An intelligent monitoring method and system for drainage of a pumped storage power station pipe gallery and a storage medium

Abstract

The application discloses an intelligent monitoring method and system for drainage of a pumped storage power station pipe gallery and a storage medium, and comprises the following steps: S1: collecting parameter information of a drainage system of a pumped storage power station under normal operation, wherein the parameter information comprises alarm threshold information and a plurality of corresponding real-time variable sets, each real-time variable set has a corresponding variable type and a time stamp, and the variable type comprises pipeline temperature, pipeline pressure, pipe gallery humidity and drainage temperature; S2: constructing a network model, using the alarm threshold information and the plurality of real-time variable sets under normal operation as a training set, and establishing a plurality of real-time variable prediction models; and S3: when an abnormal alarm is triggered, the real-time variable prediction model is used to predict real-time variables of the drainage system, the predicted real-time variables are compared with corresponding alarm threshold values, and the cause of drainage abnormality is determined; and the application solves the problems of weak timing correlation capture, strong artificial dependence and low efficiency of abnormal cause determination in traditional monitoring.

Description

Technical Field

[0001] This invention relates to the field of pumped storage power station technology, and in particular to an intelligent monitoring method, system and storage medium for drainage of the pipe gallery in a pumped storage power station. Background Technology

[0002] With the advancement of new power system construction, pumped storage power stations, as key peak-shaving energy storage facilities, are crucial to the reliability of power grid supply due to their safe and stable operation. The drainage system of the pumped storage power station's pipe gallery is a critical auxiliary system ensuring the normal operation of the core equipment of the power station, including water transmission pipelines and drainage pumps. Its core functions include draining accumulated water in the pipe gallery, regulating the humidity of the pipe gallery, and monitoring the operating status of the pipelines. If the drainage system malfunctions, such as pipe blockage leading to a sudden increase in pressure, excessive humidity in the pipe gallery causing equipment corrosion, or pump failure leading to abnormal drainage temperature, it will directly affect the water transmission efficiency of the power station and may even cause serious failures such as pipe rupture and equipment short circuits, threatening the overall operational safety of the power station.

[0003] Currently, the monitoring methods for drainage systems in pumped storage power station tunnels still have significant technical limitations. One is that traditional monitoring relies on a combination of manual inspection and fixed threshold alarms. Manual inspection requires maintenance personnel to enter the closed, narrow tunnel environment where there may be toxic and harmful gases, which is not only inefficient and unsafe, but also prone to missing intermittent anomalies due to human observation errors.

[0004] In summary, existing pumped storage power station drainage monitoring technology suffers from insufficient real-time performance, failing to meet the requirements for safe and efficient operation and maintenance of power stations. There is an urgent need for an intelligent monitoring system that can accurately capture temporal characteristics, dynamically optimize models and thresholds, and achieve accurate anomaly diagnosis. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned shortcomings and provide an intelligent monitoring method, system and storage medium for drainage in pumped storage power station tunnels. This invention addresses the problems of weak time-series correlation capture, strong reliance on manual intervention and inefficient anomaly identification in traditional monitoring methods. It achieves the effects of automatically collecting parameters across all dimensions, accurately predicting real-time variables, and quickly locating the causes of drainage anomalies, thereby ensuring the stable operation of the drainage system in pumped storage power station tunnels and reducing operation and maintenance safety risks and fault handling cycles.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an intelligent monitoring method for drainage of pumped storage power station pipe corridors, comprising the following steps:

[0007] S1: Collect parameter information of the pumped storage power station drainage system under normal operation. The parameter information includes alarm threshold information and multiple real-time variable sets. Each real-time variable set has a corresponding variable type and timestamp. The variable types include pipeline temperature, pipeline pressure, pipe gallery humidity and drainage temperature.

[0008] S2: Construct a network model, using the alarm threshold information and multiple real-time variable sets under normal operation as the training set, and establish multiple real-time variable prediction models;

[0009] S3: When an abnormal alarm is triggered, the real-time variables of the drainage system are predicted by the real-time variable prediction model, and the predicted real-time variables are compared with the corresponding alarm thresholds to determine the cause of the drainage abnormality.

