Thermal power sensor data anomaly detection method and system

Abstract

The invention discloses a thermal power sensor data anomaly detection method and system, and relates to the field of thermal power plant sensor data, and the method comprises the steps: obtaining thermal power plant sensor data, segmenting the data into single features according to time nodes, and constructing a single variable histogram; performing binning operation on the data, and calculating numerical value characteristics and values and frequencies of classified data; generating abnormal value scores under different visual angles based on the binned data; and by combining the abnormal value scores under different visual angles, completing the abnormal detection of the sensor data. According to the invention, a single-variable abnormal value can be efficiently identified, local and global anomalies can be captured, and comprehensive monitoring and early warning capabilities can be provided. The scoring function is based on the information amount, is suitable for high-dimensional and low-dimensional data, reduces the representation deviation through normalization and combination, and improves the detection accuracy and the detection capability of various abnormal modes.

Description

Technical Field

[0001] The present invention relates to the field of thermal power plant sensor data, and in particular to a thermal power plant sensor data anomaly detection method and system. Background Art

[0002] During the operation of thermal power units, outlier detection of sensor data is crucial for equipment maintenance and fault prevention. When applied to sensor data in thermal power plants, existing technologies face the following specific problems: a large number of sensors are installed in thermal power plants, and the multidimensional data generated by these sensors require complex processing. Existing high-dimensional outlier detection methods are computationally intensive and inefficient in this environment; outliers of certain key parameters of thermal power plants (such as boiler pressure, temperature, etc.) in single-variable conditions may directly indicate equipment failures or potential risks. However, existing methods often ignore the detection of these single-variable outliers, which may cause important abnormal signals to be ignored; the operation of thermal power plants requires real-time monitoring and rapid response. The complexity and computational resource requirements of existing high-dimensional outlier detection methods make it difficult to meet the requirements of real-time detection. Summary of the invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is: how to design a method that can efficiently identify univariate outliers to improve the accuracy and real-time performance of detection.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a method for detecting abnormal data of a thermal power sensor, comprising:

[0007] Obtain sensor data from thermal power plants, and segment the sensor data collected at time nodes into single features according to sensor dimensions. One sensor corresponds to one dimension, and the data is segmented into samples according to time nodes.

[0008] The acquired sensor data is binned and a univariate histogram is constructed for each single feature. If the feature contains categorical data, the value and frequency of each category are calculated. For numerical features, the data is binned by single feature.

[0009] Generate outlier scores from different perspectives based on the binned data;

[0010] Based on the outlier scores from different perspectives, all bins matching the point are listed, and the outliers of the bins are merged into the outliers of the point to complete the sensor data anomaly detection.

[0011] As a preferred solution for thermal power sensor data anomaly detection method, wherein:

[0012] Generating outlier scores from different perspectives includes:

[0013] Assume that a single feature view s,s is generated by traversing each feature, the number of features is equal to the number of types of a single feature view, and only the outlier scores in the single-dimensional features are studied;

[0014] The global perspective g,g combines all features into one view and studies the outlier scores of the data;

[0015] The local perspective l,l consists of specifying B strongly correlated bins to study the outlier scores in the local subspace, where B represents the number of potential bins for each feature.

[0016] As a preferred solution for thermal power sensor data anomaly detection method, wherein:

[0017] Generating outlier scores from different perspectives also includes:

[0018] The calculation method of outlier score under each perspective is the same: get Traverse each Calculated in The amount of information in is denoted as The calculation method is:

[0019]

[0020] in, represents the feature representation within the view, is a symbolic representation of a binning instance of a feature within a view, Represents the amount of information of each bin in the view, view∈{l,s,g}.

[0021] As a preferred solution for thermal power sensor data anomaly detection method, wherein:

[0022] Generating outlier scores from different perspectives also includes:

[0023] In each perspective Normalized, the result is recorded as It is expressed as:

[0024]

[0025] in, Represents the normalized representation of the amount of bin information, Indicates the amount of information in each bin within the view.

[0026] As a preferred solution for thermal power sensor data anomaly detection method, wherein:

[0027] Generating outlier scores from different perspectives also includes:

[0028] The dimensionless coefficient of variation V is introduced to restore the lost quantity information, V i view Defined as:

[0029]

[0030] Where V i view represents the coefficient of variation, m is T i view The total number of O in , where O represents a bin.

[0031] As a preferred solution for thermal power sensor data anomaly detection method, wherein:

[0032] The step of merging the outliers of the bins into the outliers of the points includes:

[0033] Based on V weighting, α, β, and γ are specified as hyperparameters to adjust the proportion of outlier information in different perspectives. Specifically:

[0034]

[0035] Merge all outliers for each point, expressed as:

[0036]

[0037] The more times a point is found, the higher the probability that it is an outlier.

