Frequency conversion control method and system for ventilation system of nuclear power station

Abstract

The invention relates to the technical field of frequency conversion control, and discloses a frequency conversion control method and system for a nuclear power station ventilation system, and the method comprises the steps: collecting the multi-region real-time operation data of the nuclear power station ventilation system, and carrying out the standardization processing, and obtaining a standardized data set; performing fan performance curve fitting and differential pressure field feature mapping on the standardized data set to obtain differential pressure field mapping features; the differential pressure field mapping features are input into an active neural differential pressure field network model for frequency conversion parameter prediction, and target frequency conversion control parameters are obtained; feedforward-feedback combined control is conducted on the target frequency conversion control parameters through the multi-mode frequency conversion controller, the fan rotating speed and the air valve opening value are obtained, the target frequency conversion control parameters can be generated in a self-adaptive mode, the control performance is remarkably improved, and smooth transition of the nuclear power station ventilation system is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of variable frequency control, and in particular to a variable frequency control method and system for a ventilation system of a nuclear power plant. Background Art

[0002] Traditional nuclear power plant ventilation system control methods mainly use a combination of constant speed operation and valve adjustment, which not only consumes a lot of energy, but also makes it difficult to achieve precise control when facing complex and changeable working conditions. In recent years, although frequency conversion technology has been applied in ventilation systems, the existing frequency conversion control methods still have many problems, including insufficient long-term operation stability, irregular sensor data sampling, weak data loss processing mechanism, poor control coordination between different ventilation areas, and lack of effective fault diagnosis and prediction mechanism, which limits their application in high safety requirements such as nuclear power plants.

[0003] Existing variable frequency control solutions for nuclear power plant ventilation systems usually rely on complex PID control algorithms or fuzzy logic control, which perform poorly when dealing with nonlinear, multi-variable, and strongly coupled complex systems. In particular, when the system faces sensor failures, sudden changes in operating conditions, or the need for multi-region collaborative control, it is often difficult to maintain an ideal control effect. At the same time, existing technical solutions lack an in-depth understanding and modeling of the dynamic characteristics of the ventilation system, and are unable to accurately capture the pressure difference field distribution and its evolution between different ventilation areas, making it difficult for control strategies to adapt to the strict requirements of nuclear power plant ventilation systems, especially in the case of rapid response and stable control capabilities in emergency situations. Summary of the invention

[0004] The present invention provides a frequency conversion control method and system for a ventilation system of a nuclear power plant. The present invention can adaptively generate target frequency conversion control parameters, significantly improve control performance, and achieve a smooth transition of the ventilation system of the nuclear power plant.

[0005] In a first aspect, the present invention provides a frequency conversion control method for a ventilation system of a nuclear power plant, the frequency conversion control method for a ventilation system of a nuclear power plant comprising: Collect real-time operating data from multiple areas of the nuclear power plant ventilation system and perform standardized processing to obtain a standardized data set; Performing fan performance curve fitting and pressure difference field feature mapping on the standardized data set to obtain pressure difference field mapping features; Inputting the pressure difference field mapping characteristics into the active neural pressure difference field network model to predict the variable frequency parameters and obtain the target variable frequency control parameters; The target frequency conversion control parameter is subjected to feedforward-feedback combined control through a multi-mode frequency conversion controller to obtain the fan speed and air valve opening value.

[0006] In a second aspect, the present invention provides a variable frequency control system for a nuclear power plant ventilation system, the variable frequency control system for the nuclear power plant ventilation system comprising: The acquisition module is used to collect real-time operation data of multiple areas of the nuclear power plant ventilation system and perform standardized processing to obtain a standardized data set; A feature mapping module, used to perform fan performance curve fitting and pressure difference field feature mapping on the standardized data set to obtain pressure difference field mapping features; A frequency conversion parameter prediction module is used to input the pressure difference field mapping characteristics into the active neural pressure difference field network model to predict the frequency conversion parameters and obtain the target frequency conversion control parameters; The combined control module is used to perform feedforward-feedback combined control on the target frequency conversion control parameter through a multi-mode frequency conversion controller to obtain the fan speed and air valve opening value.

[0007] In the technical solution provided by the present invention, in terms of data quality, the present invention adopts a distributed sensor network and a dual-channel redundant acquisition mechanism, combined with cubic spline interpolation to repair missing data, and applies digital filtering technology to effectively filter out power grid interference to ensure data integrity and reliability; in terms of system characteristic characterization, a complete mathematical model of ventilation system characteristics is constructed by accurately modeling fan impedance characteristics and duct network topology; in terms of dynamic characteristic capture, the pressure difference field function is constructed by using the finite element method and the core features are extracted by combining the empirical mode decomposition method to achieve a comprehensive characterization of the dynamic characteristics of the ventilation system and future state prediction; in terms of control strategy, an active neural pressure difference field network architecture is used to integrate the functions of feature extraction, pressure difference prediction and parameter generation to form an end-to-end intelligent control system, and adaptively generate target variable frequency control parameters; in terms of operation mode, intelligent switching between normal operation, emergency response and maintenance modes is realized, and the coordinated control of fans and air valves and the multi-regional collaborative mechanism are combined to ensure the stable operation of the system under various working conditions; in terms of fault management, accurate identification of common faults is achieved through a deep confidence network, and the hierarchical fault response strategy and fault prediction based on a recurrent neural network are combined to greatly improve the system reliability and safety.

[0008] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0009] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1A schematic diagram of an embodiment of a frequency conversion control method for a ventilation system of a nuclear power plant according to an embodiment of the present invention; Figure 2 It is a schematic diagram of an embodiment of a variable frequency control system of a nuclear power plant ventilation system in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0012] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0013] To facilitate understanding of this embodiment, a frequency conversion control method for a nuclear power plant ventilation system disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps: 101. Collect real-time operating data of multiple areas of the nuclear power plant ventilation system and perform standardized processing to obtain a standardized data set; It is understandable that the execution subject of the present invention may be a variable frequency control system of a nuclear power plant ventilation system, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0014] Specifically, a multi-region real-time data acquisition network suitable for the ventilation system of a nuclear power plant is constructed. The network covers multiple key plant areas in the nuclear power plant, and a series of high-precision, multi-parameter environmental monitoring sensors are deployed in each area. These sensors include temperature sensors, humidity sensors, differential pressure sensors, wind speed sensors, and radioactivity level monitors distributed in ventilation ducts, exhaust outlets, and key measurement locations in the plant. These devices together constitute a distributed data acquisition network to ensure that parameters in each area, such as temperature, humidity, airflow pressure difference, wind speed, and radiation level, can be obtained in real time. The data transmission frequency is set to a high-frequency acquisition mode to ensure data real-time and dynamic response requirements. A dual-channel redundant acquisition and transmission scheme is implemented for multi-region real-time operation data. That is, two completely independent acquisition and communication channels are configured at the deployment location of each sensor to ensure high reliability and redundant fault tolerance of the collected data. After the data acquisition is completed, the two channels independently transmit data streams to the central processing unit to form two parallel and completely independent real-time data streams. Cross-comparison verification is carried out on the two parallel data streams to identify data loss points or abnormal data points. By calculating the difference in data collected at the same location by two channels and setting a clear abnormal threshold (such as a multiple method based on the standard deviation, or a method based on the absolute error threshold), the abnormal points or missing locations of the data can be quickly identified. When it is found that the difference in data at a certain timestamp location exceeds the set threshold, a missing data mark is made at this location. After marking the missing or abnormal data points, the cubic spline interpolation function is used to reconstruct the data to repair the missing or abnormal data problems at these marked locations. The cubic spline interpolation can well meet the requirements of continuity and smoothness of the curve, and has a good fitting effect on the fluctuation trend of various operating data (such as temperature, pressure difference or wind speed) involved in the ventilation system of nuclear power plants. Using the cubic spline interpolation method, the adjacent valid data at the marked location is used as the interpolation reference point, and the data reconstruction value of the marked location is interpolated and calculated to obtain a repaired continuous data stream. The repaired continuous data stream is digitally filtered to suppress and reduce the influence of random noise and interference from the complex operating environment of the nuclear power plant, and extract the real trend and effective characteristics of the data. When performing digital filtering, the appropriate cutoff frequency is selected according to the characteristics of the nuclear power plant ventilation system data to ensure that useful change trends are not filtered out and unnecessary interference noise is effectively removed, ultimately generating a standardized data set.

