Early warning method and system for safe operation of thermal power plant

Abstract

The invention discloses a thermal power plant safety operation early warning method and system, and relates to the technical field of power plant safety early warning. The method comprises the following steps: acquiring operation parameters from a DCS (Distributed Control System) configured by a target thermal power plant, and carrying out data preprocessing; training a multivariable Gaussian distribution model as a normal operation reference; the technical key points are as follows: the whole early warning scheme has the characteristics of self-adaption, learnability and high robustness, the safety monitoring level of the thermal power plant under complex working conditions is remarkably improved, the problem of high misjudgment rate of a fixed threshold value is solved by setting a dynamic threshold value, the early warning accuracy is improved, and the safety of the thermal power plant is improved. The first correction mechanism can solve the problem that a traditional scheme lacks self-adaptive ability, so that the current state is better matched, multi-level sensitive mode division can accurately respond to different risks, the second correction mechanism solves the problem that the traditional scheme lacks learning feedback, automatic threshold optimization is achieved, and the method better meets actual requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of power plant safety early warning, and in particular to a thermal power plant safe operation early warning method and system. Background Art

[0002] Power plant safety early warning technology is a system that monitors the operating parameters of key equipment in thermal power plants (such as temperature, pressure, vibration, etc.) in real time, combines historical data with intelligent algorithms, predicts potential failures, and issues alerts in advance.

[0003] Traditional early warning methods are usually based on fixed threshold alarms for a single parameter (such as temperature exceeding the limit triggering an alarm), which has certain limitations: First, relying on manual experience to set static thresholds, it is difficult to adapt to dynamic operating conditions such as equipment aging and load fluctuations. Traditional fixed threshold settings can easily lead to a large number of false alarms or missed alarms when the environment or operating conditions change. The early warning systems currently designed on the market rely on fixed rules or a single data source, which makes it difficult to fully reflect the health status of the system and easily cause false alarms or missed alarms; second, there is a lack of equipment health status assessment and insufficient sensitivity to abnormalities of sub-healthy equipment; there is often a lack of effective feedback mechanism for early warning judgments, and it is impossible to make corresponding adjustments based on the actual health status of the equipment, which affects the effectiveness of the early warning; in summary, the current design of the safe operation early warning system for thermal power plants is difficult to meet actual needs. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: A thermal power plant safe operation early warning method, the method comprising: Obtain operating parameters from the DCS system configured in the target thermal power plant and perform data preprocessing; A multivariate Gaussian distribution model is trained as a normal operation benchmark, and a parameter trend prediction model is constructed to predict the operating parameters. After comparing the deviation percentage with the initial basic threshold as the current threshold, the prediction error analysis action is completed; A fuzzy entropy weighted health index is constructed, and a weighted health index is introduced. After normalization, the health of the unit equipment in the target thermal power plant is obtained. Under non-fault conditions, the status of the unit equipment is determined based on the health threshold. When the unit equipment is judged to be in poor condition, the first-level correction mechanism is triggered, and the deviation of the abnormal operating parameters, the initial basic threshold, and the difference between the health level and the health level threshold are input into the pre-built threshold adjustment function model. According to the divided sensitive mode, the new threshold after one correction is output and the new threshold is used as the current threshold. The simulation model is used to verify the early warning accuracy of the first-level correction mechanism. When the early warning accuracy does not exceed the standard value, the second-level correction mechanism is triggered, and the new threshold is adjusted based on the historical alarm rate feedback to obtain the final threshold as the current threshold. The final determined current threshold is compared with the deviation percentage to complete the early warning guidance action.

[0005] Furthermore, the operating parameters include at least main steam temperature, exhaust gas temperature, turbine speed, boiler pressure, bearing vibration signal and load power; the data preprocessing process includes: using linear interpolation to fill in missing data, constructing derived features: including at least sliding mean and standard deviation, and completing normalization processing.

[0006] Furthermore, the parameter trend prediction model adopts: LSTM-Attention model; input historical data of the past Q hours; output: trend prediction value in the next 1 / Q hours; where Q ≥ 6 and Q is a positive integer.

