Boiler wall temperature anomaly detection and early warning method based on approximate entropy calculation

Abstract

The invention belongs to the technical field of equipment state monitoring, and particularly relates to a boiler wall temperature anomaly detection and early warning method based on approximate entropy calculation, which comprises the following steps: dividing a large coal-fired power plant boiler wall temperature time sequence into continuously overlapped sliding windows; calculating three statistical parameters in each sliding window; the three statistical parameters comprise standard deviation, range and approximate entropy; forming a three-dimensional time sequence matrix formed by n groups of parameters based on the three statistical parameters; and performing anomaly detection and early warning on the three-dimensional time sequence matrix by using an isolated forest anomaly detection algorithm. Abnormalities are accurately recognized from different dimensions, and high-dimensional data are efficiently dealt with. The method has the real-time performance and the dynamic performance, online calculation and real-time response can be achieved, the model is flexibly updated along with new data inrush, and the anomaly detection requirement of the dynamic data environment is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of equipment status monitoring, and in particular relates to a boiler wall temperature anomaly detection and early warning method based on approximate entropy calculation. Background Art

[0002] Abnormal boiler wall temperatures in large coal-fired power plants are closely linked to safety. Abnormally high boiler wall temperatures can cause metal materials to overheat, degrading their strength, toughness, and other mechanical properties. Prolonged exposure to high temperatures can also cause creep, gradually deforming components and potentially leading to safety hazards such as ruptures and leaks. For example, when the wall temperature of a boiler's water-cooled wall tubes exceeds the design range, the tubes' strength decreases significantly, and internal pressure can lead to problems such as bulging and bursting. Abnormal wall temperatures can cause localized overheating of the heating surface, leading to uneven expansion of the metal material. This uneven expansion generates thermal stress, which, when exceeded by the material's tolerance limit, can cause cracks. Over time, these cracks propagate, eventually leading to rupture of the heating surface and steam or hot water leakage. If the leaked medium is high-temperature, high-pressure steam, serious consequences such as burns and equipment damage can occur. A boiler is a complex system with interconnected components. Abnormal wall temperatures can have a ripple effect on other components, compromising the overall stability of the boiler. For example, abnormal superheater wall temperature may cause excessive deviation in steam temperature, affecting the normal operation of the turbine and even causing damage to components such as turbine blades, thereby affecting the safety and reliability of the entire power generation system.

[0003] Monitoring boiler wall temperatures in large coal-fired power plants is crucial and holds multiple significance, value, and implications. From a safety perspective, boilers operate at high temperatures and high pressures, and wall temperature monitoring provides real-time visibility into the temperature of the boiler's heating surfaces. Abnormally high wall temperatures may indicate problems such as pipe coking or steam-water circulation failures. If not addressed promptly, these can easily lead to serious accidents such as pipe bursts, threatening the safe and stable operation of the power plant and the safety of personnel. Wall temperature monitoring enables early warning and action to nip potential accidents in the bud. Regarding equipment maintenance, accurate wall temperature monitoring data facilitates the development of appropriate maintenance plans. It helps operators understand the distribution of thermal stresses and material aging in equipment, avoiding resource waste due to excessive maintenance or equipment damage due to insufficient maintenance. This not only extends equipment life but also reduces maintenance costs. From an energy-saving perspective, wall temperature monitoring provides a key basis for optimizing combustion. Operators can adjust combustion conditions based on wall temperature fluctuations, ensuring sufficient fuel combustion, improving thermal efficiency, reducing coal consumption and pollutant emissions, and enhancing the economic and environmental benefits of the power plant. At the same time, stable wall temperature helps to improve the availability of the unit and reduce economic losses caused by downtime.

[0004] Monitoring the boiler wall temperature of large coal-fired power plants generates a vast amount of data. However, some plants currently handle this data using relatively simple methods, primarily focusing on real-time display and over-limit alarms, without sufficient in-depth data mining and analysis. This makes it difficult to extract valuable information from this massive amount of data and uncover underlying patterns and trends in wall temperature fluctuations. The lack of effective data analysis models hinders accurate prediction and diagnosis of abnormal wall temperature conditions. Consequently, alarms are often issued only after the wall temperature has clearly exceeded limits or a fault has occurred, making it difficult to implement preventive measures in a timely manner.

