Nuclear power plant emergency diesel generator risk early warning method and system

Abstract

The invention discloses a risk early warning method and system for an emergency diesel generator of a nuclear power plant, and the method comprises the steps: collecting operation parameters of a generator set in a hot standby state, and when the parameters are not in a preset normal range, indicating that the parameters are abnormal and the generator set breaks down; preprocessing the acquired operating parameters, and extracting effective features of the preprocessed operating parameters; on the basis of the generator set with the fault, combining with the effective characteristics, and utilizing a high-frequency data flow transient anomaly recognition algorithm to capture an anomaly type label to obtain fault information; evaluating the risk degree of the generator set according to the effective features and the abnormal type labels of the preprocessed operation parameters; and carrying out risk early warning and response on the generator set according to the risk degree of the generator set. According to the invention, the operation state of the emergency diesel generator set can be monitored in real time, accurate risk assessment and early warning information is provided, and the risk management capability of managers is enhanced.

Description

Technical Field

[0001] This invention relates to the field of risk warning technology, specifically to a risk warning method and system for emergency diesel generators in nuclear power plants. Background Technology

[0002] In the nuclear power plant safety system, emergency diesel generators serve as the last line of defense, and their reliability directly affects the continuous operation of critical equipment such as the reactor cooling system. A failure of an emergency diesel generator during emergency startup could lead to catastrophic consequences such as core meltdown. Therefore, real-time monitoring and risk warning of the emergency diesel generator's operating status have become core technologies in the field of nuclear power safety. However, existing technologies still have significant limitations, specifically in the following aspects:

[0003] 1. Insufficient real-time performance and sensitivity of traditional monitoring technologies: Parameter acquisition is delayed, and conventional monitoring methods mostly rely on offline or periodic data acquisition, making it difficult to meet the "≤10 seconds" rapid response requirement for emergency diesel generator startup. Limitations of single-parameter threshold alarms: Existing systems typically set static thresholds based on a single parameter, but emergency diesel generator failures are often coupled failures of mechanical, electrical, and control systems; exceeding the limit of a single parameter cannot comprehensively reflect the risk.

[0004] 2. Weak ability to identify complex fault modes: The transient anomalies during the emergency start-up phase of emergency diesel generators are short in duration, and traditional methods are difficult to detect effectively due to the low sampling frequency (usually ≤1kHz).

[0005] 3. Static limitations of risk assessment models: Existing reliability analyses rely on a large amount of historical failure data and cannot dynamically adapt to extreme conditions such as aging emergency diesel generators. When multiple parameters deteriorate, static weight allocation models struggle to dynamically adjust the assessment focus, leading to a disconnect between risk assessment results and actual safety requirements. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a risk early warning method and system for emergency diesel generators in nuclear power plants, which solves the problems of difficulty in capturing transient faults, insufficient real-time analysis capability of multiple parameters and static risk assessment in the prior art.

[0007] To solve the above technical problems, the present invention adopts the following technical solution:

[0008] A risk warning method for emergency diesel generators in nuclear power plants includes the following steps:

[0009] S1. Collect the operating parameters of the generator set in hot standby mode. When the parameter is not within the preset normal range, it indicates that the parameter is abnormal and the generator set has failed.

[0010] S2. Preprocess the operating parameters collected in step S1 and extract the effective features of the preprocessed operating parameters; based on the generator set that failed in step S1, combine the effective features and use the high-frequency data stream transient anomaly identification algorithm to capture the anomaly type label and obtain the fault information.

[0011] S3. Assess the risk level of the generator set based on the effective characteristics and anomaly type labels of the preprocessed operating parameters.

[0012] S4. Based on the risk level of the generator set, conduct risk warning and response for the generator set.

[0013] Furthermore, in step S1, the operating parameters of the generator set include starting battery voltage, starting time, oil pressure, starting speed, coolant temperature, engine room temperature, fuel flow rate, vibration signal, electrical parameters, temperature parameters, and insulation resistance.

