Wind power plant monitoring method and system based on real-time database

Abstract

The invention discloses a wind power plant monitoring method and system based on a real-time database, and relates to the technical field of wind power plants, and the wind power plant monitoring method comprises the following steps: storing obtained real-time operation characteristic data in the real-time database; identifying operation fault data of the current wind power plant; based on the identified operation fault data of the current wind power plant, key factors influencing the operation fault of the wind power plant are obtained; and dynamically adjusting operation parameters of wind power plant equipment and optimizing real-time monitoring. According to the method, the real-time operation characteristic data is analyzed based on the fault diagnosis algorithm, the current operation fault data of the wind power plant can be quickly identified, the fault can be found more quickly and accurately, the abnormal condition of equipment can be mastered in time, and the equipment damage risk caused by delayed detection is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of wind farms, and in particular to a wind farm monitoring method and system based on a real-time database. Background Art

[0002] Today, wind energy is widely known as an important clean energy source and is widely used in the production of electric energy. Wind energy is obtained through wind turbines, which convert the kinetic energy of wind into mechanical energy, and then convert the mechanical energy into electrical energy, thus achieving efficient conversion of wind energy into electrical energy. A typical wind turbine structure includes a tower, a nacelle located on the top of the tower, and a rotor supported on the nacelle by an axis. In order to improve power generation efficiency and stability, multiple wind turbines are usually grouped to form large wind farms for centralized management and energy output. However, the operating environment of wind farms is complex, and wind turbine equipment is often under high load, so real-time monitoring is particularly important. Through a monitoring method based on a real-time database, data collection, storage and analysis of the operating status of wind turbines can be achieved.

[0003] However, the existing technology is not convenient for identifying the operational fault data of the wind farm, making it impossible to grasp the abnormal situation of the equipment in time, increasing the risk of equipment damage due to delayed detection, and is not convenient for predicting the operational failures that may occur in the future, nor is it convenient to dig out the key factors affecting the failures, thereby reducing the foresight and preventive nature of the system. At the same time, it is not convenient to dynamically adjust the equipment operating parameters according to the key failure factors, and the wind farm cannot maintain efficient and stable operation under different operating conditions. It is not convenient for real-time optimization monitoring, which increases the downtime of the wind farm, thereby reducing the life of the equipment and increasing the operation and maintenance costs.

[0004] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a wind farm monitoring method and system based on a real-time database, which solves the problem mentioned in the above background technology that it is not easy to identify the operating fault data of the wind farm, making it impossible to timely grasp the abnormal conditions of the equipment, increasing the risk of equipment damage due to delayed detection, and it is not easy to predict the operating faults that may occur in the future, nor is it easy to dig out the key factors affecting the faults, thereby reducing the foresight and preventiveness of the system. At the same time, it is not easy to dynamically adjust the equipment operating parameters according to the key fault factors, and the wind farm cannot maintain efficient and stable operation under different operating conditions. It is not easy to optimize the monitoring in real time, which increases the downtime of the wind farm, thereby reducing the life of the equipment and increasing the operation and maintenance costs.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] According to one aspect of the present invention, a wind farm monitoring method based on a real-time database is provided, the wind farm monitoring method comprising the following steps:

[0008] S1. Acquire real-time operation data of the wind farm, pre-process the real-time operation data to obtain real-time operation characteristic data, and store the obtained real-time operation characteristic data in a real-time database;

[0009] S2. Analyze the real-time operation characteristic data in the real-time database based on the fault diagnosis algorithm to identify the operation fault data of the current wind farm;

[0010] S3. Based on the identified operation fault data of the current wind farm, a fault prediction model is established, and the operation faults of the wind farm at a future moment are predicted by using the fault prediction model to obtain key factors affecting the operation faults of the wind farm;

[0011] S4. Based on the obtained key factors affecting the operational failure of the wind farm, a fault warning is issued, the operational parameters of the wind farm equipment are dynamically adjusted, and real-time monitoring is optimized.

[0012] Furthermore, based on the fault diagnosis algorithm, the real-time operation characteristic data in the real-time database is analyzed to identify the operation fault data of the current wind farm, including the following steps:

[0013] S21, acquiring real-time operation feature data in the real-time database, and dividing the real-time operation feature data into a training set and a data set;

[0014] S22, initializing fault diagnosis algorithm parameters, and building an initial fault detection model based on the fault diagnosis algorithm;

[0015] S23, randomly generating an initial fault detection model parameter set, each set of parameters corresponding to a different fault detection strategy;

[0016] S24, inputting the training set and the initial fault detection model parameters into the initial fault detection model, optimizing the initial fault detection model using a model optimization algorithm, and outputting the optimal initial fault detection model parameters;

[0017] S25. Based on the optimal initial fault detection model parameters, a final fault detection model is constructed, and the data set is input into the final fault detection model for detection to identify the operating fault data of the current wind farm.

[0018] Further, initializing the fault diagnosis algorithm parameters and constructing an initial fault detection model based on the fault diagnosis algorithm includes the following steps:

[0019] S221, initializing fault diagnosis algorithm parameters and establishing an initial diagnosis matrix based on real-time operation characteristic data of the wind farm;

[0020] S222, randomly generate an initial fault detection model parameter combination, each set of parameters corresponding to a different fault detection strategy;

[0021] S223, sorting the initial fault detection model evaluation results, selecting the parameter combination with the highest recognition accuracy, and if the selected parameter combination is consistent with the current initial fault detection model parameters, then keep it unchanged; otherwise, adjust the initial fault detection model parameters to the parameter combination with the highest recognition accuracy;

[0022] S224, evaluate the unoptimized parameter combinations one by one, and insert new parameter combinations for testing as needed. After completing the evaluation of each set of parameters, return to step S222 to continue optimizing and adjusting the initial fault detection model parameters;

[0023] S225. Update the initial fault detection model structure according to the final evaluation result, calculate the diagnostic accuracy of the overall initial fault detection model, save the optimized initial fault detection model parameters and structure, and complete the construction of the initial fault detection model.

