A boiler heating surface over-temperature early warning method

Abstract

The application provides a boiler heating surface over-temperature early warning method, selects main factors affecting the wall temperature of the boiler heating surface, determines the maximum position of thermal deviation based on historical data, determines the wall temperature prediction length required by each heating surface in combination with the dynamic response characteristics of the wall temperature, selects key characteristic variables affecting the wall temperature by adopting an XGBoost model feature importance extraction method, optimizes the XGBoost model super parameter by adopting SSA, constructs a boiler heating surface wall temperature prediction model, determines the prediction interval of the wall temperature at different moments in combination with the confidence interval, determines the over-temperature early warning threshold of the heating surface by probability density estimation, and adopts a sliding window detection method to statistically determine the average value of the upper limit of the predicted wall temperature in the window, judges whether the average value exceeds the early warning threshold, and realizes over-temperature early warning of the wall temperature of the heating surface. The boiler heating surface over-temperature early warning method provided by the application can provide the wall temperature early warning information of the boiler heating surface for the operation personnel, reduces the possibility of over-temperature pipe explosion, and helps the safe operation of the boiler.

Description

Technical Field

[0001] This invention belongs to the field of boiler safety technology, and in particular relates to an over-temperature early warning method for boiler heating surfaces. Background Technology

[0002] When boilers operate under peak-shaving conditions, especially deep peak-shaving conditions, the feedwater flow rate decreases, and the water circulation dynamics deteriorate, inevitably leading to heat load deviations. This causes the wall temperature of some heating surfaces to exceed their maximum allowable temperature. Prolonged operation with excessively high temperatures can cause wall cracks or even tube ruptures, seriously affecting the safe operation of the unit. Therefore, accurately predicting the trend of boiler heating surface wall temperature changes, rationally determining the wall temperature threshold for normal operation, and detecting and issuing early warnings of excessive temperatures can effectively reduce the impact of deep peak-shaving on the heating surfaces of coal-fired boilers and decrease the possibility of tube rupture due to excessive temperatures.

[0003] Currently, research on boiler heating surface wall temperature overheating early warning in my country mainly relies on numerical simulation and data-driven methods. Numerical simulation can obtain detailed information about the working fluid flow, combustion, heat transfer processes, and temperature distribution inside the boiler. However, due to the nonlinear, hysteretic, and uncertain characteristics of coal-fired boilers, their combustion and heat transfer mechanisms are complex, making it very difficult to establish accurate numerical simulation models. Furthermore, numerical simulation methods are time-consuming, have limited simulation conditions, and struggle to obtain boiler wall temperature characteristics under continuous operating conditions, thus restricting their engineering applications. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, a method for early warning of overheating of boiler heating surfaces is provided.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A method for early warning of boiler heating surface overheating includes: selecting the main influencing factors affecting the boiler heating surface wall temperature; determining the location with the largest heat load deviation based on historical operating data of wall temperature; determining the required wall temperature prediction time for each heating surface by combining the dynamic response characteristics of wall temperature; selecting key feature variables affecting wall temperature using the XGBoost model feature importance extraction method; constructing a boiler heating surface wall temperature prediction model by optimizing the hyperparameters of the XGBoost model through SSA; determining the prediction interval of wall temperature at different times by combining the confidence interval algorithm; determining the early warning threshold of heating surface overheating by estimating the probability density distribution of data samples; and using a sliding window detection method to statistically analyze the average value of the upper limit of the predicted wall temperature within the window to determine whether it exceeds the early warning threshold, thereby realizing the detection and early warning of boiler heating surface wall temperature overheating. The specific steps include:

[0007] Step 1: Data collection and storage of key factors affecting boiler heating surface wall temperature: Real-time data collection of key factors affecting boiler heating surface wall temperature and operating data of each boiler heating surface wall temperature; Simultaneously, synchronous storage of operating data of each boiler heating surface wall temperature to establish a historical operating database; Key factors affecting boiler heating surface wall temperature include coal feed rate, furnace air volume, flue gas damper regulating valve position, main steam temperature, main steam pressure, boiler feedwater flow rate, feedwater temperature, superheating desuperheating water flow rate, superheating desuperheating water temperature, and unit load;

[0008] Step 2, Data Analysis and Selection of Key Feature Variables: Based on historical operating data, determine the location with the largest heat load deviation, and combine the wall temperature response characteristics to determine the required wall temperature prediction duration for the heated surface; use the XGBoost model feature importance extraction method to select key feature variables affecting the wall temperature, and reconstruct the runtime sequence data of the obtained key feature variables into supervised learning data;

[0009] Step 3: Construction of boiler heating surface wall temperature prediction model: The hyperparameters of the XGBoost model are optimized by the Sparrow Search Algorithm (SSA) to construct the SSA-XGBoost wall temperature prediction model; combined with the confidence interval algorithm, the predicted wall temperature point values ​​are expanded into a wall temperature prediction interval.