[0014] The preset normal range for starting battery voltage is 24V~28V, the preset normal range for starting time is less than 10s, the preset normal range for oil pressure is to rise to greater than or equal to 100kPa within 3s after starting and then stabilize to 200~600kPa, and the preset normal range for starting speed is to reach 1500RPM or 1800RPM within 5s after starting.

[0015] Furthermore, in step S2, the window size is set to 200ms and the step size is 50ms. The running parameters are segmented using a sliding window and normalized to obtain the preprocessed running parameters.

[0016] The effective characteristics of the preprocessed operating parameters include time-domain characteristics, frequency-domain characteristics, and time-frequency-domain characteristics.

[0017] Among them, the time-domain features include the voltage peak drop rate, current kurtosis, and rotational speed rise slope; the frequency-domain features include the dominant frequency amplitude; and the time-frequency-domain features include wavelet energy entropy.

[0018] An improved autoencoder model is constructed, which includes an input layer, an encoder, a decoder, and an output layer.

[0019] The effective feature vector of the preprocessed running parameters The input is fed into the improved autoencoder model. After the input layer receives the effective feature vector, it is transmitted to the encoder. The multilayer perceptron is used to compress the effective feature vector to obtain a low-dimensional latent vector. The specific expression is:

[0020] ;

[0021] in, Represents the encoder mapping function. The parameters represent the improved autoencoder model. , This represents the Dth effective feature component.

[0022] Low-dimensional hidden vectors After passing through the decoder, the reconstructed feature vector is obtained using a symmetric multilayer perceptron. The specific expression is:

[0023] ;

[0024] in, This represents the decoder mapping function.

[0025] Calculated using mean square error and The reconstruction error is specifically expressed as:

[0026] ;

[0027] in, Indicates reconstruction error, This represents the effective feature vector of the i-th preprocessed running parameter. This represents the reconstructed feature vector of the i-th preprocessed runtime parameters. This indicates the total number of preprocessed running parameters.

[0028] If the reconstruction error exceeds the set threshold, it is marked as a transient anomaly and the anomaly type label is output to obtain fault information.

[0029] Furthermore, the voltage peak drop rate is obtained using the adjacent window peak difference method, the current kurtosis value is obtained using the probability statistics method, the speed increase slope is obtained by fitting the speed data within the window using linear regression, and the main frequency amplitude is obtained using fast Fourier transform.

[0030] Furthermore, in step S3, a dynamic risk assessment model is constructed by combining the operational priority requirements of the nuclear power plant's emergency diesel generator set under different emergency conditions. This model is a random forest, consisting of N decision trees, constructed through random sampling and random feature selection. The dataset composed of the effective features of the preprocessed operating parameters and the abnormal type labels is divided into a training set and a test set in a 7:3 ratio. The dynamic risk assessment model is trained using the training set and tested using the test set to obtain the trained dynamic risk assessment model.

[0031] The effective features of the preprocessed running parameters and the anomaly type labels are input into the trained dynamic risk assessment model. Each decision tree performs a classification operation, and the key anomaly features and their importance scores are obtained through feature space mapping, and then the anomaly prediction results are output respectively.

[0032] The majority voting method is used to fuse all anomaly prediction results to obtain the final anomaly category prediction and its corresponding anomaly probability value. The anomaly feature importance score is obtained based on the split contribution of effective features in the trained dynamic risk assessment model.

[0033] The dynamic weights are calculated using the following formula:

[0034] ;

[0035] in, This represents the dynamic weight of the i-th preprocessed running parameter. This indicates the dynamic adjustment range; the baseline weight is determined by associating it with historical faults through an expert system.

[0036] Adjust the weight allocation of the trained dynamic risk assessment model according to the current working conditions.

