Hydropower station icing power generation prediction method based on integrated learning and related equipment

Abstract

The invention provides a hydropower station icing power generation prediction method based on integrated learning and related equipment, and the method comprises the steps: obtaining multi-source data of a hydropower station, the multi-source data comprising time feature data, spatial feature data and operation feature data; screening the time characteristic data based on icing factor correlation to obtain target characteristic data; based on a trained icing parameter model, determining icing parameters of a target time period for the target time feature data and the spatial feature data; and predicting power generation parameters of the target time period based on the icing parameters and the operation characteristic data.

Description

Technical Field

[0001] This disclosure relates to the power industry, and in particular to a method and related equipment for predicting icing power generation in hydropower stations based on ensemble learning. Background Technology

[0002] Ice accumulation poses multiple threats to the safety of hydropower stations: Imbalance in turbine blades caused by ice accumulation can lead to increased vibration, reduced turbine efficiency (typically by 10-30%), and decreased effective flow at the intake, potentially jeopardizing normal unit operation. Ice accumulation on critical structures such as spillways and intakes can cause structural damage. Important components like gates and hoists may malfunction due to ice accumulation. From a power grid stability perspective, during peak winter electricity demand, the impact of ice accumulation on power generation capacity directly threatens power supply reliability, weakens the peak-shaving performance of hydropower stations, and reduces the grid's regulation capacity.

[0003] Meanwhile, icing phenomena are sudden and complex, making it difficult for traditional manual observation methods to provide sufficient foresight. This increases the difficulty of decision-making between icing prevention, de-icing, and safe power generation. In particular, most of my country's hydropower stations are located in high-altitude areas where winter temperatures are low, making major river basins such as the Yangtze and Yellow Rivers prone to icing. Global climate change has further exacerbated the frequency of extreme weather events, making the icing problem even more prominent. Therefore, accurate icing forecasting is crucial for predicting the power generation capacity of hydropower stations, planning icing prevention measures in advance, reducing economic losses, and ensuring the safe and stable operation of the power grid.

[0004] Currently, regarding the reliability of predictions, icing formation is influenced by multiple factors such as temperature, humidity, wind speed, and water flow velocity, and these factors have complex interactions. Traditional single prediction models struggle to fully capture the complexity and nonlinear characteristics of icing at hydropower stations. Most icing prediction methods lack sufficient fusion processing of multi-source heterogeneous data from different sources (such as meteorology, hydrology, and photography), resulting in low data utilization efficiency and insufficient feature extraction. The lack of a comprehensive analysis of the relationship between icing and power generation capacity makes it difficult for prediction results to directly guide actual operational decisions at power plants. Summary of the Invention

[0005] In view of this, the purpose of this disclosure is to propose a method and related equipment for predicting hydropower generation under icing conditions based on ensemble learning.

[0006] The first aspect of this disclosure provides a method for predicting icing power generation in hydropower stations based on ensemble learning, including: Acquire multi-source data about the hydropower station, including time feature data, spatial feature data, and operational feature data; The time feature data is filtered based on the correlation of icing factors to obtain target feature data; Based on the trained icing parameter model, the icing parameters for the target time period are determined using the target temporal feature data and the target spatial feature data. Based on the icing parameters and the operational characteristic data, the power generation parameters for the target time period are predicted.

[0007] A second aspect of this disclosure improves a hydropower station icing power generation prediction device based on ensemble learning, comprising: The acquisition module is used to acquire multi-source data about the hydropower station, including time feature data, spatial feature data, and operational feature data. The feature filtering module is used to filter the time feature data based on the correlation of icing factors to obtain target feature data; The icing prediction module is used to determine the icing parameters for a target time period based on the target temporal feature data and the spatial feature data, using a trained icing parameter model. The power generation prediction module is used to predict the power generation parameters for the target time period based on the icing parameters and the operating characteristic data.

[0008] A third aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in the first aspect.

[0009] In a fourth aspect, this disclosure provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method described in the first aspect.

[0010] A fifth aspect of this disclosure provides a computer program product including computer program instructions that, when executed on a computer, cause the computer to perform the method described in the first aspect.

[0011] As described above, this disclosure provides a method and related equipment for predicting hydropower generation under icing conditions based on ensemble learning. It acquires multi-source data from the hydropower station, including temporal, spatial, and operational characteristics. Then, it filters temporal characteristic data based on the correlation of icing factors to obtain target characteristic data. Next, it uses a trained icing parameter model combined with the target temporal and spatial characteristic data to determine the icing parameters for the target time period. Finally, it predicts the power generation parameters for the target time period based on the icing parameters and operational characteristic data. This method effectively determines icing parameters, thereby achieving accurate prediction of power generation parameters and providing a reliable basis for hydropower station operation planning and resource allocation. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in this disclosure or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of an ensemble learning-based hydropower icing power generation prediction architecture according to an embodiment of this disclosure.

[0014] Figure 2 This is a schematic diagram of the structure of an exemplary electronic device according to an embodiment of the present disclosure.

[0015] Figure 3 This is a schematic flowchart illustrating the hydropower icing power generation prediction method based on ensemble learning, as described in this disclosure.

[0016] Figure 4 This is a schematic diagram illustrating the principle of hydropower station icing power generation prediction based on ensemble learning, according to an embodiment of this disclosure.

[0017] Figure 5 This is a schematic diagram of the icing parameter model according to an embodiment of the present disclosure.

[0018] Figure 6 This is a schematic diagram of a hydropower station icing power generation prediction device based on ensemble learning, according to an embodiment of this disclosure. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0020] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0021] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0022] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.

[0023] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0024] Figure 1 A schematic diagram of an ensemble learning-based hydropower icing power generation prediction architecture according to an embodiment of this disclosure is shown. (Reference) Figure 1 The hydropower station icing power generation prediction architecture 100 based on ensemble learning may include a server 110, a terminal 120, and a network 130 providing a communication link. The server 110 and the terminal 120 can be connected via a wired or wireless network 130. The server 110 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, security services, and CDN.

[0025] Terminal 120 can be implemented in hardware or software. For example, when terminal 120 is implemented in hardware, it can be various electronic devices with a display screen and support page display, including but not limited to smartphones, tablets, e-book readers, laptops, and desktop computers. When terminal 120 is implemented in software, it can be installed in the electronic devices listed above; it can be implemented as multiple software programs or software modules (e.g., software programs or software modules used to provide distributed services) or as a single software program or software module, without specific limitations.

[0026] It should be noted that the hydropower station icing power generation prediction method based on ensemble learning provided in this disclosure embodiment can be executed by the terminal 120 or by the server 110. It should be understood that... Figure 1 The number of terminals, networks, and servers shown is for illustrative purposes only and is not intended to be a limitation. Any number of terminals, networks, and servers can be included depending on implementation needs.

[0027] Figure 2 A schematic diagram of the hardware structure of an exemplary electronic device 200 provided in an embodiment of this disclosure is shown. Figure 2 As shown, the electronic device 200 may include: a processor 202, a memory 204, a network module 206, a peripheral interface 208, and a bus 210. The processor 202, memory 204, network module 206, and peripheral interface 208 are interconnected within the electronic device 200 via the bus 210.