[0010] Step 4: Determination of the over-temperature warning threshold for boiler heating surface: Steady-state screening of historical operating data, using the Parezn window to estimate the probability density distribution of wall temperature data samples, setting the confidence level to 85%, and obtaining the upper limit of the wall temperature confidence interval under different operating conditions as the over-temperature warning threshold.

[0011] Step 5, Boiler heating surface wall temperature over-temperature warning: Using the sliding window detection method, the average value of the upper limit of the wall temperature at different prediction times within the window is statistically analyzed. It is then determined whether the average value of the upper limit of the wall temperature is less than the over-temperature warning threshold. If the average value of the upper limit of the wall temperature is greater than the over-temperature warning threshold, an over-temperature warning is issued; otherwise, the detection continues.

[0012] Furthermore, in step 1, the parameters of coal feed rate, furnace air volume, flue gas damper regulating valve position, main steam temperature, main steam pressure, boiler feedwater flow rate, feedwater temperature, superheated desuperheating water flow rate, superheated desuperheating water temperature, unit load, and boiler heating surface wall temperature are collected and stored every 30 seconds.

[0013] Furthermore, step 2 specifically includes the following steps:

[0014] Step 201: Determining the locations of boiler heating surfaces prone to overheating: Analyze the historical operating data of the wall temperature of each heating surface of the boiler, observe the heat load distribution at each location, find the location with the largest heat load deviation, and take the wall temperature at the location with the largest heat load deviation as the main observation object for boiler heating surface overheating early warning.

[0015] Step 202: Determination of the prediction time for boiler heating surface wall temperature: Based on the historical operating data of the heating surface wall temperature and its cooling measures, analyze the dynamic response characteristics of the boiler heating surface wall temperature and determine the dynamic response time m of the wall temperature; to avoid abnormal alarms due to abnormal real-time data acquisition, a time margin of m / 5 is used as the sliding alarm window; the final required wall temperature prediction time is determined to be 6m / 5.

[0016] Step 203: Determination of key feature variables of boiler heating surface wall temperature: Using the XGBoost model feature importance extraction method, the five factors with the highest XGBoost model feature importance were selected from the main influencing factors of boiler heating surface wall temperature obtained in Step 1 as key feature variables.

[0017] Step 204, Time Series Data Reconstruction: Use the Pandas.shift function in Python to convert the time series data of key feature variables into supervised learning data that can be processed by machine learning algorithms.

[0018] Furthermore, step 3 specifically includes the following steps:

[0019] Step 301, Construction of SSA-XGBoost wall temperature prediction model: The hyperparameters of the XGBoost model are optimized by SSA to construct the wall temperature prediction model; the hyperparameters include the number of trees, the depth of the regression tree, the learning rate, and the weight of the leaf samples.

[0020] Step 302, Determining the number of input data samples for the wall temperature prediction model: By calculating and comparing the root mean square error of the SSA-XGBoost model under different input sample numbers, the number of input data samples, HistoryWindow, is determined.

[0021] Step 303: Determining the predicted range of boiler heating surface wall temperature: Calculate and compare the root mean square error of the predicted range of wall temperature under different significance levels, select a root mean square error of less than 4 to determine the confidence level, and determine the predicted wall temperature range at different times.

[0022] Furthermore, step 301 involves optimizing the hyperparameters of the XGBoost model based on SSA, including the following steps:

[0023] Step 1A: Initialize parameters, including population size, ratio of discoverers to watchers, watch threshold and safety threshold, and maximum number of iterations;

[0024] Step 1B: Initialize the location of individual populations;

[0025] Step 1C: Calculate the fitness value and corresponding position of individual sparrows, and randomly select 20% of the sparrows with high fitness as discoverers, and the remaining sparrows as joiners;

[0026] Step 1D: Update the individual sparrow positions and the global optimal position;

[0027] Step 1E: Determine and verify the termination condition: If the maximum number of iterations is reached, stop the loop and output the optimal hyperparameters of the XGBoost model; otherwise, return to step 1C to continue the calculation.