[0037] A comprehensive risk score is obtained through weighted summation. The specific formula is as follows:

[0038] ;

[0039] in, This represents the real-time monitoring value of the i-th preprocessed running parameter. This represents the baseline value of the i-th parameter under normal operating conditions. The normal baseline values ​​of each parameter can be determined based on historical data and expert experience.

[0040] Furthermore, in step S4, a visual interface is used to display risk warnings and responses. The visual interface includes real-time data charts, risk types and levels, and warning information.

[0041] When the comprehensive risk score When the risk level is low, it is displayed in green, and the operating parameter curves and risk trends are updated in real time on the visualization interface, requiring no intervention; when the comprehensive risk score... When the risk level is medium, it will be displayed in yellow, and a warning prompt box will pop up in the visualization interface to review the operating parameters; when the comprehensive risk score is... When the risk is high, it will be displayed in red, triggering an audible and visual alarm on the visual interface and shutting down the corresponding device.

[0042] Furthermore, this invention also proposes a risk early warning system for emergency diesel generators in nuclear power plants, comprising:

[0043] The fault test module is used to collect the operating parameters of the generator set in hot standby mode. When the parameter is not within the preset normal range, it indicates that the parameter is abnormal and the generator set has failed.

[0044] The fault information acquisition module is used to preprocess the operating parameters collected in the fault test module and extract the effective features of the preprocessed operating parameters. Based on the generator set that has failed in the fault test module, and combined with the effective features, the high-frequency data stream transient anomaly identification algorithm is used to capture the anomaly type label and obtain the fault information.

[0045] The fault early warning module is used to assess the risk level of the generator set based on the effective characteristics and anomaly type labels of the preprocessed operating parameters; and to provide risk warnings and responses to the generator set based on the risk level.

[0046] Furthermore, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the risk warning method for the emergency diesel generator of the nuclear power plant.

[0047] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the aforementioned risk warning method for emergency diesel generators in nuclear power plants.

[0048] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0049] 1. This invention can monitor the operating status of emergency diesel generator sets in real time, and with the rapid parallel collection of multiple parameters, it meets the strict requirements of the start-up time of emergency diesel generators in nuclear power plants, realizing a rapid response across the entire chain from data acquisition to status assessment.

[0050] 2. This invention focuses on the problem of transient faults that are difficult to detect during emergency startup of a unit and issues early warnings.

[0051] 3. This invention provides accurate risk assessment and early warning information, and combined with a dynamic risk assessment mechanism, enhances the risk management capabilities of managers. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the overall implementation of the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0054] To achieve the above objectives, this invention proposes a risk early warning method for emergency diesel generators in nuclear power plants, such as... Figure 1 As shown, the specific steps are as follows:

[0055] S1. Utilize high-speed parallel acquisition technology to collect real-time operating parameters of the generator set under hot standby status. When a parameter deviates from the preset normal range, it indicates an anomaly and a generator set failure. This step provides raw data for subsequent judgments. If the parameter is abnormal, it triggers the subsequent anomaly analysis process, specifically:

[0056] The operating parameters of the generator set include starting battery voltage, starting time, oil pressure, starting speed, coolant temperature, engine room temperature, fuel flow, vibration signal, electrical parameters, temperature parameters, and insulation resistance.

[0057] The default normal range for starter battery voltage is 24V~28V. An abnormal starter battery voltage indicates battery aging, starter motor overload, or internal short circuit. The default normal range for starter time is less than 10 seconds. An abnormal starter time indicates insufficient fuel supply, insufficient air intake, or insufficient battery capacity. The default normal range for oil pressure is to rise to at least 100 kPa within 3 seconds of starting and then stabilize at 200~600 kPa. Oil pressure less than 200 kPa indicates insufficient lubrication, which may cause bearing wear. Oil pressure greater than 600 kPa indicates a clogged filter or oil circuit malfunction. The default normal range for starter speed is to reach 1500 RPM or 1800 RPM within 5 seconds of starting. A starter speed less than 1500 RPM or 1800 RPM may cause abnormal power output frequency, leading to damage to the load equipment. Fluctuating starter speed indicates a governor malfunction or unstable fuel supply. Excessively high coolant temperature and engine room temperature may trigger a high-temperature shutdown.

