the configuration of the power consumption estimation device according to Embodiment 3. This is a schematic diagram showing an example of the correspondence between a group of device data and a sorting list. This is a flowchart showing an example of processing based on the genetic algorithm of the optimization processing unit.

[0012] The power consumption estimation device, power consumption estimation method, and power consumption estimation program related to this disclosure will be described below with reference to the drawings. The following embodiments are merely examples, and it is possible to combine the embodiments as appropriate and modify each embodiment as appropriate.

[0013] Embodiment 1 Figure 1 is a block diagram showing an example of the configuration of a power consumption estimation device 100 according to Embodiment 1. In Figure 1, the power consumption estimation device 100 includes a time sampling processing unit 13, a noise reduction processing unit 14, a process mining unit 15, and a power consumption estimation unit 16, thereby enabling the extraction of operating states related to power consumption from various data obtained from multiple devices (devices 1, 2, ..., N: where N is a positive integer) and the estimation of power consumption. In Embodiment 1, the power consumption estimation device 100 includes a time sampling processing unit 13 that converts multiple device data obtained from multiple devices 12 into time-series device data, a process mining unit 15 that analyzes the state transitions of each of the multiple devices 12 based on the time-series device data and outputs state transition time-series data indicating the operating state related to the power consumption of each of the multiple devices 12, and a power consumption estimation unit 16 that estimates the power consumption of each of the multiple devices 12 based on the total power consumption data and state transition time-series data of the multiple devices 12. Furthermore, the power consumption estimation device 100 may include a noise reduction processing unit 14 that performs noise reduction processing, which is the process of removing data that has a low correlation with total power consumption data from time-series device data.

[0014] The time sampling processing unit 13 receives multiple device data, such as log data indicating the operating status of each of the multiple devices 12, or sensor data such as temperature, vibration, or rotational speed acquired by sensors provided on each of the multiple devices 12. Depending on the type of input device data, it converts it into time-series device data, which is time-series data with the same sampling period as the total power consumption data. The noise reduction processing unit 14 removes time-series data that has a low correlation with the total power consumption data of the facility measured by the power measuring instrument 10 from the time-series data group converted by the time sampling processing unit 13. For example, in the noise reduction process, the noise reduction processing unit 14 removes data for which the correlation coefficient calculated by analyzing the correlation between the total power consumption data and the time-series device data is lower than a predetermined threshold. The process mining unit 15 analyzes the state specific to each device 12 and its transition relationships (for example, the state of the device 12 and the transitions between states) from the time-series data group of the multiple devices 12, and outputs information on what state the device 12 is in at each time point. The power consumption estimation unit 16 estimates and outputs the power consumption of each device 12 based on the information about the state of the device 12 output by the process mining unit 15 and the total power consumption data.

[0015] The power measuring instrument 10 is a device installed in the facility's power distribution panel or substation, etc., that measures the total power consumption of the facility. The sensors installed in the device 12 are, for example, a temperature sensor that detects the operating temperature of the device 12, an IMU (inertial measurement unit) that detects vibrations of the device 12, and a rotary encoder that detects the rotational speed of the rotating shaft of the device 12. The log data of the device 12 is data that records the time of occurrence of an event, an event ID to identify the event, and the name of the event each time an event occurs in the device 12.

[0016] Figure 2 is a block diagram showing an example of the hardware configuration of the power consumption estimation device 100 according to Embodiment 1. As shown in Figure 2, the power consumption estimation device 100 is composed of a computer in which each of the following components, a CPU (Central Processing Unit) 21 (which is a processing element), a main memory 22, an input / output interface (I / O interface) 23, and a storage unit 24, are connected to a system bus 25. The power consumption estimation device 100 may be composed of multiple computers connected by a network, or it may be composed of processing circuits.