[0028] Furthermore, step 303, determining the predicted wall temperature range, includes the following steps:

[0029] Step 3A: Describe the predicted wall temperature range as (x max (t),x min (t)) uses the t-distribution method to calculate the confidence interval for the input samples of size HistoryWindow+N; where HistoryWindow represents the number of data samples input to the wall temperature prediction model, and N represents the number of predicted data samples after prediction time t; the upper limit of the interval is x. max (t) and the lower limit of the interval x min (t) is calculated using the following formula:

[0030]

[0031]

[0032] Where n is the number of samples, and n = HistoryWindow + N; X is the sample mean; s is the sample variance; α represents the significance level;

[0033] Step 3B: Calculate the root mean square error at different significance levels at prediction time t; measured wall temperature x of the heated surface. true If the temperature falls within the predicted operating range, the prediction accuracy is considered 100%; otherwise, the root mean square error R of the wall temperature prediction range is calculated using the following formula. range Perform the calculation:

[0034]

[0035]

[0036] Step 3C: Set the root mean square error threshold. Under the premise that the root mean square error is less than the threshold, select the minimum confidence level for the wall temperature point value at prediction time t and determine the wall temperature prediction interval.

[0037] Furthermore, step 4 specifically includes the following steps:

[0038] Step 401, Steady-state data screening: The sliding window detection method is used to screen historical operating data based on steady-state discrimination indicators to construct a steady-state operating condition database; the indicators include main steam temperature, pressure, unit load, and reheat steam temperature and pressure;

[0039] Step 402, Wall temperature probability density distribution estimation: Using the Parzen window method and the working condition division method, the wall temperature probability density distribution under different working conditions is calculated;

[0040] Step 403: Determination of the over-temperature warning threshold for the heated surface: Set the confidence level to 85%, calculate the upper limit of the wall temperature confidence interval under different operating conditions as the over-temperature warning threshold, and establish a load-wall temperature threshold database.

[0041] Furthermore, in step 402, the Parzen window method is used to calculate the wall temperature probability density distribution under different operating conditions. The probability density estimation formula is as follows: Where S is the number of samples to be estimated, h is the window bandwidth, and r is the data dimension to be estimated. It's the Gaussian kernel function.

[0042] Furthermore, step 5 specifically includes the following steps:

[0043] Step 501: Select a wall temperature prediction time series with a window length of m / 6 from the starting time i forward, and statistically calculate the average value of the upper limit of the prediction interval of the wall temperature from time i+m / 6 to time i.

[0044] Step 502: Based on the real-time unit load, determine the early warning threshold for the boiler heating surface wall temperature under the current load in the load-wall temperature threshold database;

[0045] Step 503: Determine whether the average value of the predicted upper limit of wall temperature in the window is less than the over-temperature warning threshold under the corresponding operating condition. If the average value of the upper limit of wall temperature is greater than the warning threshold, then an over-temperature warning for the heated surface is issued.

[0046] Step 504: Increment the step size k to obtain a new time series. Repeat steps 502 and 503 to recursively deduce the result.

[0047] The beneficial effects of this invention are:

[0048] The boiler heating surface overheating early warning method of this invention identifies locations with significant thermal deviations based on historical operating data of wall temperature. Combining this with the dynamic response characteristics of wall temperature, it determines the required prediction time for each heating surface. The XGBoost model feature importance extraction method is used to select key feature variables affecting wall temperature. By optimizing the hyperparameters of the XGBoost model through SSA, a boiler heating surface wall temperature prediction model is constructed. Combined with confidence interval calculations, the prediction interval for wall temperature at different times is determined, expanding the predicted value of the boiler heating surface wall temperature to a prediction interval value. This avoids overheating misjudgments caused by prediction errors due to thermal delay and hysteresis, improving the model's robustness. The overheating warning threshold is determined by estimating the probability density distribution of the data samples. A sliding window detection method is used to statistically analyze the average value of the upper limit of the predicted wall temperature within the window to determine whether it exceeds the warning threshold, thus achieving boiler heating surface wall temperature overheating detection and early warning. Determining the overheating warning threshold based on the probability density distribution of historical data replaces the traditional method of using the boiler's factory design value, helping to detect overheating trends of the heating surface as early as possible. The boiler heating surface over-temperature early warning method proposed in this invention takes into account the variability of actual boiler operating conditions, the complexity of the influence of multiple factors on the heating surface wall temperature, and the delay and lag of wall temperature changes. Based on the boiler's operating conditions and its historical operating data, the over-temperature early warning threshold under different operating conditions is determined. It fully considers the impact of boiler peak-shaving and variable operating conditions on wall temperature, can accurately predict the trend of boiler heating surface wall temperature changes, reasonably determine its over-temperature early warning threshold, detect tube wall over-temperature trends and issue early warnings as early as possible, effectively reduce the impact of deep peak shaving on the over-temperature of coal-fired boiler heating surfaces, reduce the possibility of over-temperature tube rupture, improve the accuracy of the early warning method, and assist in achieving safe boiler operation. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0050] Figure 2 This is a flowchart illustrating the data analysis and key feature variable selection process of this invention.