[0024] Furthermore, randomly generating an initial fault detection model parameter combination, each set of parameters corresponding to a different fault detection strategy includes the following steps:

[0025] S2221, determining a parameter set of an initial fault detection model, and presetting a variation range for each parameter;

[0026] S2222. Randomly generate each group of fault detection model parameters, where each group of parameters represents a fault detection strategy, and ensure that each generated parameter combination is not repeated in structure;

[0027] S2223. Compare the fault detection strategy of each parameter combination with known fault characteristics to evaluate the adaptability of the detection strategy.

[0028] Furthermore, the training set and the initial fault detection model parameters are input into the initial fault detection model, and the initial fault detection model is optimized using a model optimization algorithm. Outputting the optimal initial fault detection model parameters includes the following steps:

[0029] S241, setting the population size and maximum number of iterations of the optimization process, and initializing the initial fault detection model parameter set and the objective function of the initial fault detection model;

[0030] S242, randomly initializing a parameter set in the initial fault detection model parameter space, defining the initial fault detection model structure and calculating the objective function;

[0031] S243, using the training set to evaluate the initial fault detection model, calculating the performance index of the initial fault detection model corresponding to the current parameter combination, and recording the performance results;

[0032] S244, based on the model optimization algorithm update rule, the current parameter combination is updated, if the new parameter combination has the best performance, the current initial fault detection model parameters are updated to the new parameter combination; if the new parameter combination performance is higher than the current optimal parameter combination performance, the global optimal parameters are updated to the new parameter combination, and the initial fault detection model structure is expanded to adapt to the new parameters;

[0033] S245. If the current optimization fails to find the optimal solution, the parameter search range is expanded again, the search strategy is dynamically adjusted, and the parameter combination with the lowest performance is eliminated;

[0034] S246, determine whether the maximum number of iterations has been reached, if so, jump to step S247; otherwise, return to S244 to step S245 to continue optimization;

[0035] S247, terminate the optimization process, and output the global optimal initial fault detection model parameters.

[0036] Furthermore, based on the identified operational fault data of the current wind farm, a fault prediction model is established, and the operational faults of the wind farm at a future moment are predicted using the fault prediction model to obtain the key factors affecting the operational faults of the wind farm, including the following steps:

[0037] S31, quantifying the identified operation fault data of the current wind farm, and dividing the quantified operation fault data set into a training set and a test set;

[0038] S32, setting the maximum number of iterations and population size of the meta-heuristic algorithm, initializing the population position using convex lens imaging reverse learning, and setting the number of training times, number of hidden layer neurons, and learning rate of the fault prediction model;

[0039] S33, updating the nonlinear escape energy according to the change mode of the logic curve characteristics, and dynamically adjusting the objective function weight of the fault prediction model in combination with the current operating fault data;

[0040] S34, evaluating the fitness value of the fault prediction model under the current fault prediction model parameter combination according to the objective function, and ranking the fitness values ​​of all fault prediction model parameter combinations;

[0041] S35, judging by nonlinear escape energy, if the escape energy is less than the threshold, adopting a local optimization strategy to update the fault prediction model parameters; otherwise, adopting a global optimization strategy to update the fault prediction model parameters, and synchronously adjusting the auxiliary parameter combination according to the update direction of the dominant parameter combination;

[0042] S36, comparing the optimization result of the current fault prediction model with the current optimal parameter combination, and recording the fault prediction model parameters with the best performance;

[0043] S37, introducing a disturbance mechanism, randomly disturbing the current fault prediction model parameters, optimizing the disturbed parameters using a local optimization algorithm, obtaining the local optimal fault prediction model parameters in the disturbance area, and using a greedy strategy mechanism for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, proceed to step S38, otherwise, return to step S32 for re-optimization;

[0044] S38. If the current number of iterations reaches the maximum number, the optimal parameters and fault prediction results of the fault prediction model are output, and the key factors affecting the operation failures of the wind farm are extracted from the fault prediction model; otherwise, the fault prediction model results are subjected to dimension-by-dimensional Gaussian mutation, and the process returns to step S31 for re-optimization.

[0045] Furthermore, the formula for updating the nonlinear escape energy according to the change of the characteristics of the logic curve is:

[0046]

[0047] Where T represents the nonlinear escape energy of the current fault prediction model parameter combination; T 0 represents the nonlinear escape energy of the initial fault prediction model parameter combination; T max and T min They represent the maximum nonlinear escape energy and the minimum nonlinear escape energy of the fault prediction model parameter combination; f s (·) represents the Sigmoid function; x represents the value range of nonlinear energy; λ represents the adjustment coefficient; M represents the maximum number of iterations; k represents the number of iterations of the current optimization process; and e represents the base of the natural logarithm.

[0048] Further, a disturbance mechanism is introduced to randomly disturb the current fault prediction model parameters, and the disturbed parameters are optimized using a local optimization algorithm to obtain the local optimal fault prediction model parameters in the disturbance area, and a greedy strategy mechanism is used for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, step S38 is entered, otherwise, step S32 is returned to re-optimization, including the following steps:

[0049] S371, setting parameters of the local optimization algorithm, initializing the wind farm operation fault data, quantifying the fault data characteristics, generating a parameter combination of the initial fault prediction model, and performing random perturbations;

[0050] S372, based on the change trend of the wind farm operation fault data, calculate the initial fitness value of each parameter combination, and evaluate the performance of the fault prediction model under the current parameter combination;

[0051] S373, using the behavior rules of the local optimization algorithm, updating the position of each fault prediction model parameter combination, and dynamically adjusting the parameters of the fault prediction model;

[0052] S374, performing local search optimization on the currently found fault prediction model parameters through a local sequence floating backward selection mechanism;

[0053] S375. When the local optimization algorithm reaches the maximum number of iterations, the local optimal fault prediction model parameters of the disturbance area are output, and the greedy strategy mechanism is used for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, step S38 is entered; otherwise, step S32 is returned to re-optimize.