[0038] As a preferred solution for thermal power sensor data anomaly detection method, wherein:

[0039] The step of merging the outliers of the bins into the outliers of the points to complete the sensor data anomaly detection includes:

[0040] Calculate the outlier score of the corresponding point of each bin, set the threshold as the product of the data set size and the specified anomaly rate; obtain the points whose bin frequency is lower than the threshold to complete the sensor data anomaly detection.

[0041] In a second aspect, an embodiment of the present invention provides a thermal power sensor data anomaly detection system, comprising:

[0042] The data acquisition module is used to acquire sensor data from the thermal power plant, and divide the sensor data collected at the time nodes into single features according to the sensor dimension. One sensor corresponds to one dimension, and the data is divided into samples according to the time nodes.

[0043] The binning module is used to bin the acquired sensor data. For each single feature, a univariate histogram is constructed. If the feature contains categorical data, the value and frequency of each category are calculated. For numerical features, the data is binned by single feature.

[0044] The generation module is used to generate outlier scores from different perspectives based on the binned data;

[0045] The integration module is used to list all bins matching the point based on the outlier scores from different perspectives, and merge the outliers of the bins into the outliers of the point to complete the sensor data anomaly detection.

[0046] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0047] Memory and processor;

[0048] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the method for detecting abnormal data of thermal power sensor data as described in any embodiment of the present invention.

[0049] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method for detecting abnormality in thermal power sensor data.

[0050] Beneficial effects of the present invention: The present invention targets key parameters of thermal power plants (such as boiler pressure, temperature, etc.), and a single feature perspective can efficiently identify abnormal values ​​of these parameters in a single variable. This is crucial for rapid detection and response to equipment failures or potential risks; considering the local correlation between different sensors in a thermal power plant, the local perspective can capture abnormal behaviors in specific equipment or areas; the global perspective integrates all sensor data and can identify system-level anomalies. Through the global perspective, the MVOD method can discover abnormal patterns across devices or regions, providing more comprehensive monitoring and early warning capabilities; the designed general outlier scoring function based on information volume can evaluate the strength of the data distribution structure and ensure accurate quantification of data anomalies. This scoring function shows good adaptability to both high-dimensional and low-dimensional data; normalizing and combining outlier scores from different perspectives can reduce representation biases from different perspectives. This method not only improves the accuracy of outlier detection, but also enhances the ability to detect multiple abnormal patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0052] Figure 1 is an overall flow chart of the thermal power sensor data anomaly detection method according to the first embodiment of the present invention;

[0053] Figure 2 It is a simplified schematic diagram of the subspace of the thermal power sensor data anomaly detection method according to the first embodiment of the present invention;

[0054] Figure 3 It is a multi-view aggregation diagram of the thermal power sensor data anomaly detection method described in the first embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0058] Example 1

[0059] Reference Figure 1-Figure 3 , which is the first embodiment of the present invention, and provides a method for detecting abnormal data of thermal power sensor, comprising:

[0060] S1: Obtain sensor data from the thermal power plant, and divide the sensor data collected at time nodes into single features according to sensor dimensions. One sensor corresponds to one dimension, and the data is divided into samples according to time nodes.

[0061] Specifically, the sensor data is represented as Among them, x represents a single sensor, n represents the number of sensors, and d represents the time series length of a single-dimensional sensor, that is, the number of samples.

[0062] S2: Bin the acquired sensor data and build a univariate histogram for each single feature. If the feature contains categorical data, calculate the value and frequency of each category. For numerical features, bin the data by single feature.

[0063] Specifically, the K-Means clustering algorithm is used for binning. A good binning strategy can effectively improve the accuracy of the multi-view outlier detection algorithm (MVOD) and reduce the complexity of the algorithm. When using the clustering algorithm for binning, it can be applied to unsupervised algorithms. The Davies-Bouldin (DB) index and the Calinski-Harabasz (CH) index can evaluate the binning effect and measure the clustering performance without true labels. Davies-Bouldin and Calinski-Harabasz are used as evaluation indicators of clustering algorithms. The lower the former, the better, and the higher the latter, the better.

[0064] S3: Generate outlier scores from different perspectives based on the binned data;

[0065] In the embodiment of the present application, it is assumed that a single feature perspective s, s is generated by traversing each feature, the number of features is equal to the number of types of a single feature view, and only the outlier scores in the single-dimensional features are studied;

[0066] The global perspective g,g combines all features into one view and studies the outlier scores of the data;

[0067] The local perspective l,l consists of specifying B strongly correlated bins to study the outlier scores in the local subspace, where B represents the number of potential bins for each feature.

[0068] The calculation method of outlier score under each perspective is the same: get Traverse each Calculated in The amount of information in is denoted as The calculation method is:

[0069]

[0070] in, represents the feature representation within the view, is a symbolic representation of a binning instance of a feature within a view, Represents the amount of information of each bin in the view, view∈{l,s,g}.