[0005] Therefore, a method for detecting and warning boiler wall temperature anomalies based on approximate entropy calculations is urgently needed. This method can accurately identify anomalies from different dimensions and efficiently handle high-dimensional data. It should be both real-time and dynamic, capable of online calculations and real-time responses, and flexible model updates as new data arrives, meeting the anomaly detection needs of dynamic data environments. Summary of the Invention

[0006] Monitoring the boiler wall temperature of large coal-fired power plants generates a vast amount of data. However, some plants currently handle this data using relatively simple methods, primarily focusing on real-time display and over-limit alarms, without sufficient in-depth data mining and analysis. This makes it difficult to extract valuable information from this massive amount of data and uncover underlying patterns and trends in wall temperature fluctuations. The lack of effective data analysis models hinders accurate prediction and diagnosis of abnormal wall temperature conditions. Consequently, alarms are often issued only after the wall temperature has clearly exceeded limits or a fault has occurred, making it difficult to implement preventive measures in a timely manner.

[0007] The present invention aims to provide a method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation, comprising the following steps:

[0008] The time series of the boiler wall temperature of a large coal-fired power plant is divided into consecutive overlapping sliding windows;

[0009] Calculate three statistical parameters within each sliding window; the three statistical parameters include: standard deviation, range, and approximate entropy;

[0010] Based on the three statistical parameters, a three-dimensional time series matrix consisting of n groups of parameters is formed;

[0011] The isolation forest anomaly detection algorithm is used to perform anomaly detection and early warning on the three-dimensional time series matrix.

[0012] The method of dividing the time series of the wall temperature of the boiler of a large coal-fired power plant into continuous overlapping sliding windows includes:

[0013] Use sliding windows to observe temperature time series data, with a data collection interval of Δt, the number of sliding windows of m, the length of the sliding window of w, and the sliding step of s;

[0014] Assume that the initial time is t0, and the temperature time series data is T={T1,T2,…,T n}(n≥(m-1)s+w), where T i Indicates that at time t i =t0+(i-1)Δt(i=1,2,…,n) the collected temperature value;

[0015] The data W in the kth (k=1,2,…,m) sliding window k ={T (k-1)s+1 ,T (k-1)s+2 ,…,T (k-1)s+w}, the time interval is [t0+((k-1)s)Δt,t0+((k-1)s+w-1)Δt].

[0016] The calculation of the three statistical parameters within each sliding window includes:

[0017] Using the mature formulas and algorithms of the corresponding industries for standard deviation, range, and approximate entropy, calculate the standard deviation σ of the data in the kth (k=1,2,…,m) sliding window k , range R k , approximate entropy ApEn k .

[0018] The forming of a three-dimensional time series matrix consisting of n groups of parameters based on the three statistical parameters includes:

[0019] Combine the standard deviation, range, and approximate entropy of m windows into an m×3 matrix M:

[0020]

[0021] Where, σ n is the nth standard deviation, R n is the nth range, ApEn n is the nth approximate entropy.

[0022] The use of the isolation forest anomaly detection algorithm to perform anomaly detection and early warning on the three-dimensional time series matrix includes:

[0023] Step S4.1: data preprocessing;

[0024] The standard deviation, range, and approximate entropy of m windows are combined into an m×3 matrix M. Each row of the m×3 matrix M represents a data sample, and each column represents a different feature. There are m samples in total, and each sample has 3 features, representing m points in three-dimensional space. The data is normalized and preprocessed to ensure that different features have similar scales to prevent a single feature from having a large impact on the results.

[0025] Step 4.2: Build an isolation forest model;

[0026] In three-dimensional space, an m×3 matrix M represents m points. The m×3 matrix M data is used as a training data set to train and construct an isolation forest model. The isolation forest model learns the normal pattern of the m points in the data, that is, the m×3 matrix M, and can identify significantly different abnormal points.

[0027] After training is complete, the isolation forest model is used to calculate the data of m points, i.e., the m×3 matrix M, to obtain an anomaly confidence score for each data point. The anomaly confidence score reflects the likelihood of each point being an anomaly, with lower scores indicating a higher likelihood of the point being an anomaly. A threshold is set based on the obtained anomaly score to identify anomalies, and a corresponding early warning mechanism is developed.

