Pumped storage power station operation trend early warning analysis method and system

Abstract

The invention discloses a pumped storage power station operation trend early warning analysis method and system. The method comprises the following specific steps: S1, data collection and preprocessing; s2, feature selection and construction: screening key features of internal operation data by combining a principal component analysis method with a Lasso regression method; s3, constructing a prediction model: taking external environment factors as weight factors, adopting a deep learning model combining an attention mechanism and a bidirectional LSTM network, integrating an XGBoost model to perform training optimization to obtain the prediction model, and predicting various operation parameters of the power station; and S4, constructing a double-early-warning mechanism, setting an early-warning threshold value of each operation parameter of the power station based on an early-warning rule, comparing a prediction result of the prediction model constructed in the step S3 with the early-warning threshold value, and establishing an instant early-warning mechanism and a long-period early-warning mechanism at the same time. Accurate early warning of the operation trend of the pumped storage power station is realized, the operation and maintenance cost is saved, and the operation intelligence and reliability of an electrical system are improved.

Description

Technical Field

[0001] The present invention belongs to the field of power station system safety design, and in particular relates to a pumped storage power station operation trend early warning analysis method and system. Background Art

[0002] With the transformation of energy structure, pumped storage power stations, as important energy storage facilities, play an increasingly important role in the power system. However, due to the complex and changeable operating environment of pumped storage power stations, how to monitor their operating status in real time and discover potential operating risks in a timely manner has become the key to ensuring the safe and stable operation of the power system. Traditional monitoring methods often rely on manual experience and are difficult to cope with the growing amount and complexity of data. Therefore, it is imperative to develop an early warning analysis method based on big data and machine learning.

[0003] Patent CN114442543A discloses a computer monitoring method suitable for fault warning of hydropower station, which monitors the operating status data of the hydropower station in real time through a computer monitoring system, completes model training based on historical data, diagnoses faults based on the operating status data and issues fault warnings. However, the invention uses a relatively simple algorithm, which leads to limitations in the scope of application, and it is impossible to update the warning rules and threshold ranges based on real-time operating data or other data sources. In addition, the entire warning detection algorithm only considers the influence of the power station's own factors, and does not consider the influence of other environmental factors.

[0004] Patent CN115977855A discloses a hydropower station fault diagnosis system based on artificial intelligence, including a server, a fault diagnosis efficiency analysis unit, a real-time physical signal monitoring unit, a diagnostic fault analysis unit, and a fault rectification control unit. Although the operational safety of the hydropower station can be diagnosed through thresholds, the degree of intelligence is low and only simple right and wrong judgments can be made based on operating parameters. Summary of the invention

[0005] In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides a pumped-storage power station operation trend early warning analysis method and system, which realizes accurate early warning of the operation trend of the pumped-storage power station, is conducive to saving operation and maintenance costs, and improving the intelligence and reliability of electrical system operation.

[0006] The main technical solutions adopted in the present invention are:

[0007] A method for early warning analysis of operation trend of pumped storage power station.

[0008] S1: Data collection and preprocessing: regularly collect power plant operation data from multiple data sources, and perform data cleaning and standardization on the power plant operation data, wherein the power plant operation data includes internal operation data and external environmental factors;

[0009] S2: Feature selection and construction: principal component analysis combined with Lasso regression method is used to screen the key features of internal operation data and construct new feature data reflecting complex operation status;

[0010] S3: Construct a prediction model: Take external environmental factors as weight factors, based on the new feature data obtained in step S2, adopt a deep learning model that combines the attention mechanism and the bidirectional LSTM network, integrate the XGBoost model for training and optimization to obtain a prediction model, identify data patterns, update the feature database in real time, and predict various operating parameters of the power plant;

[0011] S4: Construct a dual warning mechanism: Set warning thresholds for various operating parameters of the power plant based on warning rules, compare the prediction results of the prediction model constructed in step S3 with the warning thresholds, and establish an immediate warning mechanism and a long-term warning mechanism.

[0012] Preferably, in S1, the internal operation data include equipment operation parameters, environmental monitoring data and historical fault records; the external environmental factors include hydrological information of the entire river basin and weather forecast information within the basin.