[0007] Furthermore, the process of forecast error analysis is as follows: compare the actual value with the forecast value and calculate the deviation percentage E i ; and the deviation percentage E i Compare with the initial base threshold T_base as the current threshold to obtain the primary comparison result: When E i When it is greater than T_base, it indicates that the trend is abnormal and the corresponding early warning signal is triggered; When E i When ≤T_base, it means the trend is normal and no response action is taken.

[0008] Furthermore, the status of the unit equipment is determined based on the health threshold as follows: When the health H(t) exceeds the health threshold Hth, it means that the health is good; When the health H(t) does not exceed the health threshold Hth, it indicates that the health is poor.

[0009] Furthermore, based on the primary comparison results, when only a single operating parameter has an abnormal trend, it is classified as a low-sensitivity mode; when two operating parameters have abnormal trends, it is classified as a medium-sensitivity mode; when at least three operating parameters have abnormal trends, it is classified as a high-sensitivity mode.

[0010] Furthermore, a threshold adjustment function model is pre-built to output the new threshold T_new after correction. The operation process is as follows: ; Among them, α, β and γ are fine-tuning tightening coefficient, moderate tightening coefficient and strong tightening coefficient respectively, and 0<α<β<γ, Deviation k1 Deviation represents the deviation of the only abnormal operating parameter in the low sensitivity mode LSM. k2 、Deviationk3 Deviations represents the deviation of two abnormal operating parameters in the medium sensitivity mode (MSM). Deviations represents the deviation of all abnormal operating parameters in the high sensitivity mode (HSM). N represents the number of abnormal parameters, and N ≥ 3. SM represents the mode.

[0011] Furthermore, the threshold adjustment function model also includes: based on the health difference ΔH, using the Sigmoid function to dynamically generate the fine-tuning tightening coefficient α, the moderate tightening coefficient β and the strong tightening coefficient γ; where ΔH=Hth-H(t).

[0012] Furthermore, the operation process of the secondary correction mechanism is as follows: the new threshold T_new is adjusted based on the historical alarm rate feedback to obtain the final threshold T_final: ; Among them, Rh represents the historical alarm rate, δ represents the impact weight, and the value range is [0.05, 0.1]; The content of the early warning guidance action is: When E i When the current threshold is exceeded, the corresponding warning signal is triggered; when E i If the current threshold is not exceeded, no response action is taken.

[0013] A thermal power plant safe operation early warning system, the system comprising: Parameter acquisition module: obtains operating parameters from the DCS system configured in the target thermal power plant and performs data preprocessing; Modeling and Prediction Module: This module trains a multivariate Gaussian distribution model as a benchmark for normal operation, builds a parameter trend prediction model to predict operating parameters, and compares the deviation percentage with the initial base threshold, which serves as the current threshold, to complete the prediction error analysis. Health diagnosis module: Constructs a fuzzy entropy weighted health indicator, introduces a weighted health index, and normalizes it to derive the health of the target thermal power plant's equipment. Under non-fault conditions, the health threshold is used to determine the status of the equipment. Threshold Correction Module: When the unit equipment is judged to be in poor condition, the first-level correction mechanism is triggered. The deviation of the abnormal operating parameters, the initial basic threshold, and the difference between the health level and the health level threshold are input into the pre-built threshold adjustment function model. Based on the divided sensitive mode, the new threshold after one correction is output and used as the current threshold. The simulation model is used to verify the early warning accuracy of the first-level correction mechanism. When the early warning accuracy does not exceed the standard value, the second-level correction mechanism is triggered. The new threshold is adjusted based on the historical alarm rate feedback to obtain the final threshold as the current threshold. Early warning judgment module: compares the current threshold of the final judgment with the deviation percentage to complete the early warning guidance action.