[0017] The CPU 21 is an IC (Integrated Circuit) that performs arithmetic processing. In addition to the CPU 21, other arithmetic elements such as a DSP (Digital Signal Processor), GPU (Graphics Processing Unit), Network Processor, or FPGA (Field Programmable Gate Array) may also be used. The CPU 21 can realize a power consumption estimation method that includes a time sampling processing function that converts data into time series data by executing the power consumption estimation program according to Embodiment 1, a noise reduction processing function that removes time series data with low correlation to the total power consumption data, a process mining function that outputs information on the state of the device 12 at each time point, and a power consumption estimation function that estimates and outputs the power consumption of each device 12 from the information output by the process mining unit 15 and the total power consumption data. As a result, the CPU 21 functions as a time sampling processing unit 13, a noise reduction processing unit 14, a process mining unit 15, and a power consumption estimation unit 16 by executing the power consumption estimation program. The power consumption estimation program is provided, for example, on a recording medium on which these are recorded.

[0018] The main memory 22 is composed of a volatile storage device such as RAM (Random Access Memory) or a non-volatile storage device such as ROM (Read Only Memory). The storage unit 24 is composed of a non-volatile storage device such as an HDD (Hard Disk Drive) or flash memory.

[0019] The I / O interface 23 is a port to which the power meter 10, the equipment 12, and the output unit 17 are connected. The output unit 17 is an output device such as a display or printer that outputs the power consumption of each of the equipment 12 estimated by the power consumption estimation unit 16.

[0020] Figure 3 is a flowchart showing an example of processing in the time sampling processing unit 13. In step S101, the time sampling processing unit 13 determines whether the input data is time-series data or not. If the input data is time-series data, the procedure proceeds to step S102. If the input data is not time-series data, the procedure proceeds to step S104.

[0021] In step S102, the time sampling processing unit 13 performs a conversion to resample the data to the same sampling period as the total power consumption data by converting the sampling rate. In other words, if the device data is time-series data, the time sampling processing unit 13 outputs the data obtained by resampling the device data to time-series data with the same sampling period as the total power consumption data as time-series device data.

[0022] In step S103, the time sampling processing unit 13 normalizes the time series data that has been converted to a sampling rate and terminates the processing. As will be described later, normalization makes it possible to compare and convert time series data that originally had different units.

[0023] In step S104, the time sampling processing unit 13 determines whether the input data is log data from the device 12. If the input data is log data from the device 12, the procedure proceeds to step S105. If the input data is not log data from the device 12, the procedure proceeds to step S106.

[0024] In step S105, the time sampling processing unit 13 converts the log data into time-series data. In step S105, the time sampling processing unit 13 converts the log data into time-series data such that the period from the event occurrence time to the end time is 1, and the rest is 0, and outputs it. In other words, if the device data is log data, the time sampling processing unit 13 converts it into time-series data such that the period from the event occurrence time to the end time is 1, and the rest is 0, and outputs the resulting data as time-series device data. In step S105, the time sampling processing unit 13 converts the log data into time-series data so that the sampling period is the same as that of the total power consumption data, and then proceeds to step S103.

[0025] In step S106, the time sampling processing unit 13 determines that the input data is neither time-series data nor log data from the device 12, and therefore is unsuitable for estimating the power consumption of the device 12. It then removes the data and terminates the process.

[0026] Figure 4 is an example of a functional block diagram of the noise reduction processing unit 14. As shown in Figure 4, the noise reduction processing unit 14 includes a correlation analysis unit 30 that analyzes the correlation between total power consumption data and a group of data obtained from equipment 12 (equipment X), and a low-correlation data removal unit 31 that removes time-series data that has a low correlation with the total power consumption data.

[0027] The correlation analysis unit 30 analyzes the correlation between each data point in the data group and the total power consumption data, for example, by calculating the correlation coefficient between the total power consumption data and any of the data points that make up the data group. The correlation analysis unit 30 may also perform a multiple regression analysis with power consumption data as the dependent variable and the data group as the independent variables, and may remove data that does not significantly increase the regression error when removed from the independent variables.

[0028] The low-correlation data removal unit 31 removes data with a correlation lower than a predetermined threshold. For example, the low-correlation data removal unit 31 removes data where the correlation coefficient calculated by the correlation analysis unit 30 is lower than a predetermined threshold. The predetermined threshold is, as an example, the lower limit of the correlation coefficient at which the estimation result by the power consumption estimation unit 16 can be a valid value for the power consumption value when the device 12 is actually measured (for example, an agreement of 95% or more with the measured value).