[0054] Furthermore, the local search optimization of the currently found fault prediction model parameters through the local sequence floating backward selection mechanism includes the following steps:

[0055] S3741, initializing a floating backward selection mechanism, and randomly adjusting each fault prediction model parameter according to the floating mechanism to generate a current candidate fault prediction model parameter combination;

[0056] S3742, for the current candidate fault prediction model parameter combinations, the parameter combinations with the lowest fitness are eliminated one by one, and the fitness of the remaining parameter combinations is re-evaluated after each elimination until a preset stop condition is reached;

[0057] S3743. Calculate the fitness value of each retained candidate parameter combination, evaluate its influence on the fault prediction model, compare the evaluation result with the current optimal fault prediction model parameter combination, and record the parameter combination with the highest fitness;

[0058] S3744. If the parameter combination after the current local search optimization is better than the current optimal parameter combination, update it to the optimal parameter combination, otherwise retain the original optimal parameter combination;

[0059] S3745. If the maximum number of iterations of the local search is reached, the local search is terminated; otherwise, return to step S3741 to perform a new round of floating perturbation and backward selection.

[0060] According to another aspect of the present invention, a wind farm monitoring system based on a real-time database is provided. The wind farm monitoring system comprises:

[0061] A data acquisition module is used to acquire real-time operation data of the wind farm, pre-process the real-time operation data to obtain real-time operation characteristic data, and store the obtained real-time operation characteristic data in a real-time database;

[0062] A fault diagnosis module is used to analyze the real-time operation characteristic data in the real-time database based on the fault diagnosis algorithm to identify the operation fault data of the current wind farm;

[0063] A fault prediction module is used to establish a fault prediction model based on the identified operating fault data of the current wind farm, and use the fault prediction model to predict the operating faults of the wind farm at a future time to obtain key factors affecting the operating faults of the wind farm;

[0064] The early warning monitoring module is used to issue fault warnings based on the key factors affecting the operation failure of the wind farm, dynamically adjust the operating parameters of the wind farm equipment and optimize real-time monitoring.

[0065] The beneficial effects of the present invention are:

[0066] 1. The present invention analyzes real-time operation characteristic data based on a fault diagnosis algorithm, and can quickly identify the current operation fault data of the wind farm, making fault discovery more rapid and accurate, and helping to timely grasp equipment abnormalities, thereby reducing the risk of equipment damage caused by delayed detection. By establishing a fault prediction model, the operation faults that may occur in the future are predicted, and the key factors affecting the faults are excavated, which significantly enhances the foresight and preventiveness of the system. Combined with the key fault factors, the equipment operation parameters are dynamically adjusted, so that the wind farm can maintain efficient and stable operation under different operating conditions. The dynamic adjustment mechanism can extend the life of the equipment and reduce the operation and maintenance costs. Through the real-time early warning mechanism, managers can take measures before potential risks evolve into serious faults to ensure the safe operation of the wind farm. At the same time, the introduction of real-time optimization monitoring significantly reduces the downtime of the wind farm, thereby improving the power generation efficiency and economic benefits.

[0067] 2. The present invention optimizes the initial fault detection model using a model optimization algorithm to ensure that the output optimal model parameters can maximize the accuracy and reliability of fault detection. The use of the optimal model parameters significantly improves the practical application value of the model. The final fault detection model is constructed based on the optimal parameters, and the operating fault data of the current wind farm is identified by detecting the data set, ensuring the efficiency and accuracy of fault detection, and providing technical support for timely discovery of equipment abnormalities. The combination of the fault diagnosis algorithm and the real-time database transforms the analysis of wind farm operating data into an efficient and automated monitoring method, thereby reducing manual participation and improving overall operation and maintenance efficiency and power generation reliability.

[0068] 3. The present invention uses a meta-heuristic algorithm to set model parameters and performs intelligent initialization through convex lens imaging reverse learning, which provides a good starting point for the fault prediction model, significantly improves the global search capability of the algorithm, reduces the convergence time of the fault prediction model, and combines the dynamic adjustment of the logic curve characteristics, so that the fault prediction model can respond more sensitively to changes in complex fault data of wind farms, ensure the accuracy and flexibility of prediction, and finally output the fault prediction results through the optimization and dynamic adjustment of the optimal parameter combination, and extract the key influencing factors, which provides a scientific basis for fault warning and operation parameter optimization of wind farms, strengthens the forward-looking nature of monitoring, and effectively improves the overall economic benefits of wind farms by reducing fault downtime, optimizing operating parameters and improving prediction capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1 is a flow chart of a wind farm monitoring method based on a real-time database according to an embodiment of the present invention;

[0071] Figure 2 The diagram is a principle block diagram of a wind farm monitoring system based on a real-time database according to an embodiment of the present invention.

[0072] In the figure:

[0073] 1. Data acquisition module; 2. Fault diagnosis module; 3. Fault prediction module; 4. Early warning monitoring module. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0075] In the description of the present invention, unless otherwise specified, the meaning of "plurality" is two or more. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0076] According to an embodiment of the present invention, a wind farm monitoring method and system based on a real-time database are provided.

[0077] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to a wind farm monitoring method based on a real-time database according to an embodiment of the present invention, the wind farm monitoring method comprises the following steps:

[0078] S1. Acquire real-time operation data of the wind farm, pre-process the real-time operation data to obtain real-time operation characteristic data, and store the obtained real-time operation characteristic data in a real-time database;

[0079] It should be explained that obtaining the real-time operation data of the wind farm, preprocessing the real-time operation data to obtain the real-time operation characteristic data, and storing the obtained real-time operation characteristic data in the real-time database includes:

[0080] Acquire the real-time operation data of the wind farm, perform wavelet decomposition on the real-time operation data, remove noise and reconstruct the operation data to obtain pre-processed real-time operation data;

[0081] Decompose the reconstructed real-time operation data using empirical mode decomposition, extract the intrinsic mode function, calculate the mutual correlation coefficient between each intrinsic mode function and the real-time operation data, and select the intrinsic mode function with the highest mutual correlation coefficient;

[0082] Based on fast Fourier transform, feature extraction is performed on the selected intrinsic mode function with the highest mutual correlation coefficient to obtain the real-time operation characteristic data of the wind farm, and the real-time operation characteristic data is stored in the real-time database.

[0083] Specifically, the real-time operation data of wind farms include:

[0084] 1) Equipment operating parameters: wind speed, wind direction, speed (rotor speed, generator speed), power output (active power, reactive power), voltage, current, temperature (cabin temperature, generator temperature).