[0051] Figure 3 The flowchart illustrates the construction of the boiler heating surface wall temperature prediction model of this invention.

[0052] Figure 4 This is a flowchart illustrating the process for determining the over-temperature warning threshold for the boiler heating surface according to the present invention.

[0053] Figure 5 This is a flowchart of the boiler heating surface over-temperature early warning detection process of the present invention. Detailed Implementation

[0054] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0055] like Figures 1-5 As shown, this invention discloses a method for early warning of boiler heating surface overheating. The method selects the main influencing factors affecting the boiler heating surface wall temperature; determines the location of the largest thermal deviation based on historical wall temperature operating data; determines the required wall temperature prediction time for each heating surface by combining the dynamic response characteristics of the wall temperature; selects key feature variables affecting the wall temperature using the XGBoost model feature importance extraction method; constructs a boiler heating surface wall temperature prediction model by optimizing the hyperparameters of the XGBoost model through SSA; determines the prediction interval of the wall temperature at different times by combining the confidence interval algorithm; determines the early warning threshold for heating surface overheating by estimating the probability density distribution of the data samples; and uses a sliding window detection method to statistically analyze the average value of the upper limit of the predicted wall temperature within the window to determine whether it exceeds the early warning threshold, thereby realizing the detection and early warning of boiler heating surface wall temperature overheating.

[0056] Figure 1 This is the overall flowchart of the boiler heating surface over-temperature early warning method.

[0057] The boiler heating surface overheating early warning method includes the following steps:

[0058] Step 1: Data collection and storage of key factors affecting boiler heating surface wall temperature: Real-time data collection of key factors affecting boiler heating surface wall temperature and operating data of each heating surface wall temperature. Key factors affecting boiler heating surface wall temperature include coal feed rate, furnace air volume, flue gas damper regulating valve position, main steam temperature, main steam pressure, boiler feedwater flow rate, feedwater temperature, superheating desuperheating water flow rate, superheating desuperheating water temperature, and unit load. Simultaneously, the operating data of each heating surface wall temperature of the boiler are stored synchronously to establish a historical operating database.

[0059] This invention collects signals related to coal feed rate, furnace air volume, flue gas damper regulating valve position, main steam temperature, main steam pressure, boiler feedwater flow rate, feedwater temperature, superheater desuperheater water flow rate, superheater desuperheater water temperature, unit load, and boiler heating surface wall temperature. These signals are acquired from a distributed control system (DCS) via OPC or from a plant-level supervision information system (SIS) via API calls and stored. Under the premise of meeting analytical requirements, the values ​​of these signals are collected and stored every 30 seconds.

[0060] Step 2: Data Analysis and Selection of Key Feature Variables: Based on historical operating data, the location with the largest heat load deviation is determined. Combined with wall temperature response characteristics, the required prediction duration for the heated surface wall temperature is determined. The XGBoost model feature importance extraction method is used to select key feature variables affecting the wall temperature, and the runtime sequence data of the obtained key feature variables are reconstructed into supervised learning data.

[0061] This invention monitors and predicts the wall temperature at locations on the boiler's heating surface where thermal deviations are large and overheating is frequent. This allows for wall temperature control at vulnerable points on the boiler's heating surface prone to overheating, ensuring the safe operation of the heating surface. Simultaneously, because the boiler is a large thermally inertial body, the dynamic response of the wall temperature after cooling measures is implemented requires a certain amount of time; the change in wall temperature exhibits delayed and subsequent characteristics. Therefore, the determination of the wall temperature prediction time in this invention considers both the dynamic response time of the wall temperature and incorporates a certain time margin.

[0062] A coal-fired boiler is a complex, nonlinear, multi-parameter, and highly coupled system. Its heating surface wall temperature is affected by factors such as flue gas heat absorption deviation, steam flow deviation, and structural deviation. Therefore, selecting key characteristic variables that are highly correlated with the boiler heating surface wall temperature from a large number of parameters is a prerequisite for constructing a boiler heating surface wall temperature prediction model. This will improve the accuracy of the wall temperature prediction model and reduce computational costs during the modeling process.

[0063] Figure 2 This is a flowchart of data analysis and key feature variable selection for boiler heating surface overheating early warning. Step 2 includes:

[0064] Step 201: Determining the locations of boiler heating surfaces prone to overheating: Analyze the historical operating data of the wall temperature of each heating surface of the boiler, observe the heat load distribution at each location, and identify the location with the largest heat load deviation. Use the wall temperature at the location with the largest heat load deviation as the main observation target for boiler heating surface overheating early warning.

