Operation optimization method, operation optimization device, and operation optimization program
The operational optimization method addresses the complexity and cost of conventional simulation models by extracting improved numerical value sets and analyzing parameters, achieving efficient and explainable optimization of plant operations.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- IHI CORP
- Filing Date
- 2025-12-23
- Publication Date
- 2026-07-02
AI Technical Summary
Conventional simulation models for optimizing plant operations are complex, computationally expensive, and impractical for custom-made products, while AI simulations lack explanatory power.
An operational optimization method that includes acquiring a dataset, setting indicators, extracting improved numerical value sets, and analyzing parameters to optimize system operation, reducing computational costs and ensuring explainability.
Reduces computational costs and improves explainability in optimizing plant operations, enabling efficient optimization even for custom-made products and facilities with significant operational responsibility.
Smart Images
Figure JP2025044995_02072026_PF_FP_ABST
Abstract
Description
Driving optimization method, driving optimization device, and driving optimization program
[0001] This disclosure relates to a method for optimizing driving, a device for optimizing driving, and a program for optimizing driving.
[0002] Technologies for simulating the operation of plants and the like are known. Patent Document 1 describes a technology for adjusting a statistical model using measurement data from a plant. Patent Document 2 describes a technology for estimating the operating performance for a target period based on environmental data showing the operating environment and performance data showing the operating results, and for evaluating the actual values against the estimated values.
[0003] Japanese Patent Publication No. 2012-141862 Japanese Patent Publication No. 2022-169825
[0004] Simulations using conventional models tend to result in large and complex models. Conventional models also have high simulation loads and increased computational costs. Furthermore, for custom-made products, building individual models using conventional methods can be impractical.
[0005] Another example is simulation using AI (Artificial Intelligence). However, AI simulations lack explanatory power.
[0006] This disclosure describes a technology that can reduce the computational cost of data used to optimize the operation of a target system while ensuring explainability.
[0007] An operational optimization method relating to one aspect of this disclosure is performed by a computer. The operational optimization method comprises: an acquisition step of acquiring a dataset that includes multiple parameters and multiple operational data other than the multiple parameters and shows the operational performance of the target system; an indicator setting step of setting a set of numerical values for indicators using at least a portion of the multiple operational data; a first extraction step of extracting a first numerical value set that is included in a first period which is a predetermined period, and a second numerical value set that is included in a second period which is a part of the first period, from the numerical value set of indicators; a second extraction step of extracting a third numerical value set that has improved values from the first numerical value set, which is a set of numerical values that is better than the second numerical value set; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset which is a dataset corresponding to the third numerical value set and a second dataset which is a dataset corresponding to the second numerical value set; and an output step of outputting result data that includes change information for the parameters to be optimized.
[0008] This disclosure provides a technology that can reduce the computational cost of data used to optimize the operation of the target system while ensuring explainability.
[0009] Figure 1 is a block diagram showing an example of the overall configuration including the driver optimization system. Figure 2 is a diagram showing an example of the data used in the driver optimization system. Figure 3 is a diagram showing an example of determining the state of the target system. Figure 4 is a flowchart showing an example of the operation of the driver optimization device. Figure 5 is a diagram showing an example of the hardware configuration related to the driver optimization system.
[0010] An operational optimization method relating to one aspect of this disclosure is performed by a computer. The operational optimization method comprises: an acquisition step of acquiring a dataset that includes multiple parameters and multiple operational data other than the multiple parameters and shows the operational performance of the target system; an indicator setting step of setting a set of numerical values for indicators using at least a portion of the multiple operational data; a first extraction step of extracting a first numerical value set that is included in a first period which is a predetermined period, and a second numerical value set that is included in a second period which is a part of the first period, from the numerical value set of indicators; a second extraction step of extracting a third numerical value set that has improved values from the first numerical value set, which is a set of numerical values that is better than the second numerical value set; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset which is a dataset corresponding to the third numerical value set and a second dataset which is a dataset corresponding to the second numerical value set; and an output step of outputting result data that includes change information for the parameters to be optimized.
[0011] An operational optimization device relating to one aspect of this disclosure includes: an acquisition unit that acquires a dataset showing the operational performance of a target system, which includes a plurality of parameters and a plurality of operational data other than the plurality of parameters; an indicator setting unit that sets a group of numerical values for an indicator using at least a portion of the plurality of operational data; a first extraction unit that extracts a first numerical value group included in a first period, which is a predetermined period, and a second numerical value group included in a second period, which is a part of the first period, from the group of numerical values for the indicator; a second extraction unit that extracts a third numerical value group from the first numerical value group, which is a group of numerical values that have improved values compared to the second numerical value group; an analysis unit that analyzes the parameters to be optimized based on the discrepancy between a third dataset, which is a dataset corresponding to the third numerical value group, and a second dataset, which is a dataset corresponding to the second numerical value group; and an output unit that outputs result data including change information for the parameters to be optimized.
[0012] An operational optimization program relating to one aspect of this disclosure includes an acquisition step of acquiring a dataset that shows the operational performance of a target system, which includes multiple parameters and multiple operational data other than the multiple parameters; an indicator setting step of setting a set of numerical values for indicators using at least a portion of the multiple operational data; a first extraction step of extracting a first numerical value set that is included in a first period, which is a predetermined period, and a second numerical value set that is included in a second period, which is a part of the first period, from the set of numerical values for indicators; a second extraction step of extracting a third numerical value set that has improved values from the first numerical value set, which is a set of numerical values that is better than the second numerical value set; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset, which is a dataset corresponding to the third numerical value set, and a second dataset, which is a dataset corresponding to the second numerical value set; and an output step of outputting result data that includes change information for the parameters to be optimized.
[0013] In the operational optimization method, operational optimization device, and operational optimization program relating to one aspect of this disclosure, a dataset containing multiple parameters and multiple operational data is acquired. A set of numerical indicators is set based on the multiple operational data. From the set of numerical indicators, a first numerical group for the first period and a second numerical group for the second period are extracted. Then, a third numerical group having improved values compared to the second numerical group is extracted from the first numerical group. Result data is output that includes change information for the parameters targeted for optimization, analyzed based on the deviation between the third dataset and the second dataset. That is, a subset of the first numerical group, the third numerical group, having improved values compared to the second numerical group, is extracted from the overall first numerical group, and the parameters causing the deviation between the third numerical group and the second numerical group are analyzed. The result data based on the deviation can also be said to be data aiming for champion data from when efficiency was high. In other words, the result data is data that can optimize the operation of the target system. This makes it possible to generate data that optimizes the operation of the target system. Furthermore, since the processing method disclosed herein eliminates the need to create models, computational costs can be reduced, and result data that takes individual changes into account can be generated even for subjects where individual model construction is difficult (e.g., custom-made products). In addition, since the processing method disclosed herein generates result data using a rule-based method, explainability can be improved compared to simulations using only AI. Therefore, the processing method disclosed herein can be applied even to facilities with significant operational responsibility, such as plants. As a result, the computational costs of data for optimizing the operation of the target system can be reduced while ensuring explainability.
