A stall failure early warning method for parallel operation axial flow draft fan of coal-fired power plant

Abstract

The application provides a coal-fired power station parallel operation axial flow induced draft fan stall failure early warning method, data screening judgment under the steady state condition is carried out through the sliding window based on information entropy, and the normal current curve band corresponding to the unit load is given. The trend test is carried out on the key data of the induced draft fan based on the non-parametric statistical test method, and the significant trend judgment under the confidence degree is given. The collected induced draft fan data are calculated in real time based on the historical failure data and the failure early warning model, the fan stall early warning information is given by comprehensively considering the relationship between the moving blade opening and the flue gas flow and the current state, and considering the operation state of another induced draft fan in parallel operation. The system has a self-learning function, can optimize the early warning model along with the accumulation of equipment operation data, and can be integrated with the monitoring system of the power station to realize automatic early warning. Finally, the accuracy and timeliness of the failure early warning are improved, and the system has a high popularization prospect.

Description

Technical Field

[0001] This invention relates to the field of fault early warning technology for large axial flow induced draft fans, specifically a method for early warning of stall faults in parallel-operated axial flow induced draft fans in coal-fired power plants. Background Technology

[0002] Axial-flow induced draft fans are crucial large auxiliary machines in the flue gas systems of traditional coal-fired power plants. They are used to exhaust flue gas from the boiler and maintain a stable negative pressure environment required for pulverized coal combustion in the furnace. Their operating status directly affects the stability and safety of the entire thermal production system. However, due to the long-term operation of induced draft fans in a complex environment with harsh operating conditions and duct medium temperatures exceeding 100°C, changes in duct resistance characteristics, fan characteristic curve deviations, and sudden blade shut-off can occur due to factors such as flue gas ash accumulation, blade wear, and oil system malfunctions as the equipment's operating cycle lengthens. During operation, if the induced draft fan's operating point deviates from normal conditions, stall failures are likely to occur. These stall failures not only cause airflow fluctuations but can also lead to equipment vibration, worsened thermal effects, and reduced efficiency. In severe cases, they can even damage the fan blades or trigger other system failures, negatively impacting the economic and safe operation of the power plant. Currently, the monitoring and early warning of induced draft fan stall failures mainly rely on manual experience and traditional monitoring methods, which suffer from problems such as untimely warnings and low accuracy.

[0003] Traditional monitoring methods often focus on only one or a few parameters of the induced draft fan, such as vibration amplitude and bearing temperature. However, induced draft fan stall is a complex process, and changes in a single parameter may not accurately reflect the occurrence of the fault. Furthermore, traditional monitoring methods typically only detect anomalies when a fault has already occurred or is about to occur, lacking early warning capabilities. This often leaves power plant personnel in a reactive state when facing faults, making it difficult to take effective preventative measures. In addition, traditional monitoring methods are susceptible to external interference, such as ambient temperature and electromagnetic interference, leading to low accuracy and reliability of monitoring results. In summary, traditional induced draft fan stall fault monitoring systems mostly rely on offline diagnostics, exhibiting a lag and failing to detect and warn of stall faults in a timely manner, potentially leading to serious production accidents and economic losses. Therefore, developing an effective fault early warning method for large axial-flow induced draft fans is of significant practical importance.

[0004] The following is a comparison with existing technologies:

[0005] Technical Comparison with Patent CN113653607B "An Intelligent Early Warning and Diagnosis Method for Power Plant Wind Turbine Stall Based on System Efficiency Model"

[0006] This invention employs a sliding window based on information entropy and nonparametric statistical tests to perform real-time calculations and trend judgments on induced draft fan data, thereby achieving early warning of stall failure. In contrast, CN113653607B focuses on early warning of stall conditions by calculating the total volumetric flow rate and real-time efficiency of two fans. These two inventions differ fundamentally in their early warning mechanisms and data processing methods.

[0007] This invention comprehensively considers the relationship between the blade opening and the flue gas flow rate, and combines the status of the induced draft fans operating in parallel, to provide comprehensive stall warning information; while CN113653607B focuses on accurately locating the fan when a stall occurs, providing intervention time to avoid accidents. The two differ in their emphasis and methods in fault handling.

[0008] The early warning system of this invention has a self-learning function, which can optimize the early warning model; while CN113653607B focuses on real-time monitoring and early warning, and does not mention self-learning or model optimization. The two systems differ significantly in their level of intelligence.

[0009] Technical comparison with patent CN115596696A "A method for real-time online prediction of wind turbine operating status based on data mining"

[0010] This invention calculates induced draft fan data in real time and determines stall status using information entropy and nonparametric statistical methods; while CN115596696A builds multiple prediction models based on data mining, focusing on the extraction and analysis of historical operating status parameters. These two inventions differ fundamentally in their data processing methods and model construction.

[0011] This invention focuses on a comprehensive consideration of the blade opening, flue gas flow rate, and the condition of the parallel fans; while CN115596696A uses the theoretical stall safety factor, pressure margin factor, and flow margin factor for condition assessment. The two differ in their assessment indicators and emphases.

