Hydropower station flow jump alarm threshold prediction method and system based on deep learning self-adaption

Through deep learning and adaptive algorithms, the alarm threshold for hydropower station flow monitoring is dynamically adjusted, which solves the false alarm and omission problems caused by fixed thresholds in traditional systems, improves the accuracy and adaptability of flow monitoring, and improves the operation safety of hydropower stations .

CN119005753APending Publication Date: 2024-11-22THREE GORGES JINSHAJIANG CHUANYUN HYDROPOWER DEV CO LTD
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202411097216.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Traditional hydropower station flow monitoring systems rely on fixed thresholds and are difficult to adapt to dynamic environmental changes, resulting in false alarms and missed alarms, reducing the reliability and safety of the system.

Method used

Adaptive method based on deep learning is adopted to analyze the time series characteristics of fluid data through recurrent neural networks, dynamically adjust the alarm threshold, and combine Kalman filter and exponential smoothing technology to adjust the threshold in real time to cope with environmental changes.

Benefits of technology

It significantly improves the accuracy and adaptability of flow monitoring, reduces false alarms and missed alarm rates, ensures that the alarm system can respond to flow changes in a timely manner, and improves the operating stability and safety of hydropower stations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005753A_ABST
    Figure CN119005753A_ABST
Patent Text Reader

Abstract

The invention provides a deep learning self-adaption-based hydropower station flow jump alarm threshold prediction method and system, relates to the field of hydropower station operation monitoring and early warning, and solves the problems of insufficient accuracy and poor adaptability caused by a fixed threshold in a traditional flow monitoring alarm system. The method comprises the following steps: acquiring fluid monitoring data and environment monitoring data, and preprocessing to obtain a data set; a recurrent neural network (RNN) is constructed and trained, and a deep learning model adapted to fluid environment change for flow prediction is obtained; deploying the deep learning model in a traffic monitoring environment, collecting an actual traffic monitoring value, and obtaining a traffic prediction value after a preset time; based on the moment of the flow prediction value, collecting a flow monitoring value corresponding to the moment, calculating an error between the flow monitoring value and the flow prediction value, and taking the error as an adjustment basis of a flow jump alarm threshold value; according to the invention, changes of various fluid environments can be flexibly handled, and occurrence of false alarm and missing alarm can be effectively reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of hydropower station operation monitoring and early warning, and is applied to the flow monitoring process, and specifically to a method and system for predicting a hydropower station flow jump alarm threshold based on deep learning adaptation. Background Art

[0002] In the daily operation of a hydropower station, the technical water supply system plays a vital role. It is a key auxiliary equipment to ensure the efficient and stable operation of the turbine generator set. The core task of the system is to provide necessary cooling and lubrication for the thrust bearings, guide bearings, air coolers, turbine guide bearings, main transformers and other key components of the generator, while maintaining the main shaft seal, effectively preventing overheating caused by continuous operation of the equipment, and avoiding the resulting safety hazards and shortened equipment life. Therefore, precise control and real-time monitoring of water pressure, flow, temperature and other parameters are of immeasurable value in ensuring the safe and economical operation of the hydropower station.

[0003] At present, large hydropower stations generally use electromagnetic flowmeters, thermal conductivity flowmeters, ultrasonic flowmeters and other devices based on a single measurement principle to monitor liquid flow. Although these flowmeters meet basic monitoring needs to a certain extent, their inherent limitations cannot be ignored. First, these devices mostly rely on preset fixed thresholds to trigger the alarm mechanism. This static threshold setting is difficult to cope with the dynamic changes in the production site environment, especially in the face of complex working conditions, such as pipeline vibration, changes in medium characteristics, etc., which may cause abnormal fluctuations in flow readings, thereby causing false alarms or missed alarms. Secondly, frequent false alarms not only consume a lot of human resources for on-site inspections and troubleshooting, but also reduce the overall reliability of the system in the long run, posing a potential threat to industrial safety production.