[0013] Preferably, the specific steps in S2 are as follows:

[0014] S2-1: Based on the internal operation data after the data processing in step S1, the principal component analysis method is used to reduce the characteristic dimension of the internal operation data, and the original internal operation data is projected onto a new coordinate system composed of the selected principal components to obtain a reduced dimension internal operation data set;

[0015] S2-2: The Lasso regression method is used to process the internal operation data set after dimensionality reduction to screen out the key features that have a significant impact on the prediction of the power plant operation status;

[0016] S2-3: Construct new feature data based on the key features screened out in step S2-2.

[0017] Preferably, in S3, the specific structure of the deep learning model is as follows: a bidirectional LSTM network is used as the main model, allowing the model to read information from both the past and the future, and an attention mechanism is embedded in the bidirectional LSTM network, so that the model can automatically focus on key information fragments when processing long sequence data. At the same time, a Dropout layer, an early stopping mechanism and L2 regularization are introduced to prevent the model from overfitting.

[0018] Preferably, in S3, the training optimization method of the prediction model is as follows:

[0019] Use the new feature data constructed in step S2 as the input of the XGBoost model;

[0020] Select the loss function for model training based on the task requirements of the pumped storage power station;

[0021] Optimize the hyperparameters of the XGBoost model through grid search or random search methods;

[0022] The prediction results of the deep learning model are integrated as additional features of XGBoost, and the strong generalization ability of the XGBoost model is used to improve the prediction accuracy of the deep learning model.

[0023] Preferably, in S4, the immediate warning mechanism is as follows: comparing the prediction result of the prediction model within a short period with the warning threshold, and if the prediction result exceeds the warning threshold, triggering the short-period immediate warning mechanism;

[0024] The long-term warning mechanism is as follows: considering external environmental factors, obtaining the prediction results of the prediction model within a long period, comparing the prediction results with the warning threshold, and triggering a long-term warning if the prediction results exceed the warning threshold.

[0025] Preferably, the early warning threshold that is dynamically adjusted according to the uncertainty adjustment of the model prediction is a dynamic threshold, and the adjustment formula of the dynamic threshold is as follows:

[0026] Dynamic threshold = average value of operating parameters ± k * standard deviation of operating parameters

[0027] Here, k is a coefficient adjusted according to the confidence level.

[0028] A pumped-storage power station operation trend early warning analysis system, the system executes the pumped-storage power station operation trend early warning analysis method as described in any one of claims 1-7 above.

[0029] Preferably, the system also includes a visualization and interaction platform having a forecast information display function, a decision support function and a data interaction function.

[0030] Beneficial effects: The present invention provides a method and system for early warning analysis of operation trends of a pumped storage power station, which has the following advantages:

[0031] (1) The present invention adopts a deep learning model that combines the attention mechanism and the bidirectional LSTM network, which can effectively capture key information fragments in time series data. At the same time, the historical operation data of the power plant and the new data generated in real time every day are used to train the machine learning model, which enhances the model's prediction ability and realizes the intelligent update of the model, making the prediction model more applicable.

[0032] (2) The present invention fully considers the impact of external factors on power plant operation, constructs a long-term early warning mechanism and an immediate early warning mechanism, and can dynamically adjust the early warning threshold in real time according to the uncertainty of model prediction. This dual early warning mechanism greatly improves the accuracy and timeliness of the early warning, has broad application prospects and practical value, and is conducive to realizing more intelligent, safer and more efficient early warning analysis of power plant operation trends. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Schematic diagram of the construction process of the early warning analysis method of this embodiment 1. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application are clearly and completely described below. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0035] Example 1

[0036] A method for early warning analysis of pumped storage power station operation trends based on machine learning, the specific steps are as follows:

[0037] S1: Data collection and preprocessing: Collect power plant operation data from multiple data sources at regular intervals, and perform data cleaning and standardization on the power plant operation data. The power plant operation data includes internal operation data and external environmental factors.

[0038] In the present invention, the internal operation data may include, but is not limited to, equipment operation parameters, environmental monitoring data, and historical fault records; the external environmental factors may include, but are not limited to, the hydrological information of the entire river basin and the weather forecast information within the basin. Those skilled in the art may selectively collect the internal operation data and external environmental factors according to actual application requirements.