[0012] This invention aims to improve the accuracy and timeliness of fault early warning and possesses self-learning capabilities; while CN115596696A primarily enhances the operational safety and economy of wind turbines, emphasizing the rationality of control strategies. The two differ fundamentally in their ultimate goals and functional implementation. Summary of the Invention

[0013] To address the aforementioned technical problems, this invention proposes a method for early warning of stall faults in axial-flow induced draft fans operating in parallel at coal-fired power plants. This method aims to overcome the shortcomings of existing methods and improve the operational safety and reliability of coal-fired power plants.

[0014] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0015] A stall fault early warning method for axial-flow induced draft fans operating in parallel at a coal-fired power plant is proposed. This method utilizes a sliding window based on information entropy to filter and judge data under steady-state conditions, providing the normal current curve band corresponding to the unit load. It employs non-parametric statistical testing methods to perform trend tests on key fan data and provides a significant trend judgment at a certain confidence level. Based on historical fault data and a fault early warning model, the method performs real-time calculations on the collected fan data. By comprehensively considering the relationship between blade opening and flue gas flow rate, current status, and the operating status of the other induced draft fan operating in parallel, it provides fan stall early warning information. The specific steps include:

[0016] S1: Based on the historical operating data of the axial flow induced draft fan, sensor noise and outliers are removed through data cleaning. Steady-state data are selected through a sliding window steady-state judgment method based on information entropy. Then, statistical methods are used to fit the curve of induced draft fan current versus unit load. The upper and lower limits of the induced draft fan current curve corresponding to the unit load at a certain confidence level are given as the calculation function relationship between the fan current and the fitted curve of the unit.

[0017] S2: Based on the DCS or SIS system of the thermal power plant, collect the running sequence data of the induced draft fan in real time, including but not limited to the induced draft fan opening, flue gas flow, motor current, fan speed, unit load, pressure, temperature and vibration amplitude;

[0018] S3: Based on historical fault data and normal operation data, feature extraction is performed on the collected data using signal processing technology to extract key feature quantities that can characterize the stall state. Feature parameters include parameter change trends, fluctuation amplitudes, and spectral characteristics.

[0019] S4: Based on the extracted feature parameters, a stall fault early warning model for induced draft fans is established using non-parametric statistical testing methods. For axial flow induced draft fans operating in parallel, the trend of time series data of the induced draft fan opening, flue gas flow, motor current, and unit load of the two induced draft fans is tested in real time using non-parametric statistical testing methods. The trend test detects the trend of the time series data curve, including significant upward trend, significant upward trend, and no significant trend.

[0020] S5: Input the real-time collected operating parameters into the early warning model to make fault judgments. If the model judges that the induced draft fan may have a stall fault, it will issue an early warning signal.

[0021] S6: When the early warning model determines that the induced draft fan has stalled, it will promptly notify the power plant staff of the early warning signal so that corresponding preventive measures can be taken. At the same time, the early warning information will be uploaded to the power plant's monitoring system to achieve integration between the early warning system and the power plant's monitoring system, so that operation and maintenance personnel can conduct remote monitoring and decision-making.

[0022] As a further improvement of the present invention, in step S1, the upper and lower limits of the wind turbine current and the unit fitting curve are determined by the confidence interval method based on polynomial fitting based on historical operating data.

[0023] Polynomial fitting is an extension of linear regression, with the model form y = β0 + β1x + β2x 2 +L+β m x m The principle for calculating the confidence interval for +∈ is similar to that for linear regression, and the calculation steps are as follows:

[0024] S101: Estimating polynomial coefficients using the least squares method Obtain the fitting polynomial

[0025] S102: Calculate the covariance matrix Its elements Indicates the estimated coefficient and The covariance.

[0026] S103: For a given x0, the variance estimate of the predicted value is...

[0027] S104: Calculate the confidence interval as follows:

[0028] lower limit

[0029] Maximum: Where t α / 2,n-(m+1) It is the two-sided quantile of the t-distribution with n-(m+1) degrees of freedom.

[0030] As a further improvement of the present invention, step S2 involves obtaining historical operating data of the induced draft fan based on the DCS or SIS system of the thermal power plant, and obtaining a data sequence from the thermal process that needs to be judged to determine the steady state.

[0031] S201: First, data preprocessing is performed. Data cleaning removes sensor noise and outliers to ensure data accuracy. For missing values, methods such as mean filling and interpolation are used. For outliers, including values ​​that deviate significantly from the normal range, thresholds are set for identification and correction. Data points that exceed the normal operating range by ±20% are considered outliers and are remeasured or corrected according to the actual situation.

[0032] S202: Since different parameters in a thermal process may have different dimensions and numerical ranges, the data needs to be normalized to facilitate subsequent information entropy calculations. Minimum-maximum normalization is used, and the formula is as follows: Where x is the original data, x min and x maxThese are the minimum and maximum values ​​in the original data, x. new It is normalized data;

[0033] S203: Based on the characteristics of the thermal process and the time scale of the data, the time series data is divided into several sliding windows. The time width of the sliding window is greater than the sampling period of the DCS or SIS system. Furthermore, the time width of the sliding window is at least 6 times the sampling time of the DCS or SIS system.