[0058] S2. Preprocess the operating parameters collected in step S1 and extract the effective features of the preprocessed operating parameters; based on the generator set that experienced a fault in step S1, and combined with these effective features, use a high-frequency data stream transient anomaly identification algorithm to capture anomaly type labels and obtain fault information. Specifically:

[0059] The window size is set to 200ms and the step size to 50ms. The running parameters are segmented using a sliding window and normalized to obtain the preprocessed running parameters.

[0060] The effective characteristics of the preprocessed operating parameters include time-domain characteristics, frequency-domain characteristics, and time-frequency-domain characteristics.

[0061] Among them, the time-domain features include the voltage peak drop rate, current kurtosis, and rotational speed rise slope; the frequency-domain features include the dominant frequency amplitude; and the time-frequency-domain features include wavelet energy entropy.

[0062] The voltage peak drop rate is obtained using the adjacent window peak difference method, the current kurtosis value is obtained using the probability statistics method, the speed rise slope is obtained by fitting the speed data within the window using linear regression, and the main frequency amplitude is obtained using the fast Fourier transform.

[0063] An improved autoencoder model is constructed (based on the existing autoencoder architecture, the hidden layer dimension and activation function are optimized). The model includes an input layer, an encoder, a decoder and an output layer.

[0064] The effective feature vector of the preprocessed running parameters The input is fed into the improved autoencoder model. After the input layer receives the effective feature vector, it is transmitted to the encoder. The multilayer perceptron is used to compress the effective feature vector to obtain a low-dimensional latent vector. The specific expression is:

[0065] ;

[0066] in, Represents the encoder mapping function. The parameters represent the improved autoencoder model. , This represents the Dth effective feature component.

[0067] Low-dimensional hidden vectors After passing through the decoder, the reconstructed feature vector is obtained using a symmetric multilayer perceptron. The specific expression is:

[0068] ;

[0069] in, This represents the decoder mapping function.

[0070] Calculated using mean square error and The reconstruction error is specifically expressed as:

[0071] ;

[0072] in, Indicates reconstruction error, This represents the effective feature vector of the i-th preprocessed running parameter. This represents the reconstructed feature vector of the i-th preprocessed runtime parameters. This indicates the total number of preprocessed running parameters.

[0073] If the reconstruction error exceeds the set threshold, it is marked as a transient anomaly and the anomaly type label is output to obtain fault information.

[0074] S3. Based on the effective characteristics and anomaly type labels of the preprocessed operating parameters, assess the risk level of the generator set. Specifically:

[0075] When multiple generator unit parameters deteriorate, a dynamic risk assessment model is constructed based on the operational priority requirements of emergency diesel generator units in nuclear power plants under different emergency conditions, especially considering the typical causes of nuclear power plant accidents. This model is a random forest, consisting of N decision trees. The decision trees are constructed through random sampling and random feature selection to reduce the risk of overfitting and improve generalization ability. The dataset consisting of effective features of preprocessed operating parameters and abnormal type labels is divided into training and test sets in a 7:3 ratio. The dynamic risk assessment model is trained using the training set and tested using the test set to obtain the trained dynamic risk assessment model.

[0076] The effective features of the preprocessed running parameters and the anomaly type labels are input into the trained dynamic risk assessment model. Each decision tree performs a classification operation, and the key anomaly features and their importance scores are obtained through feature space mapping, and then the anomaly prediction results are output respectively.

[0077] The majority voting method is used to fuse all anomaly prediction results to obtain the final anomaly category prediction (0 or 1) and its corresponding anomaly probability value. The anomaly feature importance score is obtained based on the split contribution of effective features in the trained dynamic risk assessment model.