[0028] Step 4.3: Anomaly detection and early warning;

[0029] Based on the obtained anomaly confidence score for each data point, a threshold θ = the 5th percentile of m anomaly confidence scores is set to identify anomalies;

[0030] If a point (σ k ,R k ,ApEn k ) is ≤θ, the point is determined to be an abnormal point, and an alert is issued for abnormal temperature fluctuations within the start and end time periods of the sliding window associated with the point [t0+((k-1)s)Δt, t0+((k-1)s+w-1)Δt], where k∈[1,2,…,m].

[0031] Another object of the present invention is to provide a boiler wall temperature anomaly detection and early warning system according to the boiler wall temperature anomaly detection and early warning method based on approximate entropy calculation of the present invention, comprising: a boiler wall temperature acquisition device, a power plant production equipment SIS database, and a boiler wall temperature anomaly real-time monitoring and early warning server;

[0032] The boiler wall temperature collecting device is used to collect the wall temperature of the boiler;

[0033] The power plant production equipment SIS database is used to obtain monitoring data;

[0034] The boiler wall temperature abnormality real-time monitoring and early warning server obtains monitoring data, inputs the monitoring data into the implementation monitoring program, and realizes the draft fan bearing temperature abnormality detection and early warning method; the monitoring data includes: boiler wall temperature and other power plant equipment real-time operation data.

[0035] Another object of the present invention is to provide a computer device comprising a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the boiler wall temperature abnormality detection and early warning method based on approximate entropy calculation according to the present invention.

[0036] Another object of the present invention is to provide a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the boiler wall temperature abnormality detection and early warning method based on approximate entropy calculation according to the present invention.

[0037] The beneficial effects of the present invention are:

[0038] The present invention discloses a method for detecting and warning abnormal boiler wall temperatures based on approximate entropy calculation, which combines approximate entropy with the isolation forest algorithm to achieve abnormal boiler wall temperature detection and warning in large coal-fired power plants, with significant advantages. Approximate entropy is good at capturing the complexity of time series, sensitively sensing subtle changes in data, and providing key features based on complexity for detection. Isolation forests rely on tree structures to efficiently divide the data space and quickly identify sparsely distributed anomalies. The combination of the two complements each other and can accurately identify anomalies from different dimensions, and can also efficiently deal with high-dimensional data. At the same time, it is both real-time and dynamic, can be calculated online, respond in real time, and flexibly update the model as new data flows in, meeting the anomaly detection needs of dynamic data environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Schematic diagram of a flow chart of a method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation according to the present invention;

[0040] Figure 2 This is a schematic structural diagram of a boiler wall temperature anomaly detection and early warning system according to an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of boiler wall temperature time series data according to an embodiment of the present invention;

[0042] Figure 4 Schematic diagram of standard deviation and range two-dimensional isolation forest anomaly detection results according to an embodiment of the present invention;

[0043] Figure 5 2D schematic diagram of the standard deviation, range and approximate entropy of the three-dimensional isolation forest anomaly detection results according to an embodiment of the present invention;

[0044] Figure 6 3D schematic diagram of the standard deviation, range and approximate entropy of the three-dimensional isolation forest anomaly detection results according to an embodiment of the present invention;

[0045] Among them, 100-boiler; 200-boiler wall temperature acquisition device; 300-power plant production equipment SIS database; 400-boiler wall temperature abnormal real-time monitoring and early warning server. DETAILED DESCRIPTION

[0046] The present invention provides a method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation, which is further described in detail below with reference to the accompanying drawings.

[0047] like Figure 1 The embodiment of the present invention shown discloses a method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation, comprising the following steps:

[0048] The time series of the boiler wall temperature of a large coal-fired power plant is divided into consecutive overlapping sliding windows;

[0049] Calculate three statistical parameters within each sliding window; the three statistical parameters include: standard deviation, range, and approximate entropy;

[0050] Based on the three statistical parameters, a three-dimensional time series matrix consisting of n groups of parameters is formed;

[0051] The isolation forest anomaly detection algorithm is used to perform anomaly detection and early warning on the three-dimensional time series matrix.

[0052] In this embodiment, the calculation of statistical parameters such as approximate entropy is combined with the isolation forest anomaly detection algorithm to achieve anomaly detection and early warning of boiler wall temperature in large coal-fired power plants. The advantages of combining approximate entropy with isolation forest for anomaly detection are: approximate entropy can effectively characterize the complexity and regularity of time series, is sensitive to dynamic changes in data, and can identify anomalies from the perspective of sequence characteristics; isolation forest is based on a tree structure to quickly isolate anomalies, and is efficient and scalable when processing high-dimensional data and large-scale data. The combination of the two can integrate the advantages of each other, using approximate entropy to grasp the inherent regularity of data to discover subtle abnormal patterns, and relying on the rapid isolation mechanism of isolation forest to improve detection efficiency, thereby more comprehensively, accurately and efficiently detecting anomalies in the data.