[0010] Preferably, in step S2, when establishing multiple real-time variable prediction models, cross-validation is used to optimize the parameters of the network model to obtain the optimal parameters as the final parameters, specifically including:

[0011] Multiple real-time variable sets are divided into training and validation sets according to a preset ratio; each real-time variable set is further divided into variable subsets of the same length, each subset containing multiple variables, and each subset forming multiple input feature data; multiple variables in each set of input feature data are sequentially input into the network model, outputting the corresponding predicted variables, and simultaneously merging the variables corresponding to multiple sets of input feature data and the output predicted variables to form output feature data;

[0012] The output feature data corresponding to a set of input feature data and the validation set of the same group are input into the cross-validation method. The loss function of the validation set is used as the loss function of the network model. The parameters in the network model are updated according to the backpropagation error based on the loss function. The network model is then optimized to form an optimized network model.

[0013] Repeat the above steps to train and optimize the network model in sequence. Use cross-validation to validate the model. Calculate the square value based on the loss function of each cross-validation and select the model with the highest square value as the final network model.

[0014] Preferably, step S3 specifically includes the following steps:

[0015] After predicting a real-time variable using a network model, the continuous-time threshold and threshold weight of that real-time variable are determined. The predicted variables output by the network model are standardized to obtain standard variables. If all continuous real-time variables of the same real-time variable prediction model exceed the continuous-time threshold, an alarm is triggered. Otherwise, the threshold weight and threshold ratio are compared. If the threshold ratio is greater than the threshold weight, it is judged as an anomaly. After the alarm is triggered, the cause of the alarm is obtained based on the anomaly judgment result.

[0016] Preferably, the method for determining the continuous time threshold is as follows:

[0017] Perform a normality test on the real-time variables in the real-time variable set, and calculate the mean and standard deviation when the real-time variables in multiple real-time variable sets conform to a normal distribution;

[0018] Based on the mean and standard deviation, calculate the time threshold for real-time variables in each set of real-time variables to exceed three times the standard deviation, and use this time threshold as the continuous time threshold.

[0019] The threshold ratio of real-time variables is calculated based on the mean and standard deviation of real-time variables in multiple real-time variable sets.

[0020] Preferably, in step S1, the method for collecting the alarm threshold information includes:

[0021] Set an initial alarm threshold, calculate a safe threshold range based on the initial alarm threshold, divide the safe threshold range into multiple threshold intervals, and select a set of normal variables from each threshold interval;

[0022] The initial alarm threshold is matched with the variables in the normal variable set. When none of the variables in the normal variable set meet the initial alarm threshold, the alarm threshold is adjusted by increasing the threshold step size to obtain a new alarm threshold. The adjustment stops when at least one variable in the normal variable set meets the initial alarm threshold.

[0023] The present invention also discloses an intelligent monitoring system for drainage of pipe corridors in pumped storage power stations, which is used to implement the above-mentioned intelligent monitoring method, including a data acquisition module and a model building module connected to it in communication, as well as a result output module;

[0024] The data acquisition module is used to collect and transmit parameter information during normal operation of the drainage system. The parameter information includes alarm threshold information and multiple real-time variable sets.

[0025] The model building module is used to receive parameter information, build a real-time variable prediction model based on a long short-term memory network, and transmit the model to the result output module.

[0026] The result output module is used to receive the prediction model. When a drainage anomaly alarm is triggered, it predicts real-time variables through the model and compares them with the alarm threshold to determine the cause of the anomaly.

[0027] Furthermore, the model building module includes a network model unit and a cross-validation unit; the network model unit is used to process input feature data and output predicted variables; the cross-validation unit is used to optimize model parameters, and by comparing the loss function of the validation set with that of the model output, the model parameters are updated in reverse, and finally the model with the highest variance is selected as the output.

[0028] Furthermore, the result output module includes an anomaly judgment unit, which is used to determine the continuous time threshold and threshold weight of the real-time variable, as well as to standardize the predicted variable, and to trigger an alarm based on the comparison result of the continuous time threshold or threshold ratio and weight, and output the reason for the anomaly.