[0078] The dynamic weights are calculated using the following formula:

[0079] ;

[0080] in, This represents the dynamic weight of the i-th preprocessed running parameter. Indicates the dynamic adjustment range. The baseline weights are determined by associating them with historical faults through an expert system.

[0081] The weight allocation of the trained dynamic risk assessment model is adjusted according to the current operating conditions, prioritizing the assessment of operating states that are critical to safety, thereby achieving accurate risk quantification.

[0082] A comprehensive risk score is obtained through weighted summation. The specific formula is as follows:

[0083] ;

[0084] in, This represents the real-time monitoring value of the i-th preprocessed running parameter. This represents the baseline value of the i-th parameter under normal operating conditions. The normal baseline values ​​of each parameter can be determined based on historical data and expert experience.

[0085] S4. Based on the risk level of the generator set, conduct risk warnings and responses for the generator set. Specifically:

[0086] A visual interface is used to display risk warnings and responses. This interface includes real-time data charts, risk types and levels, and warning information. The interface is interactive, allowing users to interact with the data through clicking and dragging to gain a deeper understanding of details. It also features alarm and notification functions; when a risk exceeds a threshold or meets alarm conditions, it automatically sends alerts to relevant personnel, helping them to quickly take measures to mitigate the risk.

[0087] When the comprehensive risk score When the risk level is low, it is displayed in green, and the operating parameter curves and risk trends are updated in real time on the visualization interface, requiring no intervention; when the comprehensive risk score... When the risk level is medium, it will be displayed in yellow, a warning message will pop up on the visual interface, and a notification will be pushed to the on-duty personnel via the management terminal. It is recommended to manually review the key parameters of the fuel, lubrication, or cooling systems. When the overall risk score is... When the risk level is high, it will be displayed in red. An audible and visual alarm will be triggered on the visual interface, and the risk report and abnormal parameters will be automatically reported to the control center. At the same time, an emergency shutdown strategy or a switch to a backup power module will be executed to prevent the equipment failure from escalating.

[0088] S5 utilizes Grafana to build dashboards that dynamically display real-time parameter curves (such as engine speed and oil pressure), risk level trends (line charts), and historical anomaly statistics (bar charts). It uses the time-series database InfluxDB to store raw data and analysis results by timestamp. SQL queries are supported (e.g., "query all high-risk events in the past week"). Operators can quickly locate problems through the dashboard and analyze the causes of failures by backtracking through the database.

[0089] Example:

[0090] In a simulated operating condition of a 1,000 MW nuclear power plant emergency diesel generator (rated power 20MW, rated speed 1800RPM, in a hot standby state ready to start), during the emergency start-up process (0-10s critical response phase), a "minor leak in the fuel supply system" was artificially injected, resulting in a 25% drop in fuel flow rate (from the normal range of 8-10L / min to 6L / min) within 3 seconds after start-up, an abnormal rate of oil pressure rise (normally ≥100kPa within 3 seconds, actual measurement was 80kPa), and a delay in the time to reach the starting speed standard (normally 5 seconds to reach the rated speed, actual measurement was 7 seconds).

[0091] Multi-parameter sensing modules can be selected for acquiring operating parameters. These modules include a speed sensor (accuracy ±0.05%), an in-cylinder pressure sensor (range 0~25MPa), a lubricating oil pressure sensor (0~1MPa), a winding temperature sensor (RTD type), a vibration accelerometer, and voltage and current sensors.

[0092] Field-programmable gate array (FPGA) technology is employed, relying on an 8-channel synchronous analog-to-digital converter (ADC) for analog signal acquisition. Data from each sensor is transmitted via a dedicated DMA channel to avoid data congestion. An integrated DS3231 high-precision clock ensures a timestamp accuracy of 1μs, enabling the acquisition and processing of multi-channel parallel data. Using development tools such as Nios IIIDE, a data acquisition program is written to achieve communication with the sensors and data acquisition. A preset relationship table, showing the correspondence between the preset faults and the unit parameters, is established, as shown in Table 1.