[0028] Processor 202 may be a Central Processing Unit (CPU), a Neural Processing Unit (NPU), a Microcontroller (MCU), a programmable logic device, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits. Processor 202 can be used to perform functions related to the techniques described in this disclosure. In some embodiments, processor 202 may also include multiple processors integrated as a single logic component. For example, such as... Figure 2 As shown, processor 202 may include multiple processors 202a, 202b and 202c.

[0029] Memory 204 can be configured to store data (e.g., instructions, computer code, etc.). Figure 2 As shown, the data stored in memory 204 may include program instructions (e.g., program instructions for implementing the ensemble learning-based hydropower station icing power generation prediction method of this disclosure) and data to be processed (e.g., the memory may store configuration files of other modules, etc.). Processor 202 may also access the program instructions and data stored in memory 204 and execute the program instructions to operate on the data to be processed. Memory 204 may include volatile storage devices or non-volatile storage devices. In some embodiments, memory 204 may include random access memory (RAM), read-only memory (ROM), optical disk, magnetic disk, hard disk, solid-state drive (SSD), flash memory, memory stick, etc.

[0030] Network module 206 can be configured to provide communication with other external devices to electronic device 200 via a network. This network can be any wired or wireless network capable of transmitting and receiving data. For example, the network can be a wired network, a local wireless network (e.g., Bluetooth, WiFi, Near Field Communication (NFC), etc.), a cellular network, the Internet, or a combination thereof. It is understood that the type of network is not limited to the specific examples described above. In some embodiments, network module 206 may include any combination of any number of network interface controllers (NICs), radio frequency modules, transceivers, modems, routers, gateways, adapters, cellular network chips, etc.

[0031] The peripheral interface 208 can be configured to connect the electronic device 200 to one or more peripheral devices to enable information input and output. For example, peripheral devices may include input devices such as keyboards, mice, touchpads, touch screens, microphones, and various sensors, as well as output devices such as displays, speakers, vibrators, and indicator lights.

[0032] Bus 210 can be configured to transmit information between various components of electronic device 200 (e.g., processor 202, memory 204, network module 206, and peripheral interface 208), such as internal buses (e.g., processor-memory bus), external buses (USB port, PCI-E bus), etc.

[0033] It should be noted that although the architecture of the above-described electronic device 200 only shows the processor 202, memory 204, network module 206, peripheral interface 208, and bus 210, in specific implementations, the architecture of the electronic device 200 may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the architecture of the above-described electronic device 200 may only include the components necessary for implementing the embodiments of this disclosure, and does not necessarily include all the components shown in the figures.

[0034] In related technologies, icing poses multifaceted threats to the safety of hydropower stations. For example, the imbalance caused by icing on turbine blades can lead to increased vibration, reduced turbine efficiency (typically by 10-30%), and decreased effective flow at the intake, potentially jeopardizing normal unit operation. Icing on critical structures such as spillways and intakes can cause structural damage. Important components like gates and hoists may malfunction due to icing. From a power grid stability perspective, during peak winter electricity demand, icing directly threatens power generation reliability and weakens the peak-shaving performance of hydropower stations, reducing the grid's regulation capacity. Furthermore, icing is sudden and complex, making traditional manual observation methods insufficient for prediction, increasing the difficulty of decision-making between icing prevention, de-icing, and safe power generation. In particular, most hydropower stations in my country are located in high-altitude areas with low winter temperatures, making major river basins like the Yangtze and Yellow Rivers prone to icing. Global climate change further exacerbates the frequency of extreme weather events, making the icing problem even more prominent. Therefore, accurate icing prediction is of great significance for predicting the power generation capacity of hydropower stations, planning anti-icing measures in advance, reducing economic losses, and ensuring the safe and stable operation of the power grid.

[0035] Currently, regarding the reliability of predictions, icing formation is influenced by multiple factors such as temperature, humidity, wind speed, and water flow velocity. These factors have complex interactions, making it difficult for traditional single prediction models to fully capture the complexity and nonlinear characteristics of icing at hydropower stations. Most icing prediction methods lack sufficient fusion processing of multi-source heterogeneous data from different sources (such as meteorology, hydrology, and photography), resulting in low data utilization efficiency and insufficient feature extraction. Furthermore, traditional methods typically focus only on a specific aspect of icing, such as whether icing occurs or the thickness of the icing, lacking a comprehensive analysis of the relationship between icing and power generation capacity. This makes it difficult for prediction results to directly guide actual operational decisions at power plants.

[0036] River freezing and ice jams can cause severe flooding, property damage, and loss of life. Floods caused by freezing and ice jams disrupt critical transportation routes and impact energy production at hydroelectric power plants, resulting in significant economic losses. Freezing and ice jams place immense stress on existing hydroelectric infrastructure, including dams, powerhouses, and flood control facilities. Freezing and ice jams also severely impact the environment, affecting aquatic habitats and ecosystems. Current research suggests several approaches for developing, implementing, and operating ice cover forecasts, including statistical, empirical, numerical, machine learning, and hybrid methods. Statistical methods utilize historical data to establish correlations between hydrometeorological variables such as temperature, precipitation, and river ice conditions. For example, regression models can be used to predict the probability of ice cover occurrence based on past observations of these variables. One advantage of empirical methods is their relative ease of implementation, requiring no detailed understanding of the underlying physical processes involved in ice cover formation. However, they may be ineffective if there are significant changes in climate or other factors influencing river conditions, and they are also difficult to transfer from one location to another. Empirical methods use analytical equations to calculate the hydraulic conditions of river ice. Algebraic solutions are often sufficient to solve these equations, while numerical methods (discussed in the next paragraph) require complex algorithms to execute. Empirical models, which do not need to be location-specific, have been used to transfer river ice parameters along rivers to infer the scale of ice blockages from locations with to without measurements; however, empirical methods require long periods of data monitoring and their predictive accuracy is relatively poor. Numerical models use mathematical equations to simulate the physical processes of ice layer formation and ice sheet formation. Examples of numerical models include RIVER2D, HEC-RAS, and RIVICE. These models can be used to predict changes in ice conditions over time, taking into account factors such as river topography, hydraulics, meteorology, and river ice conditions. Compared to empirical methods, numerical modeling requires more computational resources, so its complexity must be reduced to be suitable for rapid forecasting. Machine learning uses algorithms learned from historical data to predict future ice blockage events. For example, decision tree or random forest models can be used to predict the probability of ice blockages based on past observations of hydro-meteorological variables and other factors such as river geometry and tributary inflow. Other methods include the K-nearest neighbor algorithm, neural networks, and fuzzy logic models. Machine learning methods can be very accurate if there is a large amount of high-quality data available to train the model, but they may perform poorly if the relationships between variables are complex or if there are significant changes in climate or other factors affecting river conditions.

