A cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station group

Abstract

This invention relates to the field of intelligent scheduling and control technology for hydropower station groups, and particularly to a cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups, including a remote cloud control platform and multiple edge power station nodes equipped with PSK-8 controllers. This invention employs a scheme that uses the cloud to calculate in real time the failure probability of each hydropower station device and the overall system anomaly probability. Based on the failure probability ranking, it prioritizes updating the failure judgment rules and handling methods corresponding to high-risk devices. Simultaneously, it dynamically adjusts the total number of rules stored in the edge computing node database according to the probability of system anomalies. This approach enables the edge to maintain the strongest judgment capability for the devices most likely to fail, while maximizing the conservation of limited storage resources at the edge while ensuring the coverage of fault judgment. This significantly improves the accuracy and timeliness of fault identification and effectively reduces the risk of false alarms and missed alarms.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling and control technology for hydropower station groups, and in particular to a cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups. Background Technology

[0002] With the increasing proportion of hydropower in new power systems, the coordinated and optimized scheduling and safe and stable operation of hydropower station clusters have become crucial for ensuring power supply. Traditional hydropower station cluster scheduling systems mostly adopt a centralized architecture, where all operational data is uploaded to a remote control center for unified processing and decision-making. This approach has inherent drawbacks such as large data transmission volumes, insufficient real-time response capabilities, and strong dependence on communication networks. Once communication is interrupted, the entire system will be paralyzed, failing to meet the needs of rapid control and emergency response for hydropower station equipment.

[0003] In recent years, although some hydropower station dispatching systems based on cloud-edge collaborative architecture have emerged, these systems generally suffer from problems such as insufficient utilization of edge computing capabilities, fixed fault judgment rules, lack of edge database update mechanisms, and unreasonable data transmission strategies. This results in low fault identification accuracy, high false alarm rate, and poor system fault tolerance, making it difficult to effectively solve the problem of efficient dispatching and safe control of hydropower station groups in complex operating environments. Summary of the Invention

[0004] To overcome the problems mentioned in the background art, this invention proposes a cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups.

[0005] The technical solution of the present invention is: a cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups, including a remote cloud control platform and multiple edge power station nodes equipped with PSK-8 controllers, wherein the remote cloud control platform and the multiple edge power station nodes are connected through a communication network; The remote cloud control platform includes: The big data optimization and scheduling module is used to perform global optimization calculations based on regional power load data, watershed flow situation data, and the operating status data of each edge power station node, generate hydropower station group scheduling instructions, and issue them to the corresponding edge power station nodes. The fault probability calculation module is used to calculate the fault probability and abnormal situation probability of each hydropower station equipment in real time based on the equipment operation characteristic data uploaded by the edge power station nodes. The edge database update module is used to sort the fault probabilities based on the fault probability calculation results, generate updated fault judgment rules and handling methods, and determine the number of rules that the edge computing node database needs to store based on the probability of abnormal situations. The deep fault analysis module is used to analyze the abnormal features uploaded by edge power station nodes. When a suspected fault feature is found, a raw data extraction request is sent to the corresponding edge power station node, and the obtained raw data is used for simulation and fault identification and prediction. The PSK-8 controller of the edge power station node includes: The local execution module is used to receive and execute scheduling instructions issued by the remote cloud control platform to control the operation of hydropower station equipment; The local safety fault tolerance module is used to autonomously monitor the equipment's operating status and handle emergency faults based on locally stored fault judgment rules and handling methods when communication is interrupted. The data feature extraction module is used to extract features from the raw operating data collected by hydropower station equipment, generate equipment operating feature data, and upload it to the remote cloud control platform. Edge database is used to store fault diagnosis rules, handling methods, and equipment operation history data; The raw data on-demand transmission module is used to receive raw data extraction requests sent by the remote cloud control platform and upload the corresponding raw operating data to the remote cloud control platform.

[0006] Preferably, the edge database update module performs the following steps: The calculated failure probabilities of each device are sorted in descending order; Prioritize updating fault diagnosis rules and handling methods for equipment with high failure probability; The number of rules stored in the edge database is dynamically adjusted based on the probability value of abnormal situations.

[0007] Preferably, the step of dynamically adjusting the number of rules stored in the edge database based on the probability value of abnormal situations specifically involves: When the probability of an anomaly exceeds the first preset threshold, the number of rules stored in the edge database is increased. When the probability of an anomaly is lower than the second preset threshold, the number of rules stored in the edge database is reduced.