[0053] The following is a detailed explanation of each step:

[0054] Step 1: Data preparation;

[0055] The temperature time series data is observed using a sliding window. The data collection interval is Δt, the number of sliding windows is m, the length of the sliding window is w, and the sliding step is s.

[0056] Assume that the initial time is t0, and the temperature time series data is T={T1,T2,…,T n}(n≥(m-1)s+w). Among them, T i Indicates that at time t i =t0+(i-1)Δt(i=1,2,…,n) the collected temperature value.

[0057] The data W in the kth (k=1, 2, ..., m) sliding window k ={T (k-1)s+1 , T (k-1)s+2 ,…,T (k-1)s+w}, the time interval is [t0+((k-1)s)Δt,t0+((k-1)s+w-1)Δt].

[0058] Step 2: Calculate the standard deviation σ within the sliding window k , range R k , approximate entropy ApEn k ;

[0059] Standard Deviation:

[0060] For the kth sliding window W k , its mean

[0061] Standard deviation

[0062] Very bad:

[0063] Extremely poor

[0064] R k =max{T (k-1)s+1 ,T (k-1)s+2 ,…,T (k-1)s+w}-min{T (k-1)s+1 , T (k-1)s+2 ,…,T (k-1)s+w}.

[0065] Approximate entropy:

[0066] For the kth sliding window W k ={x1, x2, ..., x w The steps to calculate the approximate entropy are as follows (assuming the embedding dimension is m0, the delay time is τ, and the tolerance is r):

[0067] οReconstruct phase space: construct vector i=1,…,w-(m0-1)τ.

[0068] οCalculate distance:

[0069] ο Calculate the probability: For each vector Calculation Satisfaction Vector Number

[0070] οCalculation

[0071] ο Increase the embedding dimension to m0+1 and repeat the above steps to obtain

[0072] ο Approximate entropy

[0073] Using the mature formulas and algorithms in the corresponding industries regarding standard deviation, range, and approximate entropy, the standard deviation σk, range Rk, and approximate entropy ApEnk of the data in the kth (k=1, 2,…, m) sliding window are calculated.

[0074] Step 3: Build the dataset;

[0075] The standard deviation, range, and approximate entropy of the m windows are combined into an m×3 matrix M.

[0076]

[0077] Step 4: Anomaly detection and warning based on the Isolation Forest algorithm

[0078] Isolation Forest is an efficient anomaly detection algorithm based on ensemble learning, specifically designed to identify outliers in datasets. Its core idea is to gradually isolate data points by randomly selecting features and split values. Because outliers reside in a sparse data space, they often require only a few splits to be isolated, while normal points require more. The algorithm constructs multiple isolation trees (iTrees), each generated by splitting a random subsample of the dataset. During the anomaly detection phase, for a given data point, the average path length of its isolation across all trees is calculated as the anomaly score. A lower score indicates a higher likelihood of an outlier.

[0079] Method for anomaly detection and early warning of m×3 matrix M using Isolation Forest algorithm:

[0080] In three-dimensional space, an m×3 matrix M represents m points. Using this m×3 matrix M as a training dataset, we train an isolation forest model. The isolation forest model learns the normal patterns within the m points (i.e., the m×3 matrix M) and identifies significantly different outliers.

[0081] After training, the isolation forest model is used to calculate the anomaly score for each data point (i.e., an m×3 matrix M). This score reflects the likelihood of each data point being an outlier, with lower scores indicating a higher likelihood of an outlier. Based on the obtained anomaly score, a threshold is set to identify anomalies and develop corresponding early warning mechanisms.

[0082] In this example, the calculation of statistical parameters such as approximate entropy and the isolation forest anomaly detection algorithm are combined to achieve anomaly detection and early warning of boiler wall temperature in large coal-fired power plants. First, the boiler wall temperature time series of a large coal-fired power plant is divided into consecutive overlapping sliding windows. Within each window, three statistical parameters, standard deviation, range, and approximate entropy, are used to form a three-dimensional time series matrix consisting of n groups of parameters. The isolation forest anomaly detection algorithm is then used to detect anomalies and provide early warnings on this coefficient matrix.