[0037] Some related technologies construct a temperature resistance value prediction framework based on AdaBoost ensemble learning. Based on temporal features, the correlation between each feature factor and the current temperature resistance value is analyzed using the maximum information coefficient, identifying features strongly correlated with the temperature resistance value. Combining the unit's temperature resistance value data and the strongly correlated features, the individual learners of the temperature resistance value prediction framework are trained, and the sample weights are adjusted through an iterative process to obtain the final strong learner. The temperature resistance value is then predicted using the strong learner to obtain the prediction result. However, this method does not consider using spatial features, i.e., it uses image data to train a weak learner of AdaBoost. In this invention, spatial features are considered, and a different model is used for ensemble prediction.

[0038] Some related technologies predict hydropower station inflow by constructing a monthly inflow prediction module based on a self-made probability distribution model, a daily inflow LSTM prediction module driven by multi-dimensional feature data, and an hourly inflow prediction module. The difference between these and the present invention lies in the selection of different feature data and the use of different prediction models. Furthermore, the present invention predicts power generation by correlating icing amount with other factors, while similar solutions only predict inflow.

[0039] Some related technologies input the real-time state of the gate into a reinforcement learning model to obtain the gate opening adjustment amount, and then control the hydropower station gate based on the opening adjustment amount at the current moment. The real-time state includes the gate opening and various environmental data, including rainfall, river flow, and upstream and downstream water levels. The difference between this and the present invention is that it does not predict the amount of ice accumulation, but rather the inflow volume; it does not consider environmental data such as temperature, nor spatial characteristics; and the learner used is also different.

[0040] Some related technologies, through monitoring modules including temperature monitoring units, humidity monitoring units, wind speed monitoring units, and icing monitoring units, can monitor various environmental parameters in real time, such as temperature, humidity, and pressure, and transmit this data to a detection module. The detection module then further analyzes this data to determine if any anomalies exist. The difference between this method and the one in this invention is that this method does not predict the degree of icing, nor does it utilize artificial intelligence ensemble learning methods.

[0041] It is evident that in data acquisition and feature extraction, traditional methods often rely solely on a single data source, such as water flow, for prediction. This results in incomplete information and an inability to fully perceive the probability of icing at hydropower stations. Furthermore, the use of manual experience in feature selection often overlooks potentially important features, leading to subjectivity in feature selection and limited prediction accuracy due to insufficient information dimensions. Regarding data processing that integrates spatiotemporal information and improves quality, existing methods and prediction models often fail to simultaneously process spatial and temporal features. While advancements in communication and sensing technologies have enabled hydropower stations to collect multi-source data such as satellite remote sensing, drone aerial photography, and on-site monitoring, some methods still rely on single-dimensional data without combining and learning from multi-dimensional data. In constructing power generation models based on icing volume, most existing prediction methods are basin-wide, establishing only a model for icing volume without establishing a mapping relationship between icing volume and power generation. This prevents a direct assessment of the impact of icing on hydropower station power generation capacity and hinders effective decision support for hydropower station operation scheduling and economic benefit optimization.

[0042] In view of this, this disclosure proposes a method and related equipment for predicting hydropower generation under icing conditions based on ensemble learning. The method involves acquiring multi-source data from the hydropower station, including temporal, spatial, and operational characteristics. Then, based on the correlation of icing factors, temporal characteristic data is filtered to obtain target characteristic data. Next, a trained icing parameter model is used in conjunction with the target temporal and spatial characteristic data to determine the icing parameters for the target time period. Finally, based on the icing parameters and operational characteristic data, the power generation parameters for the target time period are predicted. This method can effectively determine icing parameters, thereby achieving accurate prediction of power generation parameters and providing a reliable basis for hydropower station operation planning and resource allocation.

[0043] First, a spatial feature dataset of the hydropower station can be collected (including multi-scale, multi-angle spatial data such as satellite remote sensing images, UAV aerial images, and on-site monitoring videos). Simultaneously, a temporal feature dataset can be collected, including data from hydrological sensors (such as water temperature and water level), meteorological stations (such as air temperature, humidity, and wind speed), and power station operational features (including gate opening, power generation, and power output). Then, based on the temporal feature factors, wavelet multiple correlation coefficient analysis is used to analyze the correlation between historically collected data for each feature factor and historical ice accumulation data. A correlation threshold is set to filter features, selecting the feature factor with the highest correlation as the new temporal feature factor. Next, a dual-stream neural network hybrid architecture is used to predict the ice accumulation data for the next time step. The spatial feature branch of the dual-stream neural network is a convolutional neural network (CNN) used to extract spatial features and texture information from the image, while the temporal feature branch is a temporal convolutional network (TCN) used to capture long-term dependencies and periodic changes in the temporal data. Specifically, the spatial feature data is input into the CNN for training, and the temporal feature factors are input into the TCN for training, yielding predicted ice accumulation area (m²) and ice accumulation thickness (mm), respectively. A feature fusion layer is constructed to integrate the prediction results from the two parts. Then, a multi-task loss function is used to optimize the prediction results of the two-stream neural network hybrid architecture. An optimizer dynamically adjusts the weights to reduce the loss of multi-dimensional prediction results, improving the stability and accuracy of the predictions, thus obtaining the final prediction result. Finally, using the ice accretion prediction dataset as features and the power generation dataset as labels, a random forest algorithm is used to establish a mapping relationship between ice accretion and power generation to predict power generation, thereby building the final prediction model.

[0044] Ice blockage: Ice blockage refers to the accumulation of ice in rivers or hydraulic structures due to water freezing under cold weather conditions. It mainly involves two formation mechanisms: first, ice directly freezes on the water surface and then breaks and accumulates; second, suspended ice crystals in the water adhere and aggregate. When these ice blocks accumulate in narrow sections of water flow (such as inlets or sluice gates), they cause cross-sectional contraction, leading to hydraulic effects such as rising water levels and changes in flow velocity. In severe cases, this can cause overflows, equipment damage, and other safety accidents.

[0045] Multi-task loss function: A loss function that simultaneously optimizes the prediction of icing area and thickness.

[0046] Random Forest: An ensemble learning method that makes predictions by constructing multiple decision trees and taking the average of their predictions.

[0047] Adam optimizer: An optimization algorithm with an adaptive learning rate for training neural networks.

[0048] See Figure 3 , Figure 3 A schematic flowchart of an ensemble learning-based hydropower station icing power generation prediction method according to an embodiment of the present disclosure is shown. The ensemble learning-based hydropower station icing power generation prediction method according to an embodiment of the present disclosure can be deployed on a server. Figure 3 In the above, the hydropower station icing power generation prediction method 300 based on ensemble learning can further include the following steps.

[0049] In step S310, multi-source data about the hydropower station is acquired, including time feature data, spatial feature data, and operational feature data.