[0029] The low-correlation data removal unit 31 removes data that has a low correlation with the total power consumption data, while ensuring that at least one piece of data that is correlated with the total power consumption data is left for each of the devices 12.

[0030] Figure 5 is a flowchart showing an example of processing in the noise reduction processing unit 14. In step S201, the noise reduction processing unit 14 acquires the data set from the device 12 from the time sampling processing unit 13.

[0031] In step S202, the noise reduction processing unit 14 acquires total power consumption data from the power measuring instrument 10.

[0032] In step S203, the noise reduction processing unit 14 analyzes the correlation between each data item constituting the data group of the device 12 and the total power consumption data in the correlation analysis unit 30. The correlation analysis unit 30 may also analyze the correlation between each data item constituting the data group of the device 12 and the total power consumption data by multiple regression analysis with power data as the dependent variable and the data group as the independent variables.

[0033] In step S204, the noise reduction processing unit 14 determines in the correlation analysis unit 30 whether or not there is a correlation between each data item constituting the data group of the device 12 and the total power consumption data. If a correlation is found between each data item constituting the data group of the device 12 and the total power consumption data in step S204, the procedure proceeds to step S205. If no correlation is found between each data item constituting the data group of the device 12 and the total power consumption data, the procedure proceeds to step S206.

[0034] In step S205, the noise reduction processing unit 14 outputs the data group in the low-correlation data removal unit 31 that has been found to be correlated with the total power consumption data as the noise-reduced data group to the subsequent process mining unit 15, and terminates the processing.

[0035] In step S206, the noise reduction processing unit 14 terminates the process by removing data that was not found to have a correlation with the total power consumption data in the low-correlation data removal unit 31.

[0036] Figure 6 is an example of a functional block diagram of the process mining unit 15. The process mining unit 15 analyzes the state specific to each device 12 and its transition relationships from a set of time-series data of the devices, and outputs information on the state of the device 12 at each time point. To this end, the process mining unit 15 includes a state extraction unit 40 that extracts the state from the device data, a transition relationship analysis unit 41 that analyzes the transition relationships, and a state transition time-series data output unit 42 that outputs the state time-series data to the subsequent power consumption estimation unit 16. In other words, the process mining unit 15 includes a state extraction unit 40 that extracts the state of each of the multiple devices 12 from time-series device data, a transition relationship analysis unit 41 that detects the transition relationships between states, and a state transition time-series data output unit 42 that outputs state transition time-series data representing the state transitions of the multiple devices based on the states and transition relationships. The power consumption estimation unit 16 can estimate power consumption by receiving the state time-series data from the process mining unit 15.

[0037] The state extraction unit 40 identifies a unique state from the data set of the device 12 (device X). For identification, unsupervised classification such as knn (k-nearest neighbors) is used. Figure 7(A) is an explanatory diagram showing an example of state identification of a data set by the state extraction unit 40. For example, if data α is the temperature of the device 12 and data β is the rotational speed of the rotation axis of the device 12, then in Figure 7(A), the horizontal axis is temperature and the vertical axis is rotational speed. If each data is normalized as described above, it becomes easier to handle data with different units. In Figure 7(A), the rotational speed against temperature is plotted. The state extraction unit 40 then uses one of the plotted data as a sample value and identifies, for example, state A and state B by extracting a majority vote of the states indicated by k data (k is a positive integer) that approximate the sample value using the k-nearest neighbors method. For example, state A is a state where the device 12 is cold and has a low rotational speed, and state B is a state where the device 12 is hot and has a high rotational speed. The state extraction unit 40 calculates a load index, expressed as a percentage (%) or in the range of 0 to 1, which indicates how much load each extracted state for the equipment 12 represents relative to the equipment 12's rating, and associates it with that state.