[0083] In order to verify the effectiveness of the boiler wall temperature anomaly detection and early warning method based on approximate entropy calculation disclosed in the present invention, a verification experiment was conducted. The specific verification process is as follows:

[0084] Step 1: Data Preparation

[0085] The temperature time series data is observed using a sliding window. The data collection interval is Δt, the number of sliding windows is m, the length of the sliding window is w, and the sliding step is s.

[0086] Assume that the initial time is t0, and the temperature time series data is T={T1,T2,…,T n}(n≥(m-1)s+w). Among them, T i Indicates that at time t i =t0+(i-1)Δt(i=1,2,…,n) the collected temperature value.

[0087] The data W in the kth (k=1,2,…,m) sliding window k ={T (k-1)s+1 ,T (k-1)s+2 ,…,T (k-1)s+w}, the time interval is [t0+((k-1)s)Δt,t0+((k-1)s+w-1)Δt].

[0088] Step 2: Calculate the standard deviation σ within the sliding window k , range R k , approximate entropy ApEn k

[0089] Standard Deviation:

[0090] For the kth sliding window W k , its mean

[0091] Standard deviation

[0092] Very bad:

[0093] Extremely poor

[0094] R k =max{T (k-1)s+1 , T (k-1)s+2 ,…,T (k-1)s+w}-min{T (k-1)s+1 , T (k-1)s+2 ,…,T (k-1)s+w}.

[0095] Approximate entropy:

[0096] For the kth sliding window W k ={x1, x2, ..., x w}, the steps to calculate the approximate entropy are as follows (assuming the embedding dimension is m0, the delay time is τ, and the tolerance is r):

[0097] οReconstruct phase space: construct vector i=1,…,w-(m0-1)τ.

[0098] οCalculate distance:

[0099] ο Calculate the probability: For each vector Calculation Satisfaction Vector Number

[0100] οCalculation

[0101] ο Increase the embedding dimension to m0+1 and repeat the above steps to obtain

[0102] ο Approximate entropy

[0103] Using the mature formulas and algorithms of the corresponding industries on standard deviation, range, and approximate entropy, calculate the standard deviation σ of the data in the kth (k=1,2,…,m) sliding window k , range R k , approximate entropy ApEn k .

[0104] Step 3: Build the dataset

[0105] The standard deviation, range, and approximate entropy of the m windows are combined into an m×3 matrix M.

[0106]

[0107] Step 4: Anomaly detection and warning based on the Isolation Forest algorithm

[0108] 4.1 Data Preprocessing

[0109] The standard deviation, range, and approximate entropy of m windows are combined into an m×3 matrix M. Each row of this m×3 matrix M represents a data sample, and each column represents a different feature. There are m samples in total, each with three features, representing m points in three-dimensional space. Data is preprocessed by normalization to ensure that different features have similar scales to prevent any one feature from overly influencing the results.

[0110] 4.2 Building an Isolation Forest Model

[0111] The core idea behind the Isolation Forest model is to isolate outliers by leveraging their sparsity and distribution characteristics, constructing multiple decision trees. Specifically, each decision tree randomly selects a feature during construction, then randomly selects a split point within the feature's possible value range, splitting the data into two parts, and repeating this process recursively. Normal data points, due to their dense distribution, may require more splitting steps to be isolated to leaf nodes. However, outliers may only require a few splits to reach leaf nodes, resulting in a shorter average path length.

[0112] In three-dimensional space, an m×3 matrix M represents m points. Using this m×3 matrix M as a training dataset, an isolation forest model is trained to learn the normal patterns within the m points (i.e., the m×3 matrix M) and identify significantly different outliers. After training, the isolation forest model is used to calculate the anomaly confidence score for each data point (i.e., the m×3 matrix M). This score reflects the likelihood that each point is an outlier; lower scores indicate a higher likelihood of an outlier.

[0113] 4.3 Anomaly Detection and Early Warning

[0114] Based on the obtained anomaly confidence score for each data point, a threshold θ=(the 5th percentile of m anomaly confidence scores) is set to identify anomalies.

[0115] If a point (σ k ,R k ,ApEn k) has an anomaly confidence score ≤ θ, the point is considered an anomaly and an alert is issued for abnormal temperature fluctuations within the start and end time periods of the sliding window associated with the point [t0+((k-1)s)Δt, t0+((k-1)s+w-1)Δt]. Where k∈[1,2,…,m].