[0014] The present invention provides a thermal power plant safe operation early warning method and system, which has the following beneficial effects: (1) This solution uses historical data modeling combined with trend prediction models to achieve high-precision trend prediction, which not only improves the accuracy of early warning but also solves the misjudgment problem caused by a single threshold. Under the condition of establishing a normal operation benchmark model, parameter trend prediction is performed based on this benchmark model. The combination of the two enables it to detect potential anomalies at an early stage, greatly reducing the possibility of failure. (2) This solution adopts a dynamic threshold adjustment mechanism to achieve accurate monitoring under complex working conditions, which not only improves the accuracy of early warning, but also effectively addresses the problem that fixed thresholds are difficult to adapt to changing working conditions; it realizes the function of primary and secondary correction based on factors such as equipment health and the number of current abnormal parameters. This adaptive adjustment ensures that the sensitivity of the early warning system can be automatically adjusted as the equipment status changes, effectively solving the problem that traditional fixed threshold settings are prone to causing a large number of false alarms or missed alarms when the environment or working conditions change; (3) This solution introduces equipment health diagnosis, which provides real-time equipment status assessment. This not only enhances the system's self-learning ability, but also optimizes the early warning strategy and solves the problem of lack of feedback mechanism. The health of the equipment is assessed by analyzing the sensor signal and used as feedback information in the early warning decision. This enables the system to dynamically adjust the early warning logic according to the actual operating status of the equipment, further improving the effectiveness and timeliness of the early warning. (4) In summary, the entire early warning scheme has the characteristics of self-adaptation, learnability, and strong robustness, which significantly improves the safety monitoring level of thermal power plants under complex working conditions. The setting of dynamic thresholds solves the problem of high misjudgment rate of fixed thresholds and improves the accuracy of early warnings. The primary correction mechanism can solve the problem of lack of self-adaptation in traditional schemes, thereby better matching the current state. Among them, the multi-level sensitive mode division can accurately respond to different risks. The secondary correction mechanism solves the problem of lack of learning feedback in traditional schemes and realizes automatic optimization of thresholds to make them more in line with actual needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of the operation flow of a thermal power plant safety operation early warning method in the present invention; Figure 2 This is a flow chart of the dynamic threshold correction mechanism in S4 of the present invention. DETAILED DESCRIPTION

[0016] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] Example 1: See also Figure 1 and Figure 2 This embodiment provides a thermal power plant safe operation early warning method. The solution adopted in this method aims to build an intelligent early warning system that integrates multi-source industrial historical data modeling, parameter trend prediction, equipment health assessment, and dynamic threshold correction mechanism. Each core capability cooperates and influences each other to form a holistic solution. The operation of this operation method has the following core capabilities: 1. Use historical data modeling to establish an operating benchmark under normal operating conditions; 2. Identify potential anomalies through parameter trend prediction models; 3. Introduce the equipment health diagnosis module to evaluate the current equipment status; On this basis, a two-stage threshold correction mechanism is constructed: Primary correction: A preliminary adjustment is made based on the current number of abnormal parameters and sensitive patterns. The decision on whether to trigger is made based on the results of evaluating the current device status, and feedback adjustment is made to the coefficients in the primary correction. Second correction: Optimization based on simulation verification results and historical alarm rate feedback; Ultimately, accurate perception and adaptive early warning of the operating status of thermal power units can be achieved.

[0018] Specifically, the early warning method is described as follows: S1. Data collection and preprocessing: S1.1. Obtain the following key operating parameters from the DCS system: Table 1: Parameter name, meaning and number description of each operating parameter:

[0019] The recommended sampling frequency is 1 Hz, and at least 30 days of historical data should be retained for modeling. It's important to note that a DCS system stands for distributed control system. In industrial automation, especially in large-scale, continuous production processes like thermal power plants, DCS systems play a crucial role. A distributed control system is a computerized control system specifically designed for industrial process control and operation. It improves system reliability and efficiency by distributing control functions across multiple processors rather than concentrating them in a single central processing unit. This distributed architecture allows the entire system to continue operating even if a component fails, thereby enhancing the system's fault tolerance and stability.

[0020] S1.2, Data cleaning and feature engineering (data preprocessing): Missing value interpolation: Linear interpolation is used to fill missing data; Construct derived characteristics: including at least the moving mean and standard deviation; Normalization: Make different parameters comparable.