[0065] Step 202: Determination of the prediction time for boiler heating surface wall temperature: Based on the historical operating data of the heating surface wall temperature and its cooling measures, analyze the dynamic response characteristics of the boiler heating surface wall temperature and determine the dynamic response time m of the wall temperature; to avoid abnormal alarms due to abnormal real-time data acquisition, a time margin of m / 5 is used as the sliding alarm window; the final required wall temperature prediction time is determined to be 6m / 5.

[0066] Step 203: Determination of key feature variables of boiler heating surface wall temperature: Using the XGBoost model feature importance extraction method, the five factors with the highest XGBoost model feature importance were selected from the main influencing factors of boiler heating surface wall temperature obtained in Step 1 as key feature variables.

[0067] Step 204, Time Series Data Reconstruction: Use the Pandas.shift function in Python to convert the time series data of key feature variables into supervised learning data that can be processed by machine learning algorithms.

[0068] Step 3: Construction of boiler heating surface wall temperature prediction model: The hyperparameters of the XGBoost model are optimized by using the Sparrow Search Algorithm (SSA) to construct the SSA-XGBoost wall temperature prediction model; combined with the confidence interval algorithm, the predicted wall temperature point values ​​are extended to an interval value.

[0069] In this invention, considering that the variability of the unit's peak-shaving operation conditions, the delay in changes in thermodynamic parameters, the accuracy of collected data, and random interference can all affect the accuracy of the SSA-XGBoost model, this invention expands the point values ​​obtained by the wall temperature prediction model into interval values ​​to avoid misjudgments caused by normal fluctuations in operating conditions or parameters, thereby increasing the robustness of the model.

[0070] Figure 3 This is a flowchart of the boiler heating surface wall temperature prediction model construction process, wherein step 3 includes:

[0071] Step 301: Construction of the SSA-XGBoost wall temperature prediction model: The wall temperature prediction model is constructed by optimizing the hyperparameters (number of trees, regression tree depth, learning rate, and leaf sample weights) of the XGBoost model through SSA.

[0072] The hyperparameter optimization of the XGBoost model based on SSA includes the following steps:

[0073] Step 1A: Initialize parameters, including population size, the ratio of discoverers to vigilants, vigilance threshold and safety threshold, and maximum number of iterations.

[0074] Step 1B: Initialize the individual locations of the population.

[0075] Step 1C: Calculate the fitness value and corresponding position of individual sparrows, and randomly select 20% of the sparrows with high fitness as discoverers, and the remaining sparrows as joiners.

[0076] Step 1D: Update the individual sparrow positions and the global optimal position.

[0077] Step 1E: Determine and verify the termination condition: If the maximum number of iterations is reached, stop the loop and output the optimal hyperparameters of the XGBoost model; otherwise, return to step 1C to continue the calculation.

[0078] Step 302: Determining the number of input data samples for the wall temperature prediction model: By calculating and comparing the root mean square error of the SSA-XGBoost model under different input sample numbers, the number of input data samples (HistoryWindow) for the SSA-XGBoost model is determined.

[0079] Step 303: Determining the predicted range of boiler heating surface wall temperature: Calculate and compare the root mean square error of the predicted wall temperature range under different significance levels. In this embodiment, a root mean square error threshold of 4 is used to determine a suitable confidence level and determine the predicted wall temperature range at different times.

[0080] Determining the predicted wall temperature range includes the following steps:

[0081] Step 3A: Describe the predicted wall temperature range as (x max (t),x min (t)) uses the t-distribution method to calculate the confidence interval for the input samples of size HistoryWindow+N. Here, HistoryWindow represents the number of data samples input to the wall temperature prediction model, and N represents the number of prediction data samples after prediction time t. The upper limit of the interval is x. max (t) and the lower limit of the interval x min (t) is calculated from equations (9) and (10) respectively.

[0082]

[0083]

[0084] Where n is the number of samples, and n = HistoryWindow + N; X is the sample mean; s is the sample variance; and α represents the significance level.

[0085] Step B: Calculate the root mean square error of the predicted wall temperature interval at different confidence levels at prediction time t. Measured wall temperature x of the heated surface. true If the temperature falls within the predicted operating range, the prediction accuracy is considered to be 100%. Otherwise, the root mean square error R of the wall temperature prediction range is calculated according to equation (11). range Perform the calculation:

[0086]

[0087]

[0088] Step 3C: Set the root mean square error threshold. Under the premise of ensuring that the root mean square error is less than the threshold, select an appropriate confidence level for the wall temperature point value at prediction time t, and determine the wall temperature prediction interval.