[0093] Table 1. Correspondence between operating parameters and preset accidents

[0094]

[0095] In this embodiment, the operating parameters of the generator set under hot standby status are collected as shown in Table 2.

[0096] Table 2 Operating parameters of the generator set

[0097]

[0098] Based on the data in Table 2, a correlation analysis was performed: At the critical 3-second mark, the oil pressure was 80 kPa, exceeding the preset threshold of 100 kPa, while the fuel flow rate was 6 L / min, below the normal lower limit of 8 L / min, thus triggering the preset accident "Insufficient fuel supply and bearing wear risk". At 5 seconds, the engine speed remained abnormally low (1500 RPM < 1620 RPM). Combined with the aforementioned fuel and pressure anomalies, this can be identified as a "compound fault risk".

[0099] An emergency start-up test was conducted on the nuclear power plant in this embodiment, and an ignition failure transient fault occurred during the start-up phase. The collected operating data is shown in Table 3.

[0100] Table 3. Operational data collected during ignition failure.

[0101]

[0102] The fault window is 1150-1350ms, which includes the complete fault process plus 100ms of normal data before and after; the normal window is 0-200ms (standby status data before startup).

[0103] Peak voltage drop rate = The fault sensitivity is manifested by a voltage drop rate >50% when ignition fails, and a spark plug voltage peak fluctuation ≤10% during normal start-up. When ignition fails, the spark plug discharge energy drops sharply, resulting in a voltage peak drop rate >50%. The voltage peak drop rate directly reflects the discharge reliability of the ignition system by capturing the transient voltage drop amplitude.

[0104] Current kurtosis value = , This indicates the number of current data points within the window. Indicates the first One current sample value, This represents the average current within the window. This represents the standard deviation of the current within the window. Under normal operating conditions, the current signal follows an approximately normal distribution with a kurtosis value of ≈3. When faults such as bearing wear or abnormal fuel injection occur, the current signal contains impulsive components, and the kurtosis value is >3. The current kurtosis value identifies current anomalies caused by mechanical faults by quantifying the peaks in the probability distribution of the signal.

[0105] Speed ​​increase slope = The fault sensitivity is characterized by a normal speed of ≥200 RPM / s and a fault speed of <150 RPM / s. During normal startup, the speed rise slope is ≥200 RPM / s, reaching the rated speed within 5 seconds; when fuel supply is insufficient or the governor malfunctions, the slope is <150 RPM / s, causing the startup time to exceed the limit. The speed rise slope is directly related to the startup failure risk by assessing the speed response speed.

[0106] A large amount of historical data was collected, including 2000 fault-free normal startups and 300 ignition failures, as a dataset. Unsupervised learning methods, such as autoencoders or generative adversarial networks, were used to model the normal startup data and learn its latent representation. Fault data was identified by calculating the reconstruction error or discriminator loss. The mean reconstruction error of the training set was obtained through training with the 2000 normal startups. =0.03, standard deviation =0.01, threshold is set to 0.06. When MSE > 0.06, an "ignition failure risk" warning is triggered.

[0107] In this embodiment, a sudden failure in the reactor cooling system caused the emergency diesel engine to start. Under this condition, an accurate assessment of the generator's operational risks is necessary. Therefore, four key parameters were selected for monitoring: starting speed, oil pressure, fuel flow rate, and winding temperature, with data collected every second.

[0108] Set the starting speed reference value RPM, oil pressure reference value kPa, fuel flow baseline L / min, winding temperature reference value ℃. Anomaly labels are determined through analysis of historical fault data; for example, parameters deviating from a certain baseline value are marked as abnormal. Startup speed weighting is set. Oil pressure weight Fuel flow weight Winding temperature weighting .