[0085] 2) Equipment health parameters: vibration data (bearing vibration, tower vibration), noise level, lubricating oil status (temperature, pressure, particle concentration).

[0086] 3) Environmental parameters: external environment temperature, humidity, air pressure, density, etc.

[0087] S2. Analyze the real-time operation characteristic data in the real-time database based on the fault diagnosis algorithm to identify the operation fault data of the current wind farm;

[0088] Specifically, the operational fault data of wind farms include:

[0089] 1) Mechanical failure: blade failure (such as blade cracks, damage, imbalance), rotor failure (such as rotor imbalance, fatigue damage), bearing failure (such as excessive wear, insufficient lubrication, abnormal vibration), nacelle failure (such as vibration or excessive temperature).

[0090] 2) Electrical failure: generator failure (such as winding short circuit, overheating, insulation damage), electrical control system failure (such as controller failure, abnormal voltage, current fluctuation).

[0091] 3) Transmission system failure: gearbox failure (such as gear wear, cracks, overheating), lubricating oil system failure (such as lubricating oil contamination, abnormal oil pressure).

[0092] 4) Environmental-induced failures: equipment overload caused by abnormal wind speed and direction, and the impact of extreme weather on equipment operation (such as freezing and dust accumulation).

[0093] S3. Based on the identified operation fault data of the current wind farm, a fault prediction model is established, and the operation faults of the wind farm at a future moment are predicted by using the fault prediction model to obtain key factors affecting the operation faults of the wind farm;

[0094] Specifically, the key factors affecting the operational failures of wind farms include: environmental factors, equipment operation factors, equipment maintenance and aging factors, fault transmission and chain reaction factors, operation and control factors, manufacturing and installation factors, etc.

[0095] S4. Based on the obtained key factors affecting the operational failure of the wind farm, a fault warning is issued, the operational parameters of the wind farm equipment are dynamically adjusted, and real-time monitoring is optimized.

[0096] It needs to be explained that, in combination with the key factors affecting operational failures, the data of the wind farm (such as wind speed, vibration, temperature, etc.) are monitored in real time to identify abnormal signals; the current real-time feature data is compared with the key factors to determine whether it exceeds the normal threshold; a multi-level early warning mechanism (such as hidden faults, potential faults, and serious faults) is set up based on the prediction model; when the key parameters exceed the set threshold or a high risk is predicted, a fault warning is triggered; based on the warning results, the operating parameters of the wind farm equipment are automatically optimized, such as adjusting the blade angle, controlling the generator speed, adjusting the cooling system, etc., to reduce the risk of failure; the overall operation strategy of the wind farm is dynamically adjusted, such as reducing the load, adjusting the wind direction alignment, etc., to adapt to the current environment and equipment status; the adjusted operating parameters are fed back to the monitoring system in real time to ensure the effectiveness of the adjustment measures; the equipment operating status is continuously monitored through the real-time monitoring system to ensure the timeliness and accuracy of the fault warning and adjustment measures; the detailed data of each fault warning and adjustment is recorded, including the triggering conditions, adjustment measures and results; historical data is used for analysis to optimize the fault prediction model and adjustment strategy to improve the system's adaptability.

[0097] Preferably, based on the fault diagnosis algorithm, analyzing the real-time operation characteristic data in the real-time database to identify the operation fault data of the current wind farm includes the following steps:

[0098] S21, acquiring real-time operation feature data in the real-time database, and dividing the real-time operation feature data into a training set and a data set;

[0099] S22, initializing fault diagnosis algorithm parameters, and constructing an initial fault detection model (i.e., DBN model) based on the fault diagnosis algorithm;

[0100] Specifically, the DBN model is a deep learning model composed of multiple stacked restricted Boltzmann machines.

[0101] S23, randomly generating an initial fault detection model parameter set, each set of parameters corresponding to a different fault detection strategy;

[0102] S24, inputting the training set and the initial fault detection model parameters into the initial fault detection model, optimizing the initial fault detection model using a model optimization algorithm, and outputting the optimal initial fault detection model parameters;

[0103] S25. Based on the optimal initial fault detection model parameters, a final fault detection model is constructed, and the data set is input into the final fault detection model for detection to identify the operating fault data of the current wind farm.

[0104] Specifically, the fault diagnosis algorithm is a multiple sequence alignment method based on an improved ant colony algorithm, which changes the way pheromones are updated, the method of selecting characters, the round-trip search of ants between the ant nest and food, and the random assignment of ant start sequences.

[0105] Specifically, the model optimization algorithm is an improved lightning search algorithm, which is mainly implemented by a transition discharge body, a space discharge body and a guide discharge body. The present invention improves the lightning search algorithm by updating the guide discharge body position and changing the discharge body iteration mode.

[0106] In order to facilitate understanding of the above technical solution of the present invention, the present invention will be described in detail below in the actual process based on the fault diagnosis algorithm to analyze the real-time operation characteristic data in the real-time database and identify the operation fault data of the current wind farm.

[0107] Step 1: Obtain real-time operation feature data

[0108] 1) Generates simulated real-time operation characteristic data, including:

[0109] Wind speed: mean is 10m / s, standard deviation is 2m / s.

[0110] Speed: Mean is 20RPM, standard deviation is 5RPM.

[0111] Power generation: mean 150kW, standard deviation 20kW.

[0112] Temperature: The mean is 25°C and the standard deviation is 3°C.

[0113] Vibration level: mean 0.5g, standard deviation 0.1g.

[0114] 2) The data is 1000 hours in total, and the timestamp range starts from the preset year, month and day, with one record per hour.

[0115] Step 2: Initialize fault diagnosis algorithm parameters

[0116] 1) Initial parameters of fault detection algorithm:

[0117] Learning rate: 0.01.

[0118] Hidden layer configuration: [32, 16].

[0119] Activation function: ReLU.

[0120] Step 3: Generate random fault detection strategy parameters

[0121] 1) Ten groups of strategy parameters were randomly generated, each group of parameters represents a different fault detection strategy.

[0122] For example: The parameter value of strategy 7 is 0.8245.

