Operational state classification device, operational state classification method, and program

Abstract

To provide an operational state classification device capable of improving operation efficiency of a power generation facility by suppressing error alarming.SOLUTION: According to an embodiment, an operational state classification device includes: a data preprocessing part for performing data preprocessing for eliminating data related to an alarm that is erroneously emitted from time series data showing an operational state of a power generation facility; a feature amount extraction part for extracting a feature amount from the data-preprocessed data; and a clustering part for classifying the operational state of the power generation facility on the basis of the feature amount.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] An embodiment of the present invention relates to a driving state classification device, a driving state classification method, and a program. [Background technology]

[0002] Power generation facilities powered by renewable energy sources, such as wind power generation, depend on the natural environment, and their operating conditions change constantly. Currently, alarms for the operation and maintenance of such power generation facilities are set based on rules and human judgment criteria. Data acquired from a supervisory control and data acquisition (SCADA) system is typically used to determine the operating conditions of multiple pieces of equipment installed in the power generation facility. However, because the amount of data generated by such systems is typically enormous, they are unable to handle complex operating conditions. As a result, false alarms, or shutdown alarms that are issued unrelated to equipment failures and that can be immediately restored after the equipment stops operating, can occur frequently. Frequent false alarms reduce the operating efficiency of the power generation facility. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6830414 [Patent Document 2] Patent No. 5945350 [Patent Document 3] Patent Publication No. 2020-153836 Summary of the Invention [Problem to be solved by the invention]

[0004] The problem to be solved by the present invention is to provide an operating state classification device, an operating state classification method, and a program that can reduce false alarms and improve the operating efficiency of power generation equipment. [Means for solving the problem]

[0005] An operating state classification device according to one embodiment includes a data pre-processing unit that performs data pre-processing to remove data related to erroneously issued alarms from time-series data indicating the operating state of the power generation equipment, a feature extraction unit that extracts features from the pre-processed data, and a clustering unit that classifies the operating state of the power generation equipment based on the features. [Effects of the Invention]

[0006] According to this embodiment, it is possible to reduce false alarms and improve the operating efficiency of the power generation facility. [Brief explanation of the drawings]

[0007] [Figure 1] 1 is a block diagram showing a configuration of a driving state classification device according to a first embodiment. [Figure 2] 5 is a flowchart showing an example of a processing procedure of a data processing unit according to the first embodiment. [Figure 3] 4 is a flowchart of data preprocessing according to the first embodiment. [Figure 4] FIG. 2 is a diagram for explaining data preprocessing according to the first embodiment. [Figure 5] An example of classification of the operating status of wind power generation equipment is shown below. [Figure 6] FIG. 1A is a diagram showing an example of time-series data of output power of a power generation facility, and FIG. 1B is a diagram showing an example of time-series data of wind speed of wind power that drives wind turbines of the power generation facility. [Figure 7] 10 is a flowchart of data preprocessing according to the second embodiment. [Figure 8] 10 is a flowchart of data preprocessing according to a modified example of the second embodiment. [Figure 9]FIG. 10 is an image diagram of excluding operation stop data in a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0008] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. The present invention is not limited to the following embodiments.

[0009] (First embodiment) 1 is a block diagram showing the configuration of a driving state classification device according to the first embodiment. The driving state classification device 1 according to this embodiment is provided in a driving state monitoring system 100 such as SCADA. However, the driving state classification device 1 may be externally connected to the driving state monitoring system 100.

[0010] The operational status monitoring system 100 is a system that monitors the operational status of a power generation facility 200. The power generation facility 200 is a facility that uses renewable energy as a power source, such as a wind power generation facility or a solar power generation facility. The power generation facility 200 is equipped with various devices that are used for generating electricity based on renewable energy, such as wind turbines.

[0011] The driving state classification device 1 is configured, for example, by a computer. This computer has hardware resources such as a CPU, a ROM, a RAM, and a HDD. The CPU executes various programs, thereby realizing software-based information processing using the hardware resources. Furthermore, the driving state classification method of this embodiment can be realized by causing the computer to execute various programs.

[0012] The configuration of the driving state classification device 1 according to this embodiment will be described below. As shown in FIG.

[0013] The database 10 accumulates data obtained from sensors 210 installed in each device of the power generation facility 200, alarm data such as warnings and emergency alerts, etc. The data obtained from the sensors 210 includes time-series data related to the operation of each device of the power generation facility 200, such as wind speed, wind direction, temperature, acceleration, and sound.

[0014] The operation unit 20 receives input operations from a user and includes input devices such as a mouse or a keyboard.

[0015] The data processing unit 30 includes a data preprocessing unit 31, a feature extraction unit 32, and a clustering unit 33. Each unit will be described below.