[0038] The transition relationship analysis unit 41 calculates the occurrence frequency of each state extracted by the state extraction unit 40. The transition relationship analysis unit 41 may also calculate a probability distribution of state occurrences that shows the distribution of occurrence frequencies. The transition relationship analysis unit 41 then considers transitions from a low-occurrence state to another state, from another state to a low-occurrence state, and from a low-occurrence state to another low-occurrence state as errors and corrects the state label. Furthermore, the transition relationship analysis unit 41, for example, in the probability distribution of state occurrences, determines that transitions from a state whose occurrence probability is above a predetermined probability threshold to a state whose occurrence probability is also above a predetermined probability threshold are high-frequency transitions. In other words, the transition relationship analysis unit 41 calculates the probability of each state occurring based on the frequency of occurrence of each state extracted by the state extraction unit 40, determines that transitions from a state whose probability is above a predetermined probability threshold to other states whose probability is above a predetermined probability threshold are high-frequency transitions, and outputs transitions from a state whose probability is above the predetermined probability threshold to other states whose probability is above the predetermined probability threshold as transition relationships to the state transition time series data output unit 42. The predetermined probability threshold is, for example, the lower limit of the probability at which the estimation result by the power consumption estimation unit 16 can be a valid value (for example, 95% or more agreement with the measured value) when the device 12 is actually measured.

[0039] Figure 7(B) is an explanatory diagram showing an example of state label correction by the transition relationship analysis unit 41. In Figure 7(B), solid arrows represent high-frequency transition paths, and dashed arrows represent low-frequency transition paths. As shown in Figure 7(B), before correction, the state transitions were in the order of states A, D, B (state A → state D → state B) or states A, C (state A → state C). However, the transition relationship analysis unit 41 considered low-frequency transition paths as errors and corrected the state labels, resulting in the state transitions being in the order of states A, B, C (state A → state B → state C). The transition relationship analysis unit 41 outputs transitions from high-frequency states to other high-frequency states (i.e., state transitions along high-frequency transition paths shown by solid lines in Figure 7(B)) as transition relationships to the state transition time series data output unit.

[0040] The state transition time series data output unit 42 generates state transition time series data, which is time series data representing state transitions, and outputs the generated data to the power consumption estimation unit 16 at the subsequent stage. FIG. 7C is a schematic diagram showing an example of state transition time series data. As shown in FIG. 7C, the state transition time series data outputs state transitions in a time series as pulse signals in a time series.

[0041] FIG. 8 is a flowchart showing an example of processing performed by the process mining unit 15. In step S301, the state extraction unit 40 of the process mining unit 15 acquires a data group of the device 12 from the noise removal processing unit 14.

[0042] In step S302, the state extraction unit 40 of the process mining unit 15 performs state extraction to identify unique states from the data group of the device 12 (device X).

[0043] In step S303, the transition relation analysis unit 41 of the process mining unit 15 performs transition relation analysis to calculate the occurrence frequency of each state extracted by the state extraction unit 40.

[0044] In step S304, the state transition time series data output unit 42 of the process mining unit 15 generates state transition time series data, which is time series data representing state transitions, outputs the generated data to the power consumption estimation unit 16 at the subsequent stage, and then ends the processing.

[0045] The power consumption estimation unit 16 estimates the power consumption of each of the devices 12 based on the state transition time-series data related to each of the devices 12 and the total power consumption data. For example, if the load index of the device 12 in each state indicated by the state transition time-series data of the device 12 is associated in advance, the power consumption estimation unit 16 can estimate the occurrence frequency of the state of the device 12 in the state transition time-series data over a desired period (e.g., one month). The power consumption of the device 12 can be roughly calculated by integrating, for each device 12, the value obtained by multiplying the occurrence frequency by the load index of the device 12 in that state and the rated power consumption of the device 12. Then, a value δ obtained by summing the roughly calculated power consumption of each of the devices 12 is compared with the total power consumption data γ measured by the power meter 10 over the desired period. If there is a difference between the roughly calculated total power consumption δ of the devices 12 and the total power consumption data γ measured by the power meter 10 (for example, when γ / δ is 0.95 or less, or 1.05 or more), the power consumption estimation unit 16 multiplies each of the roughly calculated power consumption values of the devices 12 by a correction coefficient which is the ratio of the total power consumption data γ measured by the power meter 10 to the total roughly calculated power consumption δ of the devices 12 (γ / δ), corrects the roughly calculated power consumption of the devices 12, and outputs the corrected value as an estimation result.

[0046] FIG. 9 is a flowchart showing an example of processing performed by the power consumption estimation unit 16. In step S401, the power consumption estimation unit 16 acquires state transition time-series data from the process mining unit 15.