[0116] The existing technology generally uses a static threshold (such as 90% to 95% of the allowable temperature of steel) to provide early warning of boiler wall temperature in large power plants. However, the existing technology does not use approximate entropy technology in the field of boiler wall temperature anomaly detection.

[0117] While prior art indicates that approximate entropy excels at capturing the complexity of time series, sensitively sensing subtle changes in data, and providing key complexity-based features for detection, using only approximate entropy alone cannot achieve the same impressive results as the present invention. The present invention combines standard deviation, range, approximate entropy, and the Isolation Forest anomaly detection algorithm as an integral, integrated technical solution to achieve even better results.

[0118] The boiler wall temperature data sampling period Δt = 60 seconds. Set the parameters of the dynamic sliding window: the number of sliding windows is n, set n = 1000; set m = 300, the sliding window length is m + 1, that is, the sliding window contains m + 1 boiler wall temperature time series data; the sliding step of the sliding window is Δt × s (s ≥ 1), set s = 60. Figure 3 is the boiler wall temperature time series data.

[0119] In each sliding window, the standard deviation and range of the boiler wall temperature time series are calculated to form a two-dimensional array. The anomaly detection is implemented by combining the isolation forest anomaly detection algorithm, such as Figure 4 As shown, the red points are abnormal points.

[0120] In each sliding window, the standard deviation, range, and approximate entropy of the boiler wall temperature time series are calculated to form a three-dimensional space. Combined with the isolation forest anomaly detection algorithm, anomaly detection is implemented, such as Figure 5 and Figure 6 shown.

[0121] Comparison Figure 4 and Figure 5 (Both are XY views and can be compared), the red dots are abnormal points, Figure 5 The number of identified abnormal points in the three-dimensional space (standard deviation, range, and approximate entropy) is significantly greater than Figure 4 (Standard deviation and range form a two-dimensional array).

[0122] like Figure 2As shown, another embodiment of the present invention discloses a boiler wall temperature anomaly detection and early warning system according to the boiler wall temperature anomaly detection and early warning method based on approximate entropy calculation of the present invention, characterized by comprising: a boiler wall temperature acquisition device 200, a power plant production equipment SIS database 300, and a boiler wall temperature anomaly real-time monitoring and early warning server 400;

[0123] The boiler wall temperature collecting device 200 is used to collect the wall temperature of the boiler 100;

[0124] The power plant production equipment SIS database 300 is used to obtain monitoring data;

[0125] The boiler wall temperature abnormality real-time monitoring and early warning server 400 obtains monitoring data, inputs the monitoring data into the implementation monitoring program, and implements the induced draft fan bearing temperature abnormality detection and early warning method; the monitoring data includes: the wall temperature of the boiler 100 and real-time operating data of other power plant equipment.

[0126] Another embodiment of the present invention discloses a computer device, including a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the boiler wall temperature abnormality detection and early warning method based on approximate entropy calculation according to the present invention.

[0127] Another embodiment of the present invention discloses a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the boiler wall temperature abnormality detection and early warning method based on approximate entropy calculation according to the present invention.

[0128] The present invention discloses a method for detecting and warning abnormal boiler wall temperatures based on approximate entropy calculation, which combines approximate entropy with the isolation forest algorithm to achieve abnormal boiler wall temperature detection and warning in large coal-fired power plants, with significant advantages. Approximate entropy is good at capturing the complexity of time series, sensitively sensing subtle changes in data, and providing key features based on complexity for detection. Isolation forests rely on tree structures to efficiently divide the data space and quickly identify sparsely distributed anomalies. The combination of the two complements each other and can accurately identify anomalies from different dimensions, and can also efficiently deal with high-dimensional data. At the same time, it is both real-time and dynamic, can be calculated online, respond in real time, and flexibly update the model as new data flows in, meeting the anomaly detection needs of dynamic data environments.

Claims

1. A method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation, characterized in that: The steps include: The time series of the boiler wall temperature of a large coal-fired power plant is divided into consecutive overlapping sliding windows; Calculate three statistical parameters within each sliding window; the three statistical parameters include: standard deviation, range, and approximate entropy; Based on the three statistical parameters, a three-dimensional time series matrix consisting of n groups of parameters is formed; The isolation forest anomaly detection algorithm is used to perform anomaly detection and early warning on the three-dimensional time series matrix.