[0021] S2. Industrial historical data modeling and trend forecasting: S2.1. Establish a normal operation benchmark model: Use historical data to train a multivariate Gaussian distribution model to describe the joint probability distribution of various operating parameters under normal conditions: ; Where X represents any input operating parameter vector, μ represents the mean vector, that is, the expected value of the random variable, ∑ represents the covariance matrix, and N represents the symbol of the normal distribution; The above establishes a joint probability distribution model for the various operating parameters of the thermal power unit under normal operation, which serves as the basis for all subsequent analysis and judgment. The essence of this joint probability distribution model is to serve as the "memory bank" of the entire early warning system. It tells us: under normal circumstances, the value range of parameters such as main steam temperature, boiler pressure, and turbine speed should be; how these parameters are interrelated (covariance); and it can provide a reference for subsequent initial basic thresholds. For example, based on the historical standard deviation, the initial basic threshold can be set as: initial basic threshold = 2 × historical standard deviation (for illustration only). S2.2. Parameter trend prediction model: The LSTM-Attention model is used to predict operating parameters at several future time points. Input: Historical data from the past Q hours (i.e., operating parameters from the historical period). Output: Trend forecast for the next 1 / Q hours. Where Q is a positive integer and ≥ 6, for example, when Q = 6, the output is the trend forecast for the next hour. S2.3. Forecast error analysis: Compare the actual value with the predicted value and calculate the deviation percentage Ei : ; Among them, y i represents the actual value (i.e. the true value), represents the predicted value, represents the standard deviation of the operating parameter numbered i under normal conditions (obtained from S2.1), i=1, 2, ..., n, where n is a positive integer greater than 0. As can be seen from Table 1 above, the value of n in this embodiment is 6; E i Compare with the initial base threshold T_base corresponding to the current threshold to obtain the primary comparison result: When E i >T_base, indicating an abnormal trend, triggering the corresponding early warning signal; When E i ≤T_base, indicating that the trend is normal and no response action is taken.

[0022] Effect description: By combining historical data modeling with trend prediction models, high-precision trend prediction is achieved, which not only improves the accuracy of early warning, but also solves the misjudgment problem caused by a single threshold. Under the condition of establishing a normal operation benchmark model, parameter trend prediction is performed based on this benchmark model. The combination of the two enables it to detect potential anomalies at an early stage, greatly reducing the possibility of failure. It solves the problem that traditional early warning systems rely on fixed rules or a single data source, making it difficult to fully reflect the health status of the system and prone to false alarms or omissions. This solution significantly improves the accuracy of early warning by integrating multi-source industrial historical data modeling.