[0123] Step 4: Optimize the initial fault detection model

[0124] 1) Use the training set and strategy parameters to evaluate the performance of each strategy. The performance evaluation formula is:

[0125]

[0126] Where N is the number of samples in the training data; i represents the actual target variable value (such as power generation); i represents the characteristic variable value (such as vibration level); θ represents the current strategy parameter; (y i -(θ·z i )) 2 Represents the squared error between the predicted value and the actual value. The goal is to minimize the mean squared error of the model by adjusting θ.

[0127] 2) Strategy 7 has the best performance, and its parameter value 0.8245 minimizes the error of the training set.

[0128] Step 5: Build the final fault detection model

[0129] 1) The performance of the best strategy was verified using the test set, and the mean square error of the final model was 22888.70.

[0130] 2) Results summary: Best strategy number: 7. Best parameter value: 0.8245. Performance of the final model on the test set (MSE): 22888.70.

[0131] Preferably, initializing the fault diagnosis algorithm parameters and constructing an initial fault detection model based on the fault diagnosis algorithm includes the following steps:

[0132] S221, initializing fault diagnosis algorithm parameters and establishing an initial diagnosis matrix based on real-time operation characteristic data of the wind farm;

[0133] S222, randomly generate an initial fault detection model parameter combination, each set of parameters corresponding to a different fault detection strategy;

[0134] S223, sorting the initial fault detection model evaluation results, selecting the parameter combination with the highest recognition accuracy, and if the selected parameter combination is consistent with the current initial fault detection model parameters, then keep it unchanged; otherwise, adjust the initial fault detection model parameters to the parameter combination with the highest recognition accuracy;

[0135] S224, evaluate the unoptimized parameter combinations one by one, and insert new parameter combinations for testing as needed. After completing the evaluation of each set of parameters, return to step S222 to continue optimizing and adjusting the initial fault detection model parameters;

[0136] S225. Update the initial fault detection model structure according to the final evaluation result, calculate the diagnostic accuracy of the overall initial fault detection model, save the optimized initial fault detection model parameters and structure, and complete the construction of the initial fault detection model.

[0137] It should be explained that the operational characteristic data of the wind farm, including wind speed, rotation speed, vibration level, etc., are extracted from the real-time database. The basic parameters of the diagnostic algorithm, such as learning rate, feature weight value, error threshold, etc., are initialized. An initial diagnostic matrix is ​​established to map the relationship between the characteristic data and the fault type. For example, rows represent different operational characteristics (such as vibration, temperature). Columns represent possible fault types (such as blade cracks, bearing wear). Default weights are filled in the initial matrix to represent the initial correlation between the characteristics and the faults; multiple sets of parameter combinations are randomly generated, such as feature weights, threshold ranges, or fault type priorities. Each set of parameters corresponds to a fault detection strategy. These parameter combinations will be evaluated one by one in subsequent steps to select the strategy with the best performance; each set of parameter combinations is input into the initial fault detection model and evaluated using a portion of the training data. The accuracy of the model in identifying faults is evaluated and the performance of each set of parameters is recorded. The results of all parameter combinations are sorted and the parameter combination with the highest recognition accuracy is selected. If the parameter combination with the best recognition performance is consistent with the parameters of the current model, it remains unchanged; otherwise, the parameters of the model are updated to the optimal parameters; the performance of the parameter combinations that have not yet been optimized is evaluated one by one. If the current evaluation results show that a certain group of parameters has high potential, further fine-tune the parameters and generate new parameter combinations to insert into the test queue. Gradually eliminate parameter combinations with poor performance and retain parameters with greater potential. After completing the evaluation of all parameter combinations, return to step S222 to continue generating new parameter combinations and optimizing the initial model. This process continues until the model performance tends to stabilize or reaches the preset target; based on the final optimized parameter combination, update the structure of the initial fault detection model, such as adjusting the weight distribution or optimizing feature selection. Use test data to evaluate the optimized model, calculate the overall diagnostic accuracy, and record the final results. Save the optimized parameters and model structure for subsequent fault detection tasks.

[0138] Preferably, randomly generating an initial fault detection model parameter combination, each set of parameters corresponding to a different fault detection strategy includes the following steps:

[0139] S2221, determining a parameter set of an initial fault detection model, and presetting a variation range for each parameter;

[0140] S2222. Randomly generate each group of fault detection model parameters, where each group of parameters represents a fault detection strategy, and ensure that each generated parameter combination is not repeated in structure;

[0141] S2223. Compare the fault detection strategy of each parameter combination with known fault characteristics to evaluate the adaptability of the detection strategy.

[0142] It is necessary to explain that the core parameter set of the fault detection model is determined, for example: Feature weight: the weight of the impact of operating characteristics such as wind speed, vibration, temperature, etc. on fault detection. Threshold range: the threshold of various parameters that trigger fault warning. Learning rate: the iteration step size used for model optimization. Hidden layer structure: the size and number of hidden layers of the neural network model (such as [32, 16]). Set the variation range for each parameter, for example: Feature weight range: [0.1, 1.0]. Threshold range: [50%, 90%]. Learning rate range: [0.001, 0.1]. Make sure the parameter range covers a variety of possible fault detection strategies; example:

[0143] Parameter set and range of variation:

[0144] Feature weights: wind speed [0.2, 1.0], vibration [0.1, 0.9], temperature [0.3, 0.8]. Threshold range: [0.6, 0.95]. Learning rate: [0.01, 0.05].

[0145] Generate multiple sets of parameter combinations through random sampling. For example, use the Monte Carlo method or a random number generator. Ensure that the parameter combinations generated each time are not repeated in structure: check whether the generated parameter combinations are exactly the same as the existing combinations. If they are repeated, regenerate them. Each set of parameters represents a fault detection strategy and can reflect different diagnostic modes.

[0146] Example:

[0147] The three sets of parameter combinations generated (assuming 3 strategies are generated):

[0148] Strategy 1: wind speed weight 0.8, vibration weight 0.6, temperature weight 0.5, threshold 0.85, learning rate 0.03.

[0149] Strategy 2: wind speed weight 0.5, vibration weight 0.7, temperature weight 0.4, threshold 0.75, learning rate 0.02.

[0150] Strategy 3: wind speed weight 0.9, vibration weight 0.4, temperature weight 0.3, threshold 0.90, learning rate 0.01.