[0015] 102. Perform fan performance curve fitting and pressure difference field feature mapping on the standardized data set to obtain pressure difference field mapping features; Specifically, based on the real-time operation data of multiple regions, the fan operation parameters in the ventilation system are dynamically tested and analyzed. The dynamic test analysis is based on parameters such as fan speed, power, flow, vibration, etc. Through the fan operation data under real-time working conditions, the original performance curve data point set of the fan under different load conditions is obtained. The original performance curve data point set is fitted and calculated using the least squares method, and the pressure difference-flow relationship expression of the fan, that is, the fan impedance curve equation, is effectively fitted based on the measured data points of the fan operation. Using the least squares method, a fitting function with the fan flow as the independent variable and the pressure difference between the fan outlet and the inlet as the dependent variable is constructed, and the objective function of minimizing the sum of squares of the residual is defined. By gradually iterating and optimizing the fitting parameters in the function, the fan impedance characteristic function model is obtained, revealing the pressure difference characteristics of the fan under variable frequency control conditions. At the same time, the topological modeling work is carried out on the pipe network structure information included in the standardized data set, and a complete mathematical model of the ventilation system network is formed by clarifying the topological connection relationship between the ventilation paths in the pipe network and the resistance distribution characteristics on each path. In the process of establishing this ventilation network topology model, the topological node information such as the inlet and outlet, branch node, fan position and valve position of the system is identified, and the flow resistance characteristic parameters of each pipe section are determined according to the length, diameter, shape and resistance coefficient of the actual ventilation duct, so as to obtain a detailed expression model of the ventilation system network. Based on the ventilation system network model, the node association matrix and boundary condition vector are constructed to clearly express the connection relationship and actual operation boundary conditions between the nodes in the ventilation network, so as to form a ventilation system resistance characteristic matrix equation describing the flow characteristics of the entire ventilation network through the node association matrix and boundary condition vector. Based on the fan impedance model and the ventilation system resistance characteristic matrix equation, the pressure difference field function is comprehensively constructed to effectively characterize the dynamic interaction between the fan performance and the ventilation system structure. Through the solution and calculation of the pressure difference field function, the pressure difference distribution of each node in the ventilation network under specific variable frequency control parameters is obtained, and the pressure difference distribution characteristics that characterize the system operation status are quantitatively characterized as a whole. Through the numerical solution and analysis of the pressure difference field function, the pressure difference field mapping characteristics that can effectively reflect the real-time dynamic operation characteristics of the nuclear power plant ventilation system are extracted.

[0016] The control areas involved in the ventilation system of a nuclear power plant are spatially divided, and the boundary conditions of the control areas are clearly defined to construct a spatial model covering the entire ventilation control system. In the process of establishing this spatial model, the structural characteristics of the nuclear power plant building and the layout characteristics of the ventilation pipe network need to be comprehensively considered to ensure that the model truly reflects the linkage characteristics and independence characteristics between the actual ventilation areas. The entire ventilation control space is divided into p control areas through appropriate spatial grid refinement to accurately reflect the spatial distribution of pressure differences in different areas under operating conditions. Based on the spatial model, a set of Laplace equations describing the steady-state pressure difference distribution characteristics is established at each grid node, and the fan outlet pressure is set as the boundary condition to form a complete and solvable mathematical boundary value problem. By solving the Laplace equation boundary value problem, the pressure difference field distribution under the initial steady-state condition is obtained. Through this step, it is revealed how the fan outlet pressure affects the pressure difference change law at different spatial nodes, thereby providing a preliminary spatial pressure difference field state. The time dimension is introduced into the initial pressure difference field distribution and the empirical mode decomposition algorithm is applied to dynamically analyze and decompose the pressure difference field data. Through the empirical mode decomposition method, the complex nonlinear and non-stationary pressure difference field time series is decomposed into several sets of intrinsic mode functions with clear physical meanings. This set of intrinsic mode functions can effectively capture the dynamic change characteristics of different scales in the pressure difference field and reveal the transient response characteristics of the nuclear power plant ventilation system under actual operating conditions. The Hilbert spectrum analysis of the intrinsic mode function set is performed to obtain key dynamic information such as the instantaneous frequency and instantaneous amplitude of the pressure difference field, thereby forming a pressure difference field spectrum feature set that characterizes the dynamic characteristics of the pressure difference field, reflecting the pressure difference fluctuation characteristics, transient processes and system operation stability of each control area in the nuclear power plant ventilation system. Based on the pressure difference field spectrum feature set, a feature vector for subsequent intelligent control decision-making is constructed for each control area. The feature vector of each area contains multi-dimensional feature parameters such as the mean value, standard deviation, peak factor, pulsation intensity, and autocorrelation function attenuation rate of the regional pressure difference field to fully express the dynamic change trend and stability, periodicity or random fluctuation characteristics of the regional pressure difference. In order to achieve accurate mapping of pressure difference field characteristics to fan operating parameters (including fan speed and air valve opening), the response surface method is introduced as an important tool to construct the relationship between the pressure difference field feature vector and the fan control parameters. The response surface method is used to describe the sensitivity of the dynamic characteristics of the pressure difference field to the changes in the fan operating conditions. Based on this relationship, an autoregressive moving average model (ARMA model) is established to effectively capture the dynamic mapping law between the feature vector and the fan speed and air valve opening over time. Through the construction and parameter identification process of the ARMA model, the pressure difference field mapping characteristics of the nuclear power plant ventilation system under different operating conditions are finally obtained, and the accurate conversion of the pressure difference field state to the fan operating control parameters is realized.

[0017] 103. Input the pressure difference field mapping characteristics into the active neural pressure difference field network model to predict the variable frequency parameters and obtain the target variable frequency control parameters; Specifically, the pressure difference field mapping features are input into the feature extraction network in the active neural pressure difference field network model. The feature extraction network is designed as a deep neural network with a multi-layer fully connected structure. Through effective nonlinear mapping and dimensional compression, the input high-dimensional pressure difference field mapping features are reduced in dimension to obtain a potential feature representation that efficiently characterizes the pressure difference field state of the nuclear power plant ventilation system. The potential feature representation is input into the pressure difference field prediction network contained in the active neural pressure difference field network model. The prediction network uses a two-layer long short-term memory (LSTM) neural network structure. Since the LSTM network has the ability to analyze time series data and capture long-term dependencies, when modeling and predicting the pressure difference field of the nuclear power plant ventilation system, which is highly dynamic, nonlinear, and has strong time series correlation, it can efficiently and accurately analyze and predict the future change trend of the pressure difference field. Through the gating mechanism of the LSTM network, the time delay problem in the dynamic change of the pressure difference field is effectively solved, so that the network can accurately predict the change value of the pressure difference field in the future period of time based on the current potential feature representation, and obtain real-time pressure difference field change prediction value. The active neural pressure difference field network model simultaneously inputs the potential feature representation and the pressure difference field change prediction value into the control parameter generation network inside the network. The control parameter generation network adopts a multi-layer fully connected neural network structure. Through the nonlinear mapping transformation of the potential feature representation and the pressure difference field prediction value, it completes the efficient mapping from the pressure difference field feature space to the variable frequency control parameter space, and obtains a set of initial variable frequency control parameters. These parameters include basic operating control quantities such as the initial value of the fan speed and the initial value of the air valve opening. The determination process of the initial parameter value takes into account the current operating status of the nuclear power plant ventilation system and the future pressure difference field change trend, thereby preliminarily realizing the efficient decision-making generation of the ventilation system fan operating parameters. At the same time, in order to improve the accuracy and practical applicability of the variable frequency control parameters, a pattern recognition analysis is performed on the pressure difference field mapping characteristics previously input into the network. Through the pattern recognition technology, the current pressure difference field feature data of the system is matched with the historical data for similarity, and the historical working mode state of the system is determined. This process can effectively identify the mode characteristics of the current operating conditions of the ventilation system, and effectively integrate historical experience into the control decision process, so that the subsequent control strategy has higher environmental adaptability and reliability. Based on the historical working mode determination results obtained by the above pattern recognition analysis, the initial variable frequency control parameters are adjusted in a mode adaptive manner. By automatically retrieving the optimal control adjustment experience in the historical operating mode database, the initial variable frequency control parameters are appropriately corrected and compensated for the historical working mode determined by the current system, forming the target variable frequency control parameters that are ultimately suitable for the current operating state.