[0023] S3. Equipment health diagnosis module: S3.1. Constructing fuzzy entropy weighted health index: Define fuzzy entropy FE(x) to measure signal complexity: ; Among them, p j1 represents the probability density of the j1th fuzzy subset, j1=1, 2, ..., m1, m1 is a positive integer greater than 0, and m1 is the number of fuzzy subsets, FE(x) is the fuzzy entropy of signal x; where the signal types include at least vibration, temperature, and pressure, which are all signals from sensors configured on the corresponding equipment. The fuzzy entropy algorithm is used to evaluate the stability of the corresponding equipment's operating status; Introducing the weighted health index WHI: ; Among them, w j2 Indicates the initial weight of each sensor channel, the value range is [0, 1], FE j2represents the fuzzy entropy of the j2th sensor, j2=1, 2, ..., m2, m2 is a positive integer greater than 0; After normalization, the health degree H(t) is obtained: ; Where max(WHI) represents the maximum value among all possible WHIs and is used for normalization. This generates the health H(t) of the corresponding device, reflecting the current health status of the device. S3.2 Health Status Classification: Under non-fault conditions, set the health threshold Hth to divide the corresponding device status: Good health: the health H(t) exceeds the health threshold Hth; Poor health: the health H(t) does not exceed the health threshold Hth; The health threshold is set independently based on historical data and is used to determine whether the corresponding device is healthy. Since it is not the core technology of this solution, it will not be described in detail here. S4. Dynamic threshold correction mechanism: S4.1. Determine whether correction is needed: When the health of the corresponding equipment in the thermal power plant is good, the primary comparison result is used as the final result; On the contrary (i.e. the health of the corresponding equipment in the power plant is not good), the first-level correction mechanism is triggered; S4.2. One-time correction: Divide sensitive modes based on the number of abnormal parameters: According to the number of trend anomaly parameters detected, sensitive modes are divided and different functions are applied for threshold adjustment: According to the primary comparison results, when only a single operating parameter is abnormal (trend abnormality), it is classified as a low-sensitivity mode; When there are two abnormal operating parameters (trend anomalies), it is classified as medium-sensitive mode; When there are at least three abnormal operating parameters (abnormal trends), it is classified as a high-sensitivity mode; Input the deviation of abnormal operating parameters (which can be regarded as the deviation value), the initial basic threshold, and the difference between the health level and the health level threshold into the pre-built threshold adjustment function model. According to the divided sensitive mode, the new threshold T_new after correction is output; ; Among them, α, β and γ are fine-tuning tightening coefficient, moderate tightening coefficient and strong tightening coefficient respectively, and 0<α<β<γ, Deviation k1 Deviation represents the deviation of the only abnormal operating parameter in the low sensitivity mode LSM. k2 、Deviationk3 Deviations represents the deviation of two abnormal operating parameters in the medium sensitivity mode (MSM), Deviations represents the deviation of all abnormal operating parameters in the high sensitivity mode (HSM), N represents the number of abnormal parameters, where N ≥ 3; SM represents the mode, for example, SM∈LSM, which means that when the mode is the low sensitivity mode, the formula under this condition is used to calculate the required new threshold value T_new; The threshold adjustment function model also includes the following during operation: Based on the health difference ΔH and using the Sigmoid function (smooth transition), the fine-tuning tightening coefficient α, the moderate tightening coefficient β, and the strong tightening coefficient γ are dynamically generated; Where, ΔH=Hth-H(t); This enables the early warning system to adaptively adjust its sensitivity when the health status of the equipment deteriorates, thereby improving the accuracy of the early warning; The way to run the Sigmoid function is as follows: ; The result of running the Sigmoid function is the required fine-tuning tightening coefficient α, moderate tightening coefficient β, or strong tightening coefficient γ. A represents the upper limit of the output range, g represents the steepness of the control curve, and g>0. u represents the center point of the curve. In the above, a Sigmoid function is designed for each tightening coefficient and adjusted according to different ΔH ranges: 1. Fine-tune the tightening coefficient α: Applicable range: 0≤ΔH≤0.05; Function form: α(ΔH)=0.05 / (1+e^-10(ΔH-0.025)); Among them, A=0.05, the maximum value is 0.05; g=10, controls the steepness of the curve; u=0.025, the center point of the curve; 2. Moderate tightening coefficient β: Applicable range: 0.05<ΔH≤0.15; Function form: β(ΔH)=0.15 / (1+e^-10(ΔH-0.1)); Among them, A=0.15, the maximum value is 0.15; g=10, controls the steepness of the curve; u=0.1, the center point of the curve; 3. Strong tightening coefficient γ: Applicable range: 0.05<ΔH≤0.15; Function form: β(ΔH)=0.25 / (1+e^-5(ΔH-0.25)); Among them, A=0.25, the maximum value is 0.15; g=5, controls the steepness of the curve; u=0.25, the center point of the curve; Logical explanation and description: For the fine-tuning tightening coefficient α, when the equipment health only slightly decreases, a smaller α value is used for fine-tuning to avoid overreaction; as ΔH increases, α smoothly increases to a maximum value of 0.05; for the moderate tightening coefficient β, when the equipment shows obvious signs of degradation, a larger β value is used for moderate tightening, and as ΔH increases, β smoothly increases to a maximum value of 0.15; for the strong tightening coefficient γ, when the equipment health significantly deteriorates, the maximum γ value is used for extreme tightening to ensure timely detection of potential hidden dangers; as ΔH increases, γ smoothly increases to a maximum value of 0.25.

[0024] Effect description: The dynamic threshold adjustment mechanism is adopted to achieve precise monitoring under complex working conditions, which not only improves the accuracy of early warning, but also effectively addresses the problem that fixed thresholds are difficult to adapt to changing working conditions; it realizes the function of primary and secondary correction based on factors such as equipment health and the current number of abnormal parameters. This adaptive adjustment ensures that the sensitivity of the early warning system can be automatically adjusted as the equipment status changes, and effectively solves the problem that traditional fixed threshold settings are prone to cause a large number of false alarms or missed alarms when the environment or working conditions change. The dynamic threshold adjustment can better match the actual operating environment and reduce the occurrence of such problems.

[0025] The introduction of equipment health diagnosis provides real-time equipment status assessment, which not only enhances the system's self-learning ability, but also optimizes the early warning strategy and solves the problem of lacking a feedback mechanism. The S3 solution evaluates equipment health through analysis of sensor signals and uses this information as feedback in early warning decision-making. This enables the system to dynamically adjust the early warning logic based on the actual operating status of the equipment, further improving the effectiveness and timeliness of early warnings. Traditional early warning systems often lack effective feedback mechanisms and are unable to make corresponding adjustments based on the actual health status of the equipment, thus affecting the effectiveness of early warnings. This solution overcomes this limitation through real-time health monitoring and feedback.

[0026] S4.3. Second correction: Use the simulation model to verify the situation after the first correction: Use the new threshold value T_new after the first correction to backtest historical data, simulate the early warning process, and calculate the early warning accuracy Acc: Acc = number of correct warnings / total number of warnings; then compare the early warning accuracy Acc with the standard value θ (for example, θ is 90%). If the early warning accuracy Acc does not exceed the standard value θ, the secondary correction mechanism is triggered; otherwise, the new threshold value T_new after the first correction is used as the current threshold; The operation process of the secondary correction mechanism is as follows: Combined with the historical alarm rate feedback, the new threshold T_new is adjusted to obtain the final threshold T_final: ; Rh represents the historical alarm rate (for example, the alarm frequency in the past 7 days), and δ represents the impact weight, which ranges from [0.05 to 0.1]. Its function is to further optimize the threshold to make it more suitable for the actual operating environment. And use the final threshold as the current threshold; S5, compare the deviation percentage with the current threshold for the second time to obtain the final result; When E i If the current threshold is exceeded, the corresponding warning signal will be triggered to guide the staff to carry out the corresponding maintenance work; When E i If the current threshold is not exceeded, no response action is taken.

[0027] In summary, the current threshold changes dynamically and is determined by the following factors: The basic threshold T_base provided by historical data modeling; Current number of abnormal parameters → get T_new after one correction; Simulation verification results + historical alarm rate → T_final is obtained after secondary correction.

[0028] Effect description: A complete closed-loop process from data collection to early warning output was built, achieving a high degree of robustness in the early warning system. This not only ensured the stable operation of the system, but also solved the problem of overall failure of traditional systems due to failure of a single link. The entire solution starts from S1 (data collection and preprocessing), passes through S2 (industrial historical data modeling and trend prediction), S3 (equipment health diagnosis), and then to S4 (dynamic threshold correction), finally forming a complete closed-loop process. Each link is closely connected and supports each other, ensuring that even if a problem occurs in one part, the other parts can still maintain the basic functions of the system. This solution greatly enhances the robustness and stability of the system by building a closed-loop early warning mechanism.

[0029] This mechanism makes the entire early warning scheme adaptive, learnable, and highly robust, significantly improving the safety monitoring level of thermal power plants under complex working conditions. The setting of dynamic thresholds solves the problem of high misjudgment rate of fixed thresholds and improves the accuracy of early warnings. The primary correction mechanism can solve the problem of lack of adaptive ability of traditional schemes, thereby better matching the current status. Among them, the multi-level sensitive mode division can accurately respond to different risks. The secondary correction mechanism solves the problem of lack of learning feedback in traditional schemes, and realizes automatic optimization of thresholds to make them more in line with actual needs.