[0047] In step S402, the power consumption estimation unit 16 acquires total power consumption data from the power meter 10.

[0048] In step S403, the power consumption estimation unit 16 roughly calculates the power consumption of each device 12 based on the state transition time-series data and the total power consumption data.

[0049] In step S404, the power consumption estimation unit 16 sums the roughly calculated power consumption values of the respective devices 12.

[0050] In step S405, the power consumption estimation unit 16 determines whether the difference between the estimated total power consumption δ of the devices 12 and the total power consumption data γ measured by the power measuring instrument 10 is within an acceptable range. If the difference is within an acceptable range, the procedure proceeds to step S406; otherwise, the procedure proceeds to step S407.

[0051] In step S406, the power consumption estimation unit 16 outputs the estimated power consumption of the device 12 as the estimation result and terminates the process.

[0052] In step S407, the power consumption estimation unit 16 corrects the estimated power consumption of the device 12, outputs the corrected power consumption as the estimation result, and terminates the process.

[0053] As described above, the power consumption estimation device 100 of Embodiment 1 can estimate the power consumption of each device 12 from the total power consumption data of the facility where the device 12 is installed by generating state transition time series data that shows the operating state of the device 12. Since the state transition time series data that shows the operating state of the device 12 is automatically generated by time-series conversion and noise removal of data acquired from the device 12, it is possible to provide a power consumption estimation device, power consumption estimation method, and power consumption estimation program that allow even a person without specialized knowledge of the device 12 to estimate the power consumption of the device with little effort.

[0054] Embodiment 2 will now be described. Figure 10 is a block diagram showing an example of the configuration of the power consumption estimation device 110 according to Embodiment 2. The power consumption estimation device 110 according to Embodiment 2 differs from the power consumption estimation device 100 according to Embodiment 1 in that it includes a data expansion unit 18 that performs data expansion processing on time-sampled device data. However, the other configurations are the same as in Embodiment 1, so the same reference numerals as in Embodiment 1 are used for the same configurations as in Embodiment 1, and detailed explanations are omitted.

[0055] The data expansion unit 18 adds data that has undergone various transformations from the original equipment data as candidate equipment data for process mining.

[0056] Figure 11 is an explanatory diagram showing an example of processing in the data expansion unit 18 of the power consumption estimation device 110 according to Embodiment 2. The data expansion unit 18 performs transformation processing on time series data obtained from each device 12 by the time sampling processing unit 13, including transformation processing that can be processed on a single data (one time series data) (for example, the difference between adjacent numerical values ​​in the time series, or the cumulative sum, etc.) and transformation processing that can be processed on combinations of arbitrarily extracted multiple time series data (for example, logical operations, arithmetic operations, or principal component analysis, etc.), and adds it to the original data before outputting it. In other words, the data expansion unit 18 performs transformation processing on the time series device data to be processed, including transformation processing that can be processed on a single time series device data and transformation processing that can be processed on combinations of arbitrarily extracted multiple time series device data, and adds it to the time series device data to be processed. As described above, since the time sampling processing unit 13 normalizes the data acquired from the devices 12, a variety of transformation processing is possible even if the data originally had different units.

[0057] The logical operations in the data extension unit 18 are as follows, assuming that both α and β are logical values ​​(0, 1): X = α ∩ β means "both α and β are 1". X = α ∪ β means "either α or β is 1". X = α X OR β means "only one of α or β is 1".

[0058] Furthermore, the arithmetic operations in the data extension unit 18 are as follows, assuming that both α and β are logical values ​​(0, 1): X = α + β means that "X is the sum of α and β". X = α - β means that "X is the difference value obtained by subtracting β from α". X = α × β means that "X is the product of α and β". X = α ÷ β means that "X is the division value obtained by dividing α by β". X = β ÷ α means that "X is the division value obtained by dividing β by α".

[0059] Figure 12 is a flowchart showing an example of processing in the data expansion unit 18. In step S501, the data expansion unit 18 acquires data from the time sampling processing unit 13 to the device 12.

[0060] In step S502, the data expansion unit 18 performs a conversion process on the acquired data.

[0061] In step S503, the data expansion unit 18 outputs the converted data, along with the data from the device 12 acquired from the time sampling processing unit 13, to the subsequent noise reduction processing unit 14, thereby ending the processing.

