Intelligent monitoring and early warning method and system for small hydropower station in mountainous area

Abstract

The invention discloses an intelligent monitoring and early warning method and system for a small hydropower station in a mountainous area, and relates to the technical field of power system monitoring and early warning, and the method comprises the steps: collecting hydropower station data through a sensor, and carrying out the preprocessing to obtain key features; a plurality of support vector regression base models are trained for a plurality of data variables of the hydropower station, a self-service aggregation algorithm is adopted for integrated learning, and a plurality of base models are sampled and trained through a self-service method; using a Bayesian optimization method to optimize model parameters and an integration strategy; dynamically adjusting an early warning threshold according to real-time performance feedback; periodically updating the threshold value by adopting a sliding window technology; and deploying the optimized model in a real-time monitoring system, and monitoring and predicting the running state of the hydropower station in real time. According to the invention, comprehensive and accurate data acquisition is realized through the multi-source sensor, and the comprehensiveness and quality of monitoring data of the hydropower station are improved; and a support vector machine and an integrated learning optimization prediction model are adopted, so that the prediction accuracy and robustness are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring and early warning, and in particular to an intelligent monitoring and early warning method and system for a small hydropower station in a mountainous area. Background Art

[0002] As global energy demand continues to grow, hydropower stations, as a clean and renewable energy supply model, have occupied an important position in the global energy structure. However, the efficient and safe operation of hydropower stations is crucial to ensure the stability and security of energy supply. To achieve this goal, intelligent monitoring and early warning technologies have become an indispensable part. These technologies can detect and deal with potential safety hazards in a timely manner, thereby ensuring the sustainable and stable operation of hydropower stations.

[0003] In recent years, with the rapid development of information technology, more and more researchers have begun to explore the application of advanced communication technology, data processing technology and artificial intelligence algorithms in hydropower station monitoring and early warning systems. In related research at home and abroad, the application of power communication network systems in hydropower station monitoring and early warning has shown great potential. Some studies have pointed out that by building an efficient power communication network, real-time monitoring of various operating parameters of hydropower stations can be achieved, thereby improving the safety and operating efficiency of hydropower stations. These studies not only focus on the collection and transmission of real-time data, but also focus on how to use these data for effective early warning. However, although the current research has achieved some results, there are still some problems and shortcomings. First, the actual application effect of some studies needs to be verified, especially in the complex and changeable operating environment of hydropower stations, the stability and reliability of the technology still need to be further explored. Secondly, the existing monitoring and early warning methods of small hydropower in remote mountainous areas are mostly based on manual observation triggering, with low timeliness and considered to be error interference. How to effectively integrate multiple data sources and improve the intelligence level of the early warning system is also one of the challenges facing the current research on small hydropower in mountainous areas. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: how to address the problems existing in the existing small hydropower station monitoring system, such as incomplete data collection, insufficient model prediction accuracy, inflexible early warning mechanism, and inability to adapt to data changes in real time.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent monitoring and early warning method for small hydropower stations in mountainous areas, which comprises the following steps: collecting hydropower station data through sensors, performing preprocessing to obtain key features;

[0007] Multiple support vector regression base models are trained for multiple data variables of the hydropower station, and the bootstrap aggregation algorithm is used for ensemble learning, and multiple base models are trained by bootstrap sampling;

[0008] Use Bayesian optimization methods to optimize model parameters and ensemble strategies;

[0009] Dynamically adjust warning thresholds based on real-time performance feedback; use sliding window technology to regularly update thresholds;

[0010] The optimized model is deployed in the real-time monitoring system to conduct real-time monitoring and prediction of the operating status of the hydropower station.

[0011] As a preferred solution of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in the present invention, the collection of hydropower station data through sensors includes real-time collection of water level, flow, rainfall, temperature, and power output and voltage of generator sets from the hydropower station, and the use of hydropower station data for feature extraction and model training to predict and alarm the target variables water level and flow.

[0012] As a preferred solution of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in the present invention, the preprocessing to obtain key features is to extract key features from hydropower station data, including rolling mean and rolling standard deviation, and use recursive feature elimination to perform feature selection on all features. The recursive feature elimination performs weight evaluation on each feature through a training model, removes the features with the smallest contribution to the prediction, and performs z-score standardization on all features.

[0013] As a preferred solution of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in the present invention, wherein: multiple support vector regression base models are trained for multiple data variables of the hydropower station, including:

[0014] Select support vector machine as the basic model and combine it with ensemble learning method to improve the accuracy of model prediction by finding a regression line so that the distance between most data points and the current function is within the range ∈;

[0015] For the SVM regression model, the decision function is expressed as:

[0016]

[0017] K(x i ,x j ) = exp(-γ||x i -x|| 2 )

[0018] Among them, α i and is the Lagrange multiplier; K(x i,x) is the kernel function, b is the constant term, ||x i -x j || 2 represents the square of the Euclidean distance between vectors and , and γ is the parameter of the kernel function;

[0019] Adopting the principle of bootstrap method, random sampling with replacement is performed from the original data to generate multiple training subsets and corresponding test sets. For each training subset, the unsampled samples are used as the test set of the current subset. For the data variables of each hydropower station, the corresponding support vector machine base model is trained and integrated learning is performed. The weights are set according to the performance of each base model on the corresponding test set, and the prediction results of the SVM base model of the same target variable are weighted averaged to generate the integrated support vector machine model of the target variable.