[0109] The data collected 5 seconds after the generator started is: starting speed. Oil pressure kPa, fuel flow rate L / min, winding temperature The risk scores were calculated based on the temperature (℃). The risk score for starting speed was 3, for oil pressure 2.5, for fuel flow 3.75, and for winding temperature 0.83, resulting in a total risk score of 10.08. This fault is considered low-risk, but because the risk value is close to the upper limit of the low-risk threshold, close monitoring of each parameter is still necessary. If the total risk score continues to rise, timely measures should be taken, such as inspecting the generator's fuel system, lubrication system, and electrical system, to ensure stable generator operation and the safety of the nuclear power plant.

[0110] This invention also proposes a risk early warning system for emergency diesel generators in nuclear power plants, including a fault testing module, a fault information acquisition module, a fault early warning module, and a computer program that can run on a processor. It should be noted that each module in the above system corresponds to a specific step of the method provided in this invention embodiment, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention embodiment.

[0111] This invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0112] This invention also proposes a computer-readable storage medium storing a computer program. It should be noted that when the computer program is executed by a processor, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0113] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A risk early warning method for emergency diesel generators in nuclear power plants, characterized in that, include: S1. Collect the operating parameters of the generator set in hot standby mode. When the parameter is not within the preset normal range, it indicates that the parameter is abnormal and the generator set has failed. S2. Preprocess the operating parameters collected in step S1 and extract the effective features of the preprocessed operating parameters; Based on the generator set that failed in step S1, and combined with this effective feature, the high-frequency data stream transient anomaly identification algorithm is used to capture the anomaly type label and obtain the fault information. S3. Assess the risk level of the generator set based on the effective characteristics and anomaly type labels of the preprocessed operating parameters; S4. Based on the risk level of the generator set, conduct risk warning and response for the generator set.

2. The risk early warning method for emergency diesel generators in nuclear power plants according to claim 1, characterized in that, In step S1, the operating parameters of the generator set include starting battery voltage, starting time, oil pressure, starting speed, coolant temperature, engine room temperature, fuel flow, vibration signal, electrical parameters, temperature parameters, and insulation resistance. The preset normal range for starting battery voltage is 24V~28V, the preset normal range for starting time is less than 10s, the preset normal range for oil pressure is to rise to greater than or equal to 100kPa within 3s after starting and then stabilize to 200~600kPa, and the preset normal range for starting speed is to reach 1500RPM or 1800RPM within 5s after starting.

3. The risk early warning method for emergency diesel generators in nuclear power plants according to claim 1, characterized in that, In step S2, the window size is set to 200ms and the step size is 50ms. The running parameters are segmented using a sliding window and normalized to obtain the preprocessed running parameters. The effective characteristics of the preprocessed operating parameters include time-domain characteristics, frequency-domain characteristics, and time-frequency-domain characteristics; Among them, the time-domain features include the voltage peak drop rate, current kurtosis, and rotational speed rise slope; the frequency-domain features include the dominant frequency amplitude; and the time-frequency-domain features include wavelet energy entropy. Construct an improved autoencoder model, which includes an input layer, an encoder, a decoder, and an output layer; The effective feature vector of the preprocessed running parameters The input is fed into the improved autoencoder model. After the input layer receives the effective feature vector, it is transmitted to the encoder. The multilayer perceptron is used to compress the effective feature vector to obtain a low-dimensional latent vector. The specific expression is: ； in, Represents the encoder mapping function. The parameters represent the improved autoencoder model. , This represents the Dth effective feature component; Low-dimensional hidden vectors After passing through the decoder, the reconstructed feature vector is obtained using a symmetric multilayer perceptron. The specific expression is: ； in, This represents the decoder mapping function; Calculated using mean square error and The reconstruction error is specifically expressed as: ； in, Indicates reconstruction error, This represents the effective feature vector of the i-th preprocessed running parameter. This represents the reconstructed feature vector of the i-th preprocessed runtime parameters. This indicates the total number of preprocessed runtime parameters; If the reconstruction error exceeds the set threshold, it is marked as a transient anomaly and the anomaly type label is output to obtain fault information.