[0008] Preferably, the data feature extraction module performs the following steps: Preprocess the raw operating data, including data cleaning, normalization and noise reduction. A combination of time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis was used to extract key feature parameters from the original data. The extracted key feature parameters are packaged according to a preset format and uploaded to the remote cloud control platform on a regular basis.

[0009] Preferably, the deep fault analysis module performs the following steps: Establish digital twin models of hydropower station equipment; The acquired raw operational data is input into the digital twin model for simulation. Based on simulation results and combined with historical fault data, machine learning algorithms are used to identify faults and predict fault development trends. Generate fault warning information and handling suggestions, and send them to the corresponding edge power station nodes.

[0010] Preferably, the local security fault-tolerant module performs the following steps: It monitors the status of the communication network in real time and automatically switches to local autonomous operation mode when a communication interruption is detected. Based on the fault diagnosis rules stored in the edge database, the locally collected device operation data is analyzed in real time. When a device malfunction is detected, emergency operations are automatically executed according to the corresponding handling methods, including device shutdown, load adjustment, and alarm notification.

[0011] Preferably, the fault probability calculation module uses a pre-trained Long Short-Term Memory (LSTM) network model to calculate the fault probability and the probability of abnormal conditions for each hydropower station equipment. The LSTM model is trained and updated periodically using historical fault data and normal operation data.

[0012] Preferably, after receiving the raw data extraction request from the remote cloud control platform, the raw data on-demand transmission module uses the LZ4 compression algorithm to compress the corresponding raw operating data before uploading it to the remote cloud control platform.

[0013] Preferably, the big data optimization scheduling module adopts a genetic algorithm or a particle swarm optimization algorithm, with the objective functions of maximizing power generation benefits and optimizing water resource utilization, and takes into account constraints such as the head, flow rate, and unit output limits of the hydropower station to generate a hydropower station group scheduling plan for the next 24 hours.

[0014] Preferably, the local security fault-tolerant module continues to attempt to establish communication with the remote cloud control platform during the communication interruption. Once communication is restored, it immediately uploads the operating data and fault records during the communication interruption to the remote cloud control platform and switches back to the cloud-edge collaborative operation mode.

[0015] The beneficial effects of this invention are: 1. Compared to existing technologies that use fixed storage of fault judgment rules and handling methods in edge computing node databases, or batch update all rules at fixed time intervals, which have drawbacks such as untimely updates of fault rules for high-risk equipment, invalid rules for low-risk equipment occupying a large amount of edge storage resources, decreased fault identification accuracy as equipment operating status changes, and persistently high false alarm and false negative rates, this invention uses cloud-based real-time calculation of the fault probability of each hydropower station device and the overall system anomaly probability. Based on the fault probability ranking, it prioritizes updating the fault judgment rules and handling methods corresponding to high-risk equipment. At the same time, it dynamically adjusts the total number of rules stored in the edge computing node database according to the probability of system anomalies. This approach enables the edge to always maintain the strongest judgment capability for the equipment most likely to fail, while maximizing the conservation of limited edge storage resources while ensuring fault judgment coverage. This significantly improves the accuracy and timeliness of fault identification and effectively reduces the risk of false alarms and false negatives.

[0016] 2. Compared to existing technologies that upload all raw operating data from all devices to the cloud for unified analysis, or place all data analysis tasks at the edge, the former suffers from massive network data transmission, high bandwidth consumption, large transmission latency, severe cloud computing resource consumption, and data security risks, while the latter is limited by edge computing capabilities and cannot perform complex in-depth fault analysis. This invention employs an edge computing node to first extract multi-dimensional features from the collected raw operating data, and only uploads the extracted feature data to the cloud. The cloud performs preliminary analysis of the feature data, and only requests the corresponding raw operating data back from the edge node for in-depth analysis when suspected abnormal features are found. This approach significantly reduces network data transmission volume, lowers bandwidth consumption and transmission latency, reduces the cloud computing burden, improves data transmission security, and fully utilizes the powerful computing capabilities of the cloud to achieve high-precision in-depth fault analysis.

[0017] 3. Compared to existing technologies that use simple machine learning models trained on historical fault data for fault identification and prediction, which have drawbacks such as only being able to identify known fault patterns, being unable to reconstruct the fault occurrence and development process, having difficulty obtaining internal operating parameters that cannot be directly measured, being unable to detect potential hidden faults, and having limited prediction accuracy and lead time, this invention adopts a solution that establishes high-precision digital twin models of various equipment in the hydropower station in the cloud, inputs the transmitted raw operating data into the digital twin model for simulation, reconstructs the actual operating state and internal parameter changes of the equipment, and combines machine learning algorithms for fault identification and development trend prediction. This solution has the advantages of being able to intuitively display the occurrence and evolution process of faults, obtaining key parameters that cannot be directly measured, such as internal stress distribution, temperature field, and flow field of the equipment, not only accurately identifying various known faults, but also effectively discovering potential hidden fault hazards, significantly improving the lead time and accuracy of fault prediction, and achieving true predictive maintenance of equipment.

[0018] 4. Compared to existing technologies that employ a fully centralized control architecture, where all decisions are made in the cloud or the edge only has simple emergency shutdown protection functions, this invention addresses the drawbacks of the cloud-based approach. This approach relies heavily on communication networks, and a communication interruption can paralyze the entire dispatch and control system, preventing the normal operation of the hydropower station group and severely impacting the safety and stability of the power system. Instead, this invention utilizes a cloud-based approach for global optimization scheduling and in-depth fault analysis and decision-making, while the edge is responsible for rapid execution of dispatch instructions and local safety and fault-tolerant control. When communication is interrupted, the edge automatically switches to a local autonomous operation mode, autonomously monitoring equipment operating status and handling fault emergencies based on dynamically updated local fault judgment rules and methods. Once communication is restored, data is automatically synchronized and the system switches back to the cloud-edge collaborative operation mode. This approach ensures both the global optimization of the hydropower station group and maximizes power generation efficiency, while significantly improving the system's reliability and stability under extreme conditions such as communication interruptions and cloud failures, effectively preventing hydropower station group shutdowns caused by communication failures. Attached Figure Description

[0019] Figure 1 The diagram shows the architecture of the cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to the present invention. Figure 2 The diagram shown is a schematic representation of fault risk and edge rule update data in a specific embodiment of the present invention. Figure 3 The diagram shown illustrates the unit load maintenance before and after communication interruption in a specific embodiment of the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Please see Figure 1 This invention provides an embodiment of an intelligent scheduling and fault-tolerant control system for a cloud-edge collaborative hydropower station group, employing a two-level distributed architecture of "global decision-making in the cloud + local execution at the edge". The system consists of one remote cloud control platform and N (N≥1) edge power station nodes. The remote cloud control platform is connected to all edge power station nodes via a 5G / 4G public network or a dedicated power fiber optic communication network. The communication protocol adopts the MQTT protocol, supporting QoS=1 message transmission quality to ensure reliable transmission of critical commands and data.

[0022] The remote cloud control platform is deployed on an Alibaba Cloud ECS cloud server cluster, configured with: 8 compute nodes (each configured with an Intel Xeon Platinum 8375C CPU, 128GB of memory, and a 2TB SSD), 2 storage nodes (each configured with 4 × 16TB HDDs and RAID5), and 1 management node. The operating system is Ubuntu 22.04 LTS, containerization is achieved using Docker 24.0.6, and orchestration is performed using Kubernetes 1.28.2.

[0023] Each edge power station node corresponds to a hydropower station within the river basin, and each hydropower station is equipped with one PSK-8 industrial controller as the edge computing core. The hardware configuration of the PSK-8 controller is as follows: ARM Cortex-A53 quad-core processor (1.5GHz), 4GB DDR4 memory, 32GB eMMC storage, 8 AI analog inputs, 16 digital inputs, 16 digital outputs, 2 RS485 interfaces, 2 Ethernet interfaces, and 1 5G module interface. The controller comes pre-installed with an embedded operating system based on the Linux 5.15 kernel and supports Python 3.10 and C++17 development environments.

[0024] Specifically, the remote cloud control platform includes the following modules: 1. Big Data Optimization and Scheduling Module The big data optimization and scheduling module is developed using a microservice architecture and obtains the following three types of data in real time through API interfaces: Regional electricity load data: obtained from the EMS system of the regional power grid dispatch center, with a time resolution of 15 minutes, including load forecast data for the next 24 hours and real-time load data; Watershed flow situation data: obtained from the watershed hydrological monitoring system, including real-time water level, flow rate, and rainfall data of each hydrological station, as well as runoff forecast data for the next 72 hours; Operational status data of edge power station nodes: acquired from the PSK-8 controller of each edge power station node, including real-time output, head, flow rate, opening degree, vibration, temperature and other parameters of each unit, with a time resolution of 10 seconds.

[0025] This module uses an improved particle swarm optimization algorithm for global optimization calculations. The specific implementation steps are as follows: Establish the objective function: maximizing the total power generation benefit of the hydropower station group over 24 hours is the objective function, and its mathematical expression is: ; in, Let i be the output of the i-th hydropower station in time period t. Let t be the on-grid electricity price for the t-th time period.

[0026] Define the constraints, including water balance constraints, unit output constraints, reservoir water level constraints, outflow constraints, and unit start-up and shutdown constraints.

[0027] Initialize the particle swarm: set the particle count to 100, the number of iterations to 500, the inertia weight to decrease linearly from 0.9 to 0.4, and the learning factor... .

[0028] Iterative optimization calculation: In each iteration, the velocity and position of the particles are updated, the fitness value of each particle is calculated, and the global optimal solution is updated.

[0029] Generate scheduling plan: Convert the optimal solution into hourly output plan for each hydropower station for the next 24 hours, and decompose it into scheduling instructions every 15 minutes, which are then sent to the corresponding edge power station nodes via the MQTT protocol.

[0030] 2. Fault Probability Calculation Module The fault probability calculation module uses a Long Short-Term Memory (LSTM) network to construct a fault probability prediction model. The model structure is as follows: input layer → 2 LSTM layers → 1 Dropout layer → 1 fully connected layer → output layer. The input layer has a dimension of 32 (corresponding to 32 device operation feature parameters), each LSTM layer contains 64 neurons, the Dropout layer has a dropout rate of 0.2, the fully connected layer contains 32 neurons, and the output layer has a dimension of 2 (outputting the device fault probability and the system abnormality probability, respectively).

[0031] The model training process is as follows: Data preparation: Collect the operation data and fault records of hydropower station equipment over the past 3 years, including 1 million normal operation data samples and 10,000 fault data samples.

[0032] Data preprocessing: The data is normalized to map all feature parameters to the [0,1] interval; the SMOTE algorithm is used to oversample fault samples to balance the ratio of positive to negative samples.

[0033] Model training: The Adam optimizer was used, with a learning rate of 0.001, a batch size of 64, a training epoch of 100, and a binary cross-entropy loss function.

[0034] Model validation: The 10-fold cross-validation method is used to validate the model performance, requiring the model to have an accuracy of ≥95% and a recall of ≥90%.

[0035] Model update: The model is incrementally trained monthly using newly collected operational and fault data to maintain its prediction accuracy.

[0036] During system operation, the fault probability calculation module receives equipment operation characteristic data uploaded by each edge power station node every 10 seconds, inputs it into the trained LSTM model, and calculates the fault probability of each hydropower station equipment (turbine, generator, transformer, governor, excitation system, etc.) and the abnormal situation probability of the entire hydropower station system in real time. The calculation results are stored in the cloud time series database InfluxDB.

[0037] 3. Edge Database Update Module The edge database update module performs an update operation every 5 minutes. The specific implementation steps are as follows: Failure probability ranking: Read the current failure probability values ​​of all devices from InfluxDB, sort them in descending order, and generate a ranking list of device failure probabilities.

[0038] Rule priority is determined as follows: Devices with the highest failure probability are defined as high-risk devices, those between 30% and 70% are defined as medium-risk devices, and those in the bottom 30% are defined as low-risk devices. Fault judgment rules and handling methods for high-risk devices have the highest priority, followed by medium-risk devices, and then low-risk devices.

[0039] Rule update generation: Extract all fault judgment rules and handling methods corresponding to high-risk devices, as well as the main fault judgment rules and handling methods corresponding to medium-risk devices, from the cloud-based fault rule library, and generate an update data package. The update data package is in JSON format and includes fields such as rule ID, device type, fault type, judgment conditions, handling steps, and priority.

[0040] The number of rules is dynamically adjusted: based on the probability of system anomalies, the number of rules stored in the edge database is dynamically adjusted. When the probability of an anomaly is ≥0.7, the system is judged to be in a high-risk state, and the number of rules stored in the edge database is set to 90% of the maximum capacity, including rules for all high, medium and low risk devices. When the probability of an anomaly is less than 0.7, the system is judged to be in a risky state, and the number of rules is set to 60% of the maximum capacity, including rules for all high and medium risk devices. When the probability of an anomaly is less than 0.3, the system is judged to be in a low-risk state, and the number of rules is set to 30% of the maximum capacity, including only rules for high-risk devices.

[0041] Rule update distribution: The generated update data packet is distributed to the corresponding edge power station node via the MQTT protocol. After receiving the data packet, the edge node updates the rules in its local edge database and returns an update confirmation message.

[0042] 4. In-depth fault analysis module The deep fault analysis module is implemented based on digital twin technology and machine learning algorithms. The specific implementation steps are as follows: Digital twin model construction: Digital twin models of various equipment in the hydropower station were constructed using Unity3D and ANSYS software, including geometric models, physical models, and behavioral models. The geometric model accurately reproduces the external dimensions and structure of the equipment; the physical model embeds parameters such as the material properties, mechanical characteristics, and fluid characteristics of the equipment; the behavioral model simulates the operating status of the equipment under different working conditions based on the equipment's operating mechanism and historical data.

[0043] Anomaly detection: Real-time monitoring of feature data uploaded by edge power station nodes. When a feature parameter exceeds the preset normal range (such as unit vibration amplitude > 0.08 mm, stator temperature > 105℃) or shows an abnormal trend (such as temperature rising more than 5℃ within 10 minutes), it is judged as a suspected fault feature.

[0044] Raw data request: When a suspected fault feature is detected, a raw data extraction request is sent to the corresponding edge power station node to request the raw operating data of the past 10 minutes corresponding to the feature parameter, with a sampling frequency of 1kHz.

[0045] Simulation analysis: The acquired raw operating data is input into the digital twin model to restore the actual operating state of the equipment during the time period, simulate the occurrence and development of the fault, and obtain parameters that cannot be directly measured, such as stress distribution, temperature field, and flow field inside the equipment.

[0046] Fault Identification and Prediction: Based on simulation results and combined with a historical fault database, a random forest algorithm is used for fault identification. The random forest model contains 100 decision trees with a maximum depth of 10, and can identify more than 20 common hydropower station equipment faults with an accuracy rate of ≥92%. Simultaneously, the ARIMA time series prediction algorithm is used to predict the development trend of faults, issuing fault warnings 1-72 hours in advance.

[0047] Fault handling suggestion generation: Based on the fault type, fault location, fault severity and development trend, generate targeted fault handling suggestions, including immediate shutdown, reduced load operation, enhanced monitoring, planned maintenance, etc., and send fault warning information and handling suggestions to the corresponding edge power station nodes.

[0048] Specifically, the edge power plant node PSK-8 controller includes the following modules 1. Local execution module The local execution module is developed in C++ and runs on the real-time operating system of the PSK-8 controller, with a response time of ≤10ms. This module connects to the local control unit (LCU) of the hydropower station via an RS485 interface to control equipment such as turbines, generators, gates, governors, and excitation systems.

[0049] The specific work process is as follows: Receive scheduling instructions from the remote cloud control platform and parse parameters such as target output, target flow rate, and target water level from the instructions; Based on the current operating status of the equipment, the PID control algorithm is used to calculate control quantities such as governor opening, guide vane opening, and gate opening; The control input is converted into a standard Modbus-RTU protocol signal and sent to the local control unit to control the operation of the equipment. Real-time acquisition of equipment operation feedback data, comparison with target values, and adjustment of control inputs to achieve closed-loop control; The real-time operating status data of the equipment is uploaded to the data feature extraction module.

[0050] 2. Data Feature Extraction Module The data feature extraction module is developed using Python and processes the collected raw running data every 10 seconds. The specific implementation steps are as follows: Data preprocessing: Data cleaning: Outliers are removed using the 3σ criterion, which means that when a data point deviates from the mean by more than 3 times the standard deviation, it is identified as an outlier and deleted. Missing value imputation: Missing data points are imputed using linear interpolation. Data normalization: The minimum-maximum normalization method is used to map the data to the interval [0,1]. Data denoising: Wavelet transform denoising method is adopted, db4 wavelet basis function is selected, and the decomposition level is 5 to remove high-frequency noise.

[0051] Feature extraction: Temporal feature extraction: Calculate nine temporal features of the original data, including mean, variance, standard deviation, peak value, peak-to-peak value, kurtosis, skewness, margin index, and impulse index. Frequency domain feature extraction: The time domain data is converted into frequency domain data using Fast Fourier Transform (FFT), and four frequency domain features are calculated, including the centroid of the spectrum, mean square frequency, frequency variance, and dominant frequency amplitude. Time-frequency domain feature extraction: Wavelet packet transform is used to decompose the data into sub-bands of different frequency bands, calculate the energy value of each sub-band, and extract 8 time-frequency domain features.

[0052] Data Upload: The 21 extracted feature parameters are packaged in a preset JSON format and uploaded to the remote cloud control platform via the MQTT protocol at a frequency of once every 10 seconds.

[0053] 3. Edge Database The edge database is implemented using the embedded relational database SQLite 3.40.1, with a storage capacity of 32GB, supporting transaction processing and concurrent access. The database contains the following three main tables: Fault Rule Table: Stores fault judgment rules and handling methods. The table structure is (rule ID, device type, fault type, judgment conditions, handling steps, priority, update time). Operational Data Table: Stores the raw and characteristic data of device operation over the past 7 days, using a time-sharded storage method, with one data file generated per day; Fault Log Table: Stores fault records that occurred during communication interruptions, including fields such as fault occurrence time, fault type, fault location, handling measures, and recovery time.

[0054] Edge databases support an automatic data aging mechanism, which automatically deletes the oldest running data files when storage capacity reaches 80%, freeing up storage space.

[0055] 4. Raw data on-demand return module The raw data on-demand return module is developed using a multi-threaded architecture. One thread is responsible for listening for raw data extraction requests from the cloud, while another thread is responsible for data compression and uploading. The specific implementation steps are as follows: Real-time monitoring of raw data extraction requests sent by the remote cloud control platform, and parsing of the time range and parameter types in the requests; Based on the request parameters, query the corresponding raw runtime data from the edge database; The LZ4 lossless compression algorithm is used to compress the original data, achieving a compression ratio of 3:1 or higher and a compression speed of ≥500MB / s; The compressed data is uploaded to the remote cloud control platform in chunks, each chunk being 1MB in size, and supports resuming interrupted uploads. Once the upload is complete, send an upload completion confirmation message to the cloud.

[0056] 5. Local security fault-tolerant module The local security and fault tolerance module adopts a state machine design pattern, which includes two states: cloud-edge collaborative operation and local autonomous operation. The specific implementation steps are as follows: Communication status monitoring: A heartbeat packet is sent to the remote cloud control platform every 1 second. If no heartbeat response is received from the cloud for 3 consecutive times, it is determined that the communication is interrupted and the system automatically switches from cloud-edge collaborative operation to local autonomous operation.

[0057] Local operation strategy: In local autonomous operation mode, the local execution module operates based on a preset local scheduling strategy, that is, it operates according to the output value of the last time period of the most recently received cloud scheduling instruction, and at the same time, it automatically adjusts the discharge flow according to the reservoir water level to ensure that the reservoir water level is within the normal range.

[0058] Local fault monitoring: Based on fault diagnosis rules stored in the edge database, real-time analysis is performed on locally collected device operation data. Fault diagnosis is represented using production rules, for example: If the X-axis vibration amplitude of the unit is >0.1mm and the vibration frequency is equal to the rotational frequency × 2, then it is determined to be a rotor imbalance fault. If the stator temperature is >120℃ and the duration is >5 minutes, then the stator is considered to be overheating.

[0059] Emergency Response: When a equipment malfunction is detected, appropriate emergency procedures will be performed according to the severity of the malfunction. General fault: Issues audible and visual alarms to prompt operators to check and handle the problem, while simultaneously reducing the unit load to 50% of the rated load; Serious malfunction: Immediately perform emergency shutdown operation, close the inlet gate, disconnect the generator outlet circuit breaker, and issue an audible and visual alarm.

[0060] Communication recovery process: During the communication interruption, the local safety fault-tolerant module attempts to establish communication with the cloud every 30 seconds. Once communication is restored, all operational data and fault records from the communication interruption period are immediately uploaded to the remote cloud control platform, and the module automatically switches from local autonomous operation to cloud-edge collaborative operation to execute the latest scheduling instructions from the cloud.

[0061] In summary, the complete workflow of the system based on the aforementioned remote cloud control platform and the PSK-8 controller for edge power plant nodes is as follows: The PSK-8 controllers at each edge power station node collect real-time operating data of the hydropower station equipment through local sensors at a sampling frequency of 1kHz. The data feature extraction module preprocesses and extracts features from the raw operating data to generate equipment operating feature data, which is uploaded to the remote cloud control platform every 10 seconds. The big data optimization and scheduling module of the remote cloud control platform uses an improved particle swarm optimization algorithm to perform global optimization calculations based on regional power load data, watershed flow situation data, and the operating status data of each power station, generating a hydropower station group scheduling plan for the next 24 hours and sending scheduling instructions to each edge power station node. The fault probability calculation module inputs the uploaded feature data into the LSTM model to calculate the fault probability of each device and the probability of system abnormality in real time. Every 5 minutes, the edge database update module generates updated fault judgment rules and handling methods based on the fault probability sorting results, adjusts the rule storage volume of the edge database according to the probability of abnormal situations, and sends the update data packet to each edge power station node. The deep fault analysis module monitors the uploaded feature data in real time. When a suspected fault feature is detected, it sends a raw data extraction request to the corresponding edge power station node. The raw data transmission module of the edge power station node compresses the corresponding raw operating data and uploads it to the cloud. The deep fault analysis module inputs raw data into the digital twin model for simulation, combines the random forest algorithm for fault identification and trend prediction, generates fault warning information and handling suggestions, and issues them out. The local execution module of the edge power station node executes the scheduling instructions and fault handling suggestions issued by the cloud to control the operation of the equipment; When communication is interrupted, the local security fault tolerance module automatically switches to local autonomous operation mode to ensure the safe operation of the system based on local policies; after communication is restored, it automatically switches back to cloud-edge collaborative operation mode. Specific Implementation

[0062] See Figure 2 and Figure 3 Taking a group of hydropower stations in a river basin that includes three hydropower stations as an example, the specific application process of this invention is explained: Hydropower Station A: Installed capacity 2×50MW, reservoir capacity 120 million m³ 3 ; Hydropower Station B: Installed capacity 3×30MW, reservoir capacity 0.8 billion m³ 3 ; Hydropower Station C: Installed capacity 2×20MW, reservoir capacity 0.5 billion m³ 3 .

[0063] Under normal operating conditions, the remote cloud control platform generates an optimized scheduling plan for the hydropower station group based on the regional power grid's load forecast data for the next 24 hours and the basin runoff forecast data: Hydropower station A undertakes peak-shaving tasks, operating at full capacity during peak electricity demand periods and reducing output during off-peak periods; Hydropower stations B and C undertake base load tasks, maintaining stable output. The PSK-8 controllers at each peripheral power station node execute the scheduling instructions issued from the cloud to control the operation of the units.

[0064] At 10:00 AM on a certain day, the fault probability calculation module of the remote cloud control platform calculated that the fault probability of Unit 1 of hydropower station A was 0.85, and the probability of system abnormality was 0.75. The edge database update module immediately updated all fault judgment rules and handling methods corresponding to Unit 1 to the edge database of hydropower station A, and adjusted the rule storage capacity of the edge database to 90% of the maximum capacity.

[0065] At 10:15, the deep fault analysis module detected an anomaly in the vibration characteristic parameters of Unit 1 at Hydropower Station A and sent a raw data extraction request to the PSK-8 controller of Hydropower Station A. The PSK-8 controller compressed the raw vibration data of Unit 1 from the past 10 minutes and uploaded it to the cloud. The deep fault analysis module input the raw data into a digital twin model for simulation and, using a random forest algorithm, identified the fault type as rotor imbalance, with a moderate severity, predicting that the fault would develop into a severe fault within 24 hours. The cloud immediately issued a fault warning and handling recommendations to Hydropower Station A: reduce the load of Unit 1 to 50% and arrange maintenance personnel to perform rotor balancing within 24 hours.

[0066] At 14:00, communication between hydropower station A and the remote cloud control platform was interrupted due to the fiber optic cable being severed during construction. The local safety fault-tolerant module automatically switched to local autonomous operation mode, maintaining Unit 1 at 50% load and Unit 2 at full capacity according to the most recently received dispatch instructions, while continuing to monitor the equipment operating status based on the locally updated fault judgment rules.

[0067] At 16:30, the communication fiber optic cable was repaired, and communication was restored. The local security fault-tolerant module immediately uploaded the operational data and fault records during the communication interruption to the cloud and automatically switched back to the cloud-edge collaborative operation mode to execute the latest scheduling instructions from the cloud.

[0068] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups, characterized in that, It includes a remote cloud control platform and multiple edge power station nodes equipped with PSK-8 controllers, wherein the remote cloud control platform and the multiple edge power station nodes are connected through a communication network; The remote cloud control platform includes: The big data optimization and scheduling module is used to perform global optimization calculations based on regional power load data, watershed flow situation data, and the operating status data of each edge power station node, generate hydropower station group scheduling instructions, and issue them to the corresponding edge power station nodes. The fault probability calculation module is used to calculate the fault probability and abnormal situation probability of each hydropower station equipment in real time based on the equipment operation characteristic data uploaded by the edge power station nodes. The edge database update module is used to sort the fault probabilities based on the fault probability calculation results, generate updated fault judgment rules and handling methods, and determine the number of rules that the edge computing node database needs to store based on the probability of abnormal situations. The deep fault analysis module is used to analyze the abnormal features uploaded by edge power station nodes. When a suspected fault feature is found, a raw data extraction request is sent to the corresponding edge power station node, and the obtained raw data is used for simulation and fault identification and prediction. The PSK-8 controller of the edge power station node includes: The local execution module is used to receive and execute scheduling instructions issued by the remote cloud control platform to control the operation of hydropower station equipment; The local safety fault tolerance module is used to autonomously monitor the equipment's operating status and handle emergency faults based on locally stored fault judgment rules and handling methods when communication is interrupted. The data feature extraction module is used to extract features from the raw operating data collected by hydropower station equipment, generate equipment operating feature data, and upload it to the remote cloud control platform. Edge database is used to store fault diagnosis rules, handling methods, and equipment operation history data; The raw data on-demand transmission module is used to receive raw data extraction requests sent by the remote cloud control platform and upload the corresponding raw operating data to the remote cloud control platform.

2. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 1, characterized in that, The edge database update module performs the following steps: The calculated failure probabilities of each device are sorted in descending order; Prioritize updating fault diagnosis rules and handling methods for equipment with high failure probability; The number of rules stored in the edge database is dynamically adjusted based on the probability value of abnormal situations.

3. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 2, characterized in that, The specific steps for dynamically adjusting the number of rules stored in the edge database based on the probability value of abnormal situations are as follows: When the probability of an anomaly exceeds the first preset threshold, the number of rules stored in the edge database is increased. When the probability of an anomaly is lower than the second preset threshold, the number of rules stored in the edge database is reduced.

4. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 1, characterized in that, The data feature extraction module performs the following steps: Preprocess the raw operating data, including data cleaning, normalization and noise reduction. A combination of time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis was used to extract key feature parameters from the original data. The extracted key feature parameters are packaged according to a preset format and uploaded to the remote cloud control platform on a regular basis.

5. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 1, characterized in that, The deep fault analysis module performs the following steps: Establish digital twin models of hydropower station equipment; The acquired raw operational data is input into the digital twin model for simulation. Based on simulation results and combined with historical fault data, machine learning algorithms are used to identify faults and predict fault development trends. Generate fault warning information and handling suggestions, and send them to the corresponding edge power station nodes.

6. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 1, characterized in that, The local security fault-tolerant module performs the following steps: It monitors the status of the communication network in real time and automatically switches to local autonomous operation mode when a communication interruption is detected. Based on the fault diagnosis rules stored in the edge database, the locally collected device operation data is analyzed in real time. When a device malfunction is detected, emergency operations are automatically executed according to the corresponding handling methods, including device shutdown, load adjustment, and alarm notification.

7. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 1, characterized in that, The fault probability calculation module uses a pre-trained Long Short-Term Memory (LSTM) network model to calculate the fault probability and abnormal situation probability of each hydropower station equipment. The LSTM model is trained and updated periodically using historical fault data and normal operation data.

8. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 1, characterized in that, After receiving the raw data extraction request from the remote cloud control platform, the raw data on-demand transmission module uses the LZ4 compression algorithm to compress the corresponding raw running data before uploading it to the remote cloud control platform.

9. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 1, characterized in that, The big data optimization and scheduling module uses a genetic algorithm or a particle swarm optimization algorithm, with the objective functions of maximizing power generation efficiency and optimizing water resource utilization. It takes into account constraints such as the head, flow rate, and unit output limits of the hydropower station to generate a hydropower station group scheduling plan for the next 24 hours.

10. The cloud-edge collaborative intelligent scheduling and fault-tolerant control system for hydropower station groups according to claim 6, characterized in that, The local security fault-tolerant module continues to attempt to establish communication with the remote cloud control platform during the communication interruption. Once communication is restored, it immediately uploads the operating data and fault records during the communication interruption to the remote cloud control platform and switches back to the cloud-edge collaborative operation mode.

Images