[0039] In this embodiment 1, the data comes from the SCADA system, environmental monitoring system and historical fault database of a large pumped storage power station. The SCADA system provides real-time equipment operating parameters, including generator power, water level, temperature, etc.; the environmental monitoring system records environmental factors such as temperature, humidity, wind speed, etc.; the historical fault database contains past fault events and their detailed descriptions.

[0040] In this embodiment 1, the specific data collection and preprocessing method is: use the ETL (Extract-Transform-Load) tool to automatically pull data from the above data source at regular intervals to ensure the real-time and integrity of the data; then after the data is imported, use the ETL tool to clean and standardize the data, including removing invalid values, filling missing values, and unifying the data format. For example, for missing environmental data, interpolation can be used to fill in; for abnormal equipment operating parameters, statistical methods are used to identify and correct them. The above-mentioned use of ETL tools for data cleaning and standardization is a conventional technical means, so it is not described in detail.

[0041] S2: Feature selection and construction: principal component analysis combined with Lasso regression method is used to screen the key features of internal operation data and construct new feature data reflecting complex operation status;

[0042] In this embodiment 1, the specific method of step S2 is as follows:

[0043] S2-1: Since the internal operation data of a pumped-storage power station usually includes generator power, water level change rate, ambient temperature, bearing temperature, vibration acceleration, etc., these data are characterized by high dimensionality and potential correlation. Therefore, in this embodiment 1, when processing the internal operation data of a pumped-storage power station, the principal component analysis method is used to reduce the characteristic dimension of the internal operation data, and the original internal operation data is projected onto a new coordinate system composed of the selected principal components to obtain a reduced-dimensional internal operation data set, thereby reducing the time and resource consumption of model training, while ensuring that the prediction accuracy of the model is not significantly affected. The specific method is as follows:

[0044] Standardization: Standardize the features of all internal running data to ensure that each feature has the same scale, because PCA is very sensitive to scale. Standardization includes subtracting the mean of the feature and dividing by the standard deviation;

[0045] Reflecting the strength of linear relationships: Based on the characteristics of the preprocessed internal operating data, the covariance matrix is ​​calculated, which can reveal the strength of the linear relationship between the characteristics in the pumped storage power station operating data;

[0046] Eigendecomposition: Perform eigendecomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors, where the size of the eigenvalue represents the proportion of variance explained by each principal component, and the eigenvector indicates the direction in which the original data is projected into the new space.

[0047] Cumulative explained variance: According to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues. These eigenvectors form a new coordinate system, namely the principal component space. Generally, the principal components that can explain more than 90% of the overall variance of the data are selected.

[0048] Dataset after dimensionality reduction: The original internal operation data is projected onto a new coordinate system composed of the selected principal components to obtain a dataset of internal operation data after dimensionality reduction. The dataset after dimensionality reduction not only has a reduced dimension, but also still retains most of the important information in the pumped storage power station operation data.

[0049] S2-2: For the internal operation data set after dimensionality reduction, the Lasso regression method is used to screen out the key features that have a significant impact on the prediction of the power plant operation status. The specific method is as follows:

[0050] Minimize prediction error: The goal of Lasso regression is to minimize the prediction error (such as mean square error) while penalizing large weights through the L1 regularization term. The loss function is expressed as:

[0051]

[0052] Where n is the number of samples, i.e., the number of observations in the pumped-storage power station operation data; p is the number of features, i.e., the number of variables in the pumped-storage power station operation data; y i is the true value of the i-th sample, such as the actual measured value of a performance indicator of a pumped storage power station; is the predicted value of the i-th sample, that is, the value predicted by the model based on the given features; w j is the weight of the jth feature, that is, the model coefficient, which indicates the influence of this feature on the prediction result; λ is the regularization parameter, which is used to control the strength of the L1 regularization term.

[0053] Optimization algorithm: Since the L1 regularization term is non-smooth, an optimization algorithm for L1 regularization needs to be adopted. Therefore, the present invention uses the gradient descent method or other optimization algorithms (such as the coordinate descent method) to minimize the loss function.

[0054] Retain key features: During the training process, the L1 regularization penalty term in Lasso regression will make the weights of features that contribute less to the prediction results approach zero, while retaining those that contribute significantly to the prediction results. Ultimately, only features with non-zero weights will be retained, thereby achieving key feature selection, reducing the complexity of the model, and improving the efficiency and accuracy of prediction.