[0062] As described above, the power consumption estimation device 110 according to Embodiment 2 increases the variation of data used for process mining through conversion processing, which allows for a more detailed estimation of the device's state and improves the accuracy of the power consumption estimation of the device 12.

[0063] Embodiment 3 Next, Embodiment 3 will be described. Figure 13 is a block diagram showing an example of the configuration of the power consumption estimation device 120 according to Embodiment 3. The power consumption estimation device 120 according to Embodiment 3 differs from the power consumption estimation device 100 according to Embodiment 1 in that the power consumption estimation unit 16 feeds back the reconstruction error, which is the error between the sum of the estimated power consumption values ​​for each device 12 and the total power consumption data (i.e., the difference between the sum of the estimated power consumption values ​​and the total power consumption data), to the optimization processing unit 19 and instructs it to redo the noise reduction processing so that the error becomes smaller. However, the other configurations are the same as in Embodiment 1, so the same reference numerals as in Embodiment 1 are used for the same configurations as in Embodiment 1 and detailed explanations are omitted.

[0064] The optimization processing unit 19 creates a selection list for the group of device data, determining whether or not to adopt each piece of data, and sends it to the noise reduction processing unit 14. Figure 14 is a schematic diagram showing an example of the correspondence between the group of device data and the selection list. The selection list is a table that defines whether to adopt or reject the data for the corresponding device 12. Initially, the selection list is written to adopt all data, for example, and the optimization processing unit 19 modifies the selection list so that the reconstruction error output by the power consumption estimation unit 16 becomes smaller. The reconstruction error output by the power consumption estimation unit 16 is, for example, the correction coefficient (γ / δ) calculated by the power consumption estimation unit 16 of the power consumption estimation device 100 according to Embodiment 1.

[0065] The noise reduction processing unit 14 performs noise reduction according to the selection list. The optimization processing unit 19 modifies the selection list so that the resulting reconstruction error is small. The optimization processing unit 19 searches for the best list while modifying the selection list, for example, by a genetic algorithm that preferentially selects data with high fitness. In other words, the optimization processing unit 19 receives feedback from the power consumption estimation unit 16 regarding the error between the sum of the estimated power consumption values ​​for each device 12 and the total power consumption value shown by the total power consumption data, and instructs the noise reduction processing unit 14 to redo the noise reduction process so that this error is smaller.

[0066] Figure 15 is a flowchart showing an example of the processing based on the genetic algorithm of the optimization processing unit 19. In step S601, the optimization processing unit 19 initializes the selection list. The initialized selection list, as an example, will include all data as described above.

[0067] In step S602, the optimization processing unit 19 randomly generates multiple candidates for modifying the selection list. Then, in step S603, the optimization processing unit 19 sends the selection list to the noise reduction processing unit 14 with each of the modification candidates applied. The modification candidates are, for example, candidates for data to be rejected.

[0068] In step S604, the optimization processing unit 19 obtains the power consumption error calculated using the selection list to which each correction candidate has been applied.

[0069] In step S605, the optimization processing unit 19 selects the best selection list. The best selection list is the selection list with the smallest power consumption error.

[0070] In step S606, the optimization processing unit 19 determines whether the power consumption error based on the selection list selected in step S605 is below a predetermined reference value. The power consumption error being below a predetermined reference value means, for example, that the correction coefficient γ / δ is between 0.95 and 1.05. If, in step S606, the power consumption error is below a predetermined reference value, the selection list selected in step S605 is output to the noise reduction processing unit 14, and the process ends. If, in step S606, the power consumption error is not below a predetermined reference value, the procedure proceeds to step S602, and the procedure of randomly generating multiple candidate selection list corrections is resumed.

[0071] As described above, the power consumption estimation device 120 according to Embodiment 3 can improve the accuracy of estimating the power consumption of the device 12 by feeding back the error in the estimation result and redoing the noise reduction.