[0034] S204: For each data window, calculate its information entropy, assuming the data in the data window is x1, x2, ..., x... n , probability distribution p(x i The information entropy is determined by statistically analyzing the frequency of each data value. The formula for calculating information entropy is:

[0035] S205: When the information entropy of the unit load and current is less than the set threshold, the boiler is considered to be in a relatively stable operating state. The information entropy calculated for each data window is compared with the determined steady-state threshold. If the information entropy is less than or equal to the threshold, the data in that data window is considered to be in a steady state. If the information entropy is greater than the threshold, the data is considered to be in a non-steady state. Then, the data under steady-state conditions is obtained.

[0036] As a further improvement of the present invention, step S2 is a method for obtaining real-time data from the DCS or SIS system of a thermal power plant based on a network communication protocol, wherein the network communication protocol includes, but is not limited to, OPC UA, OPC UA, Modbus TCP, WebAPI, and WebSocket.

[0037] As a further improvement of the present invention, step S4 uses a nonparametric statistical test method based on statistical methods to determine the trend of the time series data curve. This method has no requirements on the distribution of the time series data and is suitable for processing time series data that are not normally distributed or contain outliers.

[0038] Null hypothesis H0: There is no monotonic trend;

[0039] Alternative hypothesis H1: There is a monotonic trend;

[0040] The initial assumption is that H0 is true, and the data must exceed a reasonable doubt—reach a certain level of confidence—before rejecting H0 and accepting H1.

[0041] The steps are as follows:

[0042] S401: Data preparation, given a set of time series data X = {x1, x2, ..., x...} n}, where x iThis represents the i-th observation in the time series, where n is the total number of observations.

[0043] S402: Calculate the S-statistic, defining each pair of data (x, y) in the sequence. i ,x j The difference between j and i is represented by the S-statistic, which is obtained by performing a sign calculation on all differences.

[0044]

[0045] in,

[0046]

[0047] The meaning of S is the sum of the signs of the differences between all observed pairs. If there is an upward trend in the data, the value of S should be positive; conversely, if there is a downward trend, the value of S should be negative.

[0048] S403: Calculate the variance of S. Under the assumption of no trend, calculate the variance Var(S) of the statistic S. When there are repeated data points in the time series, the formula for calculating the variance is as follows:

[0049]

[0050] in:

[0051] n is the total number of data points;

[0052] m is the number of repeating value groups;

[0053] t k It is the number of data points in each group of repeated values;

[0054] When there are no repeating data points in the time series, or when repeating data points are removed, the formula for variance simplifies to:

[0055]

[0056] S404: Calculate the standardized test statistic Z. The Z value is obtained by using the standardized S-statistic. The formula for calculating Z is as follows:

[0057]

[0058] In the formula, the Z-value is used to determine the significance of the trend;

[0059] S405: Determine the significance level, choose a significance level α, which can be 0.01, 0.05, or 0.1, and then find the corresponding critical Z-value according to the normal distribution table. α2 ;

[0060] If |Z|>Z α2If the null hypothesis is rejected, it is considered that there is a significant trend in the data;

[0061] If |Z|≤Z α2 If so, the null hypothesis cannot be rejected, and it is concluded that there is no significant trend in the data;

[0062] S406: Determine the trend direction;

[0063] When Z>0, it indicates that there is a significant upward trend in the time series;

[0064] When Z < 0, it indicates a significant downward trend in the time series;

[0065] When Z = 0, it indicates that there is no significant trend;

[0066] S407: Calculate the gradual slope using the Theil-Sen estimator;

[0067] To estimate the strength of the trend, the Theil-Sen slope estimator is used to calculate the trend slope of the time series. The formula is as follows:

[0068]

[0069] The estimator gives the median slope between each pair of data points in the time series, which indicates the magnitude of the trend.

[0070] As a further improvement of the present invention, in step S5, for two fans operating in parallel, referred to as fan A and fan B respectively, firstly, the changing trends of the opening degree of induced draft fan A and the flue gas volume of the fan are calculated based on a non-parametric statistical test method. If the opening degree of induced draft fan A shows a significant upward trend, and the flue gas volume flowing through induced draft fan A shows a significant downward trend, and the operating current of induced draft fan A exceeds the limit, then the changing trend of the current of induced draft fan B is calculated based on a non-parametric statistical test method. If the current of induced draft fan B shows a significant upward trend at this time, then fan A is considered to have stalled.