[0055] In this embodiment 1, the dimension reduction by principal component analysis and the feature selection by Lasso regression can effectively reduce the complexity of the operation data processing of the pumped storage power station, while ensuring the prediction ability of the model, which is of great significance for improving the operation efficiency and safety of the power station.

[0056] S2-3: Construct new feature data based on the key features selected in step S2-2, such as equipment aging index and environmental pressure index, to more comprehensively reflect the health status and environmental adaptability of the equipment, thereby enhancing the model's explanatory power and prediction accuracy.

[0057] S3: Construct a prediction model: Use external environmental factors as weight factors, based on the new feature data obtained in step S2, adopt a deep learning model that combines the attention mechanism and the bidirectional LSTM network, integrate the XGBoost model for training and optimization to obtain a prediction model, identify data patterns, update the feature database in real time, and predict various operating parameters of the power plant.

[0058] In order to more accurately predict the operating status of a pumped storage power station, this embodiment 1 designs a deep learning model that combines an attention mechanism and a bidirectional LSTM (Bidirectional Long Short-Term Memory) network. This model can effectively capture key information fragments in time series data and enhance prediction capabilities by using historical and future data.

[0059] Application of two-way dynamic interaction: The operation of pumped storage power stations is affected by many factors, such as the water volume of upstream and downstream reservoirs, changes in grid demand, weather conditions, etc. There is a complex two-way feedback relationship between these factors. Conventional unidirectional LSTM models can only process data in a single direction and may not be able to fully capture these two-way interactions. Therefore, this embodiment 1 adopts a bidirectional LSTM network, and the model can process data from both the past and future directions at the same time, better understand these two-way dynamic interactions, and make the prediction results closer to reality, especially in predicting water level changes, power scheduling, etc.

[0060] Identification and utilization of key information fragments: During the operation of pumped storage power stations, certain transient events (such as sudden high power demand or rapid water level changes) have a huge impact on system stability. Although the traditional LSTM model can process sequence data, it may not be able to accurately identify these key events. Therefore, in this embodiment 1, by introducing an attention mechanism in the bidirectional LSTM network, the model can automatically learn which fragments in the data are most relevant during the training process, and give higher weights to these key information fragments, so as to more accurately predict the impact of these transient events and improve the prediction accuracy.

[0061] In this embodiment 1, the specific structure of the deep learning model is as follows: a bidirectional LSTM network is used as the main model, allowing the model to read information from both the past and the future, and an attention mechanism is embedded in the bidirectional LSTM network, so that the model can automatically focus on key information fragments when processing long sequence data. At the same time, a Dropout layer, an early stopping mechanism and L2 regularization are introduced to prevent the model from overfitting.

[0062] In this embodiment 1, the specific method for constructing the deep learning model is as follows:

[0063] Data processing:

[0064] The historical operating data of the pumped storage power station is organized into a time series format, including but not limited to key indicators such as generator power, water level change rate, ambient temperature, etc., to ensure the integrity and continuity of the data.

[0065] Model construction:

[0066] The LSTM network is selected as the main model to capture the dependencies between long-term and short-term parameters, such as the dependency between equipment power and water level change rate, the relationship between ambient temperature and equipment efficiency, and the changing trend of these parameters over time. The LSTM model contains an input gate, a forget gate, and an output gate, which can memorize important information and filter irrelevant information, maintain the long-term state while avoiding the gradient vanishing problem. In this embodiment 1, the LSTM network adopts a bidirectional LSTM structure, allowing the model to read information from both the past and the future, and can capture more dimensional context information.

[0067] On this basis, an attention mechanism is embedded in the LSTM network, enabling the model to automatically focus on more critical information fragments when processing long sequence data, such as the peak period of generator power or the moment when the water level changes rapidly, thereby improving the model's sensitivity to important events and prediction accuracy.

[0068] In addition, in order to prevent overfitting, the following designs are also performed in this embodiment 1:

[0069] Dropout layer: Add a Dropout layer between LSTM layers to randomly discard some neurons, reduce the complexity of the model, and enhance the generalization ability of the model.

[0070] Early stopping mechanism: Continuously monitor the model performance on the validation set. Once the performance stagnates or begins to deteriorate, stop training immediately to prevent the model from overfitting the training data.