[0014] The second extraction step may involve selecting the groups of values that are higher than those in the second group from the first group of values, in descending order, to extract the third group of values. The extracted third group of values can be said to represent the top-performing data. By reflecting the data from when efficiency was better in the result data, the operation of the target system can be further optimized.
[0015] The operational optimization method may further include a clustering step before the analysis step, in which the third dataset is divided into clusters using multiple parameters. The analysis step may analyze the discrepancy between the third dataset and the second dataset for each cluster. Here, even if the indicator values are the same, different controls may be in place. By clustering using multiple parameters, the third dataset can be divided into clusters representing similar control conditions. Then, by reflecting the analysis of the discrepancies for each cluster in the result data, the operation of the target system can be further optimized.
[0016] The first extraction step may determine whether the target system is in a stable or transient state based on the first dataset, which is a dataset corresponding to the first set of numerical values, and the second dataset. The first extraction step does not need to acquire additional data included in the second dataset when the target system is in a stable state. The first extraction step may acquire additional past datasets and include them in the second dataset when the target system is in a transient state. The state variables of the target system have a time-dependent property, especially when it is in a transient state. When the target system is in a transient state, the second dataset is acquired by considering past data. This allows for more accurate processing, such as analysis, using the second dataset. Also, when the target system is in a stable state, the second dataset is not acquired. This reduces the storage capacity and computational cost of the second dataset.
[0017] The first extraction step may involve calculating distance values and contribution values (the degree to which they influence distance values) using the Mahalanobis-Taguchi Method, and then determining whether the target system is in a stable or transient state based on the time-series pattern of the contribution values. Depending on the state of the target system, the feature space composed of multiple parameters changes. Consequently, the rank and value of the contribution values also change over time. By using this time-series pattern of contribution values, the state of the target system can be determined with high accuracy.
[0018] The first extraction step may involve narrowing down the first dataset, which corresponds to the first set of numerical data, by using similarity to the second dataset as an extraction criterion, and then extracting the first set of numerical data. By reflecting the first set of numerical data extracted in this way in the result data, the accuracy of optimizing the operation of the target system can be improved.
[0019] The analysis step may involve calculating distance values and contributions that influence those distance values using the Mahalanobis-Taguchi Method, and then identifying parameters to be optimized by selecting parameters with high contributions within intervals where the distance values are large. By identifying parameters to be optimized based on distance values and contributions, it is possible to quickly discover items that need improvement from a large number of parameters.
[0020] The output step may output the result data to the target system for use in feedback control of the target system. By using the result data for feedback control of the target system, the operation of the target system can be quickly optimized.
[0021] Embodiments of this disclosure will be described below with reference to the drawings. In the description of the drawings, the same elements will be denoted by the same reference numerals, and redundant descriptions will be omitted.
[0022] Figure 1 is a block diagram showing an example of the overall configuration including the driver optimization system 1. For example, the driver optimization system 1 includes a driver optimization device 10. The driver optimization device 10 is connected to the target system 20, an external system 30, and a terminal 40 in a communicative manner.
[0023] The target system 20 is, for example, a plant, but is not limited to that. In this disclosure, an example is given in which the target system 20 is a coal-fired boiler plant. The target system 20 includes, for example, a plurality of devices, a control device that controls the plurality of devices, and a plurality of sensors that measure the status of the plurality of devices. The plurality of devices include, for example, a boiler, a mill, and a soot blower, but is not limited to these. The target system 20 transmits an internal dataset, which is a collection of data that can be acquired within the target system 20, to the operation optimization device 10. The data included in the internal dataset is not limited. In one example, the internal dataset may include time information, various parameters, material information, fuel information (e.g., fuel type, moisture content, quality, and input amount), measured values from the plurality of sensors, yield, and power consumption. The internal dataset may also include external environmental items such as outside temperature.
[0024] The external system 30 is, for example, a weather forecasting system, but is not limited to that. The external system 30 transmits an external dataset, which is a collection of data obtainable outside the target system 20, to the operation optimization device 10. The data included in the external dataset is not limited. For example, the external dataset may include time information, outside temperature, humidity, and weather information.
[0025] Terminal 40 is one or more computers used by the user of the driving optimization system 1. The type of terminal 40 is not limited. For example, terminal 40 may be a personal computer, a high-function mobile phone (smartphone), a tablet device, or a wearable device.
[0026] The operation optimization device 10 is a device that outputs information to optimize the operation of the target system 20. The information to optimize the operation may also be information that improves the operating efficiency of the target system 20. The operation optimization device 10 includes, as functional elements, an acquisition unit 11, a database 12 (storage unit), an indicator setting unit 13, a first extraction unit 14, a second extraction unit 15, a clustering unit 16, an analysis unit 17, and an output unit 18.
[0027] Database 12 is a non-temporary storage medium or storage device that stores various information used by the driving optimization system 1. Database 12 may be constructed as a single database or as a collection of multiple databases. The location of database 12 is not limited. For example, database 12 may be located in a computer system separate from the driving optimization system 1.
[0028] The acquisition unit 11 acquires a dataset that includes multiple parameters and multiple operational data other than the multiple parameters, showing the operating performance of the target system 20. The multiple parameters are control variables or manipulated variables for operating the target system 20. The multiple parameters may also include values calculated based on the measured values of multiple sensors.
[0029] The acquisition unit 11 acquires an internal dataset from the target system 20. The acquisition unit 11 acquires an external dataset from the external system 30. The acquisition unit 11 uses the internal dataset and the external dataset to create a dataset containing multiple parameters and multiple operational data. For example, the acquisition unit 11 may create a dataset by combining the internal dataset and the external dataset based on time information. The acquisition unit 11 stores the dataset in the database 12. When performing real-time processing, the dataset does not need to be stored in the database 12, or subsequent processing may be performed in parallel with storage in the database 12.
[0030] The acquisition period and acquisition interval of the dataset are not limited. For example, the acquisition unit 11 may acquire the dataset every 10 minutes for one year, or every minute for six months. In another example, the acquisition unit 11 may acquire the dataset at any time series, and retrospectively, without being limited to fixed intervals.
[0031] The indicator setting unit 13 sets a set of numerical values for the indicators to be improved using at least a portion of the operational data from multiple sources. The indicators may be information indicating the operating efficiency of the target system 20. Examples of indicators include heat loss, power generation, power consumption, and CO2. ２Examples include, but are not limited to, emissions or steam volume. "Indicator to be improved" refers to the indicator to be improved. The indicator setting unit 13 may accept user input to select the indicator to be improved. The indicator setting unit 13 may set a set of numerical values for the indicator for each time information. The indicator setting unit 13 may store the set of numerical values for the indicator in the database 12. The type of operational data used for calculation may vary depending on the indicator. In one example, the indicator setting unit 13 may calculate heat loss using measured values from multiple sensors, etc. In another example, the indicator setting unit 13 may use power consumption, etc., to calculate CO ２ You may calculate the emissions.
[0032] The indicator setting unit 13 may set specific operational data within the dataset as the indicator. In this case, the indicator setting unit 13 sets a set of numerical values for the specific operational data as the numerical value set for the indicator. For example, the indicator setting unit 13 may set a set of numerical values for yield, which is specific operational data, as the numerical value set for the indicator. In other words, if the indicator is specific operational data, the indicator setting unit 13 does not need to calculate a new numerical value set.
[0033] The first extraction unit 14 extracts from the set of numerical values of the indicator a first numerical group that is included in a predetermined period, the first period, and a second numerical group that is included in a second period that is a part of the first period. The first period may be the entire period (for example, one year or six months). The second period may be a period that includes the latest time information (for example, the most recent week).
[0034] The first extraction unit 14 may determine the state of the target system 20 based on the first dataset, which is a dataset corresponding to the first numerical group, and the second dataset, which is a dataset corresponding to the second numerical group. The "dataset corresponding to the numerical group" can be said to be a dataset obtained by reverse lookup from the numerical group. The first dataset is the dataset for the first period. The second dataset is the dataset for the second period.
[0035] The first extraction unit 14 may determine the state of the target system 20 based on the divergence between the first dataset and the second dataset. The first extraction unit 14 may calculate a distance value and a contribution degree using the MT method (Mahalanobis-Taguchi Method). The contribution degree is the degree to which each feature quantity (such as a plurality of parameters) affects the distance value. The first extraction unit 14 may determine the state of the target system 20 based on the time-series pattern of the contribution degree. For example, the first extraction unit 14 may determine a portion where the contribution degree with a large specific gravity has not changed over time as a stable state (steady state). The first extraction unit 14 may determine a portion where the contribution degree with a large specific gravity has changed over time as a transient state. The magnitude of the specific gravity may be determined by a predetermined threshold value. For example, the first extraction unit 14 may determine that the specific gravity is large when the ratio occupied by a specific contribution degree exceeds a predetermined threshold value or is equal to or greater than a predetermined threshold value in the total contribution degree at a certain time.
[0036] The first extraction unit 14 may additionally acquire data included in the second dataset according to the state of the target system 20. For example, at a time when the target system 20 is in a stable state, the first extraction unit 14 may use the second numerical group as it is. That is, the first extraction unit 14 may not need to additionally acquire the data of the second numerical group. At a time when the target system 20 is in a transient state, the first extraction unit 14 may additionally acquire a dataset of past data within a predetermined time width and include it in the second dataset. The predetermined time width is, for example, the past five time points, but is not limited thereto.
[0037] The first extraction unit 14 may narrow down the first data set with the condition that it is similar to the operation results in the second period as the extraction condition. "Similar" may mean that the first numerical group follows a normal distribution when specific conditions are within a certain range. The first numerical group may be data that follows a continuous distribution or a discrete distribution. For example, the first extraction unit 14 may extract the first numerical group when a specific numerical value in the first data set is within a predetermined range based on a specific numerical value in the second data set. In other words, the first extraction unit 14 may narrow down the first data set on the condition that the operation results in the first period and the operation results in the second period are similar. In one example, when the mill inlet temperature in the latest time information is z °C, the first extraction unit 14 may narrow down the first data set with the extraction condition that the mill inlet temperature is z ± 10 °C and extract the first numerical group. In other words, the first extraction unit 14 may extract past data when the mill inlet temperature is similar to the most recent mill inlet temperature.
[0038] The first extraction unit 14 may narrow down at least a part of the first data set with the condition that the first data set is normal and extract the first numerical group. In other words, the first extraction unit 14 may extract the first numerical group excluding abnormal data. In one example, the first extraction unit 14 may determine it as abnormal when at least a part of the first data set cannot be obtained. In another example, the first extraction unit 14 may determine it as abnormal when at least a part of the first data set contains outliers. In yet another example, the first extraction unit 14 may determine it as abnormal when information (such as a flag) indicating an error is associated with the first data set. In yet another example, the first extraction unit 14 may determine a certain period set by the user as abnormal when the user defines it as abnormal.
[0039] The second extraction unit 15 extracts a third numerical group, which is a numerical group having values improved compared to the second numerical group, from the first numerical group. "Improved value" means that the value of the index is improved. That is, the second extraction unit 15 extracts, as the third numerical group, the numerical group when the operation efficiency is better than that of the second numerical group from the first numerical group.
[0040] The second extraction unit 15 may extract a third set of numbers by selecting from the first set of numbers, in descending order of the values that are higher than those of the second set of numbers. For example, the second extraction unit 15 may sort the first set of numbers. The second extraction unit 15 may also extract a third set of numbers by selecting the top x items (x≧0) or the top y% (y≧0) of the first set of numbers.
[0041] The clustering unit 16 divides the third dataset, which corresponds to the third numerical group, into clusters based on multiple parameters. In other words, the clustering unit 16 clusters the dataset from when the operating efficiency was good based on the controlled amount. Each cluster can be said to be a group of datasets that underwent similar control. The clustering method is not limited and may be, for example, the k-means method. The third dataset can be said to be the dataset that serves as the target for improvement.
[0042] The analysis unit 17 analyzes the parameters to be optimized based on the discrepancy between the third dataset and the second dataset. The data that shows a discrepancy can be said to be data that can contribute to the improvement of the second numerical group. In other words, the data that shows a discrepancy can identify which parameters should be optimized and how those parameters should be changed. The target of the analysis may be some or all of multiple parameters.
[0043] The analysis unit 17 may perform the analysis using an anomaly diagnosis method that assumes the third dataset is in a normal state and the second dataset is in an abnormal state. Examples of analysis methods include, but are not limited to, the MT method, clustering, and deep neural networks. For example, the analysis unit 17 may use the MT method to analyze multiple parameters in order of their contribution. In one example, the analysis unit 17 may analyze that the deviation in the operating interval of a particular soot blower contributes most to the reduction in heat loss (improvement of fuel efficiency).
[0044] The discrepancy may be a one-dimensional distance value. Examples of distance values include, but are not limited to, the Mahalanobis distance. For example, the analysis unit 17 may calculate the distance value and contribution using the MT method. A larger distance value indicates a greater difference between the current driving performance and the driving performance when driving efficiency was good. A larger contribution indicates a greater potential for improving driving efficiency. The analysis unit 17 may narrow down the sections that can be improved by selecting sections with large distance values. The analysis unit 17 may identify the parameters to be optimized by selecting items (parameters) with large contributions from the selected sections. The analysis unit 17 may select the distance values or contributions to be selected in descending order, or it may use a threshold to make a determination.
[0045] The analysis unit 17 may analyze the parameters to be optimized based on the deviations between clusters in the third dataset. For example, the analysis unit 17 may analyze the parameters to be optimized based on the deviations between each cluster and the second dataset.
[0046] The output unit 18 generates and outputs result data for optimizing the operation of the target system 20 based on the deviation. The result data includes information about the parameters to be optimized. For example, the output unit 18 may output result data that includes parameter change information. In one example, the result data may include suggestions (such as increases or decreases) for a specific parameter (control quantity) as change information. The change information may be determined based on the degree of contribution. The result data can contribute to improving indicators by being used in the operation of the target system 20. The output unit 18 may output the result data to the target system 20 for use in feedback control of the target system 20. The target system 20 may perform feedback control based on the result data. The output unit 18 may output result data for each cluster. The output unit 18 may transmit the result data to the terminal 40 or other devices. The output unit 18 may display the result data on a display device provided by the operation optimization device 10 or other devices.
[0047] Figure 2 shows an example of data used in the operation optimization system 1. Figure 2 shows an example of a second data set for a period including the latest (most recent) time information. Database 12 stores multiple parameters P and multiple operational data F associated with time information T. The index setting unit 13 sets a set of numerical values for index K using at least a portion of the multiple operational data F. The index setting unit 13 may store the calculated set of numerical values for index K in database 12. Figure 2 shows heat loss as an example of index K.
[0048] In Figure 2, an index K is set for each time information T. In one example, when the time information T is 2022 / 8 / 5 0:00, the heat loss is set to L1 [kJ / kg]. Also in Figure 2, for each time information T, a dataset containing multiple parameters P and multiple operational data F is associated with the numerical value of index K.
[0049] Multiple operational data F may include first operational data F1 and second operational data F2. First operational data F1 may be data indicating values other than the state variables of the target system 20. Second operational data F2 may be data indicating the state variables of the target system 20. Examples of first operational data F1 include, but are not limited to, set output and air-fuel ratio. Examples of second operational data F2 include, but are not limited to, coal flow rate, outlet temperature, inlet temperature, ambient temperature, and predetermined sensor values.
[0050] Figure 2 shows data D1 from 0:00 to 0:10 on 2022 / 8 / 5. Figure 2 also shows data D2 from 2:10 to 2:20 on 2022 / 8 / 5. Data D1 and D2 include the second operational data F2 (state variable) for the corresponding time. Data D1 and D2 also include the parameter P (control variable) for the corresponding time.
[0051] In Figure 2, the control quantity for data D1 is the same as the control quantity for data D2. Specifically, the control quantity from 0:00 on 2022 / 8 / 5 to 0:10 on 2022 / 8 / 5 is the same as the control quantity from 2:10 on 2022 / 8 / 5 to 2:20 on 2022 / 8 / 5.
[0052] In FIG. 2, the state quantity of data D1 and the state quantity of data D2 are different. Specifically, the state quantity from 0:00 on August 5, 2022 to 0:10 on August 5, 2022 is different from the state quantity from 2:10 on August 5, 2022 to 2:20 on August 5, 2022. Thus, even if the control quantity of data D1 and the control quantity of data D2 are the same, the state quantity of data D1 and the state quantity of data D2 are not necessarily the same.
[0053] The state quantity may have a property that depends on time. For example, when the target system 20 is in a transient state, the state quantity may have a property that depends on time. In one example, the state quantity S of the target system 20 at time t + 1 ｔ＋１ may not be determined only by the current state quantity S ｔ and the current control quantity U. ｔ The state quantity S ｔ＋１ may need to further consider the past state quantities S ｔ－１ , …, S ｔ－ｍ and the past control quantities U ｔ－１ , …, U ｔ－ｎ as well (m ≥ 1, n ≥ 1). The state quantity S ｔ＋１ can be expressed, for example, as follows. Here, f is a function of the state quantity and the control quantity. S ｔ＋１ = f(S ｔ , …, S ｔ－ｍ , U ｔ , …, U ｔ－ｎ )
[0054] FIG. 3 is a diagram showing an example of determining the state of the target system 20. FIG. 3 shows the divergence between the first data set and the second data set in a period (the second period) including the latest (most recent) time information. The first extraction unit 14 may calculate a distance value V and a contribution degree W for each time information T. The distance value V is an example of the divergence between the first data set and the second data set. The contribution degree W includes, for example, contribution degrees W1 to W4. The contribution degrees W1 to W4 indicate the degree to which the corresponding parameters, etc. affect the distance value V. The database 12 may store the distance value V and the contribution degree W in association with the time information T. The first extraction unit 14 may determine the state of the target system 20 based on the time series pattern of the contribution degree W.
[0055] Of the time periods shown in Figure 3, the distance value V is particularly large during the periods from 0:20 to 0:40 on 8 / 5 / 2022 and from 1:30 to 1:50 on 8 / 5 / 2022. The first extraction unit 14 may use a threshold or the like to identify the time periods in which the distance value V is particularly large. The time periods in which the distance value V is particularly large may be candidate times for optimizing the operation of the target system 20.
[0056] The first extraction unit 14 may determine that the portion (range R1) in which the contribution W with a high specific gravity does not change over time is a stable state J1. For example, from 0:20 on 8 / 5 / 2022 to 0:40 on 8 / 5 / 2022, the specific gravity of contributions W2 and W3 is high, while the specific gravity of contributions W1 and W4 is low. Within range R1, the values of contributions W2 and W3 are consistently higher than the values of contributions W1 and W4. That is, the contributions W2 and W3 with a high specific gravity do not change over time. In this case, the first extraction unit 14 may determine that the state of the target system 20 is a stable state J1.
[0057] The first extraction unit 14 may determine that a portion where there are no contributions W with high specific gravity, or where the Mahalanobis distance value is small, is a stable state J1. "There are no contributions W with high specific gravity" may mean, for example, that the variance of contributions W is smaller than a predetermined threshold, or that the Mahalanobis distance value is small. There are no contributions with significantly high specific gravity from 0:00 to 0:10 on 8 / 5, 2022. Similarly, there are no contributions with high specific gravity from 0:50 to 1:20 on 8 / 5, 2022, and from 2:00 to 2:40 on 8 / 5, 2022. In this case, the first extraction unit 14 may determine that the state of the target system 20 is a stable state J1.
[0058] The first extraction unit 14 may determine that the portion (range R2) in which the contribution W with the highest specific gravity changes over time is a transient state J2. For example, at 1:30 on 8 / 5 / 2022, the specific gravity of contributions W1 and W4 is high, and the specific gravity of contributions W2 and W3 is low. At 1:40 on 8 / 5 / 2022, the specific gravity of contributions W2 and W3 is high, and the specific gravity of contributions W1 and W4 is low. At 1:50 on 8 / 5 / 2022, the specific gravity of contributions W2 and W4 is high, and the specific gravity of contributions W1 and W3 is low. In other words, from 1:30 to 1:50 on 8 / 5 / 2022, the contributions W1 to W4 with the highest specific gravity change over time. In this case, the first extraction unit 14 may determine that the state of the target system 20 is a transient state J2.
[0059] The first extraction unit 14 may acquire additional data included in the second dataset depending on the state of the target system 20. For example, at a time when the target system 20 is in a stable state J1, the first extraction unit 14 may use the second numerical data set as is. That is, the first extraction unit 14 does not need to acquire additional data from the second numerical data set. At a time when the target system 20 is in a transient state J2, the first extraction unit 14 may acquire additional past dataset data for a predetermined time range (for example, the past 5 time ranges) and include it in the second dataset. The first extraction unit 14 may classify the type of transient state J2 using a pattern of contribution W. The first extraction unit 14 may acquire additional past dataset data for a variable time range corresponding to the type of transient state J2 and include it in the second dataset.
[0060] [Driving Optimization Method] An example of the operation method (driving optimization method) of the driving optimization device 10 will be explained with reference to Figure 4. Figure 4 is a flowchart showing an example of the operation of the driving optimization device 10.
[0061] In step S1 (acquisition step), the acquisition unit 11 acquires a dataset that includes multiple parameters and multiple operational data other than the parameters, and shows the operational performance of the target system 20. For example, the acquisition unit 11 acquires an internal dataset from the target system 20. The acquisition unit 11 acquires an external dataset from the external system 30. The acquisition unit 11 uses the internal dataset and the external dataset to create a dataset that includes multiple parameters and multiple operational data. For example, the acquisition unit 11 may create a dataset by combining the internal dataset and the external dataset based on time information. The acquisition unit 11 stores the dataset in the database 12. When performing real-time processing, the dataset does not need to be stored in the database 12, or subsequent processing may be performed in parallel with storage in the database 12.
[0062] In step S2 (metric setting step), the metric setting unit 13 sets a set of numerical values for the metric using at least a portion of the multiple operational data. The metric setting unit 13 may accept user input to select an metric. The metric setting unit 13 may set a set of numerical values for the metric for each time information. The metric setting unit 13 may store the set of numerical values for the metric in the database 12.
[0063] The indicator setting unit 13 may set specific operational data within the dataset as indicators. In this case, the indicator setting unit 13 sets a set of numerical values for the specific operational data as the numerical value set for the indicator.
[0064] In step S3 (first extraction step), the first extraction unit 14 extracts from the group of index values a first group of values included in a predetermined period, the first period, and a second group of values included in a second period, which is a part of the first period. The first period may be the entire period (for example, one year or six months). The second period may be a period that includes the latest time information (for example, the most recent week).
[0065] In step S4 (determination step), the first extraction unit 14 may determine the state of the target system 20 based on the first dataset, which is a dataset corresponding to the first numerical group, and the second dataset, which is a dataset corresponding to the second numerical group. The first extraction unit 14 may also determine the state of the target system 20 based on the discrepancy between the first dataset and the second dataset (see, for example, Figure 3). Alternatively, the first extraction unit 14 may determine whether the target system 20 is in a stable state or a transient state using a pattern such as the components (contribution) of the discrepancy.
[0066] The first extraction unit 14 may acquire additional data included in the second dataset depending on the state of the target system 20. For example, at a time when the target system 20 is in a stable state, the first extraction unit 14 may use the second numerical data set as is. That is, the first extraction unit 14 does not need to acquire additional data from the second numerical data set. At a time when the target system 20 is in a transient state, the first extraction unit 14 may acquire additional past datasets of a predetermined or variable time width and include them in the second dataset.
[0067] In step S5 (second extraction step), the second extraction unit 15 extracts a third numerical group from the first numerical group, which is a numerical group having values that are better than those of the second numerical group. That is, the second extraction unit 15 extracts the numerical group from the first numerical group that represents a better operating efficiency than the second numerical group as the third numerical group. The second extraction unit 15 may also extract the third numerical group by selecting numerical groups from the first numerical group that have values that are better than those of the second numerical group, in descending order.
[0068] In step S6 (clustering step), the clustering unit 16 divides the third dataset, which is a dataset corresponding to the third numerical group, into clusters based on multiple parameters. That is, the clustering unit 16 clusters the dataset from when the operating efficiency was good based on the controlled variable.
[0069] In step S7 (analysis step), the analysis unit 17 may analyze the parameters to be optimized based on the discrepancy between the third dataset and the second dataset. The analysis may target some or all of the parameters. The analysis unit 17 may perform the analysis using an anomaly diagnosis method that assumes the third dataset is in a normal state and the second dataset is in an abnormal state. The analysis unit 17 may also analyze the discrepancy for each cluster of the third dataset. For example, the analysis unit 17 may analyze the discrepancy between each cluster and the second dataset.