[0071] In each perspective Normalized, the result is recorded as It is expressed as:

[0072]

[0073] in, Represents the normalized representation of the amount of bin information, Indicates the amount of information in each bin within the view.

[0074] Also includes:

[0075] The dimensionless coefficient of variation V is introduced to restore the lost quantity information, V i view Defined as:

[0076]

[0077] Where V i view represents the coefficient of variation, m is T i view The total number of O in , where O represents a bin.

[0078] In another possible implementation, the local perspective can be generated by the FP-Growth algorithm, the output of which includes an antecedent, a consequent, and a lift, denoted as a, c, and L, respectively. The bins in the antecedent are strongly correlated with the bins in the consequent, a usually refers to some bins in multiple features, and c usually refers to a bin in a feature. When using the bins in a as data filtering conditions and observing the feature N containing the bins in c, c When , a T can be formed in l. The combination of the binning in the antecedent and the binning in the suffix can be regarded as a kind of O in T. When the FP-Growth algorithm is used to generate highly correlated O, each O corresponds to a different T, and the m kinds of O with the same c have strong correlation, denoted as R c .

[0079] like Figure 3 As shown in Figure 2, for the same post-term, multiple redundant rules are generated. For example, both rules target B. x1 Repeated counting of outliers in these subspaces may result in repeated counting of anomaly scores. To address this problem, we select the subspace that best represents these relationships. (where r∈R c). Therefore, observe the abnormal information in c and record the corresponding T as The whole generation process is as follows Figure 2 shown.

[0080] Set the local perspective screening conditions. The conditions for screening the local perspective L should be based on practical significance: the more bin types the antecedent a has, the greater the possibility that a and c appear at the same time (that is, the greater the lift L), and the more reliable the local perspective l is. Define the number of bin types in the antecedent a as A, use the bins in c as conditions, and select l with the largest possible A and L. In order to balance A and L, use the F1-Score to integrate these two indicators, and define the screening process as follows:

[0081]

[0082] This process selects only one subspace related to c from among m different subspaces. To reduce exploration.

[0083] It should be noted that in actual situations, one bin may correspond to multiple points, and one point may correspond to multiple bins. Generally speaking, the number of bins is much smaller than the number of points, so using bins to match points, assigning outliers to these points, and merging outliers can better reduce time complexity.

[0084] S4: Based on the outlier scores from different perspectives, all bins matching the point are listed, and the outliers of the bins are merged into the outliers of the point to complete the sensor data anomaly detection.

[0085] It should be noted that a point represents a value in the feature sequence.

[0086] In the embodiment of the present application, based on V weighting, α, β and γ are specified as hyperparameters to adjust the proportion of outlier information in different perspectives, specifically:

[0087]

[0088] Merge all outliers for each point, expressed as:

[0089]

[0090] The more times a point is found, the higher the probability that it is an outlier.

[0091] Calculate the outlier score of the corresponding point of each bin, set the threshold as the product of the data set size and the specified anomaly rate; obtain the points whose bin frequency is lower than the threshold to complete the sensor data anomaly detection.

[0092] The above is a schematic scheme of the thermal power sensor data anomaly detection method of this embodiment. It should be noted that the technical scheme of the thermal power sensor data anomaly detection system and the technical scheme of the thermal power sensor data anomaly detection method above belong to the same concept. For details not described in detail in the technical scheme of the thermal power sensor data anomaly detection system in this embodiment, please refer to the description of the technical scheme of the thermal power sensor data anomaly detection method above.

[0093] The thermal power sensor data anomaly detection system in this embodiment includes:

[0094] The data acquisition module is used to acquire sensor data from the thermal power plant, and divide the sensor data collected at the time nodes into single features according to the sensor dimension. One sensor corresponds to one dimension, and the data is divided into samples according to the time nodes.

[0095] The binning module is used to bin the acquired sensor data. For each single feature, a univariate histogram is constructed. If the feature contains categorical data, the value and frequency of each category are calculated. For numerical features, the data is binned by single feature.

[0096] The generation module is used to generate outlier scores from different perspectives based on the binned data;

[0097] The integration module is used to list all bins matching the point based on the outlier scores from different perspectives, and merge the outliers of the bins into the outliers of the point to complete the sensor data anomaly detection.

[0098] This embodiment also provides a computing device, applicable to the method, including:

[0099] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the method proposed in the above embodiment.

[0100] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0101] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0102] Example 2

[0103] Referring to Tables 1 and 2, an embodiment of the present invention provides a method for detecting abnormal data of thermal power sensors. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0104] This method uses the temperature of the small unit and the heater temperature of the thermal power plant to conduct experiments on Unit 3 and Unit 4. As shown in Table 1, it is a description of the experimental implementation data.