[0004] In view of the above problems, the existing technology has obvious shortcomings in achieving accurate flow monitoring, adapting to dynamic environmental changes, and reducing human intervention. There is an urgent need for a more intelligent, flexible and high-precision flow monitoring solution to improve the stability and safety of hydropower station operation and realize intelligent operation and maintenance. Summary of the invention

[0005] Based on the current status of the background technology, the purpose of the present invention is to solve the problems of insufficient accuracy and poor adaptability caused by fixed thresholds in traditional flow monitoring alarm systems, and therefore proposes a method and system for predicting flow jump alarm thresholds of hydropower stations based on deep learning adaptation. The present invention analyzes the time series characteristics of fluid data in real time and dynamically adjusts the alarm threshold to more accurately predict flow jump events; the dynamic adjustment mechanism makes the alarm system no longer limited by the limitations of fixed thresholds, and can flexibly respond to changes in various fluid environments, effectively reducing the occurrence of false alarms and missed alarms.

[0006] The present invention adopts the following technical solutions to achieve the purpose:

[0007] A method for predicting flow jump alarm threshold of a hydropower station based on deep learning adaptation, the method comprising the following steps:

[0008] S1. Obtain the fluid monitoring data and environmental monitoring data of the hydropower station and perform data preprocessing to obtain a data set for model training;

[0009] S2. Construct a recurrent neural network RNN, use the data set to train the recurrent neural network RNN, and obtain a deep learning model that adapts to changes in the fluid environment and performs flow prediction after the training is completed;

[0010] S3. Deploy the deep learning model in the flow monitoring environment of the hydropower station, collect the actual flow monitoring value of the hydropower station, input it into the deep learning model, and obtain the flow prediction value after a preset time;

[0011] S4. Based on the moment of the flow prediction value, collect the flow monitoring value corresponding to the moment, calculate the error between the flow monitoring value and the corresponding flow prediction value, use the error as the basis for adjusting the flow jump alarm threshold, and complete the prediction of the flow jump alarm threshold.

[0012] Furthermore, in step S4, y, which was originally used as the input value of the deep learning model, is converted into i Recorded as flow monitoring value y i ; The traffic prediction value of the deep learning model is calculated using the statistical measurement methods of root mean square error RMSE and mean absolute error MAE The new flow monitoring value y monitored in real time at the corresponding moment i The calculation formula of the root mean square error RMSE is as follows:

[0013]

[0014] The calculation formula of mean absolute error MAE is as follows:

[0015]

[0016] According to the error calculation results, the flow jump alarm threshold is adjusted; the adaptive filtering algorithm of the Kalman filter is used in the adjustment process, and the state update formula is as follows:

[0017]

[0018] In the formula, is the optimal state estimate at time k, y k is the observed value, K kis the Kalman gain, which is calculated as follows:

[0019] K k =P k|k-1 H T (HP k|k-1 H T +R) -1

[0020] Where P k|k-1 is the prediction error covariance, H is the observation model matrix, and R is the observation noise covariance; in the adjustment process, according to the real-time monitoring flow monitoring value y i The exponential smoothing technology is applied to make the flow jump alarm threshold respond to the real-time changes of the fluid environment.

[0021] In summary, due to the adoption of this technical solution, the beneficial effects of the present invention are as follows:

[0022] Traditional flow alarm systems usually use static thresholds, which cannot adapt to the dynamic changes of the fluid environment, resulting in reduced alarm accuracy in different seasons and construction environments. The present invention introduces a deep learning model, especially a recurrent neural network (RNN), so that the system can learn and predict the temporal pattern of fluid changes, thereby dynamically adjusting the alarm threshold, significantly improving the adaptability and accuracy of the system.

[0023] Fixed threshold methods are prone to false positives and false negatives when faced with complex and changeable fluid data. This invention realizes accurate prediction of flow rate jumps by real-time analysis and processing of historical data and real-time monitoring data, combined with the prediction results of deep learning models, effectively reducing false positives and false negatives.

[0024] The appropriate alarm threshold varies in different seasons and environmental conditions. The threshold dynamic adjustment strategy proposed in the present invention can automatically adjust the alarm threshold according to real-time monitoring data and prediction errors, so that it is always kept at an appropriate level, ensuring that the alarm system can respond to actual flow changes in a timely manner.