[0072] 10 Power measuring instrument, 12 Equipment, 13 Time sampling processing unit, 14 Noise reduction processing unit, 15 Process mining unit, 16 Power consumption estimation unit, 17 Output unit, 18 Data expansion unit, 19 Optimization processing unit, 21 CPU, 22 Main memory, 23 I / O interface, 24 Storage unit, 25 System bus, 30 Correlation analysis unit, 31 Low correlation data removal unit, 40 State extraction unit, 41 Transition relationship analysis unit, 42 State transition time series data output unit, 100, 110, 120 Power consumption estimation device.

Claims

1. A power consumption estimation device comprising: a time sampling processing unit that converts multiple device data acquired from multiple devices into time-series device data; a process mining unit that analyzes the state transitions of each of the multiple devices based on the time-series device data and outputs state transition time-series data indicating the operating state related to the power consumption of each of the multiple devices; and a power consumption estimation unit that estimates the power consumption of each of the multiple devices based on the total power consumption data of the multiple devices and the state transition time-series data.

2. The power consumption estimation device according to claim 1, further comprising a noise reduction processing unit that performs noise reduction processing, which is the process of removing data with low correlation to the total power consumption data from the time-series equipment data.

3. The power consumption estimation device according to claim 1 or 2, wherein the time sampling processing unit outputs the data obtained by resampling the equipment data to time series data with the same sampling period as the total power consumption data if the equipment data is time series data, and outputs the data obtained by converting the equipment data to time series data such that it is 1 during the period from the event occurrence time to the end time and 0 outside of that period, if the equipment data is log data, as time series equipment data.

4. The power consumption estimation device according to claim 2, wherein the noise reduction processing unit removes data in which the correlation coefficient calculated by analyzing the correlation between the total power consumption data and the time-series equipment data is lower than a predetermined threshold.

5. The power consumption estimation device according to any one of claims 1 to 4, wherein the process mining unit includes: a state extraction unit that extracts the state of each of the plurality of devices from the time-series device data; a transition relationship analysis unit that detects the transition relationship between the states; and a state transition time-series data output unit that outputs state transition time-series data representing the state transitions of the plurality of devices in time-series data based on the states and the transition relationship.

6. The power consumption estimation device according to claim 5, wherein the transition relationship analysis unit calculates the probability of occurrence of each of the states extracted by the state extraction unit based on the frequency of occurrence of each of the states, determines that a transition from a state whose probability is equal to or greater than a predetermined probability threshold to another state whose probability is equal to or greater than the predetermined probability threshold is a high-frequency transition, and outputs the transition from a state whose probability is equal to or greater than the predetermined probability threshold to another state whose probability is equal to or greater than the predetermined probability threshold as the transition relationship to the state transition time series data output unit.

7. The power consumption estimation device according to any one of claims 1 to 6, further comprising a data expansion unit that performs a conversion process that can be processed with a single time series data and a conversion process that can be processed with a combination of arbitrarily extracted time series data on the time series equipment data to be processed, and adds the data to the time series equipment data to be processed.

8. The power consumption estimation device according to claim 2 or 4, further comprising an optimization processing unit that feeds back the error between the sum of the estimated power consumption values ​​for each device and the total power consumption data from the power consumption estimation unit, and instructs the noise reduction processing unit to redo the noise reduction processing so that the error becomes smaller.

9. The power consumption estimation device according to claim 8, wherein the optimization processing unit creates a selection list of equipment data for each of the equipment, instructs the noise reduction processing unit to perform the noise reduction processing according to the selection list, and modifies the selection list to search for a selection list in which the error is less than or equal to a predetermined reference value.

10. A power consumption estimation method performed by a computer, comprising: a step of converting multiple device data obtained from multiple devices into time-series device data; a step of analyzing the state transitions of each of the multiple devices based on the time-series device data and outputting state transition time-series data indicating the operating state related to the power consumption of each of the multiple devices; and a step of estimating the power consumption of each of the multiple devices based on the total power consumption data of the multiple devices and the state transition time-series data.

11. A power consumption estimation program that causes a computer to perform the following steps: convert multiple device data acquired from multiple devices into time-series device data; analyze the state transitions of each of the multiple devices based on the time-series device data and output state transition time-series data indicating the operating state related to the power consumption of each of the multiple devices; and estimate the power consumption of each of the multiple devices based on the total power consumption data of the multiple devices and the state transition time-series data.

Images