[0071] L2 regularization: By limiting the size of weights, the model complexity is further controlled, the generalization ability of the model is improved, and the model can maintain good prediction results on unseen data.

[0072] XGBoost is an efficient machine learning framework based on gradient boosting trees, which forms a strong learner by combining multiple weak learners (usually decision trees). In the application of the pumped storage power station in this embodiment 1, XGBoost can handle nonlinear relationships and missing values, and has excellent feature combination and generalization capabilities.

[0073] In this embodiment 1, the training optimization method of the prediction model is as follows:

[0074] Use the new feature data constructed in step S2 as the input of the XGBoost model;

[0075] Training stage: selecting a loss function for model training according to the task requirements of the pumped storage power station, wherein the loss function is preferably a cross entropy loss function or an MSE loss function; for example, in this embodiment 1, the mean square error (MSE) is used as the loss function, and the network weight is adjusted by the back propagation algorithm (BPTT) to minimize the gap between the predicted value and the actual value;

[0076] Optimize the hyperparameters of the XGBoost model, such as learning rate, maximum tree depth, regularization parameters, etc., through grid search or random search methods to achieve the best model performance;

[0077] The prediction results of the deep learning model are integrated as additional features of XGBoost, and the strong generalization ability of the XGBoost model is used to improve the prediction accuracy of the deep learning model.

[0078] S4: Construct a dual warning mechanism: Set warning thresholds for various operating parameters of the power plant based on warning rules, compare the prediction results of the prediction model constructed in step S3 with the warning thresholds, and establish an immediate warning mechanism and a long-term warning mechanism.

[0079] In this embodiment 1, the warning rules can be set based on the deviation between the model prediction results and the actual operating parameters, combined with the historical failure mode. For example, if the predicted equipment power exceeds the normal operating range, or the predicted water level change rate is abnormal, the warning should be triggered immediately. In the present invention, the warning rules can be set by those skilled in the art according to the actual application requirements.

[0080] In this embodiment 1, the immediate warning mechanism is as follows: the prediction results of the prediction model within a short period are compared with the warning threshold. If the prediction results exceed the warning threshold, the short-period immediate warning mechanism is triggered; the long-period warning mechanism is as follows: considering external environmental factors, the prediction results of the prediction model within a long period are obtained, and the prediction results are compared with the warning threshold. If the prediction results exceed the warning threshold, the long-period warning is triggered.

[0081] The early warning threshold is dynamically adjusted according to the uncertainty of the model prediction (such as the standard deviation of the prediction). In this embodiment 1, the dynamic threshold mainly monitors the deviation of the equipment power, water level change rate, ambient temperature and the standard working range. Specifically, the adjustment formula of the dynamic threshold of this embodiment 1 is as follows:

[0082] The adjustment formula of the dynamic threshold is as follows:

[0083] Dynamic threshold = average value of operating parameters ± k * standard deviation of operating parameters

[0084] Among them, k is a coefficient adjusted according to the confidence level, and the mean value and standard deviation of the operating parameters are obtained based on the operating parameters collected and recorded in real time during the operation of the equipment.

[0085] In the present invention, the early warning analysis system constructed according to the above-mentioned pumped-storage power station operation trend early warning analysis method based on machine learning is applicable to various power systems, including but not limited to pumped-storage power stations, thermal power stations, wind farms, solar power stations, etc., and can effectively improve the safety and efficiency of power system operation.

[0086] In the present invention, the early warning analysis system also includes a visualization and interaction platform, which displays the prediction results and early warning information through the design of the user interface, and realizes decision support and data interaction. The specific functional design is as follows:

[0087] Forecast information display function

[0088] The user interface displays real-time monitoring data, forecast results, warning information and historical trend charts, and intuitively shows the health level and warning level of the equipment through color coding and icons.

[0089] Decision support functions

[0090] The platform provides a detailed interpretation of the warning information, including the cause of the warning, possible consequences and recommended countermeasures. At the same time, it supports users to customize the warning notification method, such as SMS, email or APP push, to ensure that operation and maintenance personnel can respond quickly.

[0091] Data interaction function

[0092] The platform supports functions such as data query, historical trend analysis and model parameter adjustment, helping operation and maintenance personnel to deeply understand the operating status of the power station and provide data support for decision-making.