[0018] 104. The target frequency conversion control parameters are subjected to feedforward-feedback combined control through a multi-mode frequency conversion controller to obtain the fan speed and air valve opening value.

[0019] Specifically, a multi-mode variable frequency controller is used to perform real-time working scenario analysis and pattern recognition on the target variable frequency control parameters to clarify the current system operation status. Based on the real-time data collected by the system and the preset operation rules, combined with the pressure difference field mapping characteristics, fan operation status and historical operation mode, the process uses a pattern recognition algorithm to classify the current operation mode, which is mainly divided into normal operation mode, emergency response mode and maintenance mode. Among them, the normal operation mode corresponds to the stable state of the nuclear power plant ventilation system under standard operating conditions. The main goal is to maintain ventilation efficiency and system safety and minimize energy consumption; the emergency response mode is used to deal with emergencies such as pipeline leakage, nuclear accident risks or other abnormal conditions. In this mode, the control strategy focuses on quickly adjusting the fan and air valve operating parameters to maximize ventilation efficiency, enhance the plant pressure difference control capability, and inhibit the spread of radioactive materials; while the maintenance mode is suitable for equipment maintenance or fan adjustment under specific operating conditions. This mode mainly focuses on the stability of fan operation and avoids unnecessary energy waste. The target variable frequency control parameters are input into the feedforward control unit for initial parameter calculation. The main function of the feedforward control unit is to calculate the initial frequency conversion parameters based on the current operating mode, using the trained fan characteristic model and the pressure difference field prediction results, that is, to preliminarily determine the fan speed and air valve opening value suitable for the current state to form a preliminary frequency conversion control scheme. The calculation process takes into account the fan impedance characteristics, the load condition of the ventilation system, and the optimal solution of the historical optimization control strategy to ensure that the initial frequency conversion parameters can meet the current ventilation system operation requirements. The feedback control mechanism is introduced to compensate for the errors caused by external interference, load changes, or system response lag. By comparing the system output expected by the initial frequency conversion parameters with the actual measured pressure difference deviation value, the control error of the current system is calculated, and then the initial frequency conversion parameters and the error signal are input into the incremental proportional-integral-differential (PID) controller for feedback compensation calculation. The incremental PID controller dynamically adjusts the frequency conversion control parameters based on the real-time error signal, so that the system output gradually converges to the optimal state, while ensuring that there will be no excessive disturbance or oscillation during the control process, thereby improving the response speed and stability of the system. The corresponding control increment is obtained through the calculation of the PID controller. Based on the operation mode determination and control increment execution mode preset control strategy to optimize the final control instruction. According to the needs of different operation modes, the control increment is adjusted for mode adaptiveness. For example, in normal operation mode, the optimization strategy focuses on energy-saving control and stable fan operation, while in emergency response mode, priority is given to increasing ventilation rate and enhancing negative pressure control effect. In maintenance mode, the optimization control logic should appropriately reduce the fan operation intensity to reduce equipment wear and ensure the safety of maintenance work. Through the application of mode preset control strategy, the final mode optimization control output can be more in line with actual operation needs.Based on the mode optimization control output and the actual pressure difference deviation value, the coordinated control logic of the fan and the air valve is executed to obtain the synergistic control instruction. Since there is a strong coupling effect between the operation of the fan and the air valve, adjusting one of the devices alone will affect the overall pressure difference distribution of the system. The core goal of the coordinated control logic is to ensure the synergy between the fan speed adjustment and the change of the air valve opening, so that the entire ventilation system can maintain the best state under different operation modes. Based on the synergistic control instruction and the regional coupling matrix, multi-region coordinated control and load switching processing are performed to obtain the fan speed and air valve opening values. The regional coupling matrix is ​​a mathematical model that describes the air flow relationship between each ventilation control area. It can quantify the ventilation impact between different areas, so as to consider the overall dynamic response characteristics of the system when controlling the fan and the air valve. By using the regional coupling matrix, the coordinated control logic of the fan speed and the air valve opening is optimized to ensure that the ventilation needs of each plant area can be reasonably met, and in the case of load changes, the system can intelligently execute the load switching strategy to optimize the fan operation efficiency.

[0020] The system performance evaluation is performed based on the fan speed and the air valve opening value to obtain the comprehensive performance index. The operating status of the fan and the air valve is analyzed in multiple dimensions to evaluate their efficiency, safety and stability under the current operating conditions. The calculation of the comprehensive performance index needs to be combined with the actual operating data of the ventilation system, including key factors such as the fan energy consumption, the air valve adjustment response rate, the plant pressure difference stability, the air flow uniformity, the equipment vibration amplitude and the radioactive material control ability. By establishing a multi-objective weighted evaluation model, the comprehensive performance score is calculated to quantify the overall operation quality of the system. The threshold comparison analysis of the comprehensive performance index is performed to determine whether parameter optimization or fault diagnosis is required. A set of reasonable performance threshold ranges is set, and the current comprehensive performance index is compared with these preset thresholds. If the comprehensive performance index remains within the normal operating range, the system continues to operate according to the existing control strategy. If the index deviates from the preset threshold (such as abnormal increase in fan energy consumption, lag in air valve opening adjustment, and excessive fluctuation of system pressure difference), the parameter optimization or fault diagnosis program is triggered to generate a parameter optimization trigger signal. After the parameter optimization trigger signal is activated, multi-dimensional features are extracted based on the operating data of the fan speed and the air valve opening value to obtain the fault feature vector. The feature extraction process covers three aspects: time domain, frequency domain and state features. The time domain features mainly include parameters such as the fan vibration mean, standard deviation, skewness, kurtosis, and pulsation amplitude, which reflect the stability and dynamic fluctuation of the fan and air valve operation; the frequency domain features use fast Fourier transform to analyze the frequency components of the fan and air valve operation data to detect whether abnormal frequency components appear, such as abnormal signals in a specific frequency band when the fan bearing or blade fails; the state features are combined with the ventilation system operation mode, and the historical data are compared to analyze whether the fan operation state under different modes has changed significantly. For example, under the same load, the fan current fluctuates abnormally or the system response speed slows down after the air valve is adjusted, indicating that there is a potential fault in the system. Through the above feature extraction process, a fault feature vector is formed to effectively characterize the current operating health status of the fan and air valve. The fault feature vector is input into a deep belief network composed of a four-layer restricted Boltzmann mechanism for pattern recognition, and the fault type and its probability distribution are determined. Restricted Boltzmann Machine (RBM) is an unsupervised learning algorithm that can automatically extract features and learn fault modes from massive operating data. The deep belief network consists of multiple layers of RBM stacking, which can learn the hierarchical representation of ventilation system faults through layer-by-layer training. During the model training process, RBM is pre-trained in an unsupervised manner to learn typical fault features under different fan operating modes. Then, based on the trained RBM, a supervised training layer is added to enable the network to accurately map the input fault feature vector to different fault categories.During the real-time operation of the system, the deep belief network automatically identifies the possible fault types of the current system through the input feature vector, and calculates the probability distribution of each fault type, thereby providing accurate fault classification results. After completing the fault mode identification, the hierarchical fault response strategy analysis is performed according to the fault type and its probability distribution to generate the optimal fault handling solution. The fault response strategy is divided into different levels according to the severity of the fault. For example, for minor faults (such as delayed air valve adjustment or slight increase in fan power consumption), adaptive compensation is performed by adjusting control parameters or optimizing PID control strategies. For medium-level faults (such as abnormal vibration caused by fan bearing wear or changes in local resistance of the pipe network), local load adjustment, fan start-stop switching or maintenance warning measures are taken to alleviate them, and maintenance plans are arranged. For serious faults (such as damage to fan blades or blockage of key pipelines in the ventilation system), the emergency mode is immediately triggered, and safety measures such as rapid shutdown, switching to standby fans or emergency pressure relief are taken to avoid serious accidents in system operation. Through the hierarchical response strategy, it is ensured that the ventilation system can identify potential faults in real time during operation, and take the best optimization adjustment or safety protection measures to maintain the stability and efficiency of the ventilation system of the nuclear power plant.