[0020] After the training is completed, the integrated support vector machine model for each target variable is used to dynamically set the warning threshold based on real-time performance feedback. When an abnormal state is predicted, the warning is triggered in time. The threshold is updated regularly using sliding window technology. When the abnormal probability predicted by the integrated model of any target variable exceeds the dynamically set warning threshold, the warning system will automatically trigger the warning.

[0021] As a preferred solution of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in the present invention, the optimization of model parameters and integration strategies using the Bayesian optimization method includes:

[0022] For the target variables water level and flow, multiple support vector machine base models are trained, and for each target variable, multiple support vector machine base models are trained;

[0023] The prediction result of the integrated model is obtained by weighted average, and its expression is:

[0024]

[0025] in, is the predicted value of the integrated model, M is the number of base models, h i (x) is the prediction result of each base model of the ith i is the weight of each i-th base model;

[0026] The F1 score is selected as the evaluation criterion of the integrated model to optimize the accuracy of the early warning system. The expression is:

[0027]

[0028] Among them, R pre Represents precision, R rec Represents the recall rate.

[0029] As a preferred solution of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in the present invention, the dynamic adjustment of the early warning threshold according to real-time performance feedback includes:

[0030] The weight parameters in the integrated model are adjusted through Bayesian optimization. The F1 score is used as the optimization objective function to find the optimal weight combination of each sub-model in the integrated model to maximize the F1 score. The objective function expression is:

[0031] L(ω)=-F1(ω)

[0032] Where ω is the weight combination of each sub-model in the integrated model; F1(ω) is the F1 score under this weight combination;

[0033] Bayesian optimization constructs a proxy model of the objective function L(ω), uses a Gaussian process to approximate the distribution of the objective function, and uses the expected improvement (EI) to obtain the weighted combination of the function selection for the next step of evaluation. The goal of the expected improvement function is to maximize the F1 score, which is expressed as:

[0034] EI(ω)=Ε[max(F1(ω)-F1(ω + ),0)]

[0035] Among them, F1(ω + ) is the currently known best F1 score. Bayesian optimization uses this function to continuously approach the global optimal solution and maximize the F1 score.

[0036] As a preferred solution of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in the present invention, the dynamic adjustment of the early warning threshold also includes:

[0037] Using sliding window technology, a sliding window with a fixed length of L is set to perform real-time analysis on the latest L data samples. The data set in the window is:

[0038] W t ={(y' t-L+1 ,y t-L+1 ),(y' t-L+2 ,y t-L+2 ),...,(y' t ,y t )}

[0039] Among them, y' i is the predicted value at time t, y i is the actual observation at time t, and L is the length of the sliding window. As time goes by, the sliding window is continuously updated, removing the earliest data points and adding the latest data points;

[0040] The performance index F1 score of the model is calculated in the sliding window. According to the change of the performance index in the sliding window, the warning threshold θ is dynamically adjusted. The expression is:

[0041]

[0042] Among them, θ t is the warning threshold at time t; α is the learning rate, which controls the step size of threshold adjustment; It is the gradient of the F1 score to the threshold θ. By monitoring the changes in the F1 score, the warning threshold θ is adjusted.

[0043] Finally, use the updated alarm threshold θ t+1 Determine the new prediction result y' t+1 Whether to trigger an early warning.

[0044] Another object of the present invention is to provide an intelligent monitoring and early warning system for small hydropower stations in mountainous areas, which can realize real-time monitoring and efficient early warning of the operating status of hydropower stations through automated data collection, dynamic data processing, intelligent prediction analysis and early warning triggering, thereby solving the problems of strong manual dependence, slow response speed and low early warning accuracy in the monitoring process of existing hydropower stations.

[0045] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent monitoring and early warning system for small hydropower stations in mountainous areas, comprising: a sensor module, a data preprocessing module, a support vector regression base model training and integration module, and a real-time monitoring and early warning triggering module;

[0046] The sensor module collects real-time data of the operation of the hydropower station;

[0047] The data preprocessing module performs preliminary cleaning and feature extraction on the collected data, using rolling average and standard deviation calculations and recursive feature elimination to screen out important features;

[0048] The support vector regression base model training and integration module generates multiple SVR base models through a self-service aggregation algorithm, and generates an overall prediction model by weighted average based on an ensemble learning method;

[0049] The real-time monitoring and early warning triggering module uses an optimized integrated model to determine whether to trigger an early warning based on a dynamic threshold.

[0050] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned intelligent monitoring and early warning method for small hydropower stations in mountainous areas when executing the computer program.

[0051] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned intelligent monitoring and early warning method for small hydropower stations in mountainous areas.