[0105] Table 1 Experimental implementation data

[0106] Implementation Examples Number of samples / length Feature Dimension Failure rate Temperature of small unit of unit 3 13824 16 0.182 Unit 3 heater temperature 23621 12 0.2062 Temperature of small unit 4 8987 16 0.1417 Unit 4 heater temperature 17843 12 0.2634

[0107] As shown in Table 2, the performance comparison of this method and the current outlier detection method is performed using the AUCROC evaluation indicator.

[0108] Table 2 Performance comparison of outlier detection methods

[0109]

[0110] It can be seen that the anomaly detection performance of this method in different implementation examples is better than that of the comparison method. Specifically, this method is significantly higher than other detection methods in the AUCROC evaluation index, especially in the detection of the small unit temperature of Unit 4 and the heater temperature of Unit 3, showing the best detection ability. This shows that this method has high accuracy and robustness in the detection of anomaly in thermal power sensor data.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting abnormal data of thermal power sensor, characterized in that: include: Obtain sensor data from thermal power plants, and segment the sensor data collected at time nodes into single features according to sensor dimensions. One sensor corresponds to one dimension, and the data is segmented into samples according to time nodes. The acquired sensor data is binned and a univariate histogram is constructed for each single feature; If the feature contains categorical data, the value and frequency of each category are calculated; for numerical features, the data is binned by single feature; Generate outlier scores from different perspectives based on the binned data; Based on the outlier scores from different perspectives, all bins matching the point are listed, and the outliers of the bins are merged into the outliers of the point to complete the sensor data anomaly detection.

2. The thermal power sensor data anomaly detection method according to claim 1, characterized in that: Generating outlier scores from different perspectives includes: Assume that a single feature view s,s is generated by traversing each feature, the number of features is equal to the number of types of a single feature view, and only the outlier scores in the single-dimensional features are studied; The global perspective g,g combines all features into one view and studies the outlier scores of the data; The local perspective l,l consists of specifying B strongly correlated bins to study the outlier scores in the local subspace, where B represents the number of potential bins for each feature.

3. The thermal power sensor data anomaly detection method according to claim 2, characterized in that: Generating outlier scores from different perspectives also includes: The calculation method of outlier score under each perspective is the same: get Traverse each Calculated in The amount of information in is denoted as The calculation method is: in, represents the feature representation within the view, is a symbolic representation of a binning instance of a feature within a view, Represents the amount of information of each bin in the view, view∈{l,s,g}.

4. The thermal power sensor data anomaly detection method according to claim 3, characterized in that: Generating outlier scores from different perspectives also includes: In each perspective Normalized, the result is recorded as It is expressed as: in, Represents the normalized representation of the amount of bin information, Indicates the amount of information in each bin within the view.

5. The thermal power sensor data anomaly detection method according to claim 4, characterized in that: Generating outlier scores from different perspectives also includes: The dimensionless coefficient of variation V is introduced to restore the lost quantity information, V i view Defined as: Where V i view represents the coefficient of variation, m is T i view The total number of O in , where O represents a bin.

6. The thermal power sensor data anomaly detection method according to claim 5, characterized in that: The step of merging the outliers of the bins into the outliers of the points includes: Based on V weighting, α, β, and γ are specified as hyperparameters to adjust the proportion of outlier information in different perspectives. Specifically: Merge all outliers for each point, expressed as: The more times a point is found, the higher the probability that it is an outlier.

7. The thermal power sensor data anomaly detection method according to claim 6, characterized in that: The step of merging the outliers of the bins into the outliers of the points to complete the sensor data anomaly detection includes: Calculate the outlier score of the corresponding point of each bin, set the threshold as the product of the data set size and the specified anomaly rate; obtain the points whose bin frequency is lower than the threshold to complete the sensor data anomaly detection.

8. A system using the thermal power sensor data anomaly detection method according to any one of claims 1 to 7, characterized in that: include: The data acquisition module is used to acquire sensor data from the thermal power plant, and divide the sensor data collected at the time nodes into single features according to the sensor dimension. One sensor corresponds to one dimension, and the data is divided into samples according to the time nodes. The binning module is used to bin the acquired sensor data and build a univariate histogram for each single feature; If the feature contains categorical data, the value and frequency of each category are calculated; for numerical features, the data is binned by single feature; The generation module is used to generate outlier scores from different perspectives based on the binned data; The integration module is used to list all bins matching the point based on the outlier scores from different perspectives, and merge the outliers of the bins into the outliers of the point to complete the sensor data anomaly detection.

9. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for detecting abnormality in thermal power sensor data described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for detecting abnormality in thermal power sensor data as described in any one of claims 1 to 7.

Images