[0016] The data preprocessing unit 31 acquires selected data selected by a user through input operations on the operation unit 20 from the data stored in the database 10. This selected data is data that can be used to analyze and determine the operating status of each device installed in the power generation facility 200. The number of data points and the data length of the selected data are appropriately selected by the user depending on the content of the analysis and determination of the operating status, and are not particularly limited. Furthermore, the data preprocessing unit 31 performs processes such as resampling of time-series data and removal of outliers that have a negative impact on machine learning, so that machine learning can be performed efficiently while maintaining the accuracy of the selected data.

[0017] The feature extraction unit 32 analyzes the data preprocessed by the data preprocessing unit 31 and extracts features of the data.

[0018] The clustering unit 33 performs machine learning on the feature amounts extracted by the feature amount extraction unit 32 and classifies the operating states of the power generation facility 200 into clusters.

[0019] The data output unit 40 outputs the clusters of the operating states of the power generation facility 200 classified by the clustering unit 33. The data output unit 40 is configured with an image display device such as a display.

[0020] The operation of the driving state classification device 1 configured as above will be described below with reference to Fig. 2. Here, the processing contents of the data processing unit 30 will be described.

[0021] 2 is a flowchart showing an example of a processing procedure of the data processing unit 30 according to the first embodiment. In this flowchart, first, the data pre-processing unit 31 performs data pre-processing (step S1). Here, the data pre-processing will be described in detail with reference to FIG. 3.

[0022] Fig. 3 is a flowchart of data pre-processing according to the first embodiment, and Fig. 4 is a diagram for explaining data pre-processing according to the first embodiment.

[0023] 3, first, the data preprocessing unit 31 acquires selected data selected from the database 10 in response to an input operation by a user on the operation unit 20 (step S111). At this time, the selected data includes time-series data on the operating state of the equipment necessary for machine learning, which will be described later, and alarm data of the equipment.

[0024] Next, the data preprocessing unit 31 extracts alarm data from the selection data acquired in step S111 (step S112). The alarm data includes, for example, a stop alarm that warns of a stop of each device installed in the power generation facility 200, and a start alarm that warns of a restart after the stop alarm is released.

[0025] Next, it is determined whether the alarm interval is equal to or greater than a predetermined interval n (step S113). As shown in Fig. 4, the alarm interval indicates the time during which the operation of the equipment continues to be stopped from when a stop alarm is issued until a start alarm is issued. If the alarm interval is long, it is assumed that some kind of malfunction has occurred in the equipment, and therefore the stop alarm is likely to have been issued normally. Conversely, if the alarm interval is short and a certain period of time has passed since the operation resumed, it is assumed that the stop alarm was issued unexpectedly, and therefore it is likely to have been an erroneous alarm.

[0026] Therefore, in this embodiment, if the alarm interval is equal to or greater than the predetermined interval n in step S113, the data preprocessing unit 31 classifies the shutdown alarm as a true alarm (step S114). On the other hand, if the alarm interval is shorter than the predetermined interval n in step S113, the data preprocessing unit 31 classifies and labels the shutdown alarm as a false alarm (step S115). Note that the predetermined interval n is set in advance depending on the equipment to be classified as a shutdown alarm and the expected operating state of the equipment.

[0027] After the above-described classification process of stop alarms is completed, the data preprocessing unit 31 then performs a resampling process on the time-series data (step S116) to enable efficient machine learning while maintaining the accuracy of the stop alarms. This time-series data is included in the selection data acquired in step S111. Furthermore, the data preprocessing unit 31 performs the resampling process at a resampling interval shorter than the interval between false alarms, as shown in FIG. 4.

[0028] Next, the data preprocessing unit 31 determines whether the value of the resampled sampling data is within a predetermined range (step S117). The predetermined range is a range between an upper limit and a lower limit that are preset according to the type of device. If the value of the sampling data is outside the predetermined range, the data value is likely to have an adverse effect on machine learning, which will be described later.

[0029] Therefore, in this embodiment, if the value of the sampling data is not within a predetermined range, the data preprocessing unit 31 determines that the data value is an outlier. Then, the data preprocessing unit 31 removes the outlier from the preprocessed data (step S118). On the other hand, if the value of the sampling data is within the predetermined range, the data preprocessing unit 31 includes the sampling data in the preprocessed data. This completes the data preprocessing.

[0030] Next, the feature extraction unit 32 performs a feature extraction process to extract features from the preprocessed data of the data preprocessing unit 31 (step S2). In step S2, the feature extraction unit 32, for example, normalizes the preprocessed data and extracts characteristic data values ​​represented in the normalized data distribution as features.