[0029] Furthermore, the data acquisition module includes an alarm threshold submodule, which is used to set an initial alarm threshold and divide the safe threshold range, and finally dynamically adjust the threshold through the threshold step size; it also includes a normal variable submodule, which is used to filter the normal variable set and assist in threshold calibration.

[0030] Furthermore, the normal variable submodule includes a safety threshold calculation unit, which is used to calculate the safety range of three types of alarm thresholds and corresponding datasets for pipelines, pumps, and pipe racks; it performs threshold verification and dynamic adjustment on the data of pipelines, pumps, and pipe racks respectively, and outputs information on the over-limit location, detection map, and duration.

[0031] The present invention also discloses a storage medium, which, when read by a computer program, executes the above-described intelligent monitoring method.

[0032] The beneficial effects of this invention are as follows: This invention solves the problems of weak capture of time-series correlation, strong reliance on manual intervention, and inefficient anomaly identification in traditional monitoring by setting up a data acquisition module to collect normal operation parameters, a model building module to build a prediction model with a built-in long short-term memory network model, and a result output module to realize the prediction, comparison, and cause identification in case of anomalies. It achieves the effects of automatically collecting parameters in all dimensions, accurately predicting real-time variables, and quickly locating the causes of drainage anomalies, ensuring the stable operation of the pumped storage power station pipe gallery drainage system, and reducing operation and maintenance safety risks and fault handling cycles. Attached Figure Description

[0033] Figure 1 This is a flowchart of an intelligent monitoring method for drainage in the pipe gallery of a pumped storage power station;

[0034] Figure 2 This is a block diagram of an intelligent monitoring system for drainage in pumped storage power station pipe corridors. Detailed Implementation

[0035] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0036] Example 1: Please refer to Figure 1 This embodiment provides an intelligent monitoring method for drainage in pumped storage power station pipe corridors, the method comprising the following steps:

[0037] S1: Collect parameter information of the pumped storage power station drainage system under normal operation. The parameter information includes alarm threshold information and multiple real-time variable sets. Each real-time variable set has a corresponding variable type and timestamp. The variable types include pipeline temperature, pipeline pressure, pipe gallery humidity and drainage temperature.

[0038] More specifically, in step S1, the method for collecting alarm threshold information includes: setting an initial alarm threshold; calculating a safe threshold range based on the initial alarm threshold; dividing the safe threshold range into multiple threshold intervals; selecting a normal variable set from each threshold interval; matching the initial alarm threshold with the variables in the normal variable set; when none of the variables in the normal variable set meet the initial alarm threshold, adjusting the alarm threshold by increasing the threshold step size to obtain a new alarm threshold; stopping the adjustment when at least one variable in the normal variable set meets the initial alarm threshold.

[0039] In the actual design process, a monitoring point is set up every 50m along the drainage pipeline of the utility tunnel. Each monitoring point is equipped with four types of sensors (pipeline temperature, pipeline pressure, utility tunnel humidity, and drainage temperature). The sensors are connected to the data acquisition terminal via waterproof cables. The data acquisition terminal is fixed to the side wall of the utility tunnel and connected to a DC power supply to ensure communication with the power station server.

[0040] The initial alarm thresholds for four types of variables are manually entered via the terminal touchscreen, and the initial threshold file is saved.

[0041] The system automatically retrieves normal operating data over a recent period. If no variable triggers the initial threshold for 7 consecutive days, it calibrates daily until a variable triggers the threshold, at which point it stops, generates a calibrated threshold file, and overwrites the initial file.

[0042] Sensor data is acquired at the data acquisition terminal. After each acquisition, a timestamp and variable type label are automatically added to the data, generating structured data containing the monitoring point ID, timestamp +, and four types of variable values. The data is uploaded to the server in real time via Ethernet, where it is received and stored in a database, while a backup is maintained on the local terminal.

[0043] S2: Construct a network model, using alarm threshold information and multiple real-time variable sets under normal operation as training sets, and establish multiple real-time variable prediction models.