[0151] Compare the fault detection strategy corresponding to each parameter combination with known fault characteristics: Test using training data and known fault modes (e.g. bearing wear, blade cracks). Check whether the parameter combination can accurately identify the corresponding fault type.

[0152] The suitability of each set of detection strategies was evaluated using the following criteria:

[0153] Detection accuracy: whether known fault data can be correctly identified.

[0154] False alarm rate: whether there are high number of false alarms without faults.

[0155] False negative rate: whether the actual fault is not captured.

[0156] Record the evaluation results of each set of parameters and select strategies with higher adaptability.

[0157] Example:

[0158] Parameter combination evaluation results:

[0159] Strategy 1: 92% accuracy, 5% false positive rate, and 3% missed negative rate.

[0160] Strategy 2: accuracy rate 85%, false positive rate 8%, missed negative rate 7%.

[0161] Strategy 3: 90% accuracy, 6% false positive rate, and 4% missed negative rate.

[0162] According to the results, strategy 1 has the highest adaptability and is selected as the basis for further optimization.

[0163] Preferably, the training set and the initial fault detection model parameters are input into the initial fault detection model, and the initial fault detection model is optimized using a model optimization algorithm. Outputting the optimal initial fault detection model parameters includes the following steps:

[0164] S241, setting the population size and maximum number of iterations of the optimization process, and initializing the initial fault detection model parameter set and the objective function of the initial fault detection model;

[0165] S242, randomly initializing a parameter set in the initial fault detection model parameter space, defining the initial fault detection model structure and calculating the objective function;

[0166] S243, using the training set to evaluate the initial fault detection model, calculating the performance index of the initial fault detection model corresponding to the current parameter combination, and recording the performance results;

[0167] S244, based on the model optimization algorithm update rule, the current parameter combination is updated, if the new parameter combination has the best performance, the current initial fault detection model parameters are updated to the new parameter combination; if the new parameter combination performance is higher than the current optimal parameter combination performance, the global optimal parameters are updated to the new parameter combination, and the initial fault detection model structure is expanded to adapt to the new parameters;

[0168] S245. If the current optimization fails to find the optimal solution, the parameter search range is expanded again, the search strategy is dynamically adjusted, and the parameter combination with the lowest performance is eliminated;

[0169] S246, determine whether the maximum number of iterations has been reached, if so, jump to step S247; otherwise, return to S244 to step S245 to continue optimization;

[0170] S247, terminate the optimization process, and output the global optimal initial fault detection model parameters.

[0171] Preferably, based on the identified operational fault data of the current wind farm, a fault prediction model is established, and the operational faults of the wind farm at a future moment are predicted using the fault prediction model, and obtaining the key factors affecting the operational faults of the wind farm includes the following steps:

[0172] S31, quantifying the identified operation fault data of the current wind farm, and dividing the quantified operation fault data set into a training set and a test set;

[0173] S32, setting the maximum number of iterations and population size of the meta-heuristic algorithm, initializing the population position using convex lens imaging reverse learning, and setting the number of training times, number of hidden layer neurons, and learning rate of the fault prediction model;

[0174] S33, updating the nonlinear escape energy (i.e., the escape energy of the prey in the improved golden jackal optimization algorithm) according to the change mode of the logistic curve characteristics (i.e., the change mode of the nonlinear s-type characteristics of the Sigmoid function), and dynamically adjusting the objective function weight of the fault prediction model in combination with the current operating fault data;

[0175] S34, evaluating the fitness value of the fault prediction model under the current fault prediction model parameter combination according to the objective function, and ranking the fitness values ​​of all fault prediction model parameter combinations;

[0176] S35, judging by nonlinear escape energy, if the escape energy is less than the threshold, adopting a local optimization strategy to update the fault prediction model parameters; otherwise, adopting a global optimization strategy to update the fault prediction model parameters, and synchronously adjusting the auxiliary parameter combination according to the update direction of the dominant parameter combination;

[0177] S36, comparing the optimization result of the current fault prediction model with the current optimal parameter combination, and recording the fault prediction model parameters with the best performance;

[0178] S37, introducing a disturbance mechanism, randomly disturbing the current fault prediction model parameters, optimizing the disturbed parameters using a local optimization algorithm, obtaining the local optimal fault prediction model parameters in the disturbance area, and using a greedy strategy mechanism for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, proceed to step S38, otherwise, return to step S32 for re-optimization;

[0179] S38. If the current number of iterations reaches the maximum number, the optimal parameters and fault prediction results of the fault prediction model are output, and the key factors affecting the operation failures of the wind farm are extracted from the fault prediction model; otherwise, the fault prediction model results are subjected to dimension-by-dimensional Gaussian mutation, and the process returns to step S31 for re-optimization.

[0180] It should be explained that the current operating fault data identified is collected, such as abnormal blade vibration, power generation fluctuation, high temperature, etc. The fault data is quantified and the fault type and characteristics are converted into a standardized numerical form, for example:

[0181] Fault type: Indicated by classification code (such as 0 = normal, 1 = minor fault, 2 = serious fault).

[0182] Feature data: Standardized to the range of [0, 1] to ensure the stability of model training.

[0183] The quantized fault data set is divided into a training set (for model training) and a test set (for model evaluation), such as a 7:3 allocation ratio.

[0184] Set the parameters of the metaheuristic optimization algorithm:

[0185] Maximum number of iterations: such as 100.

[0186] Population size: such as 50 (i.e. the number of model parameter combinations evaluated in each iteration).

[0187] Use the convex lens imaging reverse learning method to initialize the population position:

[0188] Generate multiple random initial parameter combinations to cover the entire solution space.

[0189] Through reverse learning, the location distribution of the initial population is optimized.

[0190] Determine the architecture of the fault prediction model:

[0191] Number of training sessions: such as 10.

[0192] Number of hidden layer neurons: such as 32.

[0193] Learning rate: such as 0.01.

[0194] According to the characteristics of the logistic curve (such as the nonlinear S-shaped change of the Sigmoid function), the escape energy of the prey is adjusted to control the search behavior of the optimization algorithm: in the early stage of optimization, the escape energy is increased to enhance the global search ability. In the later stage of optimization, the escape energy is reduced to improve the local search ability.