[0052] The beneficial effects of the present invention are as follows: the present invention realizes comprehensive and accurate data collection through multi-source sensors, thereby improving the comprehensiveness and quality of monitoring data of hydropower stations; adopts support vector machines and integrated learning to optimize the prediction model, thereby improving the accuracy and robustness of the prediction; dynamically tunes the model parameters and integration strategies through Bayesian optimization, thereby effectively improving the performance of the early warning system; dynamically adjusts the early warning threshold using sliding window technology, thereby enhancing the system's ability to adapt to real-time changes and ensuring timely early warning of abnormal conditions; ultimately, through real-time monitoring and intelligent early warning, the safety and operational stability of the hydropower station are greatly improved, which has important application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0054] Figure 1 A power communication network system architecture diagram of a method for intelligent monitoring and early warning of small hydropower stations in mountainous areas provided by the first embodiment of the present invention;

[0055] Figure 2 A schematic diagram of the layout of intelligent patrol terminals in an intelligent monitoring and early warning method for a small hydropower station in a mountainous area provided by the first embodiment of the present invention;

[0056] Figure 3 An application layer information modeling diagram of an intelligent monitoring and early warning method for a small hydropower station in a mountainous area provided by the first embodiment of the present invention;

[0057] Figure 4 A flow chart of an intelligent monitoring and early warning system based on machine learning for an intelligent monitoring and early warning method for a small hydropower station in a mountainous area provided by the first embodiment of the present invention;

[0058] Figure 5 A structural diagram of an early warning model of an intelligent monitoring and early warning method for a small hydropower station in a mountainous area provided by the first embodiment of the present invention;

[0059] Figure 6 A graph showing changes in water level over time in a method for intelligent monitoring and early warning of a small hydropower station in a mountainous area provided by the first embodiment of the present invention;

[0060] Figure 7A water level prediction result diagram of a method for intelligent monitoring and early warning of a small hydropower station in a mountainous area provided by the first embodiment of the present invention;

[0061] Figure 8 A schematic diagram of monitoring abnormality alarm of a method for intelligent monitoring and early warning of a small hydropower station in a mountainous area provided by a third embodiment of the present invention;

[0062] Fig. 9 A forecast accuracy distribution diagram of an intelligent monitoring and early warning method for small hydropower stations in mountainous areas provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0064] Example 1, reference Figure 1 As one embodiment of the present invention, a method for intelligent monitoring and early warning of small hydropower stations in mountainous areas is provided, comprising:

[0065] Aiming at the demand for intelligent monitoring and early warning technology of hydropower stations, the present invention designs a hierarchical power communication network system architecture ( Figure 1 ), the architecture mainly includes four layers: perception layer, network layer, data layer and application layer.

[0066] 1) Perception layer

[0067] The perception layer is responsible for collecting the operation data of the hydropower station, including water level, flow, rainfall, temperature, power output and voltage of the generator set. This layer is mainly composed of sensors and monitoring equipment, such as water level sensors, flow meters and video terminals. These devices transmit real-time data to the network layer through the communication interface with the network layer.

[0068] For the layout of video terminals ( Figure 2 ), vertical poles / wall hanging points were arranged on the roofs of the high-rise buildings of the hydropower station to ensure stable installation and wide-angle monitoring of the cameras; secondly, poles up to 6 meters high were erected around the outdoor venue to widen the monitoring range and improve the monitoring effect; at the same time, wall hanging points were also installed in the indoor venue to achieve full coverage of key information monitoring. Inspection cameras can be authorized and accepted through remote control, without manual intervention by operators.

[0069] 2) Network layer

[0070] The network layer is the core of the power communication network system, responsible for data transmission and processing. It mainly includes the following components:

[0071] (1) Communication network equipment: including routers, switches, fiber optic transceivers, etc., used to build a stable communication network and ensure reliable data transmission.

[0072] (2) Network security equipment: such as firewalls and intrusion detection systems, used to protect the security of network systems and prevent data leakage and network attacks.

[0073] 3) Data Layer

[0074] Including water data, operation data and alarm data. As the key bridge between the network layer and the application layer, the data layer is mainly responsible for data integration, storage and management. It not only covers the real-time water data (such as water level, flow, etc.) and hydropower station operation data (including switch quantity, pulse quantity, data quantity or analog quantity, etc.) collected from the perception layer, but also includes the alarm data after analysis and processing. These data are organized through an efficient database management system to ensure the accuracy, completeness and timeliness of the data, providing solid data support for upper-level applications.

[0075] 4) Application layer

[0076] The application layer is the highest layer of the power communication network system, responsible for visual display and intelligent analysis of data transmitted by the network layer. This layer mainly includes the following functions:

[0077] (1) Real-time monitoring: The various operating indicators of the hydropower station, such as water level, flow, equipment group status, etc., are displayed in real time through a graphical interface.

[0078] (2) Early warning system: Based on the results of data analysis, early warning is issued for possible abnormal situations so that timely measures can be taken to prevent accidents.

[0079] (3) Data analysis and optimization: In-depth analysis of historical data, identification of operating rules, and proposals for optimization to improve the operating efficiency of the hydropower station.

[0080] Through the layered design of perception layer, network layer, data layer and application layer, the system realizes the separation of data collection, transmission and processing, and improves the maintainability and scalability of the system. The network layer adopts redundant design and fault recovery mechanism to ensure the reliable transmission of data and the stable operation of the system. The application layer has intelligent early warning function, which can detect abnormal situations in time and issue alarms, thereby improving the safety performance of the hydropower station.