[0050] Temporal characteristic data refers to hydropower station data closely related to the time dimension, including various information at different time scales. For example, real-time operating parameters of equipment measured in seconds or minutes, such as instantaneous values ​​like generator speed and temperature; power generation and reservoir water level changes measured in hours or days; and the division of high-water and low-water seasons measured in months or years, as well as annual power generation statistics. This data reflects the changing patterns of the hydropower station's operating status over time. Spatial characteristic data mainly describes the spatial distribution and characteristics of the hydropower station's components and surrounding environment. This includes the geographical coordinates of facilities such as the dam, powerhouse, and generating units; topographic data of the reservoir, such as reservoir capacity curves and contour lines; and the distribution of surrounding rivers and geological structure information. Spatial characteristic data helps analyze the operating conditions and potential impacts of the hydropower station from a geospatial perspective. Operational characteristic data directly reflects the operating status and performance of the hydropower station. This includes unit operating parameters such as power generation, voltage, current, and power factor; equipment status data such as vibration, swing, and temperature of the turbine and generator; reservoir operating data such as water level, flow rate, inflow, and outflow; and auxiliary equipment operating data such as gate opening, ventilation equipment wind speed, and drainage equipment flow rate. This data comprehensively reflects the operating conditions and efficiency of the hydropower station.

[0051] At the data acquisition level, for time-related data, high-precision sensors can be used to collect equipment operating parameters in real time, and the Supervisory Control and Data Acquisition (SCADA) system can record data at set time intervals. Simultaneously, real-time meteorological data from weather stations is obtained to acquire information at different time scales. For spatial data, Geographic Information System (GIS) technology is used to integrate hydropower station design drawings, topographic mapping data, etc., to establish a three-dimensional geographic model, marking the precise location and spatial relationships of each facility. Remote sensing technology is used to periodically acquire topographic images of the reservoir and surrounding area to update the spatial data. For operational data, various sensors are installed at key locations such as generating units, reservoirs, and auxiliary equipment to collect operating parameters in real time, and the data is transmitted to the data center via industrial Ethernet or wireless communication networks.

[0052] At the data processing and analysis level, the collected multi-source data is cleaned, transformed, and stored to ensure accuracy and consistency. Data mining and machine learning techniques are used to analyze temporal characteristic data, uncovering trends and periodic patterns in equipment operating parameters over time to predict equipment failures and performance degradation. Spatial analysis functions are utilized, combined with spatial characteristic data, to assess the impact of hydropower station construction on the surrounding environment and the potential threats of geological disasters to hydropower station safety. Through correlation analysis of operational characteristic data, unit operation strategies are optimized to improve power generation efficiency and achieve safe, stable, and efficient operation of the hydropower station.

[0053] By comprehensively and accurately acquiring temporal, spatial, and operational characteristic data of hydropower stations, we can gain a clear understanding of how equipment operating status changes over time, identify potential faults early, schedule maintenance and repairs promptly, reduce equipment downtime, and improve equipment reliability and availability. Spatial characteristic data provides intuitive geospatial information, which helps optimize hydropower station layout, assess geological disaster risks, and ensure the safe operation of hydropower stations. Real-time monitoring and analysis of operational characteristic data enables timely understanding of the hydropower station's operating conditions, adjustment of unit operating parameters based on actual conditions, optimization of power generation dispatching schemes, and improvement of hydropower utilization efficiency and power generation.

[0054] Specifically, see Figure 4 , Figure 4 A schematic diagram of a hydropower station icing power generation prediction method based on ensemble learning according to an embodiment of this disclosure is shown. First, multi-source data can be collected and integrated. The data sources mainly include three categories: ① Temporal characteristic factor data collection: Parameters such as water temperature, water level, and flow velocity are obtained through hydrological sensors, and data such as air temperature, humidity, precipitation, wind speed, and solar radiation intensity are collected from meteorological stations, with sampling at one data point per hour; ② Spatial characteristic factor data acquisition: Image data is obtained using satellites, drone aerial photography, and video monitoring systems, with sampling at one image per source per hour; ③ Power station operating parameter recording: Operating characteristic data such as gate opening, maximum power generation, and daily cumulative power generation are collected.

[0055] The amount of icing can include the icing area (m²). 2Two indicators are measured: (1) and (2) ice thickness (mm). Maximum generating power (kW): This represents the maximum electricity output of a hydroelectric generator at a given moment, which may be affected by factors such as head, flow rate, and equipment efficiency. Daily cumulative power generation (kWh): The total power generated within a specific time period on a given day. Water temperature (°C): Generally measured at key locations such as the inlet and tailrace, and affected by air temperature changes, water depth, flow velocity, and solar radiation. Water level: Water level refers to the vertical height of the water surface relative to a reference surface (usually mean sea level or other agreed-upon reference surface), measured in meters (m). In hydroelectric power stations, the main focus is on the upstream reservoir water level and the downstream tailrace water level; the difference is called the head, a key parameter determining the efficiency of hydropower conversion. Water level measurement typically uses equipment such as water level gauges and pressure sensors, and corrections need to be made considering seasonal variations and air pressure effects. Flow velocity: Flow velocity is a physical quantity describing the speed of water flow, representing the distance a water particle travels per unit time, usually measured in meters per second (m / s). In hydropower stations, flow velocity is a three-dimensional vector containing horizontal and vertical components. Its magnitude and direction vary with spatial location and time. Measurement is commonly performed using equipment such as acoustic Doppler current meters (ADCP), which have a significant impact on power generation efficiency and equipment safety.

[0056] First, data on relevant factors were obtained from various sensors, weather stations, satellite images, and other sources at the hydropower station, according to the sampling time. Let the time-related factor data be... It is in the form of a two-dimensional matrix. , where n is the number of time sampling points (one data point per hour). It is the time feature dimension, and the time feature factor matrix is ​​represented as follows: ,Include: Hydrological characteristic parameters: [upstream water temperature, downstream water temperature, upstream water level, downstream water level, upstream flow velocity, downstream flow velocity, tributary water temperature, tributary water level, tributary flow velocity], constructing a sub-matrix as follows: ,in Represents the dimensions of hydrological characteristics.

[0057] Meteorological parameters: [temperature, air humidity, surface humidity, precipitation, wind speed, solar radiation intensity], constructing a submatrix as follows: .in Represents the dimensions of hydrological characteristics.

[0058] Additionally, there is a special set of time-related factors used to construct the icing amount matrix, represented as follows: ,in Indicates the area covered by ice (m²) 2 ), Indicates the thickness of the ice layer (mm).

[0059] Spatial feature factor data tensor is represented as a five-dimensional tensor. , where n is the number of time sampling points (one data point per hour). It is a spatial feature dimension. It is the image height. It is the image width. Image channel count (3 for RGB, 1 for grayscale).

[0060] The power plant's operational characteristic factors data are represented as a two-dimensional matrix. , where n is the number of time sampling points (one data point per hour). It is an operational feature dimension, including: [gate opening, maximum power generation, daily cumulative power generation], and the matrix can be represented as: .

[0061] It should be understood that all the above data requires synchronized sampling intervals, that is, all data are sampled in hourly alignment with unified timestamps, and missing or poor-quality data is permissible.

[0062] In step S320, the time feature data is filtered based on the correlation of icing factors to obtain target feature data.

[0063] This involves collecting long-term time-series data on hydropower stations, encompassing equipment operating parameters (such as changes in unit temperature and vibration frequency over time), environmental meteorological data (such as temperature, humidity, and wind speed records over time), and reservoir water levels under different seasons and weather conditions. Statistical methods, such as correlation and regression analysis, are used to calculate the correlation coefficients between icing conditions (data on icing thickness and duration can be obtained through specialized icing monitoring equipment) and various time-series data, identifying which time-series data show a significant correlation with icing factors. Based on the strength of the correlation, reasonable screening thresholds are set, and time-series data with a correlation higher than the threshold are selected as target data. This method of filtering time-series data based on the correlation with icing factors allows for precise focus on key data closely related to icing, eliminating a large amount of irrelevant or weakly correlated data. This improves the efficiency and relevance of data analysis, enabling more accurate identification of the changing patterns of unit temperature, vibration, and other parameters under icing conditions, thereby enhancing the operational reliability and power generation efficiency of hydropower stations under severe weather conditions.

[0064] Specifically, in the feature selection stage, wavelet multiple correlation coefficient analysis can be used to process and filter historical time feature data. First, wavelet transform is applied to the time feature data and historical icing amount data to decompose them into different frequency components. Then, the correlation coefficient matrix between each frequency component is calculated, and a correlation threshold is set (usually 0.6 or 0.7). Feature factors significantly related to icing amount are then selected, reducing data redundancy and improving model training efficiency. Wavelet transform is a signal processing method that decomposes a signal into different frequency components for analyzing multi-scale features of time series. Wavelet multiple correlation coefficient is an index used to measure the correlation between different frequency components, and in this invention, it is used for feature selection.

[0065] In some embodiments, the time feature data is filtered based on the correlation of icing factors to obtain target time feature data, including: Wavelet transform analysis was performed on historical time characteristic data and historical icing data to obtain historical time wavelet coefficients and historical icing factor wavelet coefficients at different frequency levels. The correlation between the historical time wavelet coefficients and the historical icing factor wavelet coefficients is determined to obtain the icing factor correlation. Based on the historical time feature data corresponding to the historical time wavelet coefficients whose correlation with the icing factors is greater than or equal to a preset threshold, a target feature matrix is ​​determined. The target time feature data is obtained by filtering the time feature data based on the target feature matrix.

[0066] The process involves several steps. First, wavelet transform analysis is performed on historical time feature data and historical icing data to decompose them into different frequency layers, yielding corresponding wavelet coefficients. Next, the correlation between historical time wavelet coefficients and historical icing factor wavelet coefficients is calculated to clarify the correlation of icing factors. Then, based on preset thresholds, historical time feature data corresponding to those that meet the correlation criteria with icing factors are selected from the historical time wavelet coefficients to construct a target feature matrix. Finally, using this matrix as the selection criterion, target time feature data is obtained from the current time feature data. This method effectively extracts time feature data closely related to icing, removes irrelevant interference, improves the accuracy of data analysis, and provides a reliable basis for hydropower stations to address icing issues and formulate scientific operation and maintenance strategies, ensuring the safe and stable operation of hydropower stations.

[0067] In some embodiments, determining the correlation between the historical time wavelet coefficients and the historical icing factor wavelet coefficients to obtain the icing factor correlation includes: ; in, This represents the relationship between the i-th historical time feature and the ice cover amount under the j-th level decomposition. The correlation of icing factors This represents the wavelet coefficients at time point t under the j-th level decomposition of the i-th historical time feature. Indicates icing data The wavelet coefficients at time point t in the j-th layer and These represent the mean of the corresponding sequences.

[0068] The correlation between historical time wavelet coefficients and historical icing factor wavelet coefficients can be calculated based on the Pearson correlation coefficient principle. This correlation performance accurately reflects the correlation between the historical time wavelet coefficients and historical icing factor wavelet coefficients. i The historical time feature in the first j Layer decomposition and ice cover ym The degree of linear correlation can clearly reflect the close relationship between the wavelet coefficients of various historical time features and the wavelet coefficients of icing factors. This provides a reliable and accurate basis for subsequent screening of target time feature data based on the correlation of icing factors, and helps to analyze the impact of icing on the time feature data of hydropower stations in greater depth. In turn, it provides strong support for hydropower stations to cope with icing and optimize their operation strategies.

[0069] In some embodiments, a target feature matrix is ​​determined based on the historical time feature data corresponding to the historical time wavelet coefficients whose correlation with the icing factor is greater than or equal to a preset threshold, including: Select the i-th feature and the ice accumulation parameter in the j-th layer decomposition. Maximum correlation ; The historical time feature data corresponding to the historical time wavelet coefficients whose maximum correlation is greater than or equal to the preset threshold are determined as the target historical time feature data. The target feature matrix is ​​determined based on the target's historical time feature data.

[0070] Specifically, for each historical time feature, under its frequency layer decomposition, the feature with the largest absolute value among those correlated with different icing amount parameters is selected as the feature at the 1st epoch. jThe system first decomposes the data to determine the maximum correlation with icing amount. Then, it compares all these maximum correlations with preset thresholds, selecting historical time feature data corresponding to wavelet coefficients with maximum correlations greater than or equal to the preset thresholds as target historical time feature data. Finally, a target feature matrix is ​​constructed based on these target historical time feature data. This method accurately filters historical time feature data highly correlated with icing factors, removing irrelevant or weakly correlated data. By constructing the target feature matrix, key feature information sensitive to the impact of icing is retained, providing a high-quality data foundation for subsequent analysis, modeling, and the development of targeted response strategies based on icing factors. This helps improve the scientific rigor and effectiveness of hydropower station operation decisions in icing environments.

[0071] For example, using wavelet multiple correlation coefficient analysis to analyze historical time feature data. Feature selection requires historical icing data. The specific steps for providing assistance are as follows: (1) Wavelet transform decomposition Time feature data matrix and historical icing data matrix Perform wavelet transforms separately to decompose the signal into different frequency components: .

[0072] in, Indicates a signal (factor). The wavelet coefficients at scale a and translation b are the wavelet mother function, t represents the signal length index, a is the scale parameter (reciprocal of frequency), and b is the translation parameter. Here, the signal (factor) is... It can be a time-related factor. It could also be due to the amount of ice accumulation. .

[0073] After wavelet transform, the coefficient matrix corresponding to each signal is obtained: Time-feature wavelet coefficients: ; Wavelet coefficients of icing amount: .

[0074] Where n is the number of time sampling points (one data point per hour). It is a time-feature dimension. It is the dimension of icing features. This represents the number of decomposition layers.

[0075] Then, the correlation between the wavelet coefficients of each time feature at each frequency level and the wavelet coefficients of each icing feature is calculated: .

[0076] in, This indicates that the i-th feature is related to the ice cover amount under the j-th layer decomposition. The correlation coefficient, This represents the wavelet coefficients of the i-th feature at the j-th layer and at time t. Indicates the amount of ice accumulation The wavelet coefficients at time point t in the j-th layer and These represent the mean of the corresponding sequences.

[0077] Then, the correlation between each time feature and each icing factor is obtained, that is, the i-th feature is selected and correlated with different icing amount parameters under the j-th level decomposition. Maximum correlation: .