[0195] Dynamically adjust the objective function weight based on the current operating fault data: If a certain type of fault feature has a greater impact on the model prediction, increase its weight. The goal of dynamic adjustment is to make the model pay more attention to key fault features.

[0196] Based on the objective function, calculate the fitness value of the current fault prediction model parameter combination (i.e., the prediction error or accuracy of the model on the training set). Sort the fitness values ​​of all parameter combinations and select the parameter combination with the best performance. Determine the nonlinear escape energy: If the escape energy is less than the preset threshold, enter the local optimization stage and fine-tune the current parameter combination. If the escape energy is greater than the threshold, enter the global optimization stage to explore new areas of the solution space. At the same time, adjust the update direction of the auxiliary parameter combination to keep it synchronized with the dominant parameter combination. Compare the optimization results of the current fault prediction model with the historical optimal parameter combination. If the current result is better than the historical record, update the parameter combination with the best performance; apply random perturbations to the current fault prediction model parameters to generate new parameter combinations.

[0197] For example, randomly changing the value of a parameter or adding Gaussian noise. Using a local optimization algorithm to optimize the disturbed parameters, obtain the local optimal solution in the disturbed area.

[0198] Use the greedy strategy to judge: if the local optimal parameter combination is better than the current optimal parameter combination, then update the global optimal solution. Otherwise, return to step S32 to re-optimize. If the maximum number of iterations is reached, then output the final optimal parameter combination and fault prediction result.

[0199] Extract key factors that affect wind farm operation failures:

[0200] According to the objective function weight, filter the features that contribute most to fault prediction, for example:

[0201] Vibration level: Weighting 0.9.

[0202] Temperature: Weight 0.8.

[0203] Wind speed: weight 0.6.

[0204] If the maximum number of iterations is not reached, Gaussian mutation is performed dimension by dimension on the fault prediction result, and the process returns to step S31 to continue the optimization.

[0205] Specifically, the meta-heuristic algorithm is used to improve the golden jackal optimization algorithm. First, the convex lens imaging reverse learning strategy is integrated to explore a better initial solution position; the Sigmoid function is introduced to change the escape energy and balance the exploration and development stages; the spiral encirclement mechanism of the whale optimization algorithm is integrated to enhance the exploration ability and improve the convergence accuracy. In the traditional golden jackal optimization algorithm, in the golden jackal combination, the male golden jackal is in the first leader position, which is the optimal exploration position; the female golden jackal is a follower and is in a suboptimal position. In the golden jackal optimization algorithm, the hunting process of the golden jackal is divided into three stages: (1) searching for prey and approaching the prey; (2) encircling the prey and stimulating the prey until it stops moving; (3) attacking the prey.

[0206] Specifically, the fault prediction model is a long short-term memory network LSTM, which is developed on the basis of recurrent neural networks (RNN) to solve the gradient vanishing and gradient exploding problems when RNN processes long sequences. Compared with RNN, LSTM contains input gates, forget gates, output gates, memory units, and information transmission units, so as to memorize and update key information in the process of processing timing problems.

[0207] Specifically, in the basic convex lens imaging reverse learning, the scale factor is generally used as a fixed value, which is easy to fall into the local optimum. The present invention proposes a scale factor strategy based on dynamic decreasing, which can obtain a larger value in the early iteration, so that the golden jackal can search a large range in different dimensional areas and improve the diversity of the group; obtain a smaller value in the later iteration, and conduct a fine search near the optimal individual to improve the local optimization ability.

[0208] Specifically, the escape energy of the prey is the decisive factor controlling the update of the prey's position, which is responsible for balancing the global search and local development behavior of the algorithm. The nonlinear s-shaped characteristic change of the Sigmoid function can continuously perform global search to find the optimal area, thereby improving the disadvantage of slow convergence speed in the early stage and increasing the convergence speed.

[0209] Preferably, the formula for updating the nonlinear escape energy according to the change mode of the logistic curve characteristic is:

[0210]

[0211] Where T represents the nonlinear escape energy of the current fault prediction model parameter combination; T 0 represents the nonlinear escape energy of the initial fault prediction model parameter combination; T max and Tmin They represent the maximum nonlinear escape energy and the minimum nonlinear escape energy of the fault prediction model parameter combination; f s (·) represents the Sigmoid function; x represents the value range of nonlinear energy; λ represents the adjustment coefficient; M represents the maximum number of iterations; k represents the number of iterations of the current optimization process; and e represents the base of the natural logarithm.

[0212] Preferably, a disturbance mechanism is introduced to randomly disturb the current fault prediction model parameters, and the disturbed parameters are optimized using a local optimization algorithm to obtain the local optimal fault prediction model parameters in the disturbance area, and a greedy strategy mechanism is used for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, step S38 is entered, otherwise, returning to step S32 for re-optimization includes the following steps:

[0213] S371, setting parameters of the local optimization algorithm, initializing the wind farm operation fault data, quantifying the fault data characteristics, generating a parameter combination of the initial fault prediction model, and performing random perturbations;

[0214] S372, based on the change trend of the wind farm operation fault data, calculate the initial fitness value of each parameter combination, and evaluate the performance of the fault prediction model under the current parameter combination;

[0215] S373, using the behavior rules of the local optimization algorithm, updating the position of each fault prediction model parameter combination, and dynamically adjusting the parameters of the fault prediction model;

[0216] S374, performing local search optimization on the currently found fault prediction model parameters through a local sequence floating backward selection mechanism;

[0217] S375. When the local optimization algorithm reaches the maximum number of iterations, the local optimal fault prediction model parameters of the disturbance area are output, and the greedy strategy mechanism is used for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, step S38 is entered; otherwise, step S32 is returned to re-optimize.

[0218] It is necessary to explain that the parameters of the local optimization algorithm are set, including the maximum number of iterations, search radius, step size, etc. For example:

[0219] Maximum number of iterations: 50.

[0220] Search radius: 0.1.

[0221] Step size: 0.01.

[0222] The main features are extracted from the wind farm operation fault data and quantified.