[0089] Step 4: Determining the over-temperature warning threshold of the boiler heating surface: Steady-state screening of historical operating data, using the Parezn window to estimate the probability density distribution of the wall temperature data samples, setting the confidence level to 85%, and obtaining the upper limit of the wall temperature confidence interval under different operating conditions as the over-temperature warning threshold.

[0090] In this invention, considering the complexity of the boiler operating environment and the volatility of operating boundary conditions, the boiler heating surface wall temperature warning threshold should follow the statistical distribution law of historical operating data and be set with a certain confidence level to ensure the rationality of the wall temperature warning threshold.

[0091] Figure 4 This is a flowchart for determining the boiler heating surface wall temperature warning threshold. Step 4 includes:

[0092] Step 401, Steady-state data screening: The sliding window detection method is used to screen historical operating data based on steady-state discrimination indicators (main steam temperature, pressure, unit load and reheat steam temperature, pressure) to construct a steady-state operating condition database.

[0093] Step 402, Wall temperature probability density distribution estimation: Using the Parzen window method and the working condition division method, the wall temperature probability density distribution under different working conditions is calculated.

[0094] Step 403: Determination of overheat warning threshold for heated surface: Set the confidence level to 85%, obtain the corresponding wall temperature values ​​under different operating conditions as overheat warning thresholds, and establish a load-wall temperature threshold database.

[0095] In step 402 of this invention, the Parzen window method is used to calculate the wall temperature probability density distribution under different operating conditions. The probability density estimation formula is as follows: Where S is the number of samples to be estimated, h is the window bandwidth, and r is the data dimension to be estimated. It's the Gaussian kernel function.

[0096]

[0097] Step 5, Boiler heating surface wall temperature over-temperature warning: Using the sliding window detection method, the average value of the upper limit of the wall temperature at different prediction times within the window is statistically analyzed. It is then determined whether the average value of the upper limit of the wall temperature is less than the over-temperature warning threshold. If the average value of the upper limit of the wall temperature is greater than the over-temperature warning threshold, an over-temperature warning is issued; otherwise, the detection continues.

[0098] In this invention, a sliding window detection technology is used to determine whether the average value of the upper limit of the predicted wall temperature at different times within the window time exceeds the limit, thereby realizing an over-temperature warning for the boiler heating surface. This eliminates the influence of uncertain factors on the over-temperature warning and improves the accuracy and reliability of the over-temperature warning results.

[0099] Figure 5 This is a flowchart of the boiler heating surface over-temperature early warning detection process, and step 5 includes:

[0100] Step 501: Select a wall temperature prediction time series with a window length of m / 6 forward from the starting time i, and statistically calculate the average value of the upper limit of the prediction interval of the wall temperature from time i+m / 6 to time i.

[0101] Step 502: Based on the real-time unit load, determine the warning threshold for the boiler heating surface wall temperature under the current load in the load-wall temperature threshold database.

[0102] Step 503: Determine whether the average value of the predicted upper limit of wall temperature in the window is less than the over-temperature warning threshold under the corresponding operating condition. If the average value of the upper limit of wall temperature is greater than the warning threshold, then an over-temperature warning for the heated surface is issued.

[0103] Step 504: Increment the step size k to obtain a new time series. Repeat steps 502 and 503 to recursively deduce the result.

[0104] The method of this invention is a real-time early warning program, which will not be terminated unless absolutely necessary.

[0105] The technical solution of this invention was implemented in a 660MW supercritical unit in northern China. This unit's Π-type once-through boiler adopts a single-furnace, single-stage intermediate reheat, balanced ventilation, and fully suspended structure layout. The main steam temperature is primarily regulated through matching fuel and feedwater, with water spray desuperheaters serving as an auxiliary regulation method. Normal reheat steam temperature is regulated by tail-end flue gas dampers, with water spray desuperheaters used for emergency regulation and fine-tuning. Historical operating data for one month each from March, June, September, and December of 2022 were selected from the unit's SIS (System Integrated Circuit System), with a data sampling interval of 30 seconds. Steady-state wall temperature data was statistically analyzed using the Parzen window method, and over-temperature warning thresholds were determined based on operating condition classifications. A boiler heating surface wall temperature prediction model was constructed. The analysis results can provide effective over-temperature warning information for the unit, guiding operators to issue operational instructions in advance to avoid over-temperature of the heating surface tubes and assist in achieving safe boiler operation.