[0093] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for early warning analysis of operation trends of a pumped storage power station, characterized in that: The specific steps are as follows: S1: Data collection and preprocessing: regularly collect power plant operation data from multiple data sources, and perform data cleaning and standardization on the power plant operation data, wherein the power plant operation data includes internal operation data and external environmental factors; S2: Feature selection and construction: Principal component analysis combined with Lasso regression method is used to screen key features of internal operation data and construct new feature data reflecting complex operation status; S3: Build a prediction model: Using external environmental factors as weight factors, based on the new feature data obtained in step S2, a deep learning model combining an attention mechanism and a bidirectional LSTM network is used. The XGBoost model is integrated for training and optimization to obtain a prediction model, identify data patterns, update the feature database in real time, and predict various power plant operating parameters. S4: Construct a dual early warning mechanism: Set early warning thresholds for various operating parameters of the power plant based on early warning rules, compare the prediction results of the prediction model constructed in step S3 with the early warning thresholds, and establish an immediate early warning mechanism and a long-term early warning mechanism.

2. The method for early warning analysis of pumped storage power station operation trend according to claim 1, characterized in that: In S1, the internal operating data includes equipment operating parameters, environmental monitoring data and historical fault records; the external environmental factors include hydrological information of the entire river basin and weather forecast information within the basin.

3. The method for early warning analysis of pumped storage power station operation trend according to claim 1, characterized in that: The specific steps in S2 are as follows: S2-1: Based on the internal operation data processed in step S1, principal component analysis is used to reduce the characteristic dimensions of the internal operation data, and the original internal operation data is projected onto a new coordinate system composed of the selected principal components to obtain a reduced-dimensional internal operation data set; S2-2: Lasso regression method is used to process the internal operation data set after dimensionality reduction to screen out key features that have a significant impact on predicting the power plant operation status; S2-3: Construct new feature data based on the key features screened out in step S2-2.

4. The method for early warning analysis of pumped storage power station operation trend according to claim 1, characterized in that: In S3, the specific structure of the deep learning model is as follows: a bidirectional LSTM network is used as the main model, allowing the model to read information from both the past and future directions at the same time, and an attention mechanism is embedded in the bidirectional LSTM network, so that the model can automatically focus on key information fragments when processing long sequence data. At the same time, a Dropout layer, an early stopping mechanism, and L2 regularization are introduced to prevent the model from overfitting.

5. The method for early warning analysis of pumped storage power station operation trend according to claim 4, characterized in that: In S3, the training optimization method of the prediction model is as follows: Use the new feature data constructed in step S2 as the input of the XGBoost model; Select the loss function for model training based on the mission requirements of the pumped storage power station; Optimize the hyperparameters of the XGBoost model through grid search or random search methods; The prediction results of the deep learning model are integrated as additional features of XGBoost, and the strong generalization ability of the XGBoost model is used to improve the prediction accuracy of the deep learning model.

6. The method for early warning analysis of pumped storage power station operation trend according to claim 1, characterized in that: In S4, the immediate warning mechanism is as follows: comparing the prediction result of the prediction model in a short period with the warning threshold, if the prediction result exceeds the warning threshold, the short period immediate warning mechanism is triggered; The long-term warning mechanism is as follows: considering external environmental factors, obtaining the prediction results of the prediction model within a long period, comparing the prediction results with the warning threshold, and triggering a long-term warning if the prediction results exceed the warning threshold.

7. The method for early warning analysis of pumped storage power station operation trend according to claim 6, characterized in that: The early warning threshold that is dynamically adjusted according to the uncertainty adjustment of the model prediction is the dynamic threshold, and the adjustment formula of the dynamic threshold is as follows: Dynamic threshold = average value of operating parameters ± k × standard deviation of operating parameters Here, k is a coefficient adjusted according to the confidence level.

8. A pumped storage power station operation trend early warning analysis system, characterized in that: The system executes the pumped storage power station operation trend early warning analysis method as described in any one of claims 1 to 7 above.

9. The pumped storage power station operation trend early warning analysis system according to claim 1, characterized in that: The system also includes a visualization and interaction platform with forecast information display functions, decision support functions and data interaction functions.

Images