4. The risk early warning method for emergency diesel generators in nuclear power plants according to claim 3, characterized in that, The voltage peak drop rate is obtained using the adjacent window peak difference method, the current kurtosis value is obtained using the probability statistics method, the speed rise slope is obtained by fitting the speed data within the window using linear regression, and the main frequency amplitude is obtained using the fast Fourier transform.

5. The risk early warning method for emergency diesel generators in nuclear power plants according to claim 1, characterized in that, In step S3, a dynamic risk assessment model is constructed by combining the operational priority requirements of the nuclear power plant's emergency diesel generator set under different emergency conditions. This model is a random forest, which includes N decision trees. The decision trees are constructed through random sampling and random feature selection. The dataset consisting of the effective features of the preprocessed operating parameters and the abnormal type labels is divided into a training set and a test set in a 7:3 ratio. The dynamic risk assessment model is trained using the training set and tested using the test set to obtain the trained dynamic risk assessment model. The effective features of the preprocessed running parameters and the anomaly type labels are input into the trained dynamic risk assessment model. Each decision tree performs a classification operation, and the key anomaly features and their importance scores are obtained through feature space mapping, and then the anomaly prediction results are output respectively. The majority voting method is used to fuse all anomaly prediction results to obtain the final anomaly category prediction and its corresponding anomaly probability value. The anomaly feature importance score is obtained based on the split contribution of effective features in the trained dynamic risk assessment model. The dynamic weights are calculated using the following formula: ； in, This represents the dynamic weight of the i-th preprocessed running parameter. Indicates the dynamic adjustment range; the baseline weight is determined by associating it with historical faults through an expert system; Adjust the weight allocation of the trained dynamic risk assessment model according to the current working conditions; A comprehensive risk score is obtained through weighted summation. The specific formula is as follows: ； in, This represents the real-time monitoring value of the i-th preprocessed running parameter. This represents the baseline value of the i-th parameter under normal operating conditions. The normal baseline values ​​for each parameter can be determined based on historical data and expert experience. This indicates the total number of preprocessed running parameters.

6. The risk early warning method for emergency diesel generators in nuclear power plants according to claim 5, characterized in that, In step S4, a visual interface is used to display risk warnings and responses. The visual interface includes real-time data charts, risk types and levels, and warning information. When the comprehensive risk score When the risk level is low, it is displayed in green, and the operating parameter curves and risk trends are updated in real time on the visualization interface, requiring no intervention; when the comprehensive risk score... When the risk level is medium, it will be displayed in yellow, and a warning prompt box will pop up in the visualization interface to review the operating parameters; when the comprehensive risk score is... When the risk is high, it will be displayed in red, triggering an audible and visual alarm on the visual interface and shutting down the corresponding device.

7. A system applied to the risk early warning method for nuclear power plant emergency diesel generators as described in claim 1, characterized in that, include: The fault test module is used to collect the operating parameters of the generator set in hot standby mode. When the parameter is not within the preset normal range, it indicates that the parameter is abnormal and the generator set has failed. The fault information acquisition module is used to preprocess the operating parameters collected in the fault test module and extract the effective features of the preprocessed operating parameters. Based on the generator set that has failed in the fault test module, and combined with this effective feature, the high-frequency data stream transient anomaly identification algorithm is used to capture the anomaly type label and obtain the fault information. The fault early warning module is used to assess the risk level of the generator set based on the effective characteristics and anomaly type labels of the preprocessed operating parameters; Based on the risk level of the generator set, risk warnings and responses are carried out for the generator set.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the risk warning method for emergency diesel generators in nuclear power plants according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, The computer program, when executed by the processor, performs the risk warning method for nuclear power plant emergency diesel generators as described in any one of claims 1 to 6.

Images