[0044] Step S2 further includes the following steps: When establishing multiple real-time variable prediction models, cross-validation is used to optimize the network model parameters to obtain the optimal parameters as the final parameters. Specifically, this includes: dividing multiple real-time variable sets into training and validation sets according to a preset ratio; splitting multiple real-time variable sets into variable subsets of the same length, each set containing multiple variables, and each set of variable subsets forming multiple input feature data; sequentially inputting multiple variables from each set of input feature data into the network model, outputting the corresponding predicted variables, and simultaneously merging the variables corresponding to multiple sets of input feature data and the output predicted variables to form output feature data; inputting the output feature data corresponding to a set of input feature data and the validation set of the same group into the cross-validation method, using the loss function of the validation set as the loss function of the network model, updating the parameters in the network model based on the backpropagation error of the loss function, optimizing the network model parameters, and forming an optimized network model; repeating the above steps to train and optimize the network model sequentially, using cross-validation to validate the model, calculating the corresponding square value based on the loss function of each cross-validation, and using the model with the highest square value as the final network model.

[0045] S3: When an abnormal alarm is triggered, the real-time variables of the drainage system are predicted by the real-time variable prediction model, and the predicted real-time variables are compared with the corresponding alarm thresholds to determine the cause of the drainage abnormality.

[0046] Step S3 specifically includes the following steps: after predicting a real-time variable through a network model, determine the continuous time threshold and threshold weight of the real-time variable; standardize the predicted variable output by the network model to obtain a standard variable; if all continuous real-time variables of the same real-time variable prediction model exceed the continuous time threshold, an alarm is triggered; otherwise, compare the threshold weight and the threshold ratio. If the threshold ratio is greater than the threshold weight, it is judged as abnormal; after the alarm is triggered, the cause of the alarm is obtained based on the abnormal judgment result.

[0047] In step S3, the method for determining the continuous time threshold is as follows: perform a normality test on the real-time variables in the real-time variable set, and calculate the mean and standard deviation when the real-time variables in multiple real-time variable sets conform to a normal distribution; based on the mean and standard deviation, calculate the time threshold for each real-time variable in the real-time variable set that exceeds three times the standard deviation, and use this time threshold as the continuous time threshold; calculate the threshold ratio of the real-time variables based on the mean and standard deviation of the real-time variables in multiple real-time variable sets.

[0048] Example 2: Based on this, please refer to Figure 2 This embodiment provides an intelligent monitoring system for drainage in pumped storage power station pipe corridors, which is used to implement the monitoring method described above.

[0049] The monitoring system includes a data acquisition module, a model building module connected to it, and a result output module. The data acquisition module is used to collect and transmit parameter information during normal operation of the drainage system. The parameter information includes alarm threshold information and multiple real-time variable sets. The model building module is used to receive the parameter information, build a real-time variable prediction model based on a long short-term memory network, and transmit the model to the result output module. The result output module is used to receive the prediction model, and when a drainage anomaly alarm is triggered, it predicts the real-time variables through the model and compares them with the alarm threshold to determine the cause of the anomaly.

[0050] The model building module includes a network model unit and a cross-validation unit. The network model unit is used to process the input feature data and output the predicted variables. The cross-validation unit is used to optimize the model parameters. By comparing the loss function of the validation set with that of the model output, the model parameters are updated in reverse, and finally the model with the highest variance is selected as the output.

[0051] The results output module includes an anomaly detection unit, which is used to determine the continuous time threshold and threshold weight of real-time variables.

[0052] The predictor variables are standardized, and an alarm is triggered based on the comparison between continuous-time thresholds or threshold ratios and weights, with the cause of the anomaly output.

[0053] The data acquisition module includes an alarm threshold submodule, which is used to set the initial alarm threshold and divide the safe threshold range, and finally dynamically adjust the threshold through the threshold step size;

[0054] It also includes a normal variable submodule, used to filter the set of normal variables and assist in threshold calibration.

[0055] The model building module can output alarm information for abnormal duration exceeding the limit; the result output module can output details of drainage anomalies.

[0056] The normal variable submodule includes a safety threshold calculation unit, which is used to calculate the safety range based on three types of alarm thresholds for pipelines, pumps, and pipe racks and their corresponding datasets; it performs threshold verification and dynamic adjustment for pipeline, pump, and pipe rack data respectively, and outputs information such as over-limit location, detection map, and duration.

[0057] Example 3: Based on this, this example also provides a storage medium. When the storage medium is read by a computer program, the following steps are executed: Collecting parameter information of the pumped-storage power station drainage system under normal operation, including alarm threshold information and corresponding sets of multiple real-time variables. Each set of real-time variables has a corresponding variable type and timestamp. Variable types include pipe temperature, pipe pressure, pipe gallery humidity, and drainage temperature; constructing a network model, using the alarm threshold information and the multiple sets of real-time variables under normal operation as training sets, and establishing multiple real-time variable prediction models; when establishing multiple real-time variable prediction models, using cross-validation to optimize the network model parameters to obtain the optimal parameters as the final parameters, specifically including: dividing the multiple sets of real-time variables into training and validation sets according to a preset ratio; splitting the multiple sets of real-time variables into variable subsets of the same length, each set containing multiple variables, and each set of variable subsets forming multiple input feature data; sequentially inputting the multiple variables in each set of input feature data into the network model, outputting the corresponding predicted variables, and simultaneously merging the variables corresponding to multiple sets of input feature data and the output predicted variables to form output feature data; and then... The input feature data and the corresponding output feature data of the input feature data are used in a cross-validation method with the validation set of the same group. The loss function of the validation set is used as the loss function of the network model. The parameters in the network model are updated based on the backpropagation error of the loss function, and the parameters of the network model are optimized to form an optimized network model. The above steps are repeated to train and optimize the network model in turn. The model is validated by cross-validation. The square value of the loss function of each cross-validation is calculated, and the model with the highest square value is used as the final network model. When an abnormal alarm is triggered, the real-time variable prediction model is used to predict the real-time variable of the drainage system. The predicted real-time variable is compared with the corresponding alarm threshold to determine the cause of the drainage abnormality. After the real-time variable is predicted by the network model, the continuous time threshold and threshold weight of the real-time variable are determined. The predicted variable output by the network model is standardized to obtain the standard variable. If the continuous real-time variables of the same real-time variable prediction model all exceed the continuous time threshold, an alarm is triggered. Otherwise, the threshold weight and threshold ratio are compared. If the threshold ratio is greater than the threshold weight, it is judged as an abnormality. After the alarm is triggered, the cause of the alarm is obtained based on the abnormality judgment result.

[0058] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. An intelligent monitoring method for drainage in pumped storage power station pipe corridors, characterized in that, Includes the following steps: S1: Collect parameter information of the pumped storage power station drainage system under normal operation. The parameter information includes alarm threshold information and multiple real-time variable sets. Each real-time variable set has a corresponding variable type and timestamp. The variable types include pipeline temperature, pipeline pressure, pipe gallery humidity and drainage temperature. S2: Construct a network model, using the alarm threshold information and multiple real-time variable sets under normal operation as the training set, and establish multiple real-time variable prediction models; S3: When an abnormal alarm is triggered, the real-time variables of the drainage system are predicted by the real-time variable prediction model, and the predicted real-time variables are compared with the corresponding alarm thresholds to determine the cause of the drainage abnormality. Step S3 specifically includes the following steps: After predicting a real-time variable through a network model, the continuous time threshold and threshold weight of the real-time variable are determined; the predicted variable output by the network model is standardized to obtain a standard variable; if all continuous real-time variables of the same real-time variable prediction model exceed the continuous time threshold, an alarm is triggered; otherwise, the threshold weight and threshold ratio are compared, and if the threshold ratio is greater than the threshold weight, it is judged as an anomaly; when the alarm is triggered, the cause of the alarm is obtained based on the anomaly judgment result. The method for determining the continuous time threshold is as follows: Perform a normality test on the real-time variables in the real-time variable set, and calculate the mean and standard deviation when the real-time variables in multiple real-time variable sets conform to a normal distribution; Based on the mean and standard deviation, calculate the time threshold for real-time variables in each set of real-time variables to exceed three times the standard deviation, and use this time threshold as the continuous time threshold. The threshold ratio of real-time variables is calculated based on the mean and standard deviation of real-time variables in multiple real-time variable sets. In step S1, the method for collecting the alarm threshold information includes: Set an initial alarm threshold, calculate a safe threshold range based on the initial alarm threshold, divide the safe threshold range into multiple threshold intervals, and select a set of normal variables from each threshold interval; The initial alarm threshold is matched with the variables in the normal variable set. When none of the variables in the normal variable set meet the initial alarm threshold, the alarm threshold is adjusted by increasing the threshold step size to obtain a new alarm threshold. The adjustment stops when at least one variable in the normal variable set meets the initial alarm threshold.