[0081] Through the construction of a 3D model of a hydropower station through 3D laser point cloud, data integration and application from physical hydropower station equipment, business, management decision-making and other entities to digital hydropower can be realized. Figure 3Provides a 1:1 modeling display of the 3D model in the application layer.

[0082] The present invention designs an intelligent monitoring and early warning system based on machine learning ( Figure 4 ). The system uses real-time data transmitted by the power communication network system to establish a data-driven model for training and learning, and realizes intelligent monitoring and abnormal warning of key operating parameters of hydropower stations. The core of the algorithm is to build a model that can automatically learn and identify abnormal patterns, thereby improving the accuracy and timeliness of early warning.

[0083] like Figure 5 As shown in the figure, the present invention establishes an early warning model based on machine learning through real-time data of hydropower stations. Key parameters such as water level, flow, rainfall, temperature, power output of generator sets, voltage, etc. are selected, and the parameters are used for feature extraction and model training. A support vector machine (SVM) and an integrated learning model are constructed to predict and alarm the target variables water level and flow. Through strategy optimization and dynamic threshold setting, the system intelligently identifies abnormal patterns and provides accurate early warning signals.

[0084] Information such as water level and flow is a core variable that directly affects the safe operation of power stations. By accurately predicting variables, the risks of dam breaches and flooded plants can be effectively reduced. Although traditional physical model-based methods can explain hydrological changes to a certain extent, they have great limitations in practical applications due to their complexity and uncertainty. Machine learning technology, by learning the patterns in historical data, can provide more accurate predictions of hydrological variables. Combined with multi-source data such as meteorology and the environment, machine learning models can perform better in short-term warnings and emergency response.

[0085] In the present invention, the accuracy of data and the rationality of features play a vital role in the performance of water level and flow prediction models. In order to ensure that the model can effectively predict future changes in water level and flow and achieve early warning, we introduced multi-source data, including key parameters such as water level, flow, rainfall, temperature, power output of generator sets, voltage, etc., and provided sufficient input information for support vector machine (SVM) and ensemble learning model through feature construction methods such as hysteresis features and rolling statistics.

[0086] Taking the operation data of Liguan Power Station as an example, the core water level and flow data are sampled once an hour to record the real-time status of the reservoir. At the same time, meteorological data such as rainfall and temperature are also included to more accurately predict hydrological changes. The introduction of equipment status parameters such as power output and voltage of the generator set enhances the model's perception of the equipment's operating status. In the feature construction, we extracted the lag characteristics, rolling average and rolling standard deviation of the parameters. The lag characteristics can help the model capture the impact of past water levels and flows on current changes, while the rolling statistics can smooth short-term fluctuations and reflect long-term trends. The time characteristics capture the periodic fluctuations in hydrological data through periodic information such as hours and seasons. In addition, for the two target variables of water level and flow, we use support vector machine (SVM) models for prediction respectively. In order to further improve the prediction performance, for each target variable, multiple SVM base models are trained using an ensemble learning method, and the prediction results of the same target variable are weighted averaged to form an ensemble support vector machine model.

[0087] In this way, the model can more accurately predict future water level and flow changes, providing a reliable basis for the early warning system. By learning from historical data and related features, the model showed good learning effects, further improving the accuracy and reliability of the early warning system. The study used recursive feature elimination (RFE) to select all features to ensure that the model only retains the features that contribute most to the prediction effect. Recursive feature elimination trains the model, evaluates the weight of each feature, and then recursively removes the features that contribute the least to the prediction. First, for the initial full feature set X = {x1, x2, ..., x n} to train the model and pass the model weight ω i To measure each feature x i The importance of a feature can be calculated by the absolute value of the weight (Formula 1). Then in each iteration, remove the feature x with the smallest weight j , as shown in Formula 2. This process continues until the number of remaining features reaches the preset threshold k, at which point the feature subset X k It is considered to be the feature set with the greatest predictive value for the model. Through the RFE method, the features that contribute most to water level prediction are effectively screened out, which improves the training speed and prediction accuracy of the model.

[0088] I(x i )=|ω i | (1)

[0089] X=X\{x j},where I(x j)=min(I(x1),I(x2),...,I(x n )) (2)

[0090] In the present invention, data standardization adopts z-score standardization (zero mean standardization), the purpose of which is to convert data of different features into a standard normal distribution with a mean of 0 and a standard deviation of 1. This ensures that each feature has the same magnitude, so that the machine learning model can treat different features fairly.

[0091] The z-score standardization calculation formula is as follows:

[0092]

[0093] Where: z is the normalized value; x is the original data; μ is the average value of the original data.

[0094] The standardization method improves the training efficiency of the model, facilitates the rapid convergence of the model, and improves the stability of the model. In addition, z-score standardization eliminates the scale differences between features, making the model more balanced in feature weight distribution, thereby further enhancing the model's predictive ability.

[0095] In the present invention, support vector machine (SVM) is selected as the basic model and combined with ensemble learning method to improve the accuracy of early warning. SVM is a widely used machine learning algorithm suitable for classification and regression problems. By finding an optimal hyperplane in high-dimensional space, SVM can classify data into different categories to distinguish the normal operation state and abnormal state of the hydropower station.