[0025] As time goes by, each device will age to varying degrees, and new compliant fluid environments will emerge, requiring the alarm system to continuously learn and adapt to new conditions. The present invention uses self-learning and maintenance methods to regularly retrain the deep learning model with new monitoring data, and adjusts model parameters and threshold adjustment strategies based on system operation conditions, ensuring continuous optimization and long-term effectiveness of the system. The optimized alarm notification mechanism enables relevant personnel to more intuitively and quickly obtain accurate alarm information and possible causes, so as to take effective countermeasures. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1The figure is a schematic diagram briefly describing the overall process of the method of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0028] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0029] Example

[0030] A method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation. The process of this method can be briefly described in Figure 1 The steps in the method are summarized as follows:

[0031] S1. Obtain the fluid monitoring data and environmental monitoring data of the hydropower station and perform data preprocessing to obtain a data set for model training;

[0032] S2. Construct a recurrent neural network RNN, use the data set to train the recurrent neural network RNN, and obtain a deep learning model that adapts to changes in the fluid environment and performs flow prediction after the training is completed;

[0033] S3. Deploy the deep learning model in the flow monitoring environment of the hydropower station, collect the actual flow monitoring value of the hydropower station, input it into the deep learning model, and obtain the flow prediction value after a preset time;

[0034] S4. Based on the moment of the flow prediction value, collect the flow monitoring value corresponding to the moment, calculate the error between the flow monitoring value and the corresponding flow prediction value, use the error as the basis for adjusting the flow jump alarm threshold, and complete the prediction of the flow jump alarm threshold.

[0035] This embodiment will introduce the details of each step in detail according to the above step sequence.

[0036] In step S1, historical three-year and current fluid monitoring data and environmental monitoring data are collected from multiple monitoring points at a collection frequency at preset time intervals, ensuring that the data covers different seasons and field environments; fluid monitoring data includes flow, density and conductivity, and environmental monitoring data includes temperature, humidity, pipeline pressure and pipeline vibration frequency.

[0037] At the same time, various types of fluid monitoring data and environmental monitoring data are pre-defined, and the minimum collection frequency is determined to be once a minute; the collected fluid monitoring data and environmental monitoring data are transmitted to the central database of the hydropower station for storage after data verification and transmission anomaly detection, and then undergo a data preprocessing process.

[0038] The specific details of the data collection part in step S1 are as follows:

[0039] Equipment deployment and configuration: Deploy high-precision monitoring equipment at monitoring points, including flow meters, density meters, thermometers, conductivity sensors, humidity sensors, vibration sensors, pressure sensors, etc.; Data type and collection frequency: Define the required data type and set data collection at least once a minute; Real-time data transmission and storage: Realize real-time automatic transmission of monitoring data to the central database and ensure the efficiency and security of the storage system; Data integrity and exception handling: Ensure data integrity and accuracy through data verification and transmission anomaly detection mechanisms; Regular equipment maintenance: Develop a maintenance plan to calibrate and maintain monitoring equipment regularly; Data security and multi-source integration: Take necessary measures to protect data security and integrate multi-source data when possible.

[0040] Then, data preprocessing operations are performed. First, the collected data is cleaned, duplicate collected data points are removed, and collected data points with obvious errors or obvious deviations are corrected; then missing value processing is performed to identify the missing parts in the data, and fill them with forward filling, backward filling, interpolation or mean replacement according to the characteristics of the data; then outliers in the data are identified and eliminated, and the range of outliers is determined by calculating the mean and standard deviation in the statistical characteristics of the data, and then they are eliminated from the data to avoid adverse effects on subsequent model training.

[0041] After removing the outliers, this embodiment uses the minimum-maximum normalization method to normalize the data and scale the data to a numerical range of 0 to 1. The formula for normalization is as follows:

[0042]

[0043] In the formula, Value represents the original data point, Min and Max represent the minimum and maximum values ​​of the data in the processed data set, respectively.

[0044] After normalization, feature engineering is used to identify and extract influencing features related to traffic change prediction, such as traffic change rate, traffic change amount, etc.; then, additional derived features are obtained from the extracted influencing features through mathematical transformation and statistical analysis; based on the influencing features and derived features, correlation analysis, principal component analysis (PCA) or recursive feature elimination (RFE) technology is used to reduce feature dimensions, complete feature selection and classification, and form classified data; based on the input requirements of the recurrent neural network RNN, One-Hot Encoding is used to process the classified data, and the data set is obtained after data encoding is completed.

[0045] The data set is divided into a training set, a validation set, and a test set according to a preset ratio (such as 3:1:1) to ensure that the model can perform well on unseen data. The SMOTE algorithm is used to deal with the imbalance in the data set during division to improve the prediction accuracy of the model for minority classes. At the same time, according to the preset data update interval, new fluid monitoring data and environmental monitoring data are continuously introduced into the data set, and after each data set update, the newly obtained data set is used to perform adjustment training on the deep learning model. Through the above-mentioned orderly and specific data processing steps, this embodiment can provide high-quality and highly relevant training data for the training of deep learning models.

[0046] In step S2 of this embodiment, each layer of the constructed recurrent neural network RNN ​​is composed of a preset number of neurons, which is used to capture the complex dependencies of time series data. The recurrent neural network RNN ​​of this embodiment includes an input layer, two hidden layers and an output layer, and the number of neurons in each hidden layer is 128.

[0047] The time series data is obtained by the preparation and processing of the data set, converting the time point data in the data set into a sequence form, and defining the time interval Δt and the sequence length L; after setting the hyperparameters of the recurrent neural network RNN, the activation function ReLU of the recurrent neural network RNN ​​is as follows:

[0048] ReLU: f(x) = max(0, t)

[0049] In the formula, t is the current iteration number. After the activation function introduces nonlinearity into the recurrent neural network RNN, the mean square error MSE is applied in the corresponding loss function J, as shown in the following formula:

[0050]

[0051] In the formula, the mean square error MSE is used to quantify the predicted value With the input value y i The difference between ; m is the batch size.

[0052] In addition, based on the activation function and loss function of the recurrent neural network RNN, the Adam optimizer is used as the optimization algorithm during the model training process; the update rule of the Adam optimizer adopts the following formula:

[0053]

[0054] Where η is the learning rate, b is the decay rate, ⊙ is the Hadamard product, is the descending gradient of the model parameter θ; by adjusting the model parameter θ to minimize the loss function J and adding the L2 regularization term, the loss function J is obtained as follows:

[0055]

[0056] Where λ is the regularization coefficient, W j is the model weight;

[0057] During the model training process, K-fold cross-validation and network search technology are used to optimize the learning rate η, batch size m and total number of iterations T, and the prediction performance of the deep learning model obtained is evaluated using the validation set and test set in the data set; after the deep learning model is deployed in step S3, the deep learning model is adjusted and trained using the updated data set according to the preset data update interval.

[0058] In the above specific training, the learning rate η is set to 0.001, and the total number of iterations T is set to 5000. Under the action of the Adam optimizer, the convergence process is accelerated and the model performance is improved.

[0059] In step S4 of this embodiment, y, which is originally used as the input value of the deep learning model, is converted into i Recorded as flow monitoring value y i ; The traffic prediction value of the deep learning model is calculated using the statistical measurement methods of root mean square error RMSE and mean absolute error MAE The new flow monitoring value y monitored in real time at the corresponding moment i The calculation formula of the root mean square error RMSE is as follows:

[0060]

[0061] The calculation formula of mean absolute error MAE is as follows:

[0062]

[0063] According to the error calculation results, the flow jump alarm threshold is adjusted; the adaptive filtering algorithm of the Kalman filter is used in the adjustment process, and the state update formula is as follows:

[0064]

[0065] In the formula, is the optimal state estimate at time k, y k is the observed value, K k is the Kalman gain, which is calculated as follows:

[0066] K k =P k|k-1 H T (HP k|k-1 H T +R) -1

[0067] Where P k|k-1 is the prediction error covariance, H is the observation model matrix, and R is the observation noise covariance; in the adjustment process, according to the real-time monitoring flow monitoring value y i The exponential smoothing technology is applied to make the flow jump alarm threshold respond to the real-time changes in the fluid environment. When the exponential smoothing technology is dynamically adjusted according to the real-time data change trend, the adjustment coefficient is set to 0.5 to ensure that the threshold responds smoothly to flow changes.

[0068] In this embodiment, the flow jump alarm threshold itself is introduced as follows, involving some basic threshold setting logic and alarm-related strategies: the initial flow jump alarm threshold can be the 90% quantile of the flow jump in the historical data to reduce false alarms while ensuring sufficient warning time. When the real-time model predicts that the flow change exceeds the current threshold within the next two hours, it can be judged as a potential jump event, triggering the potential alarm-related logic.