2. The method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation according to claim 1 is characterized in that: The method of dividing the time series of the wall temperature of the boiler of a large coal-fired power plant into continuous overlapping sliding windows includes: Use sliding windows to observe temperature time series data, with a data collection interval of Δt, the number of sliding windows of m, the length of the sliding window of w, and the sliding step of s; Assume that the initial time is t0, and the temperature time series data is T={T1,T2,…,T n }(n≥(m-1)s+w), where T i Indicates that at time t i =t0+(i-1)Δt(i=1,2,…,n) the collected temperature value; The data W in the kth (k=1,2,…,m) sliding window k ={T (k-1)s+1 ,T (k-1)s+2 ,…,T (k-1)s+w }, the time interval is [t0+((k-1)s)Δt,t0+((k-1)s+w-1)Δt].

3. The method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation according to claim 1, characterized in that: The calculation of the three statistical parameters within each sliding window includes: Using the mature formulas and algorithms of the corresponding industries for standard deviation, range, and approximate entropy, calculate the standard deviation σ of the data in the kth (k=1,2,…,m) sliding window k , range R k , approximate entropy ApEn k .

4. The method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation according to claim 1, characterized in that: The forming of a three-dimensional time series matrix consisting of n groups of parameters based on the three statistical parameters includes: Combine the standard deviation, range, and approximate entropy of m windows into an m×3 matrix M: Where, σ n is the nth standard deviation, R n is the nth range, ApEn n is the nth approximate entropy.

5. The method for detecting and warning abnormal boiler wall temperature based on approximate entropy calculation according to claim 1, characterized in that: The use of the isolation forest anomaly detection algorithm to perform anomaly detection and early warning on the three-dimensional time series matrix includes: Step S4.1: data preprocessing; The standard deviation, range, and approximate entropy of m windows are combined into an m×3 matrix M. Each row of the m×3 matrix M represents a data sample, and each column represents a different feature. There are m samples in total, and each sample has 3 features, representing m points in three-dimensional space. The data is normalized and preprocessed to ensure that different features have similar scales to prevent a single feature from having a large impact on the results. Step 4.2: Build an isolation forest model; In three-dimensional space, an m×3 matrix M represents m points. The m×3 matrix M data is used as a training data set to train and construct an isolation forest model. The isolation forest model learns the normal pattern of the m points in the data, that is, the m×3 matrix M, and can identify significantly different abnormal points. After training is complete, the isolation forest model is used to calculate the data of m points, i.e., the m×3 matrix M, to obtain an anomaly confidence score for each data point. The anomaly confidence score reflects the likelihood of each point being an anomaly, with lower scores indicating a higher likelihood of the point being an anomaly. A threshold is set based on the obtained anomaly score to identify anomalies, and a corresponding early warning mechanism is developed. Step 4.3: Anomaly detection and early warning; Based on the obtained anomaly confidence score for each data point, a threshold θ = the 5th percentile of m anomaly confidence scores is set to identify anomalies; If a point (σ k ,R k ,ApEn k ) is ≤θ, the point is determined to be an abnormal point, and an alert is issued for abnormal temperature fluctuations within the start and end time periods of the sliding window associated with the point [t0+((k-1)s)Δt, t0+((k-1)s+w-1)Δt], where k∈[1,2,…,m].

6. A boiler wall temperature anomaly detection and early warning system according to the boiler wall temperature anomaly detection and early warning method based on approximate entropy calculation according to any one of claims 1 to 5, characterized in that: include: A boiler wall temperature collection device (200), a power plant production equipment SIS database (300), and a boiler wall temperature abnormality real-time monitoring and early warning server (400); The boiler wall temperature collecting device (200) is used to collect the wall temperature of the boiler (100); The power plant production equipment SIS database (300) is used to obtain monitoring data; The boiler wall temperature abnormality real-time monitoring and early warning server (400) acquires monitoring data, inputs the monitoring data into an implementation monitoring program, and implements the induced draft fan bearing temperature abnormality detection and early warning method; The monitoring data includes: the wall temperature of the boiler (100) and real-time operating data of other power plant equipment.

7. A computer device, characterized in that: The invention comprises a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the boiler wall temperature abnormality detection and early warning method based on approximate entropy calculation according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor executes the boiler wall temperature abnormality detection and early warning method based on approximate entropy calculation according to any one of claims 1 to 5.

Images