2. The intelligent monitoring method for drainage of pumped storage power station pipe corridors according to claim 1, characterized in that, In step S2, when establishing multiple real-time variable prediction models, cross-validation is used to optimize the network model parameters to obtain the optimal parameters as the final parameters. Specifically, this includes: Multiple real-time variable sets are divided into training and validation sets according to a preset ratio; each real-time variable set is further divided into variable subsets of the same length, each subset containing multiple variables, and each subset forming multiple input feature data; multiple variables in each set of input feature data are sequentially input into the network model, outputting the corresponding predicted variables, and simultaneously merging the variables corresponding to multiple sets of input feature data and the output predicted variables to form output feature data; The output feature data corresponding to a set of input feature data and the validation set of the same group are input into the cross-validation method. The loss function of the validation set is used as the loss function of the network model. The parameters in the network model are updated according to the backpropagation error based on the loss function. The network model is then optimized to form an optimized network model. Repeat the above steps to train and optimize the network model in sequence. Use cross-validation to validate the model. Calculate the square value based on the loss function of each cross-validation and select the model with the highest square value as the final network model.

3. An intelligent monitoring system for drainage in pumped-storage power station pipe corridors, used to implement the intelligent monitoring method described in claim 1 or 2, characterized in that, It includes a data acquisition module, a model building module that communicates with it, and a result output module; The data acquisition module is used to collect and transmit parameter information during normal operation of the drainage system. The parameter information includes alarm threshold information and multiple real-time variable sets. The model building module is used to receive parameter information, build a real-time variable prediction model based on a long short-term memory network, and transmit the model to the result output module. The result output module is used to receive the prediction model. When the drainage abnormality alarm is triggered, the model predicts real-time variables and compares them with the alarm threshold to determine the cause of the abnormality. The result output module includes an anomaly judgment unit, which is used to determine the continuous time threshold and threshold weight of the real-time variable, as well as to standardize the predicted variable, and to trigger an alarm based on the comparison result of the continuous time threshold or threshold ratio and weight, and output the reason for the anomaly. The data acquisition module includes an alarm threshold submodule, which is used to set an initial alarm threshold and divide the safe threshold range, and finally dynamically adjust the threshold through the threshold step size; it also includes a normal variable submodule, which is used to filter the normal variable set and assist in threshold calibration.

4. The intelligent monitoring system for drainage of pumped storage power station pipe corridors according to claim 3, characterized in that, The model building module includes a network model unit and a cross-validation unit; the network model unit is used to process input feature data and output predicted variables; the cross-validation unit is used to optimize model parameters, and by comparing the loss function of the validation set with that of the model output, the model parameters are updated in reverse, and finally the model with the highest variance is selected as the output.

5. The intelligent monitoring system for drainage of pumped storage power station pipe corridors according to claim 3, characterized in that, The normal variable submodule includes a safety threshold calculation unit, which is used to calculate the safety range of three types of alarm thresholds and corresponding datasets for pipelines, pumps, and pipe racks; it performs threshold verification and dynamic adjustment for pipeline, pump, and pipe rack data respectively, and outputs information on over-limit locations, detection maps, and durations.

6. A storage medium, characterized in that, When the storage medium is read by a computer program, the intelligent monitoring method as described in claim 1 or 2 is executed.

Images