[0030] Example 2: Based on Example 1, this embodiment further provides a thermal power plant safe operation early warning system, which includes: Parameter acquisition module: obtains operating parameters from the DCS system configured in the target thermal power plant and performs data preprocessing; Modeling and Prediction Module: This module trains a multivariate Gaussian distribution model as a benchmark for normal operation, builds a parameter trend prediction model to predict operating parameters, and compares the deviation percentage with the initial base threshold, which serves as the current threshold, to complete the prediction error analysis. Health diagnosis module: Constructs a fuzzy entropy weighted health indicator, introduces a weighted health index, and normalizes it to derive the health of the target thermal power plant's equipment. Under non-fault conditions, the health threshold is used to determine the status of the equipment. Threshold Correction Module: When the unit equipment is judged to be in poor condition, the first-level correction mechanism is triggered. The deviation of the abnormal operating parameters, the initial basic threshold, and the difference between the health level and the health level threshold are input into the pre-built threshold adjustment function model. Based on the divided sensitive mode, the new threshold after one correction is output and used as the current threshold. The simulation model is used to verify the early warning accuracy of the first-level correction mechanism. When the early warning accuracy does not exceed the standard value, the second-level correction mechanism is triggered. The new threshold is adjusted based on the historical alarm rate feedback to obtain the final threshold as the current threshold. Early warning judgment module: compares the current threshold of the final judgment with the deviation percentage to complete the early warning guidance action.

[0031] The following technical problems are solved by the above system: 1. Insufficient dynamic adaptability: Traditional fixed thresholds are difficult to adapt to changing operating conditions. This solution improves the system's adaptability through multi-level threshold adjustment. 2. A single threshold leads to false alarms and missed alarms: Targeted adjustments under different sensitivity modes reduce the possibility of false alarms and missed alarms. 3. Lack of feedback mechanism: The introduction of a secondary correction mechanism combined with historical alarm rates enhances the system's self-learning and optimization capabilities. Ultimately, significant technical results were achieved, namely improving the accuracy and reliability of the early warning system, enabling thermal power plants to operate more safely and efficiently in complex and changing operating environments.

[0032] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0033] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0034] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A thermal power plant safe operation early warning method, the method comprising: Obtain operating parameters from the DCS system configured in the target thermal power plant and perform data preprocessing; The method further comprises: training a multivariate Gaussian distribution model as a normal operation benchmark, constructing a parameter trend prediction model to predict the operating parameters, and completing a prediction error analysis action after comparing the deviation percentage with an initial basic threshold value as a current threshold value; A fuzzy entropy weighted health index is constructed, and a weighted health index is introduced. After normalization, the health of the unit equipment in the target thermal power plant is obtained. Under non-fault conditions, the status of the unit equipment is determined based on the health threshold. When the unit equipment is judged to be in poor condition, the first-level correction mechanism is triggered, and the deviation of the abnormal operating parameters, the initial basic threshold, and the difference between the health level and the health level threshold are input into the pre-built threshold adjustment function model. According to the divided sensitive mode, the new threshold after one correction is output and the new threshold is used as the current threshold. The simulation model is used to verify the early warning accuracy of the first-level correction mechanism. When the early warning accuracy does not exceed the standard value, the second-level correction mechanism is triggered, and the new threshold is adjusted based on the historical alarm rate feedback to obtain the final threshold as the current threshold. The final determined current threshold is compared with the deviation percentage to complete the early warning guidance action.

2. A thermal power plant safe operation early warning method according to claim 1, characterized in that: The operating parameters include at least main steam temperature, exhaust gas temperature, turbine speed, boiler pressure, bearing vibration signal and load power; the data preprocessing process includes: using linear interpolation method to fill in missing data, constructing derived features: including at least sliding mean and standard deviation, and completing normalization processing.

3. The method for early warning of safe operation of a thermal power plant according to claim 1, characterized in that: The parameter trend prediction model uses the LSTM-Attention model; the input is historical data of the past Q hours; the output is the trend forecast value within the next 1 / Q hours; where Q ≥ 6 and Q is a positive integer.