[0021] In the embodiment of the present invention, in terms of data quality, the present invention adopts a distributed sensor network and a dual-channel redundant acquisition mechanism, combined with cubic spline interpolation to repair missing data, and applies digital filtering technology to effectively filter out power grid interference to ensure data integrity and reliability; in terms of system characteristic characterization, a complete ventilation system characteristic mathematical model is constructed by accurately modeling fan impedance characteristics and duct network topology; in terms of dynamic characteristic capture, the pressure difference field function is constructed by using the finite element method and the core features are extracted by combining the empirical mode decomposition method to achieve a comprehensive characterization of the dynamic characteristics of the ventilation system and future state prediction; in terms of control strategy, an active neural pressure difference field network architecture is used to integrate the functions of feature extraction, pressure difference prediction and parameter generation to form an end-to-end intelligent control system, and adaptively generate target variable frequency control parameters; in terms of operation mode, intelligent switching between normal operation, emergency response and maintenance modes is realized, and the coordinated control of fans and air valves and the multi-regional collaborative mechanism are combined to ensure the stable operation of the system under various working conditions; in terms of fault management, accurate identification of common faults is achieved through a deep confidence network, and the hierarchical fault response strategy and fault prediction based on a recurrent neural network are combined to greatly improve the system reliability and safety.

[0022] In a specific embodiment, the process of executing step 101 may specifically include the following steps: Use distributed airflow sensors, differential pressure sensors, temperature sensors, and humidity sensors to build a ventilation system monitoring network, and collect multi-area real-time operation data of the nuclear power plant ventilation system through the ventilation system monitoring network; Perform dual-channel redundant collection and transmission of real-time operation data in multiple regions to obtain double-verified data streams, and perform anomaly detection on the double-verified data streams to obtain the data missing mark position; The cubic spline interpolation function is applied to reconstruct the data at the data missing mark position to obtain the repaired continuous data stream, and the repaired continuous data stream is digitally filtered to obtain a standardized data set.

[0023] Specifically, a distributed sensor network is deployed in each key area of ​​the nuclear power plant. The network consists of a variety of measuring devices such as air flow sensors, differential pressure sensors, temperature sensors and humidity sensors to ensure that the system monitors key operating parameters such as air flow, differential pressure distribution, temperature and humidity changes in each area of ​​the plant in real time. Considering the complexity of the internal ventilation system of the nuclear power plant, the layout of sensors should cover multiple typical areas, such as the main ventilation duct, the ventilation outlets of each plant, the surroundings of key equipment, and the emergency exhaust system, and the density of sensors is dynamically adjusted according to the importance of different areas to ensure that high-risk areas have higher spatial resolution, while the sampling density is appropriately reduced in low-impact areas to optimize data acquisition costs and system response speed. In order to improve the reliability and anti-interference ability of data acquisition, the ventilation system monitoring network adopts a dual-channel redundant acquisition and transmission mechanism, that is, the data of each monitoring point is collected simultaneously by two independent data acquisition channels, and transmitted to the central data processing unit through two independent communication paths, forming a double verification data stream. The dual-channel redundant design can effectively reduce the risk of data loss or abnormality caused by sensor failure, network transmission interference or external environmental influence, and ensure real-time verification and error correction during data transmission. Set the sensor measurement values ​​of the two channels to be and , if the error between the two satisfies the following conditions: in, If the error threshold is greater than the set allowable error threshold, the data at that time point is considered to be abnormal or missing, and the data is marked as missing. For data points that have been marked as missing or abnormal, data reconstruction is performed to complete the missing information and ensure the continuity of the data. Since the operating parameters of the ventilation system have a strong time correlation, it is suitable to use the cubic spline interpolation function for data reconstruction. The core idea of ​​cubic spline interpolation is to construct a smooth cubic polynomial curve at the missing point so that the interpolated data maintains the continuity of the second-order derivative within the fitting interval, thereby ensuring the physical rationality of the data. Suppose that in a certain time interval If there is data missing between them, the cubic spline interpolation function can be expressed as: in, Indicates that in the interval The interpolation function inside is the interpolation coefficient to be determined. In order to ensure the smoothness of the interpolation curve and the rationality of the data, the following conditions need to be met: Continuity of interpolation points: Continuity of the first-order derivative: ; Continuity of the second-order derivative: By solving the above constraint equations, we can obtain the interpolation coefficients , and then calculate the interpolated value of the missing data point , generating a repaired continuous data stream. After data reconstruction, the repaired data stream is digitally filtered to remove high-frequency noise and retain the true physical signal characteristics. Considering that the data changes of the ventilation system have low-frequency characteristics, and high-frequency signals are often the result of measurement noise or environmental interference, a low-pass filter is used for processing. Assume that the reconstructed data stream is , then the filtered signal It is expressed as: in, is the impulse response function of the filter, and a Butterworth filter or a Kalman filter is selected for signal smoothing. For example, if a second-order Butterworth filter is used, its transfer function is expressed as: in, The cutoff frequency is set to an appropriate value according to the data characteristics to ensure that the real dynamic characteristics of the system are retained as much as possible while removing high-frequency noise. After completing the digital filtering process, a standardized data set is finally obtained.

[0024] In a specific embodiment, the process of executing step 102 may specifically include the following steps: Perform dynamic test analysis on the fan operating parameters in the standardized data set to obtain the original performance curve data point set of the fan; The original performance curve data point set is fitted and calculated using the least squares method to obtain the fan impedance model. At the same time, the pipe network structure information in the standardized data set is topologically modeled to obtain the ventilation system network model. Based on the ventilation system network model, the node association matrix and boundary condition vector are constructed to obtain the ventilation system resistance characteristic matrix equation; The pressure difference field function is constructed based on the fan impedance model and the ventilation system resistance characteristic matrix equation, and the pressure difference field mapping characteristics that characterize the dynamic characteristics of the ventilation system are calculated.