[0078] Next, feature filtering is performed, setting a relevance threshold (usually 0.6 or 0.7) to filter for significant features: .

[0079] This is a set expression, where, Represents the i-th feature. This indicates taking the absolute value of the correlation coefficient. This indicates that the i-th feature takes the maximum value in all wavelet decomposition layers j.

[0080] This is equivalent to finding the maximum correlation between each feature i and all icing features across all wavelet decomposition layers. If this maximum correlation exceeds a threshold, the feature is retained. For example, suppose there are 3 features, 4 wavelet decomposition layers, and the correlation coefficient matrix r_xy is: Feature 1: [0.3, 0.5, 0.8, 0.4] Maximum value is 0.8; Feature 2: [0.2, 0.3, 0.1, 0.2] Maximum value is 0.3; Feature 3: [0.6, 0.4, 0.5, 0.7] Maximum value is 0.7; If threshold = 0.6: Feature 1: max(|r_xy|) = 0.8 > 0.6 reserve Feature 2: max(|r_xy|) = 0.3 < 0.6 × (rejection) Feature 3: max(|r_xy|) = 0.7 > 0.6 reserve Therefore, F_selected = [1, 3].

[0081] Then, the selected features are used to construct a new historical feature matrix. and the new feature matrix to be predicted .

[0082] ; .

[0083] This is how it is constructed. Preserving the integrity of the time series and including only the features most relevant to icing prediction can improve the efficiency and accuracy of subsequent modeling.

[0084] In step S330, based on the trained icing parameter model, the icing parameters for the target time period are determined using the target temporal feature data and the spatial feature data.

[0085] The process begins by integrating and preprocessing the target temporal feature data (reflecting time-related information closely related to icing) and spatial feature data (reflecting spatial distribution and influencing factors of icing) obtained through preliminary screening, ensuring data format and scale consistency. Subsequently, the integrated data is input into a well-trained and stable icing parameter model. This model, through its complex internal algorithm and patterns learned from extensive data, performs in-depth analysis and calculation of the input data, ultimately outputting accurate and reliable icing parameters for the target time period. This fully utilizes key information from both temporal and spatial dimensions, achieving efficient data transformation and accurate prediction through a well-trained model. It effectively overcomes the limitations of prediction from a single data source, improving the accuracy and comprehensiveness of determining icing parameters for the target time period, thereby enhancing the accuracy of power generation prediction for hydropower stations under icing conditions.

[0086] Specifically, the icing parameter model can employ a two-stream neural network architecture: a TCN model processes and learns temporal features, inputting a multivariate temporal feature data matrix and outputting hourly icing branch feature 1; a CNN model processes and learns spatial features, inputting an image data tensor and outputting hourly icing branch feature 2; then, an ensemble learning layer is used for feature fusion, inputting TCN branch feature 1 and CNN branch feature 2, and the model adaptively fuses the two types of features to predict hourly icing area and thickness. A multi-task loss function is also introduced to enhance the model's flexibility. The two-stream neural network can be a network architecture that processes and fuses temporal and spatial feature streams separately. TCN (Temporal Convolutional Network) is a deep learning model used to process time-series data and effectively capture long-term temporal dependencies. A sliding window is a data processing method that slides a fixed-size time window (e.g., 24 hours) across a time series to construct training samples. CNN (Convolutional Neural Network) is used to process data with a grid structure, such as images.

[0087] In some embodiments, based on a trained icing parameter model, icing parameters for a target time period are determined from the target temporal feature data and the spatial feature data, including: Based on the icing parameter model, time features are extracted from the target time feature data, and spatial features are extracted from the spatial feature data. Based on the fusion of the temporal and spatial features, a spatiotemporal fusion feature is obtained; The icing parameters for the target time period are determined based on the spatiotemporal fusion features.

[0088] This method utilizes a trained icing parameter model to extract targeted temporal features from the target time data, obtaining temporal features that reflect the correlation between temporal variation patterns and icing. Simultaneously, spatial feature extraction is performed to obtain spatial features reflecting spatial distribution and their impact on icing. Next, the extracted temporal and spatial features are organically fused to form a spatiotemporal fusion feature, which integrates the influence of both time and space dimensions on icing. Finally, based on the spatiotemporal fusion feature, the icing parameters for the target time period are determined using the icing parameter model. This step-by-step feature extraction and fusion fully explores the intrinsic connection between time and space dimensions and icing, effectively integrating multi-dimensional information. Compared to analysis methods that only consider a single dimension, this method can more accurately and comprehensively determine the icing parameters for the target time period, providing a reliable basis for hydropower stations to formulate anti-icing and de-icing strategies in advance, ensuring the safe and stable operation of equipment, helping to reduce the adverse effects of icing on hydropower station operation, and improving overall operational efficiency.

[0089] In some embodiments, method 300 further includes: The first initial model is trained based on historical icing training samples to obtain the icing parameter model; wherein, the historical icing training samples include: historical temporal feature data, historical icing data, and historical spatial feature data; the icing parameter model includes a first network model and a second network model; wherein... Based on the historical icing data, the historical time feature data is filtered to obtain the target historical time feature data; The first initial network is trained based on the target historical time feature data and the historical icing data to obtain time feature samples; The second initial network is trained based on the historical spatial feature data and the historical icing data to obtain spatial feature samples. Feature fusion is performed based on the temporal feature branch and the spatial feature branch to obtain fused feature samples; Based on the fused feature samples, predict the corresponding icing prediction data; Based on the mean squared error loss between the predicted icing data and the historical icing data, a loss function is determined. Adjust the model parameters of the first initial network and the second initial network to minimize the loss function, thereby obtaining the trained first network model and the second network model.

[0090] The process begins with historical icing training samples containing historical temporal, icing, and spatial features. Target historical temporal features are then selected from the historical icing data. Next, a first initial network is trained using both the target historical temporal features and the historical icing data to obtain temporal feature samples, while a second initial network is trained using both historical spatial features and the historical icing data to obtain spatial feature samples. The temporal and spatial feature branches are then fused to obtain fused feature samples, which are used to predict icing data. A loss function is determined based on the mean squared error loss of the icing prediction data and the historical icing data. Finally, the parameters of the first and second initial networks are adjusted to minimize the loss function, resulting in the trained first and second network models, which serve as the icing parameter models. This multi-dimensional data selection and separate training fully explores the intrinsic relationship between temporal and spatial features and icing. By utilizing feature fusion and loss function optimization, the model can learn more accurate patterns from complex data, effectively improving the accuracy and generalization ability of the icing parameter model. This provides more reliable icing predictions for scenarios such as hydropower stations, helping to formulate response strategies in advance and reduce the risks and losses caused by icing.

[0091] In some embodiments, a loss function is determined based on the mean squared error loss between the predicted icing data and the historical icing data, including: ;in, It is a loss function. It is the mean square error loss of the icing area prediction in the icing prediction data. α is the mean square error loss of the icing thickness prediction in the icing prediction data, and β are the task weight coefficient parameters.