4. A thermal power plant safe operation early warning method according to claim 3, characterized in that: The process of forecast error analysis is: compare the actual value with the forecast value and calculate the deviation percentage E i ; and the deviation percentage E i Compare with the initial base threshold T_base as the current threshold to obtain the primary comparison result: When E i When it is greater than T_base, it indicates that the trend is abnormal and the corresponding early warning signal is triggered; When E i When ≤T_base, it means the trend is normal and no response action is taken.

5. The method for early warning of safe operation of a thermal power plant according to claim 1, characterized in that: The results of judging the status of the unit equipment based on the health threshold are as follows: When the health H(t) exceeds the health threshold Hth, it means that the health is good; When the health H(t) does not exceed the health threshold Hth, it indicates that the health is poor.

6. A thermal power plant safe operation early warning method according to claim 4, characterized in that: According to the primary comparison results, when only a single operating parameter has an abnormal trend, it is classified as a low-sensitivity mode; when two operating parameters have abnormal trends, it is classified as a medium-sensitivity mode; when at least three operating parameters have abnormal trends, it is classified as a high-sensitivity mode.

7. A thermal power plant safe operation early warning method according to claim 5, characterized in that: The threshold adjustment function model is pre-built, and the operation process of outputting the new threshold T_new after one correction is as follows: ； Among them, α, β and γ are fine-tuning tightening coefficient, moderate tightening coefficient and strong tightening coefficient respectively, and 0<α<β<γ, Deviation k1 Deviation represents the deviation of the only abnormal operating parameter in the low sensitivity mode LSM. k2 、Deviation k3 Deviations represents the deviation of two abnormal operating parameters in the medium sensitivity mode (MSM). Deviations represents the deviation of all abnormal operating parameters in the high sensitivity mode (HSM). N represents the number of abnormal parameters, and N ≥ 3. SM represents the mode.

8. A thermal power plant safe operation early warning method according to claim 7, characterized in that: The threshold adjustment function model also includes: based on the health difference ΔH, the Sigmoid function is used to dynamically generate the fine-tuning tightening coefficient α, the moderate tightening coefficient β, and the strong tightening coefficient γ; where ΔH=Hth-H(t).

9. The method for early warning of safe operation of a thermal power plant according to claim 1, characterized in that: The operation process of the secondary correction mechanism is: adjust the new threshold T_new based on the historical alarm rate feedback to obtain the final threshold T_final: ； Among them, Rh represents the historical alarm rate, δ represents the impact weight, and the value range is [0.05, 0.1]; The content of the early warning guidance action is: When E i When the current threshold is exceeded, the corresponding warning signal is triggered; when E i If the current threshold is not exceeded, no response action is taken.

10. A thermal power plant safe operation early warning system, the system comprising: Parameter acquisition module: obtains operating parameters from the DCS system configured in the target thermal power plant and performs data preprocessing; The system is characterized in that it further comprises: a modeling and prediction module: training a multivariate Gaussian distribution model as a normal operation benchmark, and constructing a parameter trend prediction model to predict the operating parameters, and completing the prediction error analysis action after comparing the deviation percentage with the initial basic threshold as the current threshold; Health diagnosis module: Constructs a fuzzy entropy weighted health indicator, introduces a weighted health index, and normalizes it to derive the health of the target thermal power plant's equipment. Under non-fault conditions, the health threshold is used to determine the status of the equipment. Threshold Correction Module: When the unit equipment is judged to be in poor condition, the first-level correction mechanism is triggered. The deviation of the abnormal operating parameters, the initial basic threshold, and the difference between the health level and the health level threshold are input into the pre-built threshold adjustment function model. Based on the divided sensitive mode, the new threshold after one correction is output and used as the current threshold. The simulation model is used to verify the early warning accuracy of the first-level correction mechanism. When the early warning accuracy does not exceed the standard value, the second-level correction mechanism is triggered. The new threshold is adjusted based on the historical alarm rate feedback to obtain the final threshold as the current threshold. Early warning judgment module: compares the current threshold of the final judgment with the deviation percentage to complete the early warning guidance action.

Images