[0025] Specifically, the key parameters of the fan under different operating conditions are collected for a long time and at a high frequency. These parameters include the fan speed, air volume, power consumption, inlet and outlet pressure difference and environmental conditions (temperature, humidity, etc.). The fan performance curve is experimentally determined to measure the flow rate of the fan under different operating conditions. With pressure The relationship between the fan and the flow rate is not fixed, and since the fan is affected by the complex pipe network resistance and variable operating conditions during operation, its performance curve is not a single fixed one. Therefore, multiple data points under different operating conditions are dynamically collected to form the original performance data set of the fan. The original performance curve data point set is fitted and calculated using the least squares method to establish the impedance model of the fan. The core of the fan impedance model is to describe the nonlinear relationship between the flow rate and the pressure difference of the fan, and a quadratic polynomial or a cubic polynomial is used for fitting: in, represents the pressure difference of the fan, is the fan flow rate, are the fitting coefficients to be found. To determine these coefficients, construct the least squares error function: in, Represents the number of sampled data points, and is the fan flow rate and pressure difference value measured in the experiment. about The set of equations with zero partial derivatives is used to obtain the optimal fitting parameters of the fan impedance model, thereby accurately describing the flow-pressure difference characteristics of the fan. At the same time, in order to establish a complete ventilation system network model, the pipe network structure information in the standardized data set is topologically modeled. The ventilation pipe network is abstracted as a graph structure consisting of several nodes and connected edges, where the nodes represent key positions in the pipe network, such as fan inlets, outlets, branch intersections, and exhaust vents, while the edges represent ventilation ducts or local resistance elements (such as valves, filters, etc.). Based on the topological relationship, the node association matrix of the ventilation system is constructed. , where the elements Representation Node and In addition, the boundary condition vectors for the ventilation system are defined. , which includes the inlet and outlet pressure conditions of the fan, environmental factors (such as external wind pressure), etc., and describes the state of the system under external influences. After the topology modeling is completed, based on the node association matrix and the boundary condition vector Construct the resistance characteristic matrix equation of the ventilation system. Assume that the airflow in the ventilation duct obeys the Darcy-Weisbach equation: in, is the friction coefficient, is the pipe length, is the pipe diameter, is the air density, is the air velocity. Combining the fan pressure difference formula and the pipe network topology model, the resistance characteristic matrix equation is established: in, is the resistance matrix of the ventilation system, which describes the flow resistance characteristics of components such as pipes, valves, and fans. is the air volume vector, which represents the air volume distribution of each pipe in the system. is the boundary condition vector, including factors such as fan outlet pressure and external wind pressure. After obtaining the fan impedance model and the ventilation system resistance characteristic matrix equation, the pressure difference field function is constructed to characterize the overall dynamic characteristics of the ventilation system. The pressure difference field function establishes a continuous mathematical model so that the pressure difference distribution at each spatial position in the ventilation system can be calculated by analytical or numerical methods. It is assumed that the spatial area of ​​the ventilation system is discretized into a finite number of grid cells, and the air flow in each grid cell is assumed to satisfy the Poisson equation: in, The Laplace operator representing the pressure field, is the air flow rate per unit volume, is the aerodynamic viscosity coefficient. By solving this equation and combining the boundary condition vector , the pressure difference field distribution of the entire ventilation system is obtained. In order to describe the dynamic characteristics of the ventilation system more specifically, the pressure difference field mapping features are calculated, which are used to evaluate the operating status and stability of the system. For example, the energy distribution function of the pressure difference field is defined as: in, Represents the pressure difference energy of the entire system, For the space area of ​​the ventilation system. If it is too large, it means that the system has high wind resistance loss, and it is necessary to optimize the fan operation strategy or adjust the air valve opening.

[0026] In a specific embodiment, the execution step of constructing a pressure difference field function based on the fan impedance model and the ventilation system resistance characteristic matrix equation, and calculating the pressure difference field mapping characteristics characterizing the dynamic characteristics of the ventilation system may specifically include the following steps: Perform spatial division and boundary definition on the ventilation system control area to obtain a spatial model containing p control areas; Based on the spatial model, the Laplace equation and the fan outlet pressure boundary conditions are established at each grid node to obtain the initial pressure difference field distribution; The time dimension is introduced into the initial pressure difference field distribution and the empirical mode decomposition algorithm is applied to obtain the intrinsic mode function set representing the pressure difference field time series. The Hilbert spectrum analysis is performed on the intrinsic mode function set to obtain the pressure difference field spectrum feature set consisting of instantaneous frequency and instantaneous amplitude information; Based on the pressure difference field spectrum feature set, a feature vector is constructed for each control area, and the feature vector includes the pressure difference mean value, standard deviation, peak factor, pulsation intensity and autocorrelation function decay rate; The response surface method is used to establish the mapping relationship between the characteristic vector and the fan speed and the air valve opening, and the autoregressive moving average model is constructed to obtain the pressure difference field mapping characteristics.

[0027] Specifically, the entire ventilation system area is divided into blocks. There are control areas, each of which corresponds to a functional unit in the ventilation system of a nuclear power plant, such as the main ventilation duct, auxiliary vents, fuel storage room, reactor building exhaust system, etc. The division criteria are based on fan layout, air valve adjustment capacity, air flow distribution characteristics and boundary constraints to ensure that the flow characteristics of each area can be analyzed independently and can be coupled with each other during overall modeling. For the entire ventilation system space, For the control areas, then: The union of all regions covers the entire ventilation system and they do not overlap with each other. It is a physical boundary (such as a plant wall), a fan outlet, a damper adjustment port, or a shared boundary of an adjacent area. Based on this spatial model, the Laplace equation is established at each grid node to solve the static pressure difference field distribution. is the pressure difference distribution function in the ventilation system. In the absence of an external pressure source, the pressure difference field satisfies the Poisson equation: in, is the Laplace operator of the pressure field, is the air flow rate per unit volume, is the aerodynamic viscosity coefficient. For the boundary conditions of the fan outlet, the pressure at the fan outlet is , then it is necessary to meet: in is the fan outlet boundary. This set of equations is solved by the finite difference method or the finite element method to obtain the initial pressure difference field distribution. After obtaining the static pressure difference field, in order to describe the dynamic characteristics of the ventilation system, the time dimension is introduced, making the pressure difference field a time dimension. The function of The empirical mode decomposition method is applied to decompose the pressure difference field time series into a set of intrinsic mode functions (IMFs), namely: in, It is An intrinsic modal function, is the residual term, which represents the undecomposed trend term. Each modal function corresponds to the pressure difference oscillation component of different scales, which can reflect the dynamic characteristics of ventilation on different time scales. The Hilbert spectrum analysis is performed on the intrinsic modal function set to extract the instantaneous frequency and instantaneous amplitude information to form the spectral feature set of the pressure difference field. Assume The Hilbert transform of is: Then the instantaneous amplitude and instantaneous frequency are: in, represents the instantaneous intensity of the pressure difference fluctuation, Represents the local pressure difference fluctuation frequency. By analyzing the instantaneous spectrum characteristics of different areas, the pressure change patterns of different areas of the ventilation system are identified. Based on the extracted pressure difference field spectrum feature set, a feature vector is constructed for each control area. The eigenvectors of are: in, It is a region The average pressure difference, is the standard deviation, is the crest factor, a measure of extreme pressure fluctuations: is the pulsation intensity: is the decay rate of the autocorrelation function: This eigenvector can fully characterize the dynamic characteristics of the pressure difference in the control area. In order to establish the mapping relationship between the eigenvector and the fan speed and air valve opening, the response surface method is used. Assuming that the fan speed Air valve opening If the pressure difference eigenvector is affected, the second-order response surface equation is constructed: in, is an undetermined parameter obtained by fitting experimental data. In order to establish a dynamic pressure difference field prediction model, an autoregressive moving average model is constructed, and the pressure difference characteristic vector is assumed to evolve over time as follows: in, and are the autoregressive coefficient and the moving average coefficient, respectively. is a white noise term. The model can predict the pressure difference field characteristics of different control areas and optimize the fan control strategy to achieve efficient and intelligent management of the ventilation system of nuclear power plants.