[0071] As a further improvement to the present invention, the types of warning information in step S6 include, but are not limited to:

[0072] (1) Audible and visual alarm signal: The alarm emits sound and flashing light to alert the operator to the fault situation;

[0073] (2) Digital display alarm: The fault code, fault type and its severity are displayed on the control panel or monitoring system to facilitate quick problem location;

[0074] (3) Remote alarm notification: Notify relevant personnel via SMS, email or APP to ensure a rapid response when a fault occurs;

[0075] (4) Visual monitoring: The monitoring system provides real-time video monitoring to observe the operating status of the induced draft fan and detect abnormalities in a timely manner;

[0076] (5) Data recording and analysis report: After a fault occurs, the system generates data records and analysis reports to help technicians diagnose the fault and carry out subsequent processing;

[0077] (6) Status indicator lights: Install indicator lights on the equipment to indicate normal, warning or fault status through different colors or flashing patterns.

[0078] As a further improvement of the present invention, after the operation status of the fan is adjusted or repaired, the fault early warning model is automatically updated to obtain the calculation function relationship of the induced draft fan current curve with upper and lower limit curves corresponding to the unit load under a certain confidence level, and to obtain the monitoring threshold of the induced draft fan opening degree, flue gas flow relationship and current state trend curve, so as to ensure the accuracy of the early warning.

[0079] As a further improvement of the present invention, in step S6, the early warning information is uploaded to the power plant's monitoring system, so that the early warning system can be integrated with the power plant's monitoring system. The data upload methods include, but are not limited to, HttpAPI interface, WebSocket interface, and IEC104 interface.

[0080] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0081] The present invention provides a method for early warning of stall faults in parallel operation of large axial flow induced draft fans for coal-fired power plant boilers, which can effectively solve the problem that it is difficult to provide early warning of stall faults in large axial flow induced draft fans in the prior art.

[0082] This method utilizes real-time monitoring and intelligent analysis to promptly detect signs of stall faults and provide early warnings. By comprehensively considering the relationship between blade opening and flue gas flow rate, current status, and the operating status of another induced draft fan operating in parallel, it provides a more comprehensive understanding of the induced draft fan's operating status, enabling timely detection of stall signs and improving the accuracy and reliability of stall judgment for axial-flow induced draft fans. The system possesses a self-learning function, continuously optimizing the early warning model as equipment operating data accumulates, adapting to different operating conditions. The early warning system can be integrated with the power plant's monitoring system to achieve automated early warning, reducing manual intervention and improving the power plant's operational efficiency and safety. The method and system of this invention are easy to implement and can be applied to existing axial-flow induced draft fan equipment without large-scale modifications. It improves the accuracy and timeliness of fault early warning. This method has high real-time performance and accuracy, providing strong support for the safe and stable operation of coal-fired power plants, and has high practical value and promising prospects for widespread application. Attached Figure Description

[0083] Figure 1 This is a schematic diagram of the induced draft fan stall fault early warning method in this invention;

[0084] Figure 2 This is a schematic diagram of steady-state data filtering based on a sliding window in this invention;

[0085] Figure 3 This is a schematic diagram illustrating the execution of the stall fault early warning algorithm established by the nonparametric statistical test method in this invention. Detailed Implementation

[0086] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0087] This invention provides a method for early warning of stall faults in parallel operation of large axial-flow induced draft fans for coal-fired power plant boilers, such as... Figure 1 As shown, it includes the following steps:

[0088] S1: Based on the historical operating data of the axial flow induced draft fan, sensor noise and outliers are removed through data cleaning. Steady-state data are selected through a sliding window steady-state judgment method based on information entropy. Then, statistical methods are used to fit the curve of induced draft fan current versus unit load, and the calculation function relationship of the upper and lower limits of the induced draft fan current curve corresponding to the unit load at a certain confidence level is given.

[0089] S2: Based on the DCS or SIS system of the thermal power plant, collect the running sequence data of the induced draft fan in real time, including but not limited to the induced draft fan opening, flue gas flow, motor current, fan speed, unit load, pressure, temperature and vibration amplitude.

[0090] S3: Based on historical fault data and normal operation data, use signal processing technology to extract features from the collected data and extract key feature quantities that can characterize the stall state. These feature parameters may include parameter change trends, fluctuation amplitudes, spectral characteristics, etc.

[0091] S4: Based on the extracted feature parameters, a stall fault early warning model for induced draft fans is established using nonparametric statistical testing methods. For axial-flow induced draft fans operating in parallel, the trend of time-series data of the induced draft fan opening, flue gas flow, motor current, and unit load of the two induced draft fans is tested in real time using nonparametric statistical testing methods. The trend test includes significant upward trend, significant upward trend, and no significant trend.

[0092] S5: Input the real-time collected operating parameters into the early warning model for fault diagnosis. If the model determines that the induced draft fan may experience a stall fault, it will issue an early warning signal.

[0093] S6: When the early warning model determines that the induced draft fan has stalled, it will promptly notify the power plant staff of the warning signal so that appropriate preventative measures can be taken. Simultaneously, the warning information can be uploaded to the power plant's monitoring system, enabling integration between the early warning system and the power plant's monitoring system for remote monitoring and decision-making by maintenance personnel.

[0094] According to the aerodynamic model of the fan, the motor current of the axial flow fan is proportional to the cube of the flue gas flow rate through the induced draft fan, the volumetric flow rate of the fan is proportional to the fuel quantity, and the fuel quantity is proportional to the load. Therefore, the fan current is proportional to the cube of the load.