[0069] Finally, after dynamically adjusting and predicting the flow jump alarm threshold, the alarm trigger logic is set to monitor the flow monitoring value of the hydropower station in real time; when the degree of change of the flow monitoring value exceeds the dynamically updated flow jump alarm threshold within the preset time, it is determined that the flow of the hydropower station is at risk or in an abnormal situation, the alarm action is executed, and an alarm is sent to the central monitoring system of the hydropower station.

[0070] Specifically, by receiving new traffic monitoring values ​​in real time, the prediction results of the deep learning model are updated every half hour to generate traffic forecast values ​​for the next two hours, and the traffic jump alarm threshold is dynamically adjusted and updated. Based on the updated traffic jump alarm threshold, after comparing the traffic monitoring values, an alarm is triggered if the threshold is exceeded.

[0071] In addition, the deep learning model is retrained quarterly using the latest monitoring data to adapt to new fluid conditions and potential environmental changes. Model parameters and basic alarm thresholds are adjusted every six months to optimize system performance based on system operation and user feedback. Data acquisition equipment is regularly checked and updated to ensure the timeliness and accuracy of data to support the effective operation of the model.

[0072] The above method of this embodiment can be applied to a computer device system, which includes a memory, a processor and a computer program stored in the memory. The processor executes the computer program to implement the steps of the hydropower station flow jump alarm threshold prediction method based on deep learning adaptation.

[0073] Furthermore, these computer programs / instructions may be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more steps in the method. Corresponding to the content of the method of this embodiment, the system can be divided into multiple specific functional modules, including a data acquisition module, a data processing module, a deep learning model module, a threshold dynamic adjustment module, and an alarm module, so as to orderly implement the corresponding functions and execute the method flow.

Claims

1. A method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation, characterized in that: The method comprises the following steps: S1. Obtain the fluid monitoring data and environmental monitoring data of the hydropower station and perform data preprocessing to obtain a data set for model training; S2. Construct a recurrent neural network RNN, use the data set to train the recurrent neural network RNN, and obtain a deep learning model that adapts to changes in the fluid environment and performs flow prediction after the training is completed; S3. Deploy the deep learning model in the flow monitoring environment of the hydropower station, collect the actual flow monitoring value of the hydropower station, input it into the deep learning model, and obtain the flow prediction value after a preset time; S4. Based on the moment of the flow prediction value, collect the flow monitoring value corresponding to the moment, calculate the error between the flow monitoring value and the corresponding flow prediction value, use the error as the basis for adjusting the flow jump alarm threshold, and complete the prediction of the flow jump alarm threshold.

2. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 1 is characterized in that: In step S1, historical and current fluid monitoring data and environmental monitoring data are collected from multiple monitoring points at a collection frequency at preset time intervals; the fluid monitoring data includes flow, density and conductivity, and the environmental monitoring data includes temperature, humidity, pipeline pressure and pipeline vibration frequency.

3. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 2 is characterized in that: Various types of fluid monitoring data and environmental monitoring data are predefined, and the minimum collection frequency is determined to be once a minute; the collected fluid monitoring data and environmental monitoring data are transmitted to the central database of the hydropower station for storage after data verification and transmission anomaly detection, followed by data preprocessing.

4. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 3 is characterized in that: The data preprocessing process is as follows: first, clean the collected data, remove duplicate collected data points, and correct the collected data points with obvious errors or obvious deviations; then perform missing value processing, identify the missing parts in the data, and fill them in by forward filling, backward filling, interpolation or mean replacement according to the data characteristics; then identify and eliminate outliers in the data, determine the range of outliers by calculating the mean and standard deviation in the statistical characteristics of the data, and then eliminate them from the data.

5. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 4 is characterized in that: After removing outliers, the data were normalized using the minimum-maximum normalization method to scale the data to a value range of 0 to 1; After normalization, feature engineering is used to identify and extract influencing features related to traffic change prediction, and then additional derived features are obtained from the extracted influencing features through mathematical transformation and statistical analysis. Based on the influencing features and derived features, correlation analysis, principal component analysis or recursive feature elimination technology is used to reduce feature dimensions, complete feature selection and classification, and form classified data. Based on the input requirements of the recurrent neural network RNN, unique hot encoding is used to process the classified data, and the data set is obtained after data encoding is completed. The dataset is divided into training set, validation set and test set according to the preset ratio. The SMOTE algorithm is used to deal with the imbalance in the dataset during the division. New fluid monitoring data and environmental monitoring data are continuously introduced into the dataset according to the preset data update interval. After each dataset update, the newly obtained dataset is used to perform adjustment training on the deep learning model.

6. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 1 is characterized by: In step S2, each layer of the constructed recurrent neural network RNN ​​consists of a preset number of neurons to capture the complex dependencies of time series data; The time series data is obtained by the preparation and processing of the data set. The time point data in the data set is converted into a sequence form, and the time interval Δt and the sequence length L are defined. After setting the hyperparameters of the recurrent neural network RNN, the activation function ReLU of the recurrent neural network RNN ​​is as follows: ReLU: f(x) = max(0, t) In the formula, t is the current iteration number. After the activation function introduces nonlinearity into the recurrent neural network RNN, the mean square error MSE is applied in the corresponding loss function J, as shown in the following formula: In the formula, the mean square error MSE is used to quantify the predicted value With the input value y i The difference between ; m is the batch size.

7. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 6 is characterized in that: Based on the activation function and loss function of the recurrent neural network RNN, the Adam optimizer is used as the optimization algorithm during the model training process; the update rule of the Adam optimizer adopts the following formula: Where η is the learning rate, b is the decay rate, ⊙ is the Hadamard product, is the descending gradient of the model parameter θ; by adjusting the model parameter θ to minimize the loss function J and adding the L2 regularization term, the loss function J is obtained as follows: Where λ is the regularization coefficient, W j is the model weight; During the model training process, K-fold cross-validation and network search technology are used to optimize the learning rate η, batch size m and total number of iterations T, and the prediction performance of the deep learning model obtained is evaluated using the validation set and test set in the data set; after the deep learning model is deployed in step S3, the deep learning model is adjusted and trained using the updated data set according to the preset data update interval.

8. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 1 is characterized by: In step S4, y, which is originally used as the input value of the deep learning model, is converted into i Recorded as flow monitoring value y i ; The traffic prediction value of the deep learning model is calculated using the statistical measurement methods of root mean square error RMSE and mean absolute error MAE The new flow monitoring value y monitored in real time at the corresponding moment i The calculation formula of the root mean square error RMSE is as follows: The calculation formula of mean absolute error MAE is as follows: According to the error calculation results, the flow jump alarm threshold is adjusted; the adaptive filtering algorithm of the Kalman filter is used in the adjustment process, and the state update formula is as follows: In the formula, is the optimal state estimate at time k, y k is the observed value, K k is the Kalman gain, which is calculated as follows: K k =P k|k-1 H T (HP k|k-1 H T +R) -1 Where P k|k-1 is the prediction error covariance, H is the observation model matrix, and R is the observation noise covariance; in the adjustment process, according to the real-time monitoring flow monitoring value y i The exponential smoothing technology is applied to make the flow jump alarm threshold respond to the real-time changes of the fluid environment.

9. The method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation according to claim 1 is characterized by: After dynamically adjusting and predicting the flow jump alarm threshold, set the alarm trigger logic to monitor the flow monitoring value of the hydropower station in real time; when the degree of change of the flow monitoring value exceeds the dynamically updated flow jump alarm threshold within the preset time, it is determined that the flow of the hydropower station is at risk or in an abnormal situation, and the alarm action is executed to send an alarm to the central monitoring system of the hydropower station.

10. A computer device system, comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method for predicting the flow jump alarm threshold of a hydropower station based on deep learning adaptation as described in any one of claims 1 to 9.

Citation Information

Cited By

  • Flow monitoring method and device based on artificial intelligence algorithm

    CN119966846A

  • Temperature threshold prediction method based on improved Transform

    CN120101966A

  • Multi-dimensional real-time operation monitoring and alarm linkage system for urban direct drinking water plant station

    CN120257174A

  • Early warning method for flooding factory building of hydropower station based on multi-source monitoring signal and dynamic threshold value

    CN121256689A

  • Intelligent monitoring method and system for pumped storage power station pipe gallery drainage and storage medium

    CN121409322A