[0028] Among them, based on the impedance model and the ventilation system resistance characteristic matrix equation, the finite element method is used to construct the pressure difference field function, and the pressure difference field mapping characteristics that characterize the dynamic characteristics of the ventilation system are obtained. Before the pressure difference field mapping characteristics are input into the active neural pressure difference field network architecture for variable frequency parameter prediction, the following steps are also included: based on the pressure difference field mapping characteristics, the basic dynamic equations of the ventilation system are constructed to obtain a group of partial differential equations that describe the evolution of the pressure difference field. The equation group includes the mass conservation equation, the momentum conservation equation and the energy conservation equation, and introduces vortex and turbulence correction terms; the partial differential equation group is parameterized using a physical constraint neural network to obtain a neural network model embedded with physical knowledge. The forward propagation part of the neural network model represents the equation solving process, and the loss function contains a measure of violation of physical laws; the neural network model is deployed at the boundary points of each key area of ​​the nuclear power plant ventilation system to obtain a distributed environmental perception network. The distributed network consists of multiple active perception nodes, and each node has local computing and communication capabilities; based on the distributed environmental perception network, the environmental parameters of the nuclear power plant ventilation system are actively detected to obtain real-time environmental parameters. The system uses the spatiotemporal attention mechanism to perform multi-scale analysis on the real-time environmental disturbance data to obtain a characteristic map of environmental changes. The attention mechanism highlights the environmental change patterns at key time points and spatial locations through adaptive weighting. The environmental change characteristic map is used to model the inter-regional impact propagation through a graph neural network guided by physical knowledge to obtain a dynamic pressure difference field topology. The nodes in the graph neural network represent pressure measurement points and the edges represent airflow propagation paths. The dynamic pressure difference field topology is compared with the traditional CFD simulation results for verification and error correction to obtain a corrected high-precision pressure difference field dynamic model. The correction process uses the Bayesian optimization algorithm to minimize the deviation between the measured value and the predicted value. Based on the high-precision pressure difference field dynamic model, an active adaptive pressure difference control strategy library is constructed to obtain the optimal control parameter combination for different environmental changes. The control strategy library adopts a hierarchical structure, including three levels: steady-state control, transition state control and emergency response control, and the control strategy is continuously optimized through reinforcement learning methods.

[0029] In a specific embodiment, the process of executing step 103 may specifically include the following steps: The pressure difference field mapping features are input into the feature extraction network in the active neural pressure difference field network model for dimensionality reduction processing to obtain potential feature representation; The pressure difference field prediction network with a double-layer LSTM structure in the active neural pressure difference field network model is used to perform time series analysis on the potential feature representation to obtain the predicted value of the pressure difference field change. Input the potential feature representation and the predicted value of the pressure difference field change into the control parameter generation network in the active neural pressure difference field network model for mapping conversion to obtain the initial variable frequency control parameters; Perform pattern recognition analysis on the pressure difference field mapping characteristics to obtain historical working mode determination results; Based on the historical working mode determination results, the initial frequency conversion control parameters are adaptively adjusted to obtain the target frequency conversion control parameters.

[0030] Specifically, the pressure difference field mapping features are processed by dimensionality reduction to reduce data redundancy, improve subsequent calculation efficiency and extract the most representative potential feature representation. Since the pressure difference field mapping features contain multidimensional feature vectors of multiple control areas, such as pressure difference average value, standard deviation, peak factor, pulsation intensity and autocorrelation function decay rate, a nonlinear dimensionality reduction method is used to construct a feature extraction network. Assume that the original pressure difference field feature vector is ,in represents the number of samples, Represents the feature dimension of each sample, and uses a dimensionality reduction method based on an autoencoder to map high-dimensional data to a low-dimensional potential feature space. The encoder network is defined as: in, is the potential feature representation after dimensionality reduction, and are the weight matrix and bias vector of the encoder respectively, represents a nonlinear activation function, such as ReLU or Sigmoid, is the feature dimension after dimensionality reduction The dimensionality reduction network minimizes the reconstruction error: in, is the decoder network. The trained encoder can extract the main dynamic features of the ventilation system and reduce redundant information, making the subsequent time series analysis more stable and efficient. The potential features after dimensionality reduction are represented as The pressure difference field prediction network with a double-layer long short-term memory (LSTM) structure is input into the active neural pressure difference field network model to perform time series analysis and predict the future pressure difference field change trend. Since the pressure difference field characteristic of the ventilation system is a dynamic time series signal, its change is affected by the historical state. The LSTM structure can effectively model the long-term dependency in time series data. Assume that the input of the LSTM network is , the output is the predicted value of the pressure difference field at the next moment , then the LSTM recursive calculation is as follows: in, is the hidden state vector, is the memory unit state, are the weight matrices of the LSTM network, is the bias term. The network minimizes the prediction error: Optimize it to accurately predict the future changes in the pressure difference field characteristics. With the current latent feature representation The control parameters are input to the control parameter generation network of the active neural pressure difference field network to perform the mapping conversion of the variable frequency control parameters. The control parameter generation network adopts a fully connected neural network, and its input is , the output is the initial frequency conversion control parameters: in, Represents the preliminary control parameters of fan speed and air valve opening, is the weight matrix, is the bias vector. The optimization goal of this network is: in, is the optimal variable frequency control parameter. After obtaining the initial variable frequency control parameter, the pressure difference field mapping characteristics are subjected to pattern recognition analysis to determine the historical working mode of the current system. Suppose the historical mode database of the system is , where each mode It consists of the past pressure difference characteristics and the corresponding control strategies. By calculating the similarity between the current characteristics and the historical patterns: Determine the current best matching historical pattern : Then determine the current operating state of the system. According to the historical working mode determination results, the initial frequency conversion control parameters are adaptively adjusted to obtain the final target frequency conversion control parameters. Suppose the adjustment function is , then the target frequency conversion control parameters are: in, Optimize the initial parameters based on the historical mode experience data. For example, if the current mode matches a high load condition and the initial control parameters suggest reducing the fan speed, then The fan speed will be appropriately increased based on the historical best strategy to avoid system overload.

[0031] In a specific embodiment, the process of executing step 104 may specifically include the following steps: The multi-mode frequency conversion controller performs working scenario analysis and mode recognition on the target frequency conversion control parameters to obtain the current operation mode, wherein the current operation mode includes a normal operation mode, an emergency response mode and a maintenance mode; According to the current operation mode, the target frequency conversion control parameter is input into the feedforward control unit to calculate the initial parameter, so as to obtain the initial frequency conversion parameter; The initial frequency conversion parameters and the actual pressure difference deviation value are input into the incremental PID controller for feedback compensation calculation to obtain the control increment; Based on the operation mode determination and control increment, the mode preset control strategy is executed to obtain the mode optimization control output; Execute the coordinated control logic of the fan and the air valve according to the mode optimization control output and the pressure difference deviation value to obtain the coordinated control instruction; Based on the synergistic control instructions and regional coupling matrix, multi-region coordinated control and load switching processing are carried out to obtain the fan speed and air valve opening value.