[0095] In step S1, the upper and lower limits of the wind turbine current versus the unit's fitted curve are determined using a confidence interval method based on polynomial fitting, based on historical operating data. Polynomial fitting is an extension of linear regression, and the model form is y = β0 + β1x + β2x. 2 +L+β m x m +∈. The principle for calculating confidence intervals is similar to that of linear regression, and the calculation steps are as follows:

[0096] S101: Estimating polynomial coefficients using the least squares method Obtain the fitting polynomial

[0097] S102: Calculate the covariance matrix Its elements Indicates the estimated coefficient and The covariance.

[0098] S103: For a given x0, the variance estimate of the predicted value is...

[0099] S104: Calculate the confidence interval as follows:

[0100] lower limit

[0101] Maximum: Where t α / 2,n-(m+1) It is the two-sided quantile of the t-distribution with n-(m+1) degrees of freedom.

[0102] Furthermore, in step S2, historical operating data of the induced draft fan is obtained based on the DCS or SIS system of the thermal power plant, and a data sequence that needs to be determined to be steady state is obtained from the thermal process.

[0103] S201: First, data preprocessing is performed. Data cleaning removes sensor noise and outliers to ensure data accuracy. For missing values, methods such as mean imputation and interpolation can be used. For outliers (such as values ​​that significantly deviate from the normal range), thresholds can be set for identification and correction. For example, data points exceeding the normal operating range by ±20% are considered outliers and can be remeasured or corrected as needed.

[0104] S202: Since different parameters in a thermal process may have different dimensions and numerical ranges, data normalization is necessary to facilitate subsequent information entropy calculations. Minimum-maximum normalization is used, with the following formula: Where x is the original data, x min and x max These are the minimum and maximum values ​​in the original data, x. new It is the normalized data.

[0105] S203: Based on the characteristics of the thermal process and the time scale of the data, the time series data is divided into several sliding windows. The time width of the sliding window is greater than the sampling period of the DCS or SIS system. Furthermore, the time width of the sliding window is at least 6 times the sampling time of the DCS or SIS system.

[0106] S204: For each data window, calculate its information entropy. Assume the data in the data window is x1, x2, ..., x... n , probability distribution p(x i Information entropy can be determined by statistically analyzing the frequency of each data value. The formula for calculating information entropy is...

[0107] S205: When the information entropy of both the unit load and current is less than a set threshold, the boiler is considered to be in a relatively stable operating state. The information entropy calculated for each data window is compared with the determined steady-state threshold. If the information entropy is less than or equal to the threshold, the data within that data window is considered to be in a steady state; if the information entropy is greater than the threshold, the data is considered to be in a non-steady state. Data under steady-state conditions is then obtained.

[0108] like Figure 2 The diagram illustrates a sliding window, where a suitable window size (N) is selected to slide across the data sequence and extract subsequences. The window size should be adjusted based on data characteristics and analytical needs. Applying a sliding window to the data sequence progressively extracts subsequences of a fixed size. First, set the starting position by placing the window at the beginning of the data sequence. Then, calculate the statistical characteristics (such as mean, variance, etc.) within the current window. Finally, move the window forward by one data point and repeat the calculation until the entire data sequence has been traversed.

[0109] Furthermore, step S2 is a method for obtaining real-time data from the DCS or SIS system of a thermal power plant based on network communication protocols, including but not limited to OPC UA, OPC UA, Modbus TCP, Web API, and WebSocket.

[0110] Step S4 uses a nonparametric statistical test based on statistical methods to determine the trend of the time series data curve. This method has no requirements on the distribution of the time series data and is suitable for processing time series data that are not normally distributed or contain outliers.

[0111] Null hypothesis H0: There is no monotonic trend;

[0112] Alternative hypothesis H1: There is a monotonic trend;

[0113] The initial assumption is that H0 is true, and the data must exceed a reasonable doubt—reach a certain level of confidence—before rejecting H0 and accepting H1.

[0114] The steps are as follows:

[0115] S401: Data preparation, given a set of time series data X = {x1, x2, ..., x...} n}, where x i This represents the i-th observation in the time series, where n is the total number of observations.

[0116] S402: Calculate the S-statistic, defining each pair of data (x, y) in the sequence. i ,x j The difference (where j > i) is calculated. The S-statistic is obtained by symbolically calculating all differences.

[0117]

[0118] in,

[0119]

[0120] The value of S represents the sum of the signs of the differences between all observed pairs. If there is an upward trend in the data, the value of S should be positive; conversely, if there is a downward trend, the value of S should be negative.

[0121] S403: Calculate the variance of S. Under the assumption of no trend, calculate the variance Var(S) of the statistic S. When there are repeated data points in the time series, the formula for calculating the variance is as follows:

[0122]

[0123] in:

[0124] n is the total number of data points.

[0125] m is the number of repeated value groups.

[0126] t k It represents the number of data points in each group of repeated values.

[0127] When there are no repeating data points in the time series, or when repeating data points are removed, the formula for variance simplifies to:

[0128]

[0129] S404: Calculate the standardized test statistic Z. The Z value is obtained by using the standardized S-statistic. The formula for calculating Z is as follows:

[0130]

[0131] In the formula, the Z-value is used to determine the significance of the trend.

[0132] S405: Determine the significance level, select a significance level α (usually 0.01, 0.05, or 0.1), and then find the corresponding critical Z-value according to the normal distribution table. α2 .

[0133] If |Z|>Z α2 If the null hypothesis is rejected, it is considered that there is a significant trend in the data.

[0134] If |Z|≤Z α2 If the null hypothesis is not accepted, then the null hypothesis cannot be rejected, and it is concluded that there is no significant trend in the data.

[0135] S406: Determine the trend direction

[0136] When Z>0, it indicates a significant upward trend in the time series.

[0137] When Z < 0, it indicates a significant downward trend in the time series.

[0138] When Z = 0, it indicates that there is no significant trend.

[0139] S407: Calculate the gradual slope (Theil-Sen estimator)

[0140] To estimate the strength of the trend, the Theil-Sen slope estimator is used to calculate the trend slope of the time series. The formula is as follows:

[0141]

[0142] This estimator gives the median slope between each pair of data points in the time series, which can be used to represent the magnitude of the trend.

[0143] For example, for wind turbine opening time series data, taking α = 0.1 as an example, Looking up the standard normal distribution table, Z0.95 = 1.645. Therefore, when Z ≥ 1.645, the 90% significance test is passed, and the null hypothesis (H0) is not true. Z > 0 indicates an upward trend in the series; that is, when Z ≥ 1.645, there is a 90% probability that the wind turbine operating rate shows a significant upward trend. If Z < 0, the series shows a downward trend; that is, when Z ≤ -1.645, there is a 90% probability that the wind turbine operating rate shows a significant downward trend. For example, for wind turbine current time series data, taking α = 0.05 as an example... Looking up the standard normal distribution table, Z0.975 = 1.96. Therefore, when Z ≥ 1.96, the significance test is passed at 95%, and the null hypothesis (H0) is not true. If Z > 0, the sequence shows an upward trend; that is, when Z ≥ 1.96, there is a 95% probability that the wind turbine current shows a significant upward trend. If Z < 0, the sequence shows a downward trend; that is, when Z ≤ -1.96, there is a 95% probability that the wind turbine current shows a significant upward trend.

[0144] like Figure 3 The diagram shows the execution of the stall fault early warning algorithm established by the nonparametric statistical test method in this invention. For an adjustable-blade axial flow fan, under normal conditions, increasing the blade opening will increase the fan flow rate, and there is a strong linear relationship between the fan's air volume and the blade opening. Based on the above fitting results, it can be considered that under normal fan operation, the blade opening and fan flow rate change linearly. However, when the fan is about to stall, the relationship between its air volume and blade opening will deviate from the relationship under normal operation. Before the fan stalls, its operating current will gradually increase. Based on the comprehensive criterion of air volume-blade opening relationship and current, a stall warning is issued for the fan. In addition, when the "automatic" mode is activated, the current and volume ratio of another fan operating in parallel with the stalled fan can increase significantly.

[0145] Increased output can easily cause motor overload. When a fan stalls, the air volume and pressure drop significantly, causing drastic changes in combustion in the furnace and increasing the risk of fire extinguishing accidents. When another fan operating in parallel is put into "automatic" mode, the increased output can easily cause motor overload. The vibration of a stalled fan increases significantly, potentially causing damage to the fan equipment and ductwork. Incorrect handling can easily trigger fan surge, damaging the equipment. Based on these characteristics...

[0146] In step S5, for two parallel-operating fans, referred to as fan A and fan B, the changing trends of the opening degree and flue gas volume of induced draft fan A are first calculated using a nonparametric statistical test method. If the opening degree of induced draft fan A shows a significant upward trend, and the flue gas volume flowing through induced draft fan A shows a significant downward trend, and the operating current of induced draft fan A exceeds its limit, then the changing trend of the current of induced draft fan B is calculated using a nonparametric statistical test method. If the current of induced draft fan B shows a significant upward trend at this time, then fan A is considered to have stalled.

[0147] Furthermore, the types of warning information in step S6 include, but are not limited to:

[0148] (1) Audible and visual alarm signal: The alarm emits sound and flashing light to alert the operator to the fault.

[0149] (2) Digital display alarm: Displays fault codes, fault types and their severity on the control panel or monitoring system to facilitate quick problem location.

[0150] (3) Remote alarm notification: Notify relevant personnel via SMS, email or APP to ensure a rapid response when a fault occurs.

[0151] (4) Visual monitoring: The monitoring system provides real-time video monitoring to observe the operating status of the induced draft fan and detect abnormalities in a timely manner.

[0152] (5) Data recording and analysis report: After a fault occurs, the system generates data records and analysis reports to help technicians diagnose the fault and carry out subsequent processing.

[0153] (6) Status indicator lights: Install indicator lights on the equipment to indicate normal, warning or fault status through different colors or flashing patterns.

[0154] Furthermore, after the fan's operating status is adjusted or it is overhauled, the fault warning model is automatically updated to obtain the calculation function relationship of the induced draft fan current curve with upper and lower limit curves corresponding to the unit load under a certain confidence level, and to obtain the monitoring thresholds of the induced draft fan opening, flue gas flow relationship, and current state trend curve to ensure the accuracy of the warning.

[0155] Furthermore, in step S6, the early warning information is uploaded to the power plant's monitoring system, enabling the early warning system to be integrated with the power plant's monitoring system. The data upload methods include, but are not limited to, HttpAPI interface, WebSocket interface, and IEC104 interface.

[0156] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A method for early warning of stall faults in parallel-operated axial-flow induced draft fans of a coal-fired power plant, characterized in that: Includes the following steps: S1: Based on the historical operating data of the axial flow induced draft fan, sensor noise and outliers are removed through data cleaning. Steady-state data are selected through the sliding window steady-state judgment method based on information entropy. Then, statistical methods are used to fit the curve of induced draft fan current versus unit load, and the calculation function relationship of the upper and lower limits of the induced draft fan current curve corresponding to the unit load at a certain confidence level is given. S2: Based on the DCS or SIS system of the thermal power plant, collect the running sequence data of the induced draft fan in real time, including but not limited to the induced draft fan opening, flue gas flow, motor current, fan speed, unit load, pressure, temperature and vibration amplitude; S3: Based on historical fault data and normal operation data, signal processing technology is used to extract features from the collected data to extract characteristic parameters that can characterize the stall state. The characteristic parameters include the parameter change trend, fluctuation amplitude and spectral characteristics. S4: Based on the extracted feature parameters, a non-parametric statistical test method is used to establish an early warning model for stall failure of induced draft fans. For axial flow induced draft fans operating in parallel, the trend test method is used to perform real-time trend test on the induced draft fan opening, flue gas flow, motor current, and unit load of the two induced draft fans. The trend test detects the trend of the time series data curve, including a significant upward trend and no significant trend. S5: Input the real-time collected operating parameters into the early warning model to determine the fault. If the model determines that the induced draft fan may experience a stall fault, it will issue an early warning signal. S6: When the early warning model determines that the induced draft fan has stalled, it will promptly notify the power plant staff of the early warning signal so that corresponding preventive measures can be taken. At the same time, the early warning information will be uploaded to the power plant's monitoring system to achieve integration between the early warning system and the power plant's monitoring system, so that operation and maintenance personnel can conduct remote monitoring and decision-making.

2. The method for early warning of stall fault of parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 1, characterized in that: In step S1, the upper and lower limits of the wind turbine current and the unit fitting curve are determined by using the confidence interval method based on polynomial fitting based on historical operating data. Polynomial fitting is an extension of linear regression, and the model has the following form: The principle for calculating confidence intervals is similar to that of linear regression, and the calculation steps are as follows: S101: Estimating polynomial coefficients using the least squares method The fitting polynomial is obtained. ; S102: Calculate the covariance matrix Its elements Indicates the estimated coefficient and covariance; S103: For a given The variance estimate of the predicted value is ; S104: Calculate the confidence interval as follows: lower limit ; Maximum: ,in It is the two-sided quantile of a t-distribution with n-(m+1) degrees of freedom.

3. The method for early warning of stall fault of parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 1, characterized in that: In step S2, historical operating data of the induced draft fan is obtained based on the DCS or SIS system of the thermal power plant, and the data sequence that needs to be judged for steady state is obtained from the thermal process. S201: First, data preprocessing is performed. Data cleaning removes sensor noise and outliers to ensure data accuracy. For missing values, methods such as mean filling and interpolation are used. For outliers, including values ​​that deviate significantly from the normal range, thresholds are set for identification and correction. Data points that exceed the normal operating range by ±20% are considered outliers and are remeasured or corrected according to the actual situation. S202: Since different parameters in a thermal process may have different dimensions and numerical ranges, the data needs to be normalized to facilitate subsequent information entropy calculations. Minimum-maximum normalization is used, and the formula is as follows: ,in It is the raw data. and These are the minimum and maximum values ​​in the original data, respectively. It is normalized data; S203: Based on the characteristics of the thermal process and the time scale of the data, the time series data is divided into several sliding windows. The time width of the sliding window is greater than the sampling period of the DCS or SIS system. Furthermore, the time width of the sliding window is at least 6 times the sampling time of the DCS or SIS system. S204: For each data window, calculate its information entropy, assuming the data in the data window is... probability distribution The information entropy is determined by statistically analyzing the frequency of each data value. The formula for calculating information entropy is: ; S205: When the information entropy of the unit load and current is less than the set threshold, the boiler is considered to be in a relatively stable operating state. The information entropy calculated for each data window is compared with the determined steady-state threshold. If the information entropy is less than or equal to the threshold, the data in that data window is considered to be in a steady state. If the information entropy is greater than the threshold, the data is considered to be in a non-steady state. Then, the data under steady-state conditions is obtained.

4. The method for early warning of stall fault of parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 3, characterized in that: Step S2 is a method for obtaining real-time data from the DCS or SIS system of a thermal power plant based on network communication protocols, including but not limited to OPC UA, OPC UA, Modbus TCP, Web API, and WebSocket.

5. The method for early warning of stall fault of parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 1, characterized in that: Step S4 uses a nonparametric statistical test method based on statistical methods to determine the trend of the time series data curve. It has no requirements on the distribution of the time series data and is suitable for processing time series data that are not normally distributed or contain outliers. Null hypothesis H0: There is no monotonic trend; Alternative hypothesis H1: There is a monotonic trend; The initial assumption is that H0 is true, and the data must exceed a reasonable doubt—reach a certain level of confidence—before rejecting H0 and accepting H1. The steps are as follows: S401: Data preparation, a set of time series data is available. ,in Represents the first in the time series One observation value, It is the total number of observations; S402: Calculate the S statistic, defining each pair of data in the sequence. ,in The differences are calculated by performing sign calculations on all differences to obtain the S statistic; ； in, ； The meaning of S is the sum of the signs of the differences between all observed pairs. If there is an upward trend in the data, the value of S should be positive; conversely, if there is a downward trend, the value of S should be negative. S403: Calculate the variance of statistic S. Under the assumption of no trend, calculate the variance of statistic S. When there are repeated data points in a time series, the variance is calculated using the following formula: ； in: It is the total number of data points; It is the number of duplicate value groups; It is the number of data points in each group of repeated values; When there are no repeating data points in the time series, or when repeating data points are removed, the formula for variance simplifies to: ； S404: Calculate the standardized test statistic Z. The Z value is obtained through the standardized S statistic. The formula for calculating Z is as follows: ； In the formula, the Z value is used to determine the significance of the trend; S405: Determine the significance level and select the significance level. The value is 0.01, 0.05, or 0.

1. Then, the corresponding critical value is found according to the normal distribution table. ; if If the null hypothesis is rejected, it is considered that there is a significant trend in the data; if If so, the null hypothesis cannot be rejected, and it is concluded that there is no significant trend in the data; S406: Determine the trend direction; When Z>0, it indicates that there is a significant upward trend in the time series; When Z < 0, it indicates a significant downward trend in the time series; When Z=0, it indicates that there is no significant trend; S407: Calculate the gradual slope using the Theil-Sen estimator; To estimate the strength of the trend, the Theil-Sen slope estimator is used to calculate the trend slope of the time series. The formula is as follows: ； The estimator gives the median slope between each pair of data points in the time series, which indicates the magnitude of the trend.

6. The method for early warning of stall fault of parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 1, characterized in that: In step S5, for two fans operating in parallel, referred to as fan A and fan B, the changing trends of the opening degree of induced draft fan A and the flue gas volume of the fan are first calculated based on the nonparametric statistical test method. If the opening degree of induced draft fan A shows a significant upward trend, and the flue gas volume flowing through induced draft fan A shows a significant downward trend, and the operating current of induced draft fan A exceeds the limit, then the changing trend of the current of induced draft fan B is calculated based on the nonparametric statistical test method. If the current of induced draft fan B shows a significant upward trend at this time, then fan A is considered to have stalled.

7. The method for early warning of stall fault of parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 1, characterized in that: The types of warning information in step S6 include, but are not limited to: (1) Audible and visual alarm signal: The alarm emits sound and flashing light to alert the operator to the fault situation; (2) Digital display alarm: The fault code, fault type and its severity are displayed on the control panel or monitoring system to facilitate quick problem location; (3) Remote alarm notification: Notify relevant personnel via SMS, email or APP to ensure a rapid response when a fault occurs; (4) Visual monitoring: The monitoring system provides real-time video monitoring to observe the operating status of the induced draft fan and detect abnormalities in a timely manner; (5) Data recording and analysis report: After a fault occurs, the system generates data records and analysis reports to help technicians diagnose the fault and carry out subsequent processing; (6) Status indicator lights: Install indicator lights on the equipment to indicate normal, warning or fault status through different colors or flashing patterns.

8. The method for early warning of stall fault of parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 7, characterized in that: After the fan's operating status is adjusted or it is overhauled, the fault warning model is automatically updated to obtain the calculation function relationship of the induced draft fan current curve with upper and lower limit curves corresponding to the unit load under a certain confidence level. The monitoring thresholds of the induced draft fan opening, flue gas flow relationship, and current state trend curve are also obtained to ensure the accuracy of the warning.

9. A method for early warning of stall fault of a parallel-operated axial-flow induced draft fan in a coal-fired power plant according to claim 7, characterized in that: In step S6, the early warning information is uploaded to the power plant's monitoring system, enabling the early warning system to be integrated with the power plant's monitoring system. Data upload methods include, but are not limited to, Http API interface, WebSocket interface, and IEC104 interface.

Images