[0092] In constructing the loss function for the icing parameter model, considering that icing conditions include two key dimensions—icing area and icing thickness—the mean squared error loss for icing area prediction and the mean squared error loss for icing thickness prediction are calculated separately. Then, task weight coefficients are introduced to combine the losses from these two dimensions according to certain weights, ultimately determining the total loss function. 。 By comprehensively considering the prediction errors of both icing area and icing thickness and assigning them different weights, this approach allows for a more comprehensive and accurate measurement of the differences between predicted icing data and historical icing data. It avoids overemphasizing one dimension while neglecting another, thereby improving the accuracy and reliability of the model's predictions of the overall icing situation and providing more effective icing prediction support for scenarios such as hydropower stations.

[0093] For example, constructing model training samples using historical spatial feature data. Directly used as image input features Where N is the number of samples. The filtered historical time feature matrix... and historical ice cover sequence matrix Construct training samples using the sliding window method, and input the window length. (e.g., 24 hours), prediction window length (e.g., the next hour), the sliding step size is u (e.g., 1 hour), and the input feature is... The output labels are

[0094] The dataset is divided into training, validation, and test sets in an 8:1:1 ratio. A two-stream neural network hybrid architecture is employed, simultaneously... and Input a CNN neural network, train the CNN model, and output the spatial feature branch matrix. On the other side and Input a TCN neural network, train the TCN model, and output a temporal feature branch matrix. Then the feature fusion layer concatenates the branch matrices. Then it is fed into the fully connected layer for learning and processing. See the specific network structure diagram. Figure 5 Meanwhile, a new multi-task loss function is introduced during network training: .in, This is the mean square error loss in the prediction of icing area. α is the mean squared error loss for icing thickness prediction, and β are the task weighting coefficient parameters.

[0095] The Adam optimizer is used to train and optimize the parameters of the entire neural network, and the final output prediction result is... , in Represents the predicted icing area (m²). This represents the predicted icing thickness (mm).

[0096] In step S340, the power generation parameters for the target time period are predicted based on the icing parameters and the operating characteristic data.

[0097] The method employs a random forest model to correlate predicted ice accumulation with hydropower station power generation. First, historical ice accumulation data (including area and thickness) and gate opening information (1 representing adjustable, 0 representing fixed) are used as input features, while power generation data (including maximum power output and daily cumulative power generation) is used as the target variable. Multiple decision trees are constructed, each trained using a randomly selected subset of features. The predictions from all decision trees are then integrated using a voting or averaging method to obtain the final predicted power generation value. This combination of multi-source data fusion and deep learning methods enables accurate prediction of hydropower station ice accumulation and accurate assessment of power generation.

[0098] In some embodiments, method 300 further includes: Based on the ice accretion prediction data, historical operating characteristic data, and historical power generation characteristic data, a random forest model is trained to obtain a trained power generation prediction model; the power generation prediction model is used to predict the maximum power generation and daily cumulative power generation for the target time period.

[0099] This process involves collecting data on icing predictions, historical operational characteristics, and historical power generation characteristics, encompassing various factors that may influence power generation and output. This data is used as input samples to train a random forest model. During training, the random forest model constructs multiple decision trees and synthesizes their results, continuously learning the inherent patterns and relationships within the data. After thorough training, a well-trained power generation prediction model is obtained, capable of predicting the maximum power generation and daily cumulative power generation for a target time period using the input data. This fully utilizes multi-source data and leverages the powerful data processing and learning capabilities of the random forest model to effectively uncover complex relationships between data points. Compared to prediction methods based on a single data source or a simple model, this significantly improves the accuracy and reliability of power generation prediction, providing strong data support for hydropower station power generation planning, resource allocation, and operational optimization, thus contributing to improved economic efficiency and operational stability of hydropower stations.

[0100] For example, when constructing training samples for the model, the input features are the icing prediction feature matrix obtained from historical data via a two-stream neural network. Historical data on power plant operation characteristics Gate control feature vector in Where N is the number of samples. Label features are historical data on the operating characteristics of the power plant. Power generation characteristic matrix ,in Represents the historical maximum power generation (kW), This represents the historical daily cumulative power generation (kWh).

[0101] Then, construct a random forest model and input... , and Train the model.

[0102] Finally, using the trained prediction model, the future prediction window length of the power plant can be predicted. Maximum power generation and daily cumulative power generation within the range.

[0103] As can be seen, this disclosure integrates three types of data sources through a multi-source data acquisition and fusion mechanism: temporal feature data (such as water temperature, water level, and flow velocity) acquired by hydrological sensors, spatial feature data (image data) acquired by satellites and drones, and power plant operating parameter data. This multi-dimensional data acquisition scheme lays the foundation for subsequent accurate prediction. It also employs a TCN model to process temporal features and a CNN model to process spatial features, achieving adaptive fusion of the two types of features through a feature fusion layer. Furthermore, a multi-task loss function is introduced. This allows the model to simultaneously predict both ice area and thickness, improving the comprehensiveness and accuracy of the predictions. This disclosure also uses wavelet multiple correlation coefficient analysis for feature selection, effectively reducing data redundancy, and establishes a correlation prediction between ice accumulation and power generation through a random forest model. This method chain design not only improves computational efficiency but also ensures the reliability of the prediction results, providing strong support for hydropower station operation decisions.

[0104] This disclosure establishes a complete multi-source data acquisition system, including multiple data sources such as hydrological sensors, meteorological stations, satellites, and drones, enabling comprehensive monitoring of the ice formation process. Through real-time data acquisition and forecasting, potential ice threats can be detected in advance, providing early warnings for the safe operation of critical equipment such as turbines, spillways, and gates. The forecast results can directly guide the timely deployment of anti-icing measures, effectively reducing the risk of damage to equipment caused by ice.

[0105] In response to the suddenness and complexity of icing phenomena, this disclosure employs wavelet multiple correlation coefficient analysis for feature selection, effectively identifying key factors influencing icing formation; it innovatively designs a dual-stream neural network architecture, combining TCN and CNN to simultaneously process temporal and spatial features, enhancing the modeling capability for complex icing processes; and it introduces a multi-task loss function to achieve simultaneous prediction of icing area and thickness, significantly improving the comprehensiveness and accuracy of prediction.

[0106] To address the issues of prediction reliability and data fusion, this disclosure significantly improves data utilization efficiency and overcomes the limitations of traditional single prediction models by effectively fusing multi-source heterogeneous data; it establishes a correlation prediction between ice cover and power generation using a random forest model, providing a direct and usable quantitative basis for power plant operation decisions; and the application of ensemble learning methods improves the model's generalization ability, giving the prediction system good transferability and making it suitable for hydropower plants in different geographical locations and climatic conditions. This disclosure proposes a hybrid architecture-based method for predicting hydropower generation under icing conditions. First, it improves the scientific rigor and accuracy of feature selection by collecting multi-source features and incorporating wavelet multiple correlation coefficient analysis. Second, the method designs a dual-stream neural network hybrid architecture, combining the advantages of CNN and TCN, and introduces a feature fusion layer and a multi-task loss function to achieve unified management and fusion of spatial and temporal features, improving the model's ability to process multi-dimensional information and ultimately enhancing prediction accuracy. Third, this invention combines icing prediction results with power generation prediction results to construct correlations, thereby improving prediction accuracy and providing effective decision support for hydropower station operation scheduling and economic benefit optimization.