[0032] Specifically, the target frequency conversion control parameters are analyzed and pattern recognized to determine the current operation mode. In the ventilation system of a nuclear power plant, the operation modes include normal operation mode, emergency response mode and maintenance mode, and the key to determining the current mode is to monitor the fan speed in real time. , air valve opening , Plant pressure difference and its deviation , and combined with the system's historical status, use the pattern recognition algorithm for classification. Define the pattern classification function at the current moment as: in, Represents the current most matching pattern. is the pattern similarity, is the current system state vector, For Mode The historical state vector, pattern library The typical operating characteristics of different modes are stored. After confirmation, the system identifies the current operating mode and enters the corresponding control process. After the mode recognition is completed, the target frequency conversion control parameters are set according to the current operating mode. Input the feedforward control unit to calculate the initial frequency conversion parameters. The function of the feedforward control unit is to use the pressure difference field dynamic model to predict the optimal fan speed and air valve opening, thereby providing a near-optimal control initial value. Assume that the state variables of the ventilation system are , the fan characteristic function is , then the feedforward control model is expressed as: in, is the initial frequency conversion parameter, Represents the feedforward control mapping relationship obtained based on historical data fitting. This function is established through neural network or response surface method and minimizes the prediction error: Train it so that its output is as close as possible to the optimal frequency conversion control parameters. Use feedback control mechanism to compensate for errors. Suppose the current pressure difference deviation is ,in is the target pressure difference, is the currently measured pressure difference, then the initial frequency conversion parameters and Enter the incremental PID controller to calculate the control increment: in, is the PID control increment, are the proportional, integral and differential gain coefficients respectively, is the sampling period. The goal of PID control is to minimize the steady-state error and improve the dynamic response capability of the system. Based on the operation mode determination and control increment, the mode preset control strategy is executed to obtain the mode optimized control output. Different operation modes correspond to different control strategies: in the normal operation mode, the system maintains the optimal energy consumption, while in the emergency response mode, the system gives priority to ensuring the stability of the negative pressure of the plant, that is: in, Represents a mode-based optimization function. For example, in the emergency mode with high pressure differential demand, the fan speed should be increased appropriately, while in the maintenance mode with low load, the damper should be partially closed to reduce energy consumption. Optimize control output according to mode And differential pressure deviation , execute the coordinated control logic of the fan and the air valve to obtain the synergistic control instructions. Since there is nonlinear coupling between the operation of the fan and the air valve, the control strategy is solved through collaborative optimization: ; in, For control instructions, For the optimized fan speed and air valve opening, is the weight coefficient to balance the pressure difference control, energy consumption and system response rate. Control command based on synergy and the regional coupling matrix , perform multi-zone coordinated control and load switching processing to optimize fan speed and air valve opening. regions, and the coupling relationship between regions is represented by the matrix express: in, represents the pressure difference field distribution after adjustment, Describe the airflow coupling relationship between zones. By solving this equation, the load demand of each zone is determined, and the fan operating mode is adjusted when necessary, such as shutting down some auxiliary fans or increasing the fan speed in high-load areas.

[0033] In a specific embodiment, the frequency conversion control method for executing the ventilation system of a nuclear power plant further includes the following steps: Perform system performance evaluation based on fan speed and air valve opening value to obtain comprehensive performance indicators; Perform threshold comparison and analysis on comprehensive performance indicators to obtain parameter optimization trigger signals; Based on the parameter optimization trigger signal, the time domain, frequency domain and state features are extracted from the operating data of the fan speed and the air valve opening value to obtain the fault feature vector; The fault feature vector is input into a deep belief network composed of four layers of restricted Boltzmann mechanism for pattern recognition to obtain the fault type and its probability distribution; Perform hierarchical fault response strategy analysis based on fault types and their probability distribution to generate fault handling plans.

[0034] Specifically, the comprehensive performance indicators of the system are defined to quantify the current operating status of the ventilation system. These comprehensive performance indicators are composed of multiple key parameters including fan energy consumption , fan efficiency , System pressure difference stability , Air valve adjustment sensitivity and air flow uniformity In order to form a unified comprehensive performance index, a weighted scoring function is defined: in, is the weight coefficient, which is obtained through historical data optimization, so that the comprehensive performance index Can effectively evaluate the current operating status of the system. Perform threshold comparison analysis to determine whether parameter optimization or fault detection is required. Set the normal operation threshold range , then the generation rule of parameter optimization trigger signal is: when When the parameter optimization trigger signal is activated, the fan speed is based on the fan speed. Air valve opening The operating data is used to extract the time domain, frequency domain and state features to form a fault feature vector . Time domain features include mean , Standard Deviation , skewness and crest factor : in, Represents the time series of fan speed or air valve opening, is the observation window length. Frequency domain features are obtained through Fourier transform, including the main frequency and spectrum energy : in, is the spectrum component after Fourier transformation. The state characteristics are described by autoregressive models, such as the ARMA model of wind turbine operation: in, and are the autoregressive and moving average coefficients, is the noise term. By fitting the ARMA model, the dynamic stability of the fan and the air valve is detected and the potential failure mode is identified. After extracting the fault features, the feature vector The input is a deep belief network (DBN) consisting of a four-layer restricted Boltzmann machine (RBM) for pattern recognition to obtain the fault type and its probability distribution. The energy function of RBM is defined as: in, is the input feature vector, is the hidden layer neuron state, is the connection weight. DBN is composed of multiple RBM stacks, which are trained layer by layer to learn the fault feature representation. The output layer uses the Sottmax classifier to calculate the probability distribution of different fault types: in, Representative fault type The probability of is the parameter of the Softmax classifier. After identifying the fault type and its probability distribution, a hierarchical fault response strategy analysis is performed to generate a fault handling plan. According to the severity of the fault, it is divided into minor faults, moderate faults and major faults. Minor faults (such as air valve adjustment lag) can be handled by adjusting the control parameters To make an adaptive correction: in, is the adjustment factor. Moderate faults (such as fan bearing wear) require load switching, and the fan operation is adjusted through the regional coupling matrix: in, is the fan speed distribution after adjustment, is the fan speed distribution before adjustment, A load distribution matrix is ​​created. Serious failures (such as fan blade damage) require emergency shutdown and activation of the backup system. At this time, the emergency mode is triggered, allowing the backup fan to take over the load of the main fan and quickly adjust the system pressure difference through the boost exhaust mode.

[0035] The frequency conversion control method of the ventilation system of a nuclear power plant in the embodiment of the present invention is described above. The frequency conversion control system of the ventilation system of a nuclear power plant in the embodiment of the present invention is described below. Figure 2 , an embodiment of the variable frequency control system of the nuclear power plant ventilation system in the embodiment of the present invention includes: The acquisition module 201 is used to collect real-time operation data of multiple areas of the nuclear power plant ventilation system and perform standardization processing to obtain a standardized data set; A feature mapping module 202 is used to perform fan performance curve fitting and pressure difference field feature mapping on the standardized data set to obtain pressure difference field mapping features; The frequency conversion parameter prediction module 203 is used to input the pressure difference field mapping characteristics into the active neural pressure difference field network model to predict the frequency conversion parameters and obtain the target frequency conversion control parameters; The combined control module 204 is used to perform feedforward-feedback combined control on the target variable frequency control parameters through a multi-mode variable frequency controller to obtain the fan speed and the air valve opening value.

[0036] Through the collaborative cooperation of the above-mentioned components, in terms of data quality, the present invention adopts a distributed sensor network and a dual-channel redundant acquisition mechanism, combined with cubic spline interpolation to repair missing data, and applies digital filtering technology to effectively filter out power grid interference to ensure data integrity and reliability; in terms of system characteristic characterization, a complete mathematical model of ventilation system characteristics is constructed by accurately modeling the fan impedance characteristics and the duct network topology; in terms of dynamic characteristic capture, the pressure difference field function is constructed using the finite element method and the core features are extracted in combination with the empirical mode decomposition method to achieve a comprehensive characterization of the dynamic characteristics of the ventilation system and future state prediction; in terms of control strategy, an active neural pressure difference field network architecture is used to integrate the functions of feature extraction, pressure difference prediction and parameter generation to form an end-to-end intelligent control system, and adaptively generate target variable frequency control parameters; in terms of operation mode, intelligent switching between normal operation, emergency response and maintenance modes is achieved, and the coordinated control of fans and air valves and the multi-regional collaborative mechanism are combined to ensure the stable operation of the system under various working conditions; in terms of fault management, accurate identification of common faults is achieved through a deep confidence network, and the hierarchical fault response strategy and fault prediction based on a recurrent neural network are combined to greatly improve the system reliability and safety.