[0096] For a binary SVM, its decision function can be expressed as:

[0097]

[0098] Where: α i is the Lagrange multiplier; y i is the label of the sample (+1 or -1); K(x,x i ) is a kernel function used to map data into a high-dimensional space; b is a constant term; sign is a sign function and returns +1 or -1 according to the positive or negative value of the function.

[0099] In SVM, commonly used kernel functions include polynomial kernel, Gaussian radial basis kernel (RBF kernel), etc. Taking Gaussian radial basis kernel as an example, its formula is:

[0100]

[0101] In the formula, ||x i -x j|| 2 Represents vectors x and x i The square of the Euclidean distance between them; σ is the width parameter of the Gaussian kernel, which controls the radial range of the function.

[0102] Support vector regression (SVR) is an extension of support vector machine (SVM) and is used to solve regression problems. Unlike classification problems, which distinguish samples of different categories by finding the optimal hyperplane, support vector regression aims to find a regression line so that the distance between most data points and the line is within a tolerable range, which is called the ∈-insensitive loss function. For SVR, the continuous value is predicted by the decision function described by formula (6):

[0103]

[0104] Where: α i and is the Lagrange multiplier, which represents the weight of the support vector; x i is the support vector (key samples in the training data); x is the input feature vector (samples to be predicted); K(x i ,x) is the kernel function (as shown in formula (5)); b is the constant term.

[0105] The kernel function in SVR is the same as that in SVM, such as the RBF kernel, which is used to map the input to a high-dimensional feature space to handle nonlinear regression problems.

[0106] In order to improve the generalization ability and accuracy of the model, the present invention adopts an integrated learning method to strengthen the established water level and flow SVM base model through the bootstrap aggregating algorithm (Bagging for short). The algorithm is based on the bootstrap principle, and randomly samples with replacement from the original data set to generate multiple training subsets. According to the bootstrap principle in statistics, for a data set containing n samples, the probability that each sample is selected at least once is shown in formula (7), which means that when n is large enough, in each training subset, about 63.2% of the independent samples appear at least once, and the unselected samples constitute the validation set of the base model, which is used to evaluate the model performance and estimate the generalization error. For each such training subset, the present invention uses it to train an SVM base model, and sets the weight according to the performance of each base model on the validation set. Finally, as shown in formula (8), by weighted averaging the prediction results of multiple models, the variance of the model is effectively reduced, overfitting is reduced, and the accuracy of the model is improved.

[0107]

[0108] In the formula is the predicted value of the integrated model, M is the number of base models, h i (x) is the prediction result of each base model of the ith i is the weight of each i-th basis model.

[0109] After the model training is completed, the warning threshold is dynamically set according to the real-time performance feedback to trigger the warning in time when the model predicts an abnormal state. The basis for adjusting the threshold includes the balance between the false alarm rate and the missed alarm rate and the data collected in the actual operation. The sliding window technology is used to regularly update the threshold to ensure that the model can adapt to the changes in the data stream and maintain the accuracy of the warning mechanism. When the abnormal probability predicted by the model exceeds the dynamically set warning threshold, the algorithm will automatically trigger the warning, thereby responding to potential risks and anomalies more effectively. As shown in formula (9), the sliding window technology is used to set a sliding window with a fixed length of L, and the latest L data samples are analyzed in real time. The data set in the window is:

[0110] W t ={(y' t-L+1 ,y t-L+1 ),(y' t-L+2 ,y t-L+2 ),...,(y' t ,y t )} (9)

[0111] where y' i is the predicted value at time t, y i is the actual observation at time t, and L is the length of the sliding window. As time goes on, the sliding window is continuously updated, removing the earliest data points and adding the latest data points.

[0112] The performance index F1 score of the model is calculated in the sliding window. According to the change of the performance index in the sliding window, the warning threshold θ is dynamically adjusted as shown in Formula 10:

[0113]

[0114] Among them, θ t is the warning threshold at time t; α is the learning rate, which controls the step size of threshold adjustment; is the gradient of the F1 score to the threshold θ. By monitoring the changes in the F1 score, the warning threshold θ is adjusted to optimize the warning performance of the model and reduce false positives and negatives.

[0115] In the actual operating environment of a hydropower station, each parameter has large volatility and uncertainty. This uncertainty requires that the monitoring and early warning algorithm must have a high degree of adaptability and accuracy to flexibly respond to various complex and changing operating conditions. Traditional fixed parameter algorithms are prone to inaccurate early warnings or delayed responses when processing such dynamic data. Therefore, continuously improving the adaptability and accuracy of the algorithm so that it can better adapt to the actual operating conditions of the hydropower station has become the key to ensuring the safe and stable operation of the hydropower station.

[0116] Taking water level as an example, the trend over time shows the complexity and dynamics of the hydrological system ( Figure 6 ). The superposition of such seasonal changes and random fluctuations requires that relevant influencing factors be fully considered in the process of monitoring and early warning of hydropower stations to ensure the accuracy of monitoring and early warning, thereby further preventing the occurrence of events such as overtopping of dams, flooding of power plants, or unplanned shutdowns caused by sudden drops in flow. Therefore, after completing the initial construction and training of the model, it is very important to further improve the predictive ability and generalization effect of the model. The present invention adopts certain optimization strategies to achieve the tuning of single model hyperparameters and the optimization of integrated model weights, thereby improving the overall performance of the model.

[0117] First, the core hyperparameters of the support vector regression (SVR) model, the penalty coefficient C, the parameter γ of the kernel function, and the ∈ value of the ∈-insensitive loss function, are optimized by combining grid search and cross-validation. Through a series of parameter tuning steps, the performance of the support vector regression model in water level prediction has been significantly improved, the mean square error (MSE) has been significantly reduced, and the coefficient of determination (R 2 ) is significantly improved, and the model's ability to capture water level change trends is significantly enhanced.

[0118] Secondly, in order to further optimize the accuracy of the early warning system and enable the model to better identify and respond to potential abnormal situations, the F1 score is selected as the evaluation criterion of the integrated model, and its formula is as follows:

[0119]

[0120] Where: R pre It represents the precision, which measures the proportion of actual anomalies among all the cases predicted to be anomalies. rec Recall stands for recall, which measures the proportion of all actual anomalies that the model can detect.

[0121] The F1 score is the harmonic average of precision and recall, which comprehensively considers the model's ability to reduce false positives and false negatives. Since the main task of the early warning system is to determine whether an alarm needs to be triggered, the F1 score can effectively balance the system's accuracy and sensitivity in detecting anomalies, making it the most representative indicator when optimizing integrated models.

[0122] At the same time, in order to further improve the performance of model ensemble learning and achieve accurate early warning effects, the present invention adjusts the weight parameters in the integrated model through Bayesian optimization, thereby further improving the predictive ability of the model. By optimizing the weights of the entire integrated model, Bayesian optimization ensures that all data and prediction models in different directions can be reasonably integrated, thereby improving the performance of the overall early warning system. In the Bayesian optimization process, the F1 score is used as the optimization objective function. Bayesian optimization aims to find the optimal weight combination of each sub-model in the integrated model to maximize the F1 score. The objective function can be expressed as follows:

[0123] L(ω)=-F1(ω) (9)

[0124] Where: ω is the weight combination of each sub-model in the integrated model; F1(ω) is the F1 score under this weight combination. Bayesian optimization constructs a proxy model of the objective function L(ω), usually using a Gaussian process (GP) to approximate the distribution of the objective function, and uses the acquisition function to select the weight combination for the next evaluation. The goal of the expected improvement function is to maximize the possible F1 score, as shown in the following formula:

[0125] EI(ω)=Ε[max(F1(ω)-F1(ω + ),0)] (10)

[0126] Among them, F1(ω + ) is the current optimal F1 score. Bayesian optimization uses this function to continuously approach the global optimal solution and maximize the F1 score.

[0127] Taking the water level prediction model as an example, in the water level prediction model, the support vector regression model is used to train the historical water level data (the measured water level of the Liguan Hydropower Station in a certain period) to predict the water level changes. The model uses the RBF kernel function (radial basis function) to capture the nonlinear change trend in the water level data, and combines key features such as lag characteristics and rolling statistics to improve the model's ability to predict time series. In addition, the Bayesian optimization method is used to optimize the model parameters and integration strategies, which improves the ability of the early warning system to identify anomalies in actual operation.

[0128] The performance of the model is measured by the mean square error (MSE), the coefficient of determination (R 2), mean absolute percentage error (MAPE) and other indicators. Figure 7 As shown in the figure, on the test set, the model has a high accuracy in short-term predictions. The SVR model shows excellent prediction ability, and its accuracy gradually decreases as the prediction step increases. The MSE in the first half of the day is 3.601e -3 , indicating that the average error of the model in prediction is small; R 2 It reaches 0.988, which is close to 1, which means that the model can well explain the fluctuation of water level changes; and MAPE is only 0.0046%, showing the high accuracy of the model in water level prediction. Overall, the support vector regression (SVR) model effectively captures the trend of water level changes and builds a solid foundation for subsequent integrated models and early warning systems. By learning from historical water level data and related features, the model can accurately predict water level changes in the next half day, showing the good learning effect of the model.

[0129] The optimization improved the early warning system's ability to identify anomalies in actual operation and reduced false positives and missed positives. Ultimately, the Bayesian optimized system showed higher early warning accuracy, further proving the effectiveness of this optimization strategy in the hydropower station early warning system.

[0130] Embodiment 2 is an embodiment of the present invention, which provides a system for intelligent monitoring and early warning method of small hydropower stations in mountainous areas, including: a sensor module, a data preprocessing module, a support vector regression base model training and integration module, and a real-time monitoring and early warning trigger module;

[0131] The sensor module collects real-time data on the operation of the hydropower station;

[0132] The data preprocessing module performs preliminary cleaning and feature extraction on the collected data, using rolling mean and standard deviation calculations and recursive feature elimination to filter out important features;

[0133] The support vector regression base model training and integration module generates multiple SVR base models through a self-service aggregation algorithm, and uses weighted average to generate an overall prediction model based on an ensemble learning method;

[0134] The real-time monitoring and early warning triggering module uses an optimized integrated model to determine whether to trigger an early warning based on dynamic thresholds.

[0135] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0136] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0137] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0138] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0139] Embodiment 3: In this embodiment, in order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments. This embodiment experiments the existing traditional method and the method of this embodiment respectively.

[0140] The experimental object is Qianneng Company's small hydropower station and its power communication network system. Due to geographical characteristics, Qianneng Company's small hydropower stations are distributed throughout the province. Qianneng Company has 80 power stations, including 66 operating power stations. The age structure of power station operation staff is relatively old, and there is an urgent need to solve the problem of shortage of substation professionals through intelligent monitoring and early warning.

[0141] In the experiment, an intelligent monitoring and early warning algorithm based on machine learning was used to intelligently monitor the key operating parameters of the hydropower station, and to issue early warnings for abnormal situations through early warning algorithms. First, real-time data from the operation of the hydropower station was collected, including key parameters such as water level, flow, and equipment status. Then, this data was used to train a model that can automatically learn and identify abnormal patterns. After the model training is completed, the model is applied to the real-time monitoring system to monitor the operating status of the hydropower station in real time, and to issue early warnings for abnormal situations through early warning algorithms. Figure 8 As shown, when an alarm is triggered, the result is visualized and intelligently analyzed through the light screen (Note: the figure is an alarm diagram, not real data).

[0142] Table 1 records the warnings in different time periods, including the total number of warnings, the number of correct warnings, the number of false alarms, and the number of missed alarms. From the overall data, the warning accuracy of the intelligent monitoring and warning algorithm based on machine learning is 87.6%. This means that the vast majority of all warnings are correct, indicating the effectiveness of the algorithm in identifying abnormal situations.

[0143] Table 1 Warning accuracy

[0144]

[0145] like Fig. 9As shown in the figure, the warning accuracy of each time period fluctuates in different time periods. For example, from January 1 to January 10, 2024, the warning accuracy is 88.2%, while in the following period from February 1 to February 10, the accuracy drops to 86.7%. This fluctuation may be caused by a variety of factors, such as changes in data quality, model training, or the actual operating environment.

[0146] Despite the fluctuations, the warning accuracy rate in all time periods remained above 85%, which shows that the algorithm is stable and reliable under different conditions, and can effectively identify abnormal situations in various situations and issue warnings in a timely manner. The number of false positives reflects the sensitivity of the algorithm in identifying abnormal situations, while the number of false negatives reflects the coverage of the algorithm. As can be seen from Table 1, the number of false positives and false negatives in each time period is relatively low, which further proves the effectiveness and accuracy of the algorithm.

[0147] Aiming at the demand for intelligent monitoring and early warning of small hydropower stations in mountainous areas, the present invention develops and implements an intelligent monitoring and early warning technology based on an electric power communication network system. By constructing a layered (including perception layer, network layer, data layer and application layer) electric power communication network system architecture, the real-time collection, transmission, processing and analysis of hydropower station operation data are realized. At the same time, an intelligent monitoring and early warning algorithm based on machine learning is proposed. After model training and learning, real-time intelligent monitoring and abnormal early warning of key operating parameters of hydropower stations are realized. The algorithm is exhaustively designed in terms of data preprocessing, feature extraction, model selection and parameter optimization, which improves the prediction accuracy and generalization ability of the model, and significantly improves the accuracy and timeliness of early warning.

[0148] The experimental results show that the intelligent monitoring and early warning system can effectively monitor the key operating parameters of the hydropower station, such as the water level and flow, and issue early warnings in a timely manner under abnormal conditions. The early warning accuracy rate reaches more than 85%, proving the effectiveness and accuracy of the algorithm. At the same time, through visual display methods such as light-emitting screens, operation and maintenance personnel can intuitively understand the status of the hydropower station and take timely measures to deal with potential risks. Experimental data show that the system has played an important role in reducing the accident rate and improving operation and maintenance efficiency.

[0149] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. An intelligent monitoring and early warning method for small hydropower stations in mountainous areas, characterized in that: include: Collect hydropower station data through sensors and perform preprocessing to obtain key features; Multiple support vector regression base models are trained for multiple data variables of the hydropower station, and the bootstrap aggregation algorithm is used for ensemble learning, and multiple base models are trained by bootstrap sampling; Use Bayesian optimization methods to optimize model parameters and ensemble strategies; Dynamically adjust warning thresholds based on real-time performance feedback; Use sliding window technology to update thresholds regularly; The optimized model is deployed in the real-time monitoring system to conduct real-time monitoring and prediction of the operating status of the hydropower station.

2. The intelligent monitoring and early warning method for small hydropower stations in mountainous areas according to claim 1, characterized in that: The collection of hydropower station data through sensors includes real-time collection of water level, flow, rainfall, temperature, and power output and voltage of generator sets from the hydropower station, using the hydropower station data for feature extraction and model training, and predicting and alarming the target variables water level and flow.

3. The intelligent monitoring and early warning method for small hydropower stations in mountainous areas as claimed in claim 2, characterized in that: The preprocessing to obtain key features is to extract key features from hydropower station data, including rolling mean and rolling standard deviation, and use recursive feature elimination to perform feature selection on all features. Recursive feature elimination evaluates the weight of each feature through a training model, removes the feature with the smallest contribution to prediction, and performs z-score standardization on all features.

4. The intelligent monitoring and early warning method for small hydropower stations in mountainous areas as claimed in claim 3 is characterized by: The multiple support vector regression models trained for multiple data variables of the hydropower station include: Select support vector machine as the basic model and combine it with ensemble learning method to improve the accuracy of model prediction by finding a regression line so that the distance between most data points and the current function is within the range ∈; For the SVM regression model, the decision function is expressed as: K(x i ,x j )=exp(-γ||x i -x|| 2 ) Among them, α i and is the Lagrange multiplier; K(x i ,x) is the kernel function, b is the constant term, ||x i -x j || 2 represents the square of the Euclidean distance between vectors and , and γ is the parameter of the kernel function; Adopting the principle of bootstrap method, random sampling with replacement is performed from the original data to generate multiple training subsets and corresponding test sets. For each training subset, the unsampled samples are used as the test set of the current subset. For the data variables of each hydropower station, the corresponding support vector machine base model is trained and integrated learning is performed. The weights are set according to the performance of each base model on the corresponding test set, and the prediction results of the SVM base model of the same target variable are weighted averaged to generate the integrated support vector machine model of the target variable. After the training is completed, the integrated support vector machine model for each target variable is used to dynamically set the warning threshold based on real-time performance feedback. When an abnormal state is predicted, the warning is triggered in time. The threshold is updated regularly using sliding window technology. When the abnormal probability predicted by the integrated model of any target variable exceeds the dynamically set warning threshold, the warning system will automatically trigger the warning.

5. The intelligent monitoring and early warning method for small hydropower stations in mountainous areas as claimed in claim 4, characterized in that: The use of the Bayesian optimization method to optimize the model parameters and the integration strategy includes: For the target variables water level and flow, multiple support vector machine base models are trained, and for each target variable, multiple support vector machine base models are trained; The prediction result of the integrated model is obtained by weighted average, and its expression is: in, is the predicted value of the integrated model, M is the number of base models, h i (x) is the prediction result of each base model of the ith i is the weight of each i-th base model; The F1 score is selected as the evaluation criterion of the integrated model to optimize the accuracy of the early warning system. The expression is: Among them, R pre Represents precision, R rec Represents the recall rate.

6. The intelligent monitoring and early warning method for small hydropower stations in mountainous areas as claimed in claim 5, characterized in that: The dynamically adjusting the warning threshold according to the real-time performance feedback includes: The weight parameters in the integrated model are adjusted through Bayesian optimization. The F1 score is used as the optimization objective function to find the optimal weight combination of each sub-model in the integrated model to maximize the F1 score. The objective function expression is: L(ω)=-F1(ω) Where ω is the weight combination of each sub-model in the integrated model; F1(ω) is the F1 score under this weight combination; Bayesian optimization constructs a proxy model of the objective function L(ω), uses a Gaussian process to approximate the distribution of the objective function, and uses the expected improvement (EI) to obtain the weighted combination of the function selection for the next step of evaluation. The goal of the expected improvement function is to maximize the F1 score, which is expressed as: EI(ω)=Ε[max(F1(ω)-F1(ω + ),0)] Among them, F1(ω + ) is the currently known best F1 score. Bayesian optimization uses this function to continuously approach the global optimal solution and maximize the F1 score.

7. The intelligent monitoring and early warning method for small hydropower stations in mountainous areas as claimed in claim 6, characterized in that: The dynamic adjustment of the warning threshold also includes: Using sliding window technology, a sliding window with a fixed length of L is set to perform real-time analysis on the latest L data samples. The data set in the window is: W t ={(and' t-L+1 ,and t-L+1 ),(and' t-L+2 ,and t-L+2 ),...,(and' t ,and t )} Among them, y' i is the predicted value at time t, y i is the actual observation at time t, and L is the length of the sliding window. As time goes on, the sliding window is continuously updated, removing the earliest data points and adding the latest data points; The performance index F1 score of the model is calculated in the sliding window. According to the change of the performance index in the sliding window, the warning threshold θ is dynamically adjusted. The expression is: Among them, θ t is the warning threshold at time t; α is the learning rate, which controls the step size of threshold adjustment; It is the gradient of the F1 score to the threshold θ. By monitoring the changes in the F1 score, the warning threshold θ is adjusted. Finally, use the updated alarm threshold θ t+1 Determine the new prediction result y' t+1 Whether to trigger an early warning.

8. A system using the intelligent monitoring and early warning method for small hydropower stations in mountainous areas as claimed in any one of claims 1 to 7, characterized in that: It includes sensor module, data preprocessing module, support vector regression model training and integration module, and real-time monitoring and early warning trigger module; The sensor module collects real-time data of the operation of the hydropower station; The data preprocessing module performs preliminary cleaning and feature extraction on the collected data, using rolling mean and standard deviation calculations and recursive feature elimination to screen out important features; The support vector regression base model training and integration module generates multiple SVR base models through a self-service aggregation algorithm, and generates an overall prediction model by weighted average based on an ensemble learning method; The real-time monitoring and early warning triggering module uses an optimized integrated model to determine whether to trigger an early warning based on a dynamic threshold.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent monitoring and early warning method for small hydropower stations in mountainous areas described in any one of claims 1 to 7 are implemented.

Images