[0107] Therefore, this scheme enables systematic prediction of icing at hydropower stations. Using an ensemble learning prediction model, it improves migration performance under different geographical locations and climatic conditions. Furthermore, it designs a correlation prediction between icing conditions and power generation capacity to more accurately reflect the impact of icing on electricity production. Through the proposed algorithms and techniques, it effectively improves data utilization efficiency and feature extraction capabilities, enhancing prediction accuracy and reliability, thereby providing more reliable support and guidance for the safe operation of hydropower stations.

[0108] It should be noted that the method of this embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this embodiment, and the multiple devices will work together to generate video to complete the method described.

[0109] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0110] Based on the same technical concept, corresponding to any of the above embodiments, this disclosure also provides a hydropower station icing power generation prediction device based on ensemble learning, see [link to relevant documentation]. Figure 6 The hydropower station icing power generation prediction device based on ensemble learning includes: The acquisition module is used to acquire multi-source data about the hydropower station, including time feature data, spatial feature data, and operational feature data. The feature filtering module is used to filter the time feature data based on the correlation of icing factors to obtain target feature data; The icing prediction module is used to determine the icing parameters for a target time period based on the target temporal feature data and the spatial feature data, using a trained icing parameter model. The power generation prediction module is used to predict the power generation parameters for the target time period based on the icing parameters and the operating characteristic data.

[0111] For ease of description, the above apparatus is described in terms of its functions, divided into various modules. Of course, in implementing this disclosure, the functions of each module can be implemented in one or more software and / or hardware.

[0112] The apparatus described above is used to implement the corresponding ensemble learning-based hydropower station icing power generation prediction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0113] Based on the same technical concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the hydropower station icing power generation prediction method based on ensemble learning as described in any of the above embodiments.

[0114] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0115] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the hydropower station icing power generation prediction method based on ensemble learning as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0116] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this disclosure as described above, which are not provided in detail for the sake of brevity.

[0117] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this disclosure, the provided drawings may or may not show well-known power / ground connections to integrated circuit (IC) chips and other components. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this disclosure, and this also takes into account the fact that the details of implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this disclosure will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that the embodiments of this disclosure can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0118] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0119] This disclosure is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for predicting icing power generation in hydropower stations based on ensemble learning, comprising: Acquire multi-source data about the hydropower station, including time feature data, spatial feature data, and operational feature data; The time feature data is filtered based on the correlation of icing factors to obtain target feature data; Based on the trained icing parameter model, the icing parameters for the target time period are determined using the target temporal feature data and the target spatial feature data. Based on the icing parameters and the operational characteristic data, the power generation parameters for the target time period are predicted.

2. The method according to claim 1, wherein, The time feature data is filtered based on the correlation of icing factors to obtain target time feature data, including: Wavelet transform analysis was performed on historical time characteristic data and historical icing data to obtain historical time wavelet coefficients and historical icing factor wavelet coefficients at different frequency levels. The correlation between the historical time wavelet coefficients and the historical icing factor wavelet coefficients is determined to obtain the icing factor correlation. Based on the historical time feature data corresponding to the historical time wavelet coefficients whose correlation with the icing factors is greater than or equal to a preset threshold, a target feature matrix is ​​determined. The target time feature data is obtained by filtering the time feature data based on the target feature matrix.

3. The method according to claim 1, wherein, Based on the trained icing parameter model, the icing parameters for the target time period are determined using the target temporal feature data and the target spatial feature data, including: Based on the icing parameter model, time features are extracted from the target time feature data, and spatial features are extracted from the spatial feature data. Based on the fusion of the temporal and spatial features, a spatiotemporal fusion feature is obtained; The icing parameters for the target time period are determined based on the spatiotemporal fusion features.

4. The method according to claim 2, wherein, Determining the correlation between the historical time wavelet coefficients and the historical icing factor wavelet coefficients yields the icing factor correlation, including: ； in, This represents the relationship between the i-th historical time feature and the ice cover amount under the j-th level decomposition. The correlation of icing factors This represents the wavelet coefficients at time point t under the j-th level decomposition of the i-th historical time feature. Indicates icing data The wavelet coefficients at time point t in the j-th layer and These represent the mean of the corresponding sequences; Then, based on the historical time feature data corresponding to the historical time wavelet coefficients whose correlation with the icing factor is greater than or equal to a preset threshold, a target feature matrix is ​​determined, including: Select the i-th feature and the ice accumulation parameter in the j-th layer decomposition. Maximum correlation ; The historical time feature data corresponding to the historical time wavelet coefficients whose maximum correlation is greater than or equal to the preset threshold are determined as the target historical time feature data. The target feature matrix is ​​determined based on the target's historical time feature data.

5. The method according to claim 4, further comprising: The first initial model is trained based on historical icing training samples to obtain the icing parameter model; wherein, the historical icing training samples include: historical temporal feature data, historical icing data, and historical spatial feature data; the icing parameter model includes a first network model and a second network model; wherein... Based on the historical icing data, the historical time feature data is filtered to obtain the target historical time feature data; The first initial network is trained based on the target historical time feature data and the historical icing data to obtain time feature samples; The second initial network is trained based on the historical spatial feature data and the historical icing data to obtain spatial feature samples. Feature fusion is performed based on the temporal feature branch and the spatial feature branch to obtain fused feature samples; Based on the fused feature samples, predict the corresponding icing prediction data; Based on the mean squared error loss between the predicted icing data and the historical icing data, a loss function is determined. Adjust the model parameters of the first initial network and the second initial network to minimize the loss function, thereby obtaining the trained first network model and the second network model.

6. The method according to claim 5, wherein, Based on the mean squared error loss between the predicted icing data and the historical icing data, a loss function is determined, including: ;in, It is a loss function. It is the mean square error loss of the icing area prediction in the icing prediction data. α is the mean square error loss of the icing thickness prediction in the icing prediction data, and β are the task weight coefficient parameters.

7. The method according to claim 5, further comprising: Based on the ice accretion prediction data, historical operating characteristic data, and historical power generation characteristic data, a random forest model is trained to obtain a trained power generation prediction model; the power generation prediction model is used to predict the maximum power generation and daily cumulative power generation for the target time period.

8. A hydropower station icing power generation prediction device based on ensemble learning, comprising: The acquisition module is used to acquire multi-source data about the hydropower station, including time feature data, spatial feature data, and operational feature data. The feature filtering module is used to filter the time feature data based on the correlation of icing factors to obtain target feature data; The icing prediction module is used to determine the icing parameters for a target time period based on the target temporal feature data and the spatial feature data, using a trained icing parameter model. The power generation prediction module is used to predict the power generation parameters for the target time period based on the icing parameters and the operating characteristic data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as claimed in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1 to 7.

Images