[0037] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0038] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0039] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A frequency conversion control method for a nuclear power plant ventilation system, characterized in that: include: Collect real-time operating data from multiple areas of the nuclear power plant ventilation system and perform standardized processing to obtain a standardized data set; Performing fan performance curve fitting and pressure difference field feature mapping on the standardized data set to obtain pressure difference field mapping features; Inputting the pressure difference field mapping characteristics into the active neural pressure difference field network model to predict the variable frequency parameters and obtain the target variable frequency control parameters; The target frequency conversion control parameter is subjected to feedforward-feedback combined control through a multi-mode frequency conversion controller to obtain the fan speed and air valve opening value.

2. The frequency conversion control method of the nuclear power plant ventilation system according to claim 1, characterized in that: The method collects real-time operation data of multiple areas of the nuclear power plant ventilation system and performs standardization processing to obtain a standardized data set, including: A ventilation system monitoring network is constructed using distributed airflow sensors, differential pressure sensors, temperature sensors, and humidity sensors, and multi-region real-time operation data of the ventilation system of a nuclear power plant is collected through the ventilation system monitoring network; Performing dual-channel redundant collection and transmission on the multi-region real-time operation data to obtain a double-verified data stream, and performing anomaly detection on the double-verified data stream to obtain a data missing mark position; A cubic spline interpolation function is applied to reconstruct data at the data missing mark position to obtain a repaired continuous data stream, and the repaired continuous data stream is digitally filtered to obtain a standardized data set.

3. The frequency conversion control method of the nuclear power plant ventilation system according to claim 1, characterized in that: The step of performing fan performance curve fitting and pressure difference field feature mapping on the standardized data set to obtain pressure difference field mapping features includes: Performing dynamic testing and analysis on the fan operating parameters in the standardized data set to obtain a data point set of the fan original performance curve; The original performance curve data point set is fitted and calculated using the least square method to obtain a fan impedance model, and the pipe network structure information in the standardized data set is topologically modeled to obtain a ventilation system network model; Based on the ventilation system network model, a node association matrix and a boundary condition vector are constructed to obtain a ventilation system resistance characteristic matrix equation; A pressure difference field function is constructed based on the fan impedance model and the ventilation system resistance characteristic matrix equation, and pressure difference field mapping features that characterize the dynamic characteristics of the ventilation system are calculated.

4. The frequency conversion control method of the nuclear power plant ventilation system according to claim 3, characterized in that: The pressure difference field function is constructed based on the fan impedance model and the ventilation system resistance characteristic matrix equation, and the pressure difference field mapping characteristics characterizing the dynamic characteristics of the ventilation system are calculated, including: Perform spatial division and boundary definition on the ventilation system control area to obtain a spatial model containing p control areas; Based on the spatial model, a Laplace equation and a fan outlet pressure boundary condition are established at each grid node to obtain an initial pressure difference field distribution; Introducing a time dimension into the initial pressure difference field distribution and applying an empirical mode decomposition algorithm to obtain a set of intrinsic mode functions representing a time series of the pressure difference field; Performing Hilbert spectrum analysis on the intrinsic modal function set to obtain a pressure difference field spectrum feature set consisting of instantaneous frequency and instantaneous amplitude information; Constructing a feature vector for each control area based on the pressure difference field spectrum feature set, wherein the feature vector includes a pressure difference average value, a standard deviation, a peak factor, a pulsation intensity, and an autocorrelation function decay rate; The response surface method is used to establish the mapping relationship between the characteristic vector and the fan speed and the air valve opening, and an autoregressive moving average model is constructed to obtain the pressure difference field mapping characteristics.

5. The frequency conversion control method of the nuclear power plant ventilation system according to claim 1, characterized in that: The step of inputting the pressure difference field mapping feature into the active neural pressure difference field network model to predict the variable frequency parameters and obtain the target variable frequency control parameters comprises: Inputting the pressure difference field mapping feature into the feature extraction network in the active neural pressure difference field network model for dimensionality reduction processing to obtain a potential feature representation; The pressure difference field prediction network with a double-layer LSTM structure in the active neural pressure difference field network model performs a time series analysis on the potential feature representation to obtain a pressure difference field change prediction value; Inputting the potential feature representation and the pressure difference field change prediction value into the control parameter generation network in the active neural pressure difference field network model for mapping conversion to obtain initial variable frequency control parameters; Performing pattern recognition analysis on the pressure difference field mapping characteristics to obtain a historical working mode determination result; The initial frequency conversion control parameter is adaptively adjusted based on the historical working mode determination result to obtain a target frequency conversion control parameter.

6. The frequency conversion control method of the nuclear power plant ventilation system according to claim 1, characterized in that: The method of performing feedforward-feedback combined control on the target variable frequency control parameter by the multi-mode variable frequency controller to obtain the fan speed and the air valve opening value includes: Performing work scenario analysis and mode recognition on the target frequency conversion control parameter through a multi-mode frequency conversion controller to obtain a current operation mode, wherein the current operation mode includes a normal operation mode, an emergency response mode, and a maintenance mode; According to the current operation mode, the target frequency conversion control parameter is input into a feedforward control unit to perform initial parameter calculation to obtain an initial frequency conversion parameter; Inputting the initial frequency conversion parameter and the actual pressure difference deviation value into an incremental PID controller for feedback compensation calculation to obtain a control increment; Based on the operation mode determination and the control increment execution mode preset control strategy, a mode optimization control output is obtained; Execute the coordinated control logic of the fan and the air valve according to the mode optimization control output and the pressure difference deviation value to obtain a coordinated control instruction; Based on the synergistic control instruction and the regional coupling matrix, multi-region coordinated control and load switching processing are performed to obtain the fan speed and air valve opening value.

7. The frequency conversion control method of a nuclear power plant ventilation system according to claim 1, characterized in that: The frequency conversion control method of the nuclear power plant ventilation system also includes: Performing system performance evaluation based on the fan speed and the air valve opening value to obtain a comprehensive performance index; Performing threshold comparison analysis on the comprehensive performance index to obtain a parameter optimization trigger signal; Based on the parameter optimization trigger signal, extracting time domain, frequency domain and state features from the operating data of the fan speed and the air valve opening value to obtain a fault feature vector; Inputting the fault feature vector into a deep belief network composed of a four-layer restricted Boltzmann machine for pattern recognition to obtain the fault type and its probability distribution; A hierarchical fault response strategy analysis is performed based on the fault type and its probability distribution to generate a fault handling plan.

8. A variable frequency control system for a nuclear power plant ventilation system, characterized in that: A method for executing a variable frequency control method for a nuclear power plant ventilation system according to any one of claims 1 to 7, wherein the variable frequency control system for the nuclear power plant ventilation system comprises: The acquisition module is used to collect real-time operation data of multiple areas of the nuclear power plant ventilation system and perform standardized processing to obtain a standardized data set; A feature mapping module, used to perform fan performance curve fitting and pressure difference field feature mapping on the standardized data set to obtain pressure difference field mapping features; A frequency conversion parameter prediction module is used to input the pressure difference field mapping characteristics into the active neural pressure difference field network model to predict the frequency conversion parameters and obtain the target frequency conversion control parameters; The combined control module is used to perform feedforward-feedback combined control on the target variable frequency control parameter through a multi-mode variable frequency controller to obtain the fan speed and air valve opening value.

Images