[0106] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for early warning of overheating of boiler heating surface, characterized in that, The main influencing factors affecting the boiler heating surface wall temperature are selected; based on historical operating data of wall temperature, the location with the largest heat load deviation is determined; combined with the dynamic response characteristics of wall temperature, the required wall temperature prediction time for each heating surface is determined; the XGBoost model feature importance extraction method is used to select key feature variables affecting wall temperature; the hyperparameters of the XGBoost model are optimized through SSA to construct a boiler heating surface wall temperature prediction model; combined with the confidence interval algorithm, the prediction interval of wall temperature at different times is determined; by estimating the probability density distribution of data samples, the warning threshold for heating surface overheating is determined; a sliding window detection method is used to statistically analyze the average value of the upper limit of the predicted wall temperature within the window and determine whether it exceeds the warning threshold, thus realizing the detection and warning of boiler heating surface wall temperature overheating. The specific steps include: Step 1: Data collection and storage of key factors affecting boiler heating surface wall temperature: Real-time data collection of key factors affecting boiler heating surface wall temperature and operating data of each boiler heating surface wall temperature; Simultaneously, synchronous storage of operating data of each boiler heating surface wall temperature to establish a historical operating database; Key factors affecting boiler heating surface wall temperature include coal feed rate, furnace air volume, flue gas damper regulating valve position, main steam temperature, main steam pressure, boiler feedwater flow rate, feedwater temperature, superheating desuperheating water flow rate, superheating desuperheating water temperature, and unit load; Step 2, Data Analysis and Selection of Key Feature Variables: Based on historical operating data, determine the location with the largest heat load deviation, and combine the wall temperature response characteristics to determine the required wall temperature prediction duration for the heated surface; use the XGBoost model feature importance extraction method to select key feature variables affecting the wall temperature, and reconstruct the runtime sequence data of the obtained key feature variables into supervised learning data; Step 3: Construction of boiler heating surface wall temperature prediction model: The hyperparameters of the XGBoost model are optimized by the Sparrow Search Algorithm (SSA) to construct the SSA-XGBoost wall temperature prediction model; combined with the confidence interval algorithm, the predicted wall temperature point values ​​are expanded into a wall temperature prediction interval. Step 4: Determining the over-temperature warning threshold of the boiler heating surface: Steady-state screening of historical operating data, using the Parzen window to estimate the probability density distribution of the wall temperature data samples, setting the confidence level to 85%, and obtaining the upper limit of the wall temperature confidence interval under different operating conditions as the over-temperature warning threshold. Step 5, Boiler heating surface wall temperature over-temperature warning: Using the sliding window detection method, the average value of the upper limit of the wall temperature at different prediction times within the window is statistically analyzed to determine whether the average value of the upper limit of the wall temperature is less than the over-temperature warning threshold. When the average value of the upper limit of the wall temperature is greater than the over-temperature warning threshold, an over-temperature warning is issued. Otherwise, continue with the testing.

2. The boiler heating surface over-temperature early warning method according to claim 1, characterized in that: In step 1, the parameters of coal feed rate, furnace air volume, flue gas damper regulating valve position, main steam temperature, main steam pressure, boiler feedwater flow rate, feedwater temperature, superheated desuperheating water flow rate, superheated desuperheating water temperature, unit load, and boiler heating surface wall temperature are collected and stored every 30 seconds.

3. The boiler heating surface over-temperature early warning method according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 201: Determining the locations of boiler heating surfaces prone to overheating: Analyze the historical operating data of the wall temperature of each heating surface of the boiler, observe the heat load distribution at each location, find the location with the largest heat load deviation, and take the wall temperature at the location with the largest heat load deviation as the main observation object for boiler heating surface overheating early warning. Step 202: Determination of Boiler Heating Surface Wall Temperature Prediction Duration: Based on historical operating data of the heating surface wall temperature and its cooling measures, analyze the dynamic response characteristics of the boiler heating surface wall temperature and determine the dynamic response duration of the wall temperature. m To avoid abnormal alarms due to anomalies in real-time data acquisition, a time margin is adopted. m / 5 serves as the sliding alarm window; the final required wall temperature prediction duration is determined to be 6. m / 5; Step 203: Determination of key feature variables of boiler heating surface wall temperature: Using the XGBoost model feature importance extraction method, the five factors with the highest XGBoost model feature importance were selected from the main influencing factors of boiler heating surface wall temperature obtained in Step 1 as key feature variables. Step 204, Time Series Data Reconstruction: Use the Pandas.shift function in Python to convert the time series data of key feature variables into supervised learning data that can be processed by machine learning algorithms.