[0223] For example, characteristics may include:

[0224] Real-time operating data such as wind speed, vibration level, temperature, etc.

[0225] Classification information of fault type (such as blade crack, bearing wear, etc.).

[0226] Generate the initial fault prediction model parameter combination and apply random disturbance to it: add random noise or offset to the original parameters to make them jump out of the local area of ​​the current solution and explore the new parameter space. Based on the changing trend of the wind farm operation fault data, calculate the initial fitness value of each set of disturbed parameter combination. The fitness value measures the performance of the model under the current parameter combination (such as prediction accuracy, false alarm rate, etc.). Record the fitness value of each parameter combination for subsequent behavioral decision-making of the optimization algorithm.

[0227] Example:

[0228] Fitness value for parameter combination 1: 90%.

[0229] Fitness value for parameter combination 2: 85%.

[0230] Fitness value for parameter combination 3: 88%.

[0231] According to the behavior rules of the local optimization algorithm, the position of each set of fault prediction model parameter combinations is updated:

[0232] Use the fitness value to guide the update direction of the parameters, so that they are adjusted in the direction of better performance. Dynamically adjust the step size and weight of the parameters to balance exploration and utilization. If the contribution of some parameters is high, increase the adjustment range; otherwise, reduce the adjustment frequency.

[0233] Use the local sequence floating backward selection mechanism to perform a fine search for the currently found parameter combination: check the boundary changes of each parameter and identify the key parameters that have the greatest impact on model performance. In the search process, gradually exclude parameters that contribute less to the optimization. Through the rules of the local optimization algorithm, narrow the search range and further improve the accuracy of the local optimal solution.

[0234] Example:

[0235] Slightly adjusted wind speed weight from 0.8 to 0.85.

[0236] Minor adjustment of vibration level weight: from 0.6 to 0.62.

[0237] When the local optimization algorithm reaches the maximum number of iterations, the local optimal parameter combination in the current disturbance area is output.

[0238] Use the greedy strategy to judge: if the current local optimal parameter combination is better than the global optimal solution, then update the global optimal solution and enter step S38 (complete optimization). If the current local optimal solution fails to reach the global optimal, then return to step S32 and regenerate the parameter combination for optimization.

[0239] Example:

[0240] The fitness value of the local optimal parameter combination is 92%, which is better than the previous global optimal solution (91%), so the global optimal parameters are updated.

[0241] If the fitness value of the local optimal solution is 89% and does not exceed the global optimal solution (91%), re-optimization is performed.

[0242] Preferably, performing local search optimization on the currently found fault prediction model parameters through the local sequence floating backward selection mechanism includes the following steps:

[0243] S3741, initializing a floating backward selection mechanism, and randomly adjusting each fault prediction model parameter according to the floating mechanism to generate a current candidate fault prediction model parameter combination;

[0244] S3742, for the current candidate fault prediction model parameter combinations, the parameter combinations with the lowest fitness are eliminated one by one, and the fitness of the remaining parameter combinations is re-evaluated after each elimination until a preset stop condition is reached;

[0245] S3743. Calculate the fitness value of each retained candidate parameter combination, evaluate its influence on the fault prediction model, compare the evaluation result with the current optimal fault prediction model parameter combination, and record the parameter combination with the highest fitness;

[0246] S3744. If the parameter combination after the current local search optimization is better than the current optimal parameter combination, update it to the optimal parameter combination, otherwise retain the original optimal parameter combination;

[0247] S3745. If the maximum number of iterations of the local search is reached, the local search is terminated; otherwise, return to step S3741 to perform a new round of floating perturbation and backward selection.

[0248] Specifically, the local optimization algorithm is an improved dragonfly algorithm. The traditional dragonfly algorithm is a swarm intelligence optimization algorithm. The traditional dragonfly algorithm mainly learns two behavior modes of dragonfly groups - static groups and dynamic groups, and approximately corresponds these two group behaviors to the global exploration and local development of the swarm intelligence algorithm. The improved dragonfly algorithm of the present invention inherits the good convergence of the traditional dragonfly algorithm and introduces a local sequence floating backward selection mechanism, which increases the global search capability of the algorithm while improving the solution accuracy of the algorithm.

[0249] Specifically, the local sequence floating backward selection mechanism is used to perform local search on the currently found optimal sugar content change trend. This mechanism further improves the accuracy of local search by making small parameter adjustments near the optimal solution. This mechanism can effectively avoid falling into the local optimum and find a more accurate optimization solution by fine-tuning the search results each time.

[0250] According to another embodiment of the present invention, Figure 2 As shown, a wind farm monitoring system based on a real-time database is also provided, and the wind farm monitoring system includes:

[0251] The data acquisition module 1 is used to acquire the real-time operation data of the wind farm, pre-process the real-time operation data to obtain real-time operation characteristic data, and store the obtained real-time operation characteristic data in the real-time database;

[0252] Fault diagnosis module 2, used to analyze the real-time operation characteristic data in the real-time database based on the fault diagnosis algorithm, and identify the operation fault data of the current wind farm;

[0253] The fault prediction module 3 is used to establish a fault prediction model based on the identified operation fault data of the current wind farm, and use the fault prediction model to predict the operation faults of the wind farm at a future time, and obtain the key factors affecting the operation faults of the wind farm;

[0254] The early warning monitoring module 4 is used to issue fault early warnings based on the key factors affecting the operational faults of the wind farm, dynamically adjust the operational parameters of the wind farm equipment and optimize real-time monitoring.

[0255] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A wind farm monitoring method based on real-time database, characterized in that: The wind farm monitoring method comprises the following steps: S1. Acquire real-time operation data of the wind farm, pre-process the real-time operation data to obtain real-time operation characteristic data, and store the obtained real-time operation characteristic data in a real-time database; S2. Analyze the real-time operation characteristic data in the real-time database based on the fault diagnosis algorithm to identify the operation fault data of the current wind farm; S3. Based on the identified operation fault data of the current wind farm, a fault prediction model is established, and the operation faults of the wind farm at a future moment are predicted by using the fault prediction model to obtain key factors affecting the operation faults of the wind farm; S4. Based on the obtained key factors affecting the operational failure of the wind farm, a fault warning is issued, the operational parