Wind power plant remote monitoring and intelligent scheduling method and system

Abstract

The invention relates to the field of wind power plants, and discloses a wind power plant remote monitoring and intelligent scheduling method and system, and the method comprises the steps: obtaining vibration data, voltage and wind speed, and forming a fusion preparation data set; classifying and merging data in the fusion preparation data set to obtain a wind power plant operation state model; performing anomaly detection to obtain real-time data marked as potential anomaly; abnormal feature extraction is carried out on the real-time data marked as potential anomalies, and an abnormal feature set is obtained; on the basis of the abnormal feature set, weighted summation is carried out according to the influence range and the fault priority to calculate each abnormal severity score, and a graded abnormal list is generated; matching a high-risk response template according to the classification exception list; generating an initial scheduling instruction based on the matched response template; taking the initial scheduling instruction as a final scheduling instruction; and transmitting an execution signal to a remote control center according to the final scheduling instruction. The method has the following effect that the accuracy of wind power plant anomaly detection is improved.

Description

Technical Field

[0001] This invention relates to the field of wind farms, and more particularly to a method and system for remote monitoring and intelligent scheduling of wind farms. Background Technology

[0002] As an important component of renewable energy, the stable operation of wind farms is crucial for ensuring energy supply and promoting green development.

[0003] Remote monitoring and intelligent dispatching technologies can monitor the real-time operating status of wind farms and respond promptly to anomalies, ensuring equipment safety and power generation efficiency. However, the complex operating environment of wind farms, involving wind turbines, power grids, and weather, makes it difficult for wind farms to respond quickly and appropriately to sudden anomalies, leading to reduced efficiency or even equipment damage.

[0004] Solving the above-mentioned technical problems is a technical challenge that needs to be overcome by those skilled in the art. Summary of the Invention

[0005] This invention provides a method and system for remote monitoring and intelligent scheduling of wind farms, which at least partially solves the above-mentioned technical problems.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for remote monitoring and intelligent scheduling of wind farms, comprising: Vibration data, voltage, and wind speed are acquired using accelerometers, voltage sensors, and ultrasonic anemometers, and then used to construct a fusion preparation dataset. A grouping and integration method was used to classify and merge the data in the fusion preparation dataset to obtain a wind farm operation status model. The real-time collected vibration data, voltage, and wind speed are input into the wind farm operation status model anomaly detection to obtain real-time data marked as potential anomalies; anomaly features are extracted from the real-time data marked as potential anomalies to obtain an anomaly feature set. Based on the set of abnormal features, the severity scores of each abnormality are calculated by weighted summation according to the scope of impact and fault priority, and a graded abnormality list is generated; the abnormalities in the graded abnormality list are divided into high-risk, medium-risk and low-risk. Based on the tiered anomaly list, match high-risk response templates; based on the matched response templates, analyze the power grid status corresponding to the anomaly data; if the power grid status indication is unstable, generate initial dispatch instructions. Acquire current meteorological data and obtain meteorological influencing factors; analyze the matching degree between the initial dispatching command and the meteorological influencing factors. If the matching degree is lower than the preset threshold, adjust the initial dispatching command and use the adjusted dispatching command as the final dispatching command; if the matching degree is higher than the preset threshold, use the initial dispatching command as the final dispatching command. The execution signal is transmitted to the remote control center according to the final scheduling instruction.

[0007] Secondly, the present invention provides a remote monitoring and intelligent dispatching system for wind farms, comprising: The first processing module is used to: acquire vibration data, voltage and wind speed based on the accelerometer, voltage sensor and ultrasonic anemometer respectively, and form a fusion preparation dataset; The second processing module is used to classify and merge the data in the fusion preparation dataset using a grouping and integration method to obtain a wind farm operation status model. The third processing module is used to: input the real-time collected vibration data, voltage and wind speed into the wind farm operation status model anomaly detection to obtain real-time data marked as potential anomalies; and extract anomaly features from the real-time data marked as potential anomalies to obtain an anomaly feature set. The fourth processing module is used to: calculate the severity score of each anomaly based on the set of anomaly features, weighted by the scope of impact and fault priority, and generate a graded anomaly list; the anomalies in the graded anomaly list are divided into high-risk, medium-risk and low-risk. The fifth processing module is used to: match high-risk response templates according to the hierarchical anomaly list; analyze the power grid status corresponding to the anomaly data based on the matched response templates; and generate initial dispatch instructions if the power grid status indication is unstable. The sixth processing module is used to: acquire current meteorological data and obtain meteorological influencing factors; analyze the matching degree between the initial dispatching command and the meteorological influencing factors; if the matching degree is lower than a preset threshold, adjust the initial dispatching command and use the adjusted dispatching command as the final dispatching command; if the matching degree is higher than a preset threshold, use the initial dispatching command as the final dispatching command. The seventh processing module is used to transmit execution signals to the remote control center according to the final scheduling instruction.

[0008] Compared with existing technologies, the present invention has at least the following beneficial effects: it improves the accuracy of wind farm anomaly detection and ensures grid stability and equipment safety. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a remote monitoring and intelligent scheduling method for wind farms provided in the first embodiment of the present invention; Figure 2 This is a block diagram of a wind farm remote monitoring and intelligent dispatching system provided in the second embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Reference Figure 1 The first embodiment of the present invention provides a method for remote monitoring and intelligent scheduling of wind farms, including the following steps: S101, based on the accelerometer, voltage sensor and ultrasonic anemometer, respectively, vibration data, voltage and wind speed are acquired and a fusion preparation dataset is formed; S102, The grouping and integration method is used to classify and merge the data in the fusion preparation dataset to obtain the wind farm operation status model; S103, The real-time collected vibration data, voltage and wind speed are input into the wind farm operation status model for anomaly detection to obtain real-time data marked as potential anomalies; anomaly features are extracted from the real-time data marked as potential anomalies to obtain an anomaly feature set. S104. Based on the set of abnormal features, calculate the severity score of each abnormality by weighted summation according to the scope of impact and fault priority, and generate a graded abnormality list; in the graded abnormality list, the abnormalities are divided into high-risk, medium-risk and low-risk. S105, Match high-risk response templates according to the hierarchical anomaly list; Analyze the power grid status corresponding to the anomaly data based on the matched response templates; If the power grid status indication is unstable, generate an initial dispatch instruction. S106: Obtain current meteorological data and obtain meteorological influencing factors; analyze the matching degree between the initial dispatching command and the meteorological influencing factors. If the matching degree is lower than the preset threshold, adjust the initial dispatching command and use the adjusted dispatching command as the final dispatching command; if the matching degree is higher than the preset threshold, use the initial dispatching command as the final dispatching command. S107, transmit the execution signal to the remote control center according to the final scheduling instruction.

[0012] In one implementation, a grouping and integration method is used to classify and merge the data in the fusion preparation dataset to obtain a wind farm operation status model, including: The vibration data, voltage data, and wind speed data in the fusion preparation dataset will be converted into a standardized dataset. Collect wind farm power output data and standardize the power output data to obtain a standardized power dataset. Clustering algorithms were used to group and integrate the standardized dataset and the standardized power dataset. Using vibration frequency, voltage fluctuation amplitude, wind speed, and power output as feature dimensions, the data in the standardized dataset and the standardized power data were divided into normal, slightly abnormal, and severely abnormal categories to obtain the operating status data. Vibration feature data of blades is extracted from the operating status data to construct a vibration feature set. If there are feature values ​​in the vibration feature set that exceed the preset vibration feature threshold, a clustering algorithm is used to regroup the vibration feature set to obtain the basic vibration data; otherwise, the vibration feature set is used as the basic vibration data. Based on the basic vibration data and the standardized voltage, wind speed and power data in the operation status data, the support vector machine algorithm is used to construct the wind farm operation status model.

[0013] Specifically, the vibration, voltage, and wind speed data in the fusion preparation dataset were converted into a standardized dataset. The vibration data in the fusion preparation dataset originated from triaxial vibration signals collected by accelerometers deployed at the root of wind turbine blades; the voltage data came from minute voltage fluctuations recorded by voltage sensors deployed in the power grid; and the wind speed data came from wind speed changes collected by ultrasonic anemometers. To eliminate the impact of differences in data dimensions and magnitudes on subsequent analysis, standardization was employed: vibration data was standardized using z-scores; voltage and wind speed data were normalized. To improve data quality, noise filtering was performed on the raw data before standardization: wavelet transform was used to separate high-frequency noise from the vibration data; Kalman filtering was used for the voltage data; and median filtering was used to remove abrupt noise from the wind speed data. The standardized dataset was obtained after these processing steps.

[0014] The power output data of each wind turbine is collected in real time by the power monitoring device of the wind farm. In order to ensure that the power data is of the same magnitude as the aforementioned standardized datasets of vibration, voltage and wind speed, a standardized power dataset is obtained by processing with a standardization method.

[0015] Clustering algorithms are used to group and integrate standardized datasets and standardized power datasets. Vibration frequency, voltage fluctuation amplitude, wind speed, and power output values ​​are used as feature dimensions to classify the data into normal, slightly abnormal, and severely abnormal categories, thus obtaining operational status data. In this embodiment, the K-means clustering algorithm is selected, with the number of clusters set to K=3, corresponding to the normal, slightly abnormal, and severely abnormal categories, respectively. Vibration frequency, voltage fluctuation amplitude, wind speed, and power output values ​​are used as four-dimensional features to perform clustering calculations on the merged data from the two datasets. For example, if the feature vector of the merged data at a certain moment is: vibration frequency 8Hz, voltage fluctuation amplitude 0.05V, wind speed 15m / s, and power output value 0.3, the Euclidean distance between this vector and the cluster centers of the normal, slightly abnormal, and severely abnormal categories is calculated. If the Euclidean distance to the cluster center of the slightly abnormal category is the smallest, then the data point is classified into the slightly abnormal category.

[0016] After clustering and grouping, all data points and their category labels are integrated to form operational status data. Blade vibration characteristic data is extracted from the operational status data to construct a vibration characteristic set. If any characteristic value in the vibration characteristic set exceeds a preset vibration characteristic threshold, a clustering algorithm is used to further group the vibration characteristic set to obtain basic vibration data; otherwise, the vibration characteristic set is used as the basic vibration data. This can be understood as filtering all data related to blade vibration from the operational status data and integrating them to form a vibration characteristic set—a collection of all blade vibration characteristic data—used for centralized analysis of blade vibration status. The preset vibration characteristic threshold is a pre-set upper limit or normal range boundary value for vibration characteristics based on the design parameters, historical operational data, and safety operation standards of the wind turbine equipment in the wind farm. Threshold judgment is performed on each characteristic value in the vibration characteristic set: if any characteristic value exceeds the preset threshold, the K-means clustering algorithm is used again to further group the vibration characteristic set, removing abnormal interference data to obtain basic vibration data; if all characteristic values ​​in the vibration characteristic set are within the preset threshold range, the vibration characteristic set is directly used as the basic vibration data.

[0017] Based on the basic vibration data combined with the standardized voltage, wind speed, and power data from the operational status data, a support vector machine algorithm is used to construct a wind farm operational status model. The basic vibration data is integrated with the standardized voltage, wind speed, and power data from the operational status data to form a sample set for model training. Each sample contains basic vibration characteristics, standardized voltage, standardized wind speed, and standardized power, as well as the corresponding operational status label.

[0018] The Support Vector Machine (SVM) algorithm was selected to construct the model, and the radial basis function (RBF) was chosen as the kernel function to adapt to the nonlinear characteristics of the multi-dimensional monitoring data of the wind farm. The algorithm regularization parameter C=1.0 and the kernel function parameter γ=0.1 were set. The sample set was divided into a training set and a test set in an 8:2 ratio. The SVM algorithm was trained using the training set, and the model performance was verified using the test set.

[0019] This solution addresses the issues of inconsistent data magnitudes and noise interference from multiple data sources by standardizing vibration, voltage, and wind speed data.

[0020] In one implementation, real-time collected vibration data, voltage, and wind speed are input into the wind farm operation state model for anomaly detection to obtain real-time data marked as potential anomalies; anomaly features are extracted from the real-time data marked as potential anomalies to obtain an anomaly feature set, including: Real-time acquisition of vibration data, voltage, and wind speed; The mean and standard deviation of vibration data, voltage, and wind speed for each window were calculated using the sliding window method to obtain the variation trend of each parameter; The changing trend is input into the wind farm operation status model for anomaly detection. If the rate of change of one of the parameters in a certain window exceeds the corresponding threshold, the real-time data in that window is marked as a potential anomaly. For real-time data marked as potential anomalies, principal component analysis algorithm is used to extract anomaly features, resulting in a set of anomaly features including vibration frequency change rate, voltage change rate, and wind speed change rate.

[0021] Specifically, the sliding window method was used to process the real-time collected vibration data, voltage, and wind speed. The sliding window size was set to 60 seconds, and the mean and standard deviation of vibration data, voltage, and wind speed within each window were calculated.

[0022] The changing trend is input into the wind farm operation status model for anomaly detection. If the rate of change of a certain parameter exceeds the corresponding threshold, the real-time data in that window is marked as a potential anomaly. For the real-time data marked as potential anomalies, principal component analysis is used to extract anomaly features, resulting in a set of anomaly features including the rate of change of vibration frequency, voltage, and wind speed.

[0023] In one implementation, based on the set of abnormal features, a severity score for each abnormality is calculated by weighted summation according to its impact range and fault priority, and a graded abnormality list is generated, including: Extract the impact range and fault priority of each anomaly from the set of anomaly features; Calculate the initial severity score for each anomaly; where the initial severity score = scope of impact × first weight + fault priority × second weight; The initial severity score is normalized to obtain a normalized severity score; Anomalies with a normalized severity score greater than the first score are marked as high-risk anomalies, anomalies with a normalized severity score greater than the second score and less than the first score are marked as intermediate-risk anomalies, and anomalies with a normalized severity score less than the second score are marked as low-risk anomalies. A tiered anomaly list is generated based on anomalies classified by anomaly level, normalized score, scope of impact, and fault priority.

[0024] Specifically, the impact range and fault priority of each anomaly are extracted from the set of abnormal features. The impact range refers to the number of wind farm systems or equipment that the anomaly may affect. For example, if a voltage anomaly may affect 3 downstream substations and 10 wind turbines, its impact range is 10 wind turbines and 3 substations. This indicator is based on the anomaly features and the topology relationship of the wind farm equipment, such as the connection relationship between wind turbines and the grid, and the coverage area of ​​substations. Fault priority refers to the urgency of the anomaly requiring handling. It is set according to the degree of threat the anomaly poses to equipment safety and grid stability, and is divided into 1-5 levels, with level 1 being the lowest and level 5 the highest. This indicator is determined with reference to wind farm safety operation specifications and historical fault handling experience. For example, if the anomaly feature set of a certain anomaly shows a voltage change rate of 200V / minute and a vibration frequency change rate of 1.5Hz / minute, combined with the equipment topology, its impact range is 10 systems, and its fault priority is determined to be 3 according to safety specifications; another anomaly has an impact range of 5 systems, with a fault priority of 5; and a third anomaly has an impact range of 20 systems, with a fault priority of 2.

[0025] Calculate the initial severity score for each anomaly; where the initial severity score = impact range × first weight + fault priority × second weight. Based on the extracted impact range and fault priority, weight coefficients are introduced to balance their contributions to anomaly severity. The first weight is the proportion of impact range in the score calculation, and the second weight is the proportion of fault priority in the score calculation; their sum is 1. Calculate the initial severity score for each anomaly using the above formula.

[0026] The initial severity score is normalized to obtain a normalized severity score. In this embodiment, a linear normalization formula is used: Normalized severity score = (current initial anomaly score - minimum of all initial anomaly scores) / (maximum of all initial anomaly scores - minimum of all initial anomaly scores) × 100.

[0027] Anomalies with a normalized severity score greater than the first score are marked as high-risk anomalies; those with a normalized severity score greater than the second score but less than the first score are marked as medium-risk anomalies; and those with a normalized severity score less than the second score are marked as low-risk anomalies. The first and second scores are grading thresholds set based on wind farm fault handling resources and equipment tolerance limits. The first score is set to 80, and the second score to 50; that is, a normalized score ≥ 80 is high-risk, 50 ≤ normalized score < 80 is medium-risk, and a normalized score < 50 is low-risk. A graded anomaly list is generated based on the anomaly level, normalized score, impact range, and fault priority.

[0028] This solution solves the problem of incomplete judgment by traditional single indicators by extracting two core indicators, the scope of impact and the fault priority, from the set of abnormal features.

[0029] In one implementation, a high-risk response template is matched according to a tiered anomaly list; based on the matched response template, the power grid status corresponding to the anomaly data is analyzed; if the power grid status indicates instability, an initial dispatch instruction is generated, including: Filter out high-risk abnormal data from the graded abnormality list; A rule engine is built using the Drools framework. The rule engine predefines voltage, frequency, and power rules. The rule engine extracts real-time voltage, frequency, and power parameters from the abnormal feature set, substitutes them into the rules, and outputs classification results that include abnormality type and parameter deviation. The rule engine calls a predefined response template library; the response templates in the response template library contain exception types, response operations, and applicable scenarios, and the corresponding response template is matched based on the classification results. Based on the matched response template, the grid node type, number of downstream affected substations, power fluctuation value and fluctuation duration of abnormal fluctuations are extracted from the wind farm grid monitoring system, and it is determined whether the grid instability judgment criteria are met; the result of the judgment of stability or instability is output. If the power grid is determined to be unstable, an emergency isolation protocol is triggered to disconnect the abnormal node and switch to a backup line. Using basic response instructions as a framework, an initial scheduling instruction is generated using a linear programming model. The objective function of the linear programming model is to minimize the sum of power loss and scheduling cost. The constraints include load balancing, the adjusted power of the main line being less than the first power, and the power adjustment range of a single substation being less than the first percentage of the rated power. The initial dispatch command is sent to the power grid control system.

[0030] In one implementation, high-risk anomaly data is selected from the graded anomaly list, and a rule engine is built using the Drools framework. The rule engine predefines voltage, frequency, and power rules. The rule engine extracts real-time voltage, frequency, and power parameters from the anomaly feature set, substitutes them into the rules, and outputs a classification result including anomaly type and parameter deviation. The Drools framework is an open-source business rule management system that can automatically analyze and classify input data through preset rules; here, it is used to build a rule engine for wind farm anomaly classification. The predefined rules are based on the wind farm grid safety operation specifications. The voltage rule is: if the real-time voltage > 1.1 × rated voltage or the real-time voltage < 0.9 × rated voltage, it is determined to be a voltage anomaly; the frequency rule is: if the real-time grid frequency > 50Hz + 0.5Hz or < 50Hz - 0.5Hz, it is determined to be a frequency anomaly; the power rule is: if the real-time power fluctuation value > 4MW and the duration > 5 seconds, it is determined to be a power anomaly.

[0031] The rules engine calls a predefined response template library. This library contains response templates that include anomaly types, response actions, and applicable scenarios. The engine matches the corresponding response template based on the classification results. The predefined response template library is a database storing handling solutions for various anomalies in the wind farm; the templates are stored in JSON format. Based on the classification results output above, such as an anomaly type of voltage anomaly, the rules engine searches the template library for templates of the same anomaly type that match the parameter deviation range. For example, for the classification results of voltage deviation, a response template that reduces load is matched.

[0032] Based on a matched response template, the system extracts the grid node type, number of downstream affected substations, power fluctuation value, and fluctuation duration of abnormal fluctuations from the wind farm grid monitoring system and determines whether they meet the grid instability judgment criteria; it then outputs a stable or unstable judgment result. The wind farm grid monitoring system is a monitoring platform that collects and stores grid operation data in real time, and data can be extracted through the SCADA interface. The grid instability judgment criteria are based on the grid capacity and safety redundancy settings of the wind farm. Specifically, if the abnormal node is a main grid node, the number of downstream affected substations is ≥3, the power fluctuation value is >4MW, and the fluctuation duration is >5 seconds, then the grid is judged to be unstable; if none of these conditions are met, it is judged to be stable. For example, if the grid is determined to be unstable, an emergency isolation protocol is triggered, executing the operation of disconnecting the abnormal node and switching to a backup line. The emergency isolation protocol is a pre-set fault isolation scheme for grid instability scenarios, used to quickly disconnect the abnormal area from the normal grid to prevent the fault from spreading. The rules engine triggers this protocol by calling the SCADA system interface. The specific operations include: sending disconnect commands to the circuit breakers of abnormal nodes and sending closing commands to the circuit breakers of backup lines to ensure that the power supply to downstream substations is not interrupted.

[0033] Using the basic response command as a framework, an initial dispatch command is generated using a linear programming model. The objective function of the linear programming model is to minimize the sum of power loss and dispatch cost. Constraints include load balancing, the adjusted power of the main line being less than the first power, and the power adjustment range of a single substation being less than a first percentage of its rated power. The basic response command is the core operation in the aforementioned matched response template, serving as the basic framework for the initial dispatch command. The linear programming model is an optimization tool that solves for optimal solutions through mathematical modeling; here, it is used to generate the optimal dispatch scheme while satisfying grid security constraints. Power loss refers to the power loss caused by load adjustment during dispatch, and dispatch cost refers to the operational cost of switching lines and adjusting loads. Among the constraints, the first power is the maximum allowable power of the wind farm's main line; the first percentage is the safe upper limit for substation power adjustment; and the load balancing constraint states that the total power demand of all substations after adjustment equals the total power supplied by the grid.

[0034] The initial dispatch command is sent to the power grid control system. The generated initial dispatch command is then sent to the wind farm's power grid control system via MQTT. This system is responsible for receiving and executing the dispatch command and controlling the operating status of the power grid equipment.

[0035] This solution solves the problem of chaotic anomaly handling priorities by accurately filtering high-risk data from a tiered anomaly list, avoiding low-risk anomalies from occupying processing resources. It also uses the Drools framework to build a rule engine and predefines voltage, frequency, and power rules to classify anomaly types, solving the problems of low efficiency and error-proneness in manual judgment, effectively preventing fault propagation, reducing power loss and scheduling costs, and ensuring the safe and stable operation of the power grid.

[0036] In one implementation, current meteorological data is acquired and meteorological influencing factors are obtained; the matching degree between the initial dispatch instruction and the meteorological influencing factors is analyzed; if the matching degree is lower than a preset threshold, the initial dispatch instruction is adjusted, and the adjusted dispatch instruction is used as the final dispatch instruction; if the matching degree is higher than the preset threshold, the initial dispatch instruction is used as the final dispatch instruction, including: Real-time acquisition of current meteorological data of the wind farm; the meteorological data includes wind speed, wind direction and temperature; calculation of the theoretical power of a single wind turbine based on wind speed and temperature and conversion of the theoretical power into actual output power by combining the efficiency loss of wind farm equipment; the mapping relationship between the actual output power of a single turbine, wind speed and temperature is used as a meteorological influencing factor. Calculate the deviation rate between the power of a single wind turbine in the initial dispatch command and the actual output power of a single turbine in the meteorological influence factors; if the deviation rate is greater than the preset deviation rate, the matching degree is determined to be lower than the preset threshold; if the deviation rate is less than the preset deviation rate, the matching degree is determined to be higher than the preset threshold. If the matching degree is lower than the preset threshold, time series analysis and iterative optimization algorithms are used to adjust the scheduling instruction parameters and generate the adjusted scheduling instruction. If the matching degree is higher than the preset threshold, the initial scheduling instruction will be used as the final scheduling instruction.

[0037] Specifically, real-time meteorological data of the wind farm is acquired; this meteorological data includes wind speed, wind direction, and temperature. The theoretical power of a single wind turbine is calculated based on wind speed and temperature, and then converted into actual output power by considering the efficiency losses of the wind farm equipment. The mapping relationship between the actual output power of a single turbine, wind speed, and temperature is used as a meteorological influencing factor. Specifically, real-time meteorological data can be collected through the API interface of the meteorological station deployed at the wind farm. The standardized data transmission interface between the meteorological station and the wind farm dispatch system enables real-time synchronization of meteorological data. The formula for the theoretical power of a single wind turbine is... in, Where A is the air density and A is the swept area of ​​the wind turbine blades. For real-time wind speed, The power factor is the efficiency coefficient of a wind turbine in converting wind energy into electrical energy. Combined with the efficiency losses of wind farm equipment, the actual output power of a single turbine is calculated. The aforementioned meteorological influence factors represent the set of relationships between the actual output power of a single turbine and real-time wind speed and temperature, reflecting the constraints of meteorological conditions on the wind turbine's power generation capacity.

[0038] The deviation rate between the power output of a single wind turbine in the initial dispatch command and the actual output power of a single turbine in the meteorological influence factors is calculated. If the deviation rate is greater than a preset deviation rate, the matching degree is determined to be lower than a preset threshold; if the deviation rate is less than the preset deviation rate, the matching degree is determined to be higher than the preset threshold. Specifically, the actual output power of a single turbine under the corresponding meteorological conditions is extracted from the meteorological influence factors.

[0039] If the matching degree is lower than a preset threshold, time series analysis and iterative optimization algorithms are used to adjust the scheduling command parameters and generate adjusted scheduling commands. The time series analysis employs the ARIMA model, predicting wind speed for the next 10 minutes based on historical wind speed data from the past hour, determining the dynamic trend of meteorological conditions, and preventing the adjusted commands from becoming mismatched again due to sudden weather changes. The iterative optimization algorithm uses a genetic algorithm, with the optimization objective function set as follows: ,in This refers to the adjusted power of a single fan. For a single wind turbine load factor, The weighting coefficients are used to balance the total power generation and equipment load. The constraints include: the adjusted power of a single unit ≤ the actual output power, and the total power ≤ the grid's capacity. The algorithm parameters are set as follows: number of particles 50, number of iterations 50, inertia weight w decreasing linearly from 0.9 to 0.4, and learning factors c1=c2=2. Based on the wind speed prediction results of the ARIMA model, the adjusted dispatch command is finally determined after 50 iterations of optimization.

[0040] If the matching degree is higher than the preset threshold, the initial scheduling command will be directly used as the final scheduling command. For example, if the initial scheduling command has a single unit power of 3.8MW and the actual output power is 3.92MW, the deviation rate is 3.06% < the preset threshold, which means that the initial command has adapted to the current weather conditions and does not need to be adjusted. It will be directly used as the final scheduling command, reducing unnecessary consumption of computing resources and the impact of frequent command adjustments on the equipment.

[0041] By using the ARIMA model to predict weather trends and the genetic algorithm to iteratively optimize parameters, we can avoid the failure of adjusted instructions due to sudden weather changes, and find the optimal solution while balancing power generation efficiency and equipment load. This improves the power generation efficiency of wind farms and avoids resource misallocation and failure risks.

[0042] In one embodiment, after transmitting the execution signal to the remote control center according to the final scheduling instruction, the method further includes: Collect feedback data; the feedback data includes wind turbine speed, power output, grid load and grid frequency, forming a monitoring dataset; The Z-score algorithm is used to detect anomalies in the monitoring dataset; if the feedback data exceeds the corresponding threshold, it is determined that the anomaly has not been mitigated. If the anomaly is determined to be unresolved, an updated severity score is calculated based on frequency deviation, power deviation, and response time. If the updated severity score is greater than the preset severity score threshold, the particle swarm optimization algorithm is used to re-optimize the scheduling scheme and generate a new execution signal. The objective function of the particle swarm optimization algorithm is: .

[0043] Specifically, feedback data is collected; this feedback data includes wind turbine speed, power output, grid load, and grid frequency, forming a monitoring dataset. In particular, after transmitting the final dispatch command to the remote control center, feedback data is collected in real time through the sensor network deployed at the wind farm and the grid monitoring system.

[0044] The Z-score algorithm is used to detect anomalies in the monitoring dataset. If the feedback data exceeds the corresponding threshold, the anomaly is determined to be unresolved. The Z-score algorithm is a quantitative anomaly detection algorithm based on the normal distribution characteristics of data. It determines whether the data is abnormal by calculating the deviation of a single data point from the mean of the dataset. For example, based on historical normal operation data of a wind farm, a preset Z-score threshold of ±3 is used, and corresponding actual thresholds are set for each feedback data point. Each data point in the monitoring dataset is substituted into the Z-score formula for calculation. If the Z-score of a feedback data point exceeds the corresponding threshold, the initial anomaly is determined to be unresolved.

[0045] When the Z-score algorithm determines that the abnormality has not been alleviated, the severity of the abnormality needs to be reassessed. The updated severity score is calculated using a weighted summation formula, as follows: The frequency deviation is the absolute difference between the feedback grid frequency and the rated frequency; the power deviation is the absolute difference between the feedback power output and the expected power of the dispatch command; the response time is the time interval from the transmission of the final dispatch command to the acquisition of abnormal feedback data; each weight is set based on the degree of impact of each indicator on grid stability, with frequency deviation having the greatest impact on grid security, hence its highest weight. The updated severity score is calculated by substituting example data.

[0046] If the updated severity score is greater than the preset severity score threshold, the particle swarm optimization algorithm is used to re-optimize the scheduling scheme and generate a new execution signal. The objective function of the particle swarm optimization algorithm is: , For the actual load of the power grid as feedback, The expected load for the safe operation of the wind farm grid is determined. If the updated severity score is greater than the corresponding threshold, the particle swarm optimization algorithm is activated to minimize grid load fluctuations. Combined with constraints including the adjusted single wind turbine power ≤ the actual output power in the meteorological influence factor and the total power ≤ the grid capacity, the optimal scheduling scheme is obtained through multiple iterations.

[0047] In one embodiment, the method further includes: The abnormal handling process corresponding to the new execution signal is used as a training sample, and the abnormal feature set, initial scheduling instructions, meteorological influence factors, feedback data and final control effect are recorded. Construct a response template knowledge graph; the knowledge graph uses anomaly type and handling strategy as nodes, and edges represent the mapping relationship from anomaly to handling strategy; record the historical success rate, response time and power recovery rate of the handling path as the weights of the edges; The response template knowledge graph is updated based on the training samples, and the Q-value of the response strategy is iteratively learned using a reinforcement learning algorithm. The state space includes the severity of the anomaly, power grid stability, and meteorological influence factors. The action space is a combination of optional scheduling operations. The reward function consists of power recovery speed, equipment safety margin, and scheduling cost. When the number of successful responses of a certain type of exception handling strategy in the knowledge graph exceeds a preset threshold and the Q value converges, a new response template is generated and injected into the response template library. Regularly mark and discard templates in the response template library that have not been matched for a long time or whose failure rate exceeds the third threshold.

[0048] Specifically, by recording the complete anomaly handling process corresponding to the new execution signal as training samples, the problem of fragmented historical handling data and inability to provide effective data support for strategy optimization is solved. By constructing a response template knowledge graph with anomaly type and handling strategy as nodes and success rate as a weight, the problem of lack of quantitative evaluation of handling strategies and inability to intuitively judge their merits is solved. By updating the knowledge graph based on training samples and iterating the Q value through reinforcement learning, the problem of strategies being unable to dynamically adapt to complex scenarios and struggling to balance power recovery speed, equipment safety, and scheduling costs is solved. When the cumulative number of successes reaches the target and the Q value converges, a new template is generated, solving the problem of no suitable template for new types of anomalies, thereby improving the accuracy, success rate, and response efficiency of anomaly handling.

[0049] In one embodiment, before determining that the abnormality has not been alleviated, the method further includes: An anomaly cause-effect graph model is constructed, wherein the scheduling instruction execution status, wind turbine mechanical status, and meteorological forecast deviation are used as potential dependent variables, and the continued existence of anomalies is used as the outcome variable. The parameters of the causal graph model are learned based on historical operation and maintenance data and real-time feedback data to obtain the contribution of each dependent variable to the result. When the anomaly is determined to be unresolved, the current feedback data is input into the cause-effect graph model for backward reasoning analysis, and the most likely cause of the anomaly's persistence is output. Before re-optimizing the scheduling scheme using the particle swarm optimization algorithm, adjust the optimization constraints or objective function weights based on the most likely cause.

[0050] Specifically, by constructing a causal graph model with scheduling command execution status, wind turbine mechanical status, and meteorological forecast deviation as potential dependent variables and anomaly persistence as the outcome variable, and combining historical operation and maintenance data with real-time feedback data to learn the contribution of each dependent variable to anomaly persistence, the problem of accurately locating the root cause of anomalies is solved. When the anomaly is not alleviated, the most likely root cause is output through backward reasoning, and the constraints or objective function weights of the particle swarm optimization algorithm are adjusted accordingly. This solves the problem of directional deviation caused by blind optimization, reduces the risk caused by improper optimization, and ensures the stable operation of the wind farm.

[0051] Reference Figure 2 The second embodiment of the present invention provides a remote monitoring and intelligent dispatching system for wind farms, comprising: The first processing module 201 is used to: acquire vibration data, voltage and wind speed based on the accelerometer, voltage sensor and ultrasonic anemometer respectively, and form a fusion preparation dataset; The second processing module 202 is used to: classify and merge the data in the fusion preparation dataset using a grouping and integration method to obtain a wind farm operation status model; The third processing module 203 is used to: input the real-time collected vibration data, voltage and wind speed into the wind farm operation status model for anomaly detection, and obtain real-time data marked as potential anomalies; extract anomaly features from the real-time data marked as potential anomalies to obtain an anomaly feature set. The fourth processing module 204 is used to: calculate the severity score of each anomaly based on the set of anomaly features by weighted summation according to the scope of influence and fault priority, and generate a graded anomaly list; the anomalies in the graded anomaly list are divided into high-risk, medium-risk and low-risk. The fifth processing module 205 is used to: match high-risk response templates according to the hierarchical anomaly list; analyze the power grid status corresponding to the anomaly data based on the matched response templates; and generate initial dispatch instructions if the power grid status indication is unstable. The sixth processing module 206 is used to: acquire current meteorological data and obtain meteorological influencing factors; analyze the matching degree between the initial scheduling instruction and the meteorological influencing factors; if the matching degree is lower than a preset threshold, adjust the initial scheduling instruction and use the adjusted scheduling instruction as the final scheduling instruction; if the matching degree is higher than a preset threshold, use the initial scheduling instruction as the final scheduling instruction. The seventh processing module 207 is used to transmit an execution signal to the remote control center according to the final scheduling instruction.

[0052] It should be noted that the wind farm remote monitoring and intelligent scheduling system provided in this embodiment of the invention is used to execute all the process steps of the wind farm remote monitoring and intelligent scheduling method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0053] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for remote monitoring and intelligent scheduling of wind farms, characterized in that, include: Vibration data, voltage, and wind speed are acquired using accelerometers, voltage sensors, and ultrasonic anemometers, and then used to construct a fusion preparation dataset. A grouping and integration method was used to classify and merge the data in the fusion preparation dataset to obtain a wind farm operation status model. The real-time collected vibration data, voltage, and wind speed are input into the wind farm operation status model for anomaly detection, and real-time data marked as potential anomalies are obtained; anomaly features are extracted from the real-time data marked as potential anomalies to obtain an anomaly feature set. Based on the set of abnormal features, the severity scores of each abnormality are calculated by weighted summation according to the scope of impact and fault priority, and a graded abnormality list is generated; in the graded abnormality list, the abnormalities are divided into high-risk, medium-risk and low-risk. Based on the tiered anomaly list, match high-risk response templates; based on the matched response templates, analyze the power grid status corresponding to the anomaly data; if the power grid status indication is unstable, generate initial dispatch instructions. Acquire current meteorological data and obtain meteorological influencing factors; analyze the matching degree between the initial dispatching command and the meteorological influencing factors. If the matching degree is lower than the preset threshold, adjust the initial dispatching command and use the adjusted dispatching command as the final dispatching command; if the matching degree is higher than the preset threshold, use the initial dispatching command as the final dispatching command. The execution signal is transmitted to the remote control center according to the final scheduling instruction.

2. The method for remote monitoring and intelligent scheduling of wind farms according to claim 1, characterized in that, A grouping and integration method was used to classify and merge the data in the fusion preparation dataset to obtain the wind farm operation status model, including: The vibration data, voltage data, and wind speed data in the fusion preparation dataset will be converted into a standardized dataset. Collect wind farm power output data and standardize the power output data to obtain a standardized power dataset. Clustering algorithms were used to group and integrate the standardized dataset and the standardized power dataset. Using vibration frequency, voltage fluctuation amplitude, wind speed, and power output as feature dimensions, the data in the standardized dataset and the standardized power data were divided into normal, slightly abnormal, and severely abnormal categories to obtain the operating status data. Vibration feature data of blades is extracted from the operating status data to construct a vibration feature set. If there are feature values ​​in the vibration feature set that exceed the preset vibration feature threshold, a clustering algorithm is used to regroup the vibration feature set to obtain the basic vibration data; otherwise, the vibration feature set is used as the basic vibration data. Based on the basic vibration data and the standardized voltage, wind speed and power data in the operation status data, the support vector machine algorithm is used to construct the wind farm operation status model.

3. The method for remote monitoring and intelligent scheduling of wind farms according to claim 2, characterized in that, Real-time collected vibration data, voltage, and wind speed are input into the wind farm operation state model for anomaly detection, resulting in real-time data marked as potential anomalies. Anomaly features are extracted from this marked real-time data to obtain an anomaly feature set, including: Real-time acquisition of vibration data, voltage, and wind speed; The mean and standard deviation of vibration data, voltage, and wind speed for each window were calculated using the sliding window method to obtain the variation trend of each parameter; The changing trend is input into the wind farm operation status model for anomaly detection. If the rate of change of one of the parameters in a certain window exceeds the corresponding threshold, the real-time data in that window is marked as a potential anomaly. For real-time data marked as potential anomalies, principal component analysis algorithm is used to extract anomaly features, resulting in a set of anomaly features including vibration frequency change rate, voltage change rate, and wind speed change rate.

4. The method for remote monitoring and intelligent scheduling of wind farms according to claim 3, characterized in that, Based on the aforementioned set of abnormal features, a severity score for each abnormality is calculated by weighting and summing according to its impact range and fault priority, and a graded abnormality list is generated, including: Extract the impact range and fault priority of each anomaly from the set of anomaly features; Calculate the initial severity score for each anomaly; where the initial severity score = scope of impact × first weight + fault priority × second weight; The initial severity score is normalized to obtain a normalized severity score; Anomalies with a normalized severity score greater than the first score are marked as high-risk anomalies, anomalies with a normalized severity score greater than the second score and less than the first score are marked as intermediate-risk anomalies, and anomalies with a normalized severity score less than the second score are marked as low-risk anomalies. A tiered anomaly list is generated based on anomalies classified by anomaly level, normalized score, scope of impact, and fault priority.

5. The method for remote monitoring and intelligent scheduling of wind farms according to claim 4, characterized in that, Based on the tiered anomaly list, match high-risk response templates; based on the matched response templates, analyze the power grid status corresponding to the anomaly data; if the power grid status indicates instability, generate initial dispatch instructions, including: Filter out high-risk abnormal data from the graded abnormality list; A rule engine is built using the Drools framework. The rule engine predefines voltage, frequency, and power rules. The rule engine extracts real-time voltage, frequency, and power parameters from the abnormal feature set, substitutes them into the rules, and outputs classification results that include abnormality type and parameter deviation. The rule engine calls a predefined response template library; the response templates in the response template library contain exception types, response operations, and applicable scenarios, and the corresponding response template is matched based on the classification results. Based on the matched response template, the grid node type, number of downstream affected substations, power fluctuation value and fluctuation duration of abnormal fluctuations are extracted from the wind farm grid monitoring system, and it is determined whether the grid instability judgment criteria are met; the result of the judgment of stability or instability is output. If the power grid is determined to be unstable, an emergency isolation protocol is triggered to disconnect the abnormal node and switch to a backup line. Using basic response instructions as a framework, an initial scheduling instruction is generated using a linear programming model. The objective function of the linear programming model is to minimize the sum of power loss and scheduling cost. The constraints include load balancing, the adjusted power of the main line being less than the first power, and the power adjustment range of a single substation being less than the first percentage of the rated power. The initial dispatch command is sent to the power grid control system.

6. The method for remote monitoring and intelligent scheduling of wind farms according to claim 5, characterized in that, Acquire current meteorological data and obtain meteorological impact factors; analyze the matching degree between the initial dispatch instructions and the meteorological impact factors; if the matching degree is lower than the preset threshold, adjust the initial dispatch instructions and use the adjusted dispatch instructions as the final dispatch instructions. If the matching degree is higher than the preset threshold, the initial scheduling instruction will be used as the final scheduling instruction, including: Real-time acquisition of current meteorological data of the wind farm; the meteorological data includes wind speed, wind direction and temperature; calculation of the theoretical power of a single wind turbine based on wind speed and temperature and conversion of the theoretical power into actual output power by combining the efficiency loss of wind farm equipment; the mapping relationship between the actual output power of a single turbine, wind speed and temperature is used as a meteorological influencing factor. Calculate the deviation rate between the power of a single wind turbine in the initial dispatch command and the actual output power of a single turbine in the meteorological influence factors; if the deviation rate is greater than the preset deviation rate, the matching degree is determined to be lower than the preset threshold; if the deviation rate is less than the preset deviation rate, the matching degree is determined to be higher than the preset threshold. If the matching degree is lower than the preset threshold, time series analysis and iterative optimization algorithms are used to adjust the scheduling instruction parameters and generate the adjusted scheduling instruction. If the matching degree is higher than the preset threshold, the initial scheduling instruction will be used as the final scheduling instruction.

7. The method for remote monitoring and intelligent scheduling of wind farms according to claim 6, characterized in that, After transmitting the execution signal to the remote control center according to the final scheduling instruction, the method further includes: Collect feedback data; the feedback data includes wind turbine speed, power output, grid load and grid frequency, forming a monitoring dataset; The Z-score algorithm is used to detect anomalies in the monitoring dataset; if the feedback data exceeds the corresponding threshold, it is determined that the anomaly has not been mitigated. If the anomaly is determined to be unresolved, an updated severity score is calculated based on frequency deviation, power deviation, and response time. If the updated severity score is greater than the preset severity score threshold, the particle swarm optimization algorithm is used to re-optimize the scheduling scheme and generate a new execution signal. The objective function of the particle swarm optimization algorithm is: .

8. The method for remote monitoring and intelligent scheduling of wind farms according to claim 7, characterized in that, The method further includes: The abnormal handling process corresponding to the new execution signal is used as a training sample, and the abnormal feature set, initial scheduling instructions, meteorological influence factors, feedback data and final control effect are recorded. Construct a response template knowledge graph; the knowledge graph uses anomaly type and handling strategy as nodes, and edges represent the mapping relationship from anomaly to handling strategy; record the historical success rate, response time and power recovery rate of the handling path as the weights of the edges; The response template knowledge graph is updated based on the training samples, and the Q-value of the response strategy is iteratively learned using a reinforcement learning algorithm. The state space includes the severity of the anomaly, power grid stability, and meteorological influence factors. The action space is a combination of optional scheduling operations. The reward function consists of power recovery speed, equipment safety margin, and scheduling cost. When the number of successful responses of a certain type of exception handling strategy in the knowledge graph exceeds a preset threshold and the Q value converges, a new response template is generated and injected into the response template library. Regularly mark and discard templates in the response template library that have not been matched for a long time or whose failure rate exceeds the third threshold.

9. A method for remote monitoring and intelligent scheduling of wind farms according to claim 8, characterized in that, Before determining that the abnormality has not been alleviated, the method further includes: An anomaly cause-effect graph model is constructed, wherein the scheduling instruction execution status, wind turbine mechanical status, and meteorological forecast deviation are used as potential dependent variables, and the continued existence of anomalies is used as the outcome variable. The parameters of the causal graph model are learned based on historical operation and maintenance data and real-time feedback data to obtain the contribution of each dependent variable to the result. When the anomaly is determined to be unresolved, the current feedback data is input into the cause-effect graph model for backward reasoning analysis, and the most likely cause of the anomaly's persistence is output. Before re-optimizing the scheduling scheme using the particle swarm optimization algorithm, adjust the optimization constraints or objective function weights based on the most likely cause.

10. A remote monitoring and intelligent dispatching system for wind farms, characterized in that, include: The first processing module is used to: acquire vibration data, voltage and wind speed based on the accelerometer, voltage sensor and ultrasonic anemometer respectively, and form a fusion preparation dataset; The second processing module is used to classify and merge the data in the fusion preparation dataset using a grouping and integration method to obtain a wind farm operation status model. The third processing module is used to: input the real-time collected vibration data, voltage and wind speed into the wind farm operation status model for anomaly detection, and obtain real-time data marked as potential anomalies; extract anomaly features from the real-time data marked as potential anomalies to obtain an anomaly feature set. The fourth processing module is used to: calculate the severity score of each anomaly based on the set of anomaly features, weighted by the scope of impact and fault priority, and generate a graded anomaly list; the anomalies in the graded anomaly list are divided into high-risk, medium-risk and low-risk. The fifth processing module is used to: match high-risk response templates according to the hierarchical anomaly list; analyze the power grid status corresponding to the anomaly data based on the matched response templates; and generate initial dispatch instructions if the power grid status indication is unstable. The sixth processing module is used to: acquire current meteorological data and obtain meteorological influencing factors; analyze the matching degree between the initial dispatching command and the meteorological influencing factors; if the matching degree is lower than a preset threshold, adjust the initial dispatching command and use the adjusted dispatching command as the final dispatching command; if the matching degree is higher than a preset threshold, use the initial dispatching command as the final dispatching command. The seventh processing module is used to transmit execution signals to the remote control center according to the final scheduling instruction.

Images