4. The boiler heating surface over-temperature early warning method according to claim 1, characterized in that: Step 3 specifically includes the following steps: Step 301, Construction of SSA-XGBoost wall temperature prediction model: The hyperparameters of the XGBoost model are optimized by SSA to construct the wall temperature prediction model; the hyperparameters include the number of trees, the depth of the regression tree, the learning rate, and the weight of the leaf samples. Step 302: Determining the number of input data samples for the wall temperature prediction model: By calculating and comparing the root mean square error of the SSA-XGBoost model under different input sample numbers, the number of input data samples for the SSA-XGBoost model is determined. HistoryWindow ; Step 303: Determining the predicted range of boiler heating surface wall temperature: Calculate and compare the root mean square error of the predicted range of wall temperature under different significance levels, select a root mean square error of less than 4 to determine the confidence level, and determine the predicted wall temperature range at different times.

5. The boiler heating surface over-temperature early warning method according to claim 4, characterized in that: Step 301, which optimizes the hyperparameters of the XGBoost model based on SSA, includes the following steps: Step 1A: Initialize parameters, including population size, ratio of discoverers to watchers, watch threshold and safety threshold, and maximum number of iterations; Step 1B: Initialize the location of individual populations; Step 1C: Calculate the fitness value and corresponding position of individual sparrows, and randomly select 20% of the sparrows with high fitness as discoverers, and the remaining sparrows as joiners; Step 1D: Update the individual sparrow positions and the global optimal position; Step 1E: Determine and verify the termination condition: If the maximum number of iterations is reached, stop the loop and output the optimal hyperparameters of the XGBoost model; otherwise, return to step 1C to continue the calculation.

6. The boiler heating surface over-temperature early warning method according to claim 4, characterized in that: Determining the predicted wall temperature range in step 303 includes the following steps: Step 3A: Describe the predicted wall temperature range as ( x max ( t ), x min ( t )),use t The number of calculations for the distribution method is HistoryWindow+N The confidence interval of the input sample; where, HistoryWindow This indicates the number of data samples input to the wall temperature prediction model. N Indicates the predicted time t The number of subsequent prediction data samples; the upper limit of the interval. x max ( t ) and lower limit of the interval x min ( t It is obtained by the following formula: （1） （2） in, n Let be the number of samples, and n = HistoryWindow + N ; This is the sample mean; s This represents the sample variance. α Indicates the significance level; Step 3B: Calculate the predicted time. t At the same time, the root mean square error at different significance levels; the measured wall temperature of the heated surface. x true If the temperature falls within the predicted operating range, the prediction accuracy is considered 100%; otherwise, the root mean square error of the wall temperature prediction range is calculated using the following formula. R range Perform the calculation: （3） （4） Step 3C: Set the root mean square error threshold. The prediction time is determined while ensuring the root mean square error is less than the threshold. t The wall temperature point value with the lowest confidence level is selected to determine the wall temperature prediction interval.

7. The boiler heating surface over-temperature early warning method according to claim 1, characterized in that: Step 4 specifically includes the following steps: Step 401, Steady-state data screening: The sliding window detection method is used to screen historical operating data based on steady-state discrimination indicators to construct a steady-state operating condition database; the indicators include main steam temperature, pressure, unit load, and reheat steam temperature and pressure; Step 402, Wall temperature probability density distribution estimation: Using the Parzen window method and the working condition division method, the wall temperature probability density distribution under different working conditions is calculated; Step 403: Determination of the over-temperature warning threshold for the heated surface: Set the confidence level to 85%, calculate the upper limit of the wall temperature confidence interval under different operating conditions as the over-temperature warning threshold, and establish a load-wall temperature threshold database.

8. The boiler heating surface over-temperature early warning method according to claim 7, characterized in that: In step 402, the Parzen window method is used to calculate the probability density distribution of wall temperature under different operating conditions. The probability density estimation formula is as follows: ,in, S It is the number of samples to be estimated. h It is window bandwidth. r The data dimension to be estimated. It's the Gaussian kernel function. .

9. The boiler heating surface over-temperature early warning method according to claim 1, characterized in that: Step 5 specifically includes the following steps: Step 501, from the start time i Select window length forward m The wall temperature prediction time series of / 6 was statistically analyzed and calculated. i+m / 6 o'clock to i The average value of the upper limit of the predicted range of wall temperature at any given time, where m represents the dynamic response time of wall temperature; Step 502: Based on the real-time unit load, determine the early warning threshold for the boiler heating surface wall temperature under the current load in the load-wall temperature threshold database; Step 503: Determine whether the average value of the predicted upper limit of wall temperature in the window is less than the over-temperature warning threshold under the corresponding operating condition. If the average value of the upper limit of wall temperature is greater than the warning threshold, then an over-temperature warning for the heated surface is issued. Step 504, Increment the step size k This yields a new time series. Steps 502 and 503 are repeated, and so on, to obtain the new time series.

Images