[0031] Next, the clustering unit 33 performs a clustering process to cluster the operating states of the power generation facility 200 by machine learning the features extracted by the feature extraction unit 32 (step S3). This completes the data processing by the data processing unit 30. For the machine learning in step S3, unsupervised learning such as the k-means method, the x-means method, the spectral clustering method, and the hierarchical clustering method can be applied.

[0032] In general, it is difficult to directly determine false alarms from alarm data contained in the operation data used for machine learning. Furthermore, the learning data contains both normal and abnormal data. Therefore, the clusters classified by machine learning using this data contain both normal and abnormal categories.

[0033] Figure 5 shows an example of classification of the operating status of wind power generation equipment. In this classification example, the operating status is broadly classified into two categories: normal and abnormal. Furthermore, abnormalities here are classified based on alarms, and alarm types are classified into warnings (continue operation), normal stops, and emergency stops. Normal stops are further classified into two categories: false alarms and normal alarms. Generally, whether or not to issue an alarm is determined based on preset criteria, so there are cases where an alarm that should actually be classified as a false alarm is classified as a normal alarm.

[0034] In contrast to this, in this embodiment, in the data pre-processing stage, it is determined whether the shutdown alarm is a true alarm or a false alarm. Furthermore, in the data pre-processing stage, outliers that may adversely affect the clustering process of the operating state of the power generation facility 200 are removed.

[0035] Therefore, the classification method according to this embodiment makes it possible to determine false alarms that are difficult to distinguish, thereby preventing unnecessary shutdowns due to false alarms, thereby improving the operational efficiency of the power generation facility 200.

[0036] (Second embodiment) A second embodiment will be described. In this embodiment, the above-described operational state classification device 1 is also used to classify the operational state of a power generation facility 200. In this embodiment, the power generation facility 200 is a wind power generation facility.

[0037] Fig. 6(a) is a diagram showing an example of time-series data of the output power of the power generation facility 200. Fig. 6(b) is a diagram showing an example of time-series data of the wind speed of the wind that drives the wind turbines of the power generation facility 200. The time-series data shown in Fig. 6(a) and Fig. 6(b) show the operating state over a 10-day period. According to the time-series data shown in Fig. 6(a), the output of the power generation facility 200 (wind power generation facility) is frequently and repeatedly stopped.

[0038] On the other hand, according to the time-series data shown in FIG. 6(b), when the output of the power generation facility 200 is stopped, the wind speed is either high or low, close to zero. In this way, the wind speed not only shows different data values ​​when the output of the power generation facility 200 is stopped, but also fluctuates even during the period when the output is stopped. These frequently occurring stopped states of the power generation facility 200 may have a negative impact on the clustering process.

[0039] Therefore, in this embodiment, in order to reduce the above adverse effects, the data preprocessing unit 31 performs data preprocessing in accordance with the flowchart shown in FIG.

[0040] 7 is a flowchart of data preprocessing according to the second embodiment. In this flowchart, first, the data preprocessing unit 31 acquires selected data from the database 10 (step S121). The processing content of step S121 is the same as the processing content of step S111 described in the first embodiment, and therefore a detailed description thereof will be omitted.

[0041] Subsequently, the data preprocessing unit 31 performs a false alarm labeling process (step S122). In step S122, the processes of steps S112 to S115 described in the first embodiment are executed.

[0042] Next, the data pre-processing unit 31 performs a long-term shutdown data removal process (step S123). In step S123, the data pre-processing unit 31 removes, for example, long-term shutdown data in which the operation shutdown time is longer than a predetermined time from the time-series data shown in Fig. 6(a) together with the false alarm data that has been subjected to the false alarm labeling process.

[0043] Next, the data preprocessing unit 31 performs a resampling process (step S125). The process content of step S125 is the same as the resampling process of step S116 described in the first embodiment, and therefore a detailed description thereof will be omitted.

[0044] Next, the data preprocessing unit 31 determines whether the value of the resampled sampling data is within a predetermined range (step S126), and performs outlier removal processing in accordance with the determination processing (step S127). The processing contents of steps S126 and S127 are similar to the processing contents of steps S117 and S118 described in the first embodiment, respectively, and therefore detailed description thereof will be omitted.

[0045] Thereafter, similarly to the first embodiment, the feature extraction unit 32 performs feature extraction processing on the data preprocessed by the data preprocessing unit 31, and the clustering unit 33 performs clustering processing using the data that has been subjected to the feature extraction processing. This completes the data processing.

[0046] In the present embodiment described above, similarly to the first embodiment, it is possible to determine false alarms that are difficult to distinguish, and therefore unnecessary shutdowns due to false alarms can be avoided, thereby improving the operational efficiency of the power generation facility 200.

[0047] In addition, in this embodiment, long-term stop data that may adversely affect the clustering process is removed in the data pre-processing stage, thereby making it possible to avoid a decrease in the accuracy of the clustering process.

[0048] (Variation) Fig. 8 is a flowchart of data preprocessing according to a modification of the second embodiment. In the flowchart shown in Fig. 8, the processing contents of steps S131 to S133 are the same as the processing contents of steps S121 to S123 described in the second embodiment, and therefore detailed explanations will be omitted. Furthermore, the processing contents of steps S136 to S138 according to this modification are the same as the processing contents of steps S124 to S126 described in the second embodiment, and therefore detailed explanations of these steps will also be omitted.

[0049] In this modification, the data preprocessing unit 31 performs the short-stop data removal process after the long-stop data removal process (step S123). While the frequent shutdown states of the power generation equipment 200 may adversely affect the clustering process, the operating data immediately before the shutdown may contain data indicating a condition that may lead to the issuance of an alarm. Therefore, simply removing the shutdown data may result in the loss of important information for determining the shutdown conditions.

[0050] FIG. 9 is an illustration of the exclusion of shutdown data in this modification. As shown in FIG. 9, in this modification, data from a certain period immediately before the issuance of a shutdown alarm is retained from the time-series data, which may include a situation that may cause the shutdown alarm. For example, the data preprocessing unit 31 determines whether the time-series data is data from the period up to m minutes immediately before the shutdown alarm (see step S134 in FIG. 8). If the data corresponds to data from the period, as shown in FIG. 9, the data preprocessing unit 31 retains the data as data necessary for clustering, similar to a normal operating state, without removing it, even if the period includes a shutdown state.

[0051] On the other hand, if the time-series data does not meet the conditions for data retention, the data preprocessing unit 31 performs processing to remove the data in the stopped state. By performing such processing, it is possible to retain information on the cause of the stop alarm while avoiding the influence of the stopped state, and therefore it is possible to obtain more effective data.

[0052] After the data preprocessing unit 31 performs the data preprocessing as described above, the feature extraction unit 32 performs feature extraction processing, and then the clustering unit 33 performs clustering processing using the data that has been subjected to the feature extraction processing, as in the first and second embodiments. This completes the data processing.

[0053] According to the modified example described above, it is possible to determine false alarms that are difficult to distinguish. Furthermore, by removing data that may have a negative impact on clustering, it is possible to perform analysis using more effective data. This makes it possible to further improve the accuracy of determining false alarms and to further increase the rate at which unnecessary shutdowns are avoided. Therefore, it is possible to further improve the operational efficiency of the power generation facility 200.

[0054] Although several embodiments have been described above, these embodiments are presented only as examples and are not intended to limit the scope of the invention. The novel system described in this specification can be embodied in various other forms. Furthermore, various omissions, substitutions, and modifications can be made to the forms of the system described in this specification without departing from the spirit of the invention. The appended claims and their equivalents are intended to cover such forms and modifications that fall within the scope and spirit of the invention. [Explanation of symbols]

[0055] 1: Driving state classifier 31: Data preprocessing section 32: Feature extraction unit 33: Clustering section

Claims

1. a data preprocessing unit that performs data preprocessing to remove data related to erroneously issued alarms from time-series data indicating the operating status of the power generation facility; a feature extraction unit that extracts features from the pre-processed data; a clustering unit that classifies the operating state of the power generation facility based on the feature amount; A driving state classification device comprising:

2. 2. The operational state classification device according to claim 1, wherein the data preprocessing unit determines data to be removed from the time-series data in accordance with an interval between a shutdown alarm that warns of a shutdown of the power generation equipment and a startup alarm that warns of a restart of the power generation equipment after the shutdown alarm is released.

3. The operating state classification device according to claim 1 or 2, wherein the data preprocessing unit removes data in which the operation of the power generation facility is stopped for a period longer than a predetermined period from the time-series data.

4. The driving state classification device according to claim 2 , wherein the data preprocessing unit holds data of the time-series data for a certain period immediately before the issuance of the shutdown alarm.

5. Perform data preprocessing to remove data related to false alarms from time-series data showing the operating status of power generation facilities. extracting features from the pre-processed data; classifying an operating state of the power generation facility based on the feature amount; Driving state classification method.

6. data preprocessing to remove data related to false alarms from time series data indicating the operating status of the power generation facility; A process of extracting features from the pre-processed data; a process of classifying an operating state of the power generation facility based on the feature amount; A program that causes a computer to execute the following.

Images