[0070] In step S8 (output step), the output unit 18 generates and outputs result data for optimizing the operation of the target system 20 based on the deviation. The result data includes information about the parameters to be optimized. For example, the output unit 18 may output result data that includes parameter change information. In one example, the result data may include suggestions (such as increases or decreases) for a specific parameter (control quantity) as change information. The change information may be determined based on the degree of contribution. The result data can contribute to improving the indicators by being used in the operation of the target system 20. The output unit 18 may output the result data to the target system 20 for use in feedback control of the target system 20. The target system 20 may perform feedback control based on the result data. The output unit 18 may output result data for each cluster. The output unit 18 may transmit the result data to the terminal 40 or other devices. The output unit 18 may display the result data on a display device provided by the operation optimization device 10 or other devices.
[0071] [Hardware Configuration] Figure 5 shows an example of the hardware configuration related to the driving optimization system 1. Figure 5 shows a computer 100 that functions as a driving optimization device 10. The computer 100 has a processor 101, a main memory unit 102, an auxiliary memory unit 103, a communication control unit 104, an input device 105, and an output device 106. The driving optimization device 10 is composed of one or more computers 100 which consist of this hardware and software such as programs.
[0072] If the driving optimization device 10 is composed of multiple computers 100, these computers 100 may be connected locally or via a communication network such as the Internet or an intranet. This connection logically constructs a single driving optimization device 10.
[0073] The processor 101 is a CPU (Central Processing Unit) that executes the operating system and application programs. The main memory unit 102 consists of ROM (Read Only Memory) and RAM (Random Access Memory). The auxiliary memory unit 103 is a storage medium consisting of a hard disk and flash memory. The auxiliary memory unit 103 generally stores a larger amount of data than the main memory unit 102. The communication control unit 104 consists of a network card or a wireless communication module. At least a part of the communication function with other devices in the driving optimization device 10 may be realized by the communication control unit 104. The input device 105 consists of a keyboard, mouse, touch panel, and microphone for voice input. The output device 106 consists of a display and printer.
[0074] The auxiliary storage unit 103 stores the program 110 (driving optimization program) and data necessary for processing in advance. The program 110 causes the computer 100 to execute each functional element of the driving optimization device 10. The program 110 causes the computer 100 to execute, for example, the processing related to the driving optimization method described above. For example, the program 110 is read by the processor 101 or the main memory unit 102 and operates at least one of the processor 101, the main memory unit 102, the auxiliary storage unit 103, the communication control unit 104, the input device 105, and the output device 106. For example, the program 110 reads and writes data to the main memory unit 102 and the auxiliary storage unit 103.
[0075] The program 110 may be provided on a computer-readable storage medium. Examples of storage mediums include, but are not limited to, CD-ROMs, DVD-ROMs, and semiconductor memory. The program 110 may also be provided as a data signal via a communication network.
[0076] As described above, the operational optimization method relating to one aspect of this disclosure is performed by a computer. The operational optimization method includes: an acquisition step of acquiring a dataset that shows the operational performance of the target system 20, which includes a plurality of parameters and a plurality of operational data other than the plurality of parameters; an indicator setting step of setting a set of numerical values for indicators using at least a portion of the plurality of operational data; a first extraction step of extracting a first numerical value set that is included in a first period which is a predetermined period, and a second numerical value set that is included in a second period which is a part of the first period, from the set of numerical values for indicators; a second extraction step of extracting a third numerical value set that has improved values from the first numerical value set, which is a set of numerical values that is better than the second numerical value set; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset which is a dataset corresponding to the third numerical value set and a second dataset which is a dataset corresponding to the second numerical value set; and an output step of outputting result data that includes change information for the parameters to be optimized.
[0077] An operation optimization device 10 relating to one aspect of this disclosure includes: an acquisition unit 11 that acquires a dataset showing the operating performance of a target system 20, which includes a plurality of parameters and a plurality of operational data other than the plurality of parameters; an indicator setting unit 13 that sets a group of numerical values for indicators using at least a portion of the plurality of operational data; a first extraction unit 14 that extracts a first numerical value group included in a first period, which is a predetermined period, and a second numerical value group included in a second period, which is a part of the first period, from the group of numerical values for indicators; a second extraction unit 15 that extracts a third numerical value group from the first numerical value group, which is a group of numerical values that have improved values compared to the second numerical value group; an analysis unit 17 that analyzes the parameters to be optimized based on the discrepancy between a third dataset, which is a dataset corresponding to the third numerical value group, and a second dataset, which is a dataset corresponding to the second numerical value group; and an output unit 18 that outputs result data including change information for the parameters to be optimized.
[0078] An operational optimization program relating to one aspect of this disclosure includes an acquisition step of acquiring a dataset showing the operational performance of a target system 20, which includes multiple parameters and multiple operational data other than the multiple parameters; an indicator setting step of setting a set of numerical values for indicators using at least a portion of the multiple operational data; a first extraction step of extracting a first numerical value set included in a first period, which is a predetermined period, and a second numerical value set included in a second period, which is a part of the first period, from the set of numerical values for indicators; a second extraction step of extracting a third numerical value set, which is a set of numerical values that has improved values compared to the second numerical value set, from the first numerical value set; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset, which is a dataset corresponding to the third numerical value set, and a second dataset, which is a dataset corresponding to the second numerical value set; and an output step of outputting result data including change information for the parameters to be optimized.
[0079] In the operational optimization method, operational optimization device, and operational optimization program relating to one aspect of this disclosure, a dataset containing multiple parameters and multiple operational data is acquired. A set of numerical indicators is set based on the multiple operational data. From the set of numerical indicators, a first numerical group for the first period and a second numerical group for the second period are extracted. Then, a third numerical group having improved values compared to the second numerical group is extracted from the first numerical group. Result data is output that includes change information for the parameters targeted for optimization, analyzed based on the deviation between the third dataset and the second dataset. That is, a subset of the first numerical group, the third numerical group, having improved values compared to the second numerical group, is extracted from the overall first numerical group, and the parameters causing the deviation between the third numerical group and the second numerical group are analyzed. The result data based on the deviation can also be said to be data aiming for champion data from when efficiency was good. In other words, the result data is data that can optimize the operation of the target system 20. This makes it possible to generate data that optimizes the operation of the target system 20. Furthermore, since the processing method disclosed herein eliminates the need to create a model, computational costs can be reduced, and result data that takes individual changes into account can be generated even for subjects where individual model construction is difficult (e.g., custom-made products). In addition, since the processing method disclosed herein generates result data using a rule-based method, explainability can be improved compared to simulations using AI alone. Therefore, the processing method disclosed herein can be applied even to facilities with significant operational responsibility, such as plants. As a result, the computational costs of data for optimizing the operation of the target system 20 can be reduced while ensuring explainability.
[0080] The second extraction step extracts a third set of numerical values by selecting from the first set of numerical values, in descending order of their value compared to the second set of numerical values. The extracted third set of numerical values can be said to represent the top-performing data. By reflecting the data from when efficiency was better in the result data, the operation of the target system 20 can be further optimized.
[0081] The operation optimization method further includes a clustering step before the analysis step, in which the third dataset is divided into clusters using multiple parameters. The analysis step analyzes the discrepancy between the third dataset and the second dataset for each cluster. Here, even if the indicator values are the same, different controls may be in place. By clustering using multiple parameters, the third dataset can be divided into clusters representing similar control conditions. Then, by reflecting the analysis of the discrepancies for each cluster in the result data, the operation of the target system 20 can be further optimized.
[0082] The first extraction step determines whether the target system 20 is in a stable state or a transient state based on the first dataset, which is a dataset corresponding to the first set of numerical values, and the second dataset. The first extraction step does not acquire additional data included in the second dataset when the target system 20 is in a stable state. The first extraction step acquires additional past datasets and includes them in the second dataset when the target system 20 is in a transient state. The state variables of the target system 20 have a time-dependent property, especially when it is in a transient state. When the target system 20 is in a transient state, the second dataset is acquired by considering past data. This allows for accurate processing such as analysis using the second dataset. Also, when the target system 20 is in a stable state, the second dataset is not acquired. This reduces the storage capacity and computational cost of the second dataset.
[0083] The first extraction step uses the Mahalanobis-Taguchi Method to calculate distance values and contribution values, which represent the degree to which they influence distance values. Based on the time-series pattern of contribution values, it is determined whether the target system 20 is in a stable or transient state. Depending on the state of the target system 20, the feature space composed of multiple parameters changes. Consequently, the rank and value of the contribution values also change over time. By using this time-series pattern of contribution values, the state of the target system 20 can be determined with high accuracy.
[0084] The first extraction step narrows down the first dataset, which corresponds to the first set of numerical data, by using similarity to the second dataset as an extraction criterion, and extracts the first set of numerical data. By reflecting the first set of numerical data extracted in this way in the result data, the accuracy of optimizing the operation of the target system 20 can be improved.
[0085] The analysis step involves calculating distance values and contributions that influence those distance values using the Mahalanobis-Taguchi Method, and then identifying parameters to be optimized by selecting parameters with high contributions within intervals where the distance value is large. By identifying parameters to be optimized based on distance values and contributions, it is possible to quickly discover items that need improvement from a large number of parameters.
[0086] The output step outputs the result data to the target system 20 for use in feedback control of the target system 20. By using the result data for feedback control of the target system 20, the operation of the target system 20 can be quickly optimized.
[0087] [Modifications] This disclosure is not necessarily limited to the embodiments described above, and various modifications are possible without departing from the spirit of the disclosure.
[0088] In the above embodiment, an example was described in which a statistical method such as the MT method was used to analyze the dataset, but the invention is not limited to this. For example, AI may be used to analyze the dataset. In one example, XAI (Explainable AI) may calculate the contribution of multiple parameters (control variables) to the superiority or inferiority of an indicator. XAI can accurately calculate the contribution without clustering even when the dataset corresponding to the third numerical group (the dataset when the operating efficiency was good) is non-normally distributed.
[0089] In the above embodiment, an example was described in which the operational optimization method is applied to improve the heat loss (or fuel consumption) of a coal-fired boiler plant, but it is not limited to this. The operational optimization method is applied to the CO2 of the plant. ２The method may be applied to improving emissions, minimizing environmentally regulated substances, optimizing the operation of glass melting furnaces in nuclear power plants, optimizing the operation of wing processing lines for aircraft, etc., or improving the performance of aircraft engines (e.g., improving fuel efficiency). The operation optimization method may also be applied to improving the efficiency of each production facility or line in a factory (e.g., minimizing power consumption), improving the yield of batch production (e.g., castings, vacuum furnaces, or heat-processed products), or improving the energy consumption of general-purpose compressors, etc., in a factory.
[0090] [Note] The gist of this disclosure is as follows: [1] A method for optimizing operation performed by a computer, comprising: an acquisition step of acquiring a dataset that includes a plurality of parameters and a plurality of operational data other than the plurality of parameters and shows the operational performance of a target system; an indicator setting step of setting a group of numerical values for an indicator using at least a part of the plurality of operational data; a first extraction step of extracting a first group of numerical values that are included in a first period which is a predetermined period, and a second group of numerical values that are included in a second period which is a part of the first period, from the group of numerical values for the indicator; a second extraction step of extracting a third group of numerical values that are improved from the first group of numerical values, which is a group of numerical values that are improved from the second group of numerical values; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset which is a dataset corresponding to the third group of numerical values and a second dataset which is a dataset corresponding to the second group of numerical values; and an output step of outputting result data that includes change information for the parameters to be optimized. [2] The method for optimizing operation according to [1], wherein the second extraction step extracts the third numerical group by selecting numerical groups having values that are better than the second numerical group from the first numerical group in descending order. [3] The method for optimizing operation according to [2], wherein, before the analysis step, the third dataset is further divided into clusters using the plurality of parameters, and the analysis step analyzes the discrepancy between the third dataset and the second dataset for each cluster. [4] The method for optimizing operation according to any one of [1] to [3], wherein the first extraction step determines whether the target system is in a stable state or a transient state based on the first dataset, which is a dataset corresponding to the first numerical group, and the second dataset, and does not acquire additional data included in the second dataset at times when the target system is in a stable state, and acquires additional past datasets at times when the target system is in a transient state and includes them in the second dataset.[5] The first extraction step is the method for optimizing operation according to [4], wherein the first extraction step is the method for optimizing operation according to [4], wherein the first extraction step is the method for optimizing operation according to any one of [1] to [5], wherein the first extraction step is the method for optimizing operation according to any one of [1] to [5], wherein the first extraction step is the method for optimizing operation according to any one of [1] to [5], wherein the first extraction step is the method for optimizing operation according to any one of [1] to [6], wherein the analysis step is the method for optimizing operation according to any one of [1] to [6], wherein the analysis step is the method for optimizing operation according to any one of [1] to [6], wherein the analysis step is the method for optimizing operation according to any one of [1] to [7 [9] An operation optimization device comprising: an acquisition unit that acquires a dataset showing the operating performance of a target system, including a plurality of parameters and a plurality of operational data other than the plurality of parameters; an indicator setting unit that sets a group of numerical values for an indicator using at least a portion of the plurality of operational data; a first extraction unit that extracts a first group of numerical values included in a first period, which is a predetermined period, and a second group of numerical values included in a second period, which is a part of the first period, from the group of numerical values for the indicator; a second extraction unit that extracts a third group of numerical values from the first group of numerical values, which is a group of numerical values that has improved values compared to the second group of numerical values; an analysis unit that analyzes the parameters to be optimized based on the discrepancy between a third dataset, which is a dataset corresponding to the third group of numerical values, and a second dataset, which is a dataset corresponding to the second group of numerical values; and an output unit that outputs result data including change information for the parameters to be optimized.
[10] An operation optimization program that causes a computer to perform the following steps: an acquisition step of acquiring a dataset that includes multiple parameters and multiple operational data other than the multiple parameters and shows the operational performance of the target system; an indicator setting step of setting a set of numerical values for an indicator using at least a part of the multiple operational data; a first extraction step of extracting from the set of numerical values a first numerical value that is included in a first period which is a predetermined period, and a second numerical value that is included in a second period which is a part of the first period; a second extraction step of extracting from the first numerical value a third numerical value that is a set of numerical values that is improved compared to the second numerical value; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset which is a dataset corresponding to the third numerical value and a second dataset which is a dataset corresponding to the second numerical value; and an output step of outputting result data that includes change information for the parameters to be optimized.
[0091] 1 Driving Optimization System 10 Driving Optimization Device 11 Acquisition Unit 12 Database 13 Indicator Setting Unit 14 First Extraction Unit 15 Second Extraction Unit 16 Clustering Unit 17 Analysis Unit 18 Output Unit 20 Target System 30 External System 40 Terminal T Time Information F Operation Data K Indicator F1 First Operation Data F2 Second Operation Data V Distance Value W Contribution J1 Stable State J2 Transient State 100 Computer 101 Processor 102 Main Memory Unit 103 Auxiliary Memory Unit 104 Communication Control Unit 105 Input Device 106 Output Device 110 Program
Claims
1. A computer-based method for optimizing driving, comprising: an acquisition step of acquiring a dataset that includes a plurality of parameters and a plurality of operational data other than the plurality of parameters, and shows the operational performance of a target system; an indicator setting step of setting a set of numerical values for an indicator using at least a portion of the plurality of operational data; a first extraction step of extracting a first numerical value set that is included in a first period which is a predetermined period, and a second numerical value set that is included in a second period which is a part of the first period, from the numerical value set of the indicator; a second extraction step of extracting a third numerical value set that is a set of numerical values that is improved from the first numerical value set, from the first numerical value set; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset which is a dataset corresponding to the third numerical value set and a second dataset which is a dataset corresponding to the second numerical value set; and an output step of outputting result data that includes change information for the parameters to be optimized.
2. The method for optimizing driving according to claim 1, wherein the second extraction step extracts the third numerical group by selecting from the first numerical group the numerical groups having values that are better than the second numerical group, in descending order.
3. The method for optimizing operation according to claim 2, further comprising a clustering step of dividing the third dataset into clusters using the plurality of parameters before the analysis step, wherein the analysis step analyzes the discrepancy between the third dataset and the second dataset for each cluster.
4. The method for optimizing operation according to claim 1, wherein the first extraction step determines whether the target system is in a stable state or a transient state based on a first dataset, which is a dataset corresponding to the first set of numerical values, and the second dataset; when the target system is in a stable state, no additional data included in the second dataset is acquired; and when the target system is in a transient state, past datasets are acquired and included in the second dataset.
5. The method for optimizing operation according to claim 4, wherein the first extraction step involves calculating a distance value and a contribution value which is the degree to which the distance value is influenced, using the Mahalanobis-Taguchi Method, and determining whether the target system is in a stable state or a transient state based on the time-series pattern of the contribution value.
6. The method for optimizing operation according to claim 1, wherein the first extraction step narrows down the first dataset, which is a dataset corresponding to the first set of numerical values, using similarity to the second dataset as the extraction criterion, and extracts the first set of numerical values.
7. The driving optimization method according to claim 1, wherein the analysis step involves calculating distance values and contributions that influence the distance values using the Mahalanobis-Taguchi Method, and identifying the parameters to be optimized by selecting parameters with large contributions within the intervals where the distance values are large.
8. The method for optimizing operation according to claim 1, wherein the output step outputs the result data to the target system for use in feedback control of the target system.
9. An operation optimization device comprising: an acquisition unit that acquires a dataset showing the operating performance of a target system, including multiple parameters and multiple operational data other than the multiple parameters; an indicator setting unit that sets a group of numerical values for an indicator using at least a portion of the multiple operational data; a first extraction unit that extracts a first group of numerical values included in a first period, which is a predetermined period, and a second group of numerical values included in a second period, which is a part of the first period, from the group of numerical values for the indicator; a second extraction unit that extracts a third group of numerical values from the first group of numerical values, which is a group of numerical values that are improved compared to the second group of numerical values; an analysis unit that analyzes the parameters to be optimized based on the discrepancy between a third dataset, which is a dataset corresponding to the third group of numerical values, and a second dataset, which is a dataset corresponding to the second group of numerical values; and an output unit that outputs result data including change information for the parameters to be optimized.
10. An operation optimization program that causes a computer to execute the following steps: an acquisition step of acquiring a dataset that includes multiple parameters and multiple operational data other than the multiple parameters and shows the operational performance of the target system; an indicator setting step of setting a set of numerical values for an indicator using at least a portion of the multiple operational data; a first extraction step of extracting from the set of numerical values a first numerical value that is included in a first period which is a predetermined period, and a second numerical value that is included in a second period which is a part of the first period; a second extraction step of extracting from the first numerical value set a third numerical value that is a set of numerical values that is improved compared to the second numerical value set; an analysis step of analyzing the parameters to be optimized based on the discrepancy between a third dataset which is a dataset corresponding to the third numerical value set and a second dataset which is a dataset corresponding to the second numerical value set; and an output step of outputting result data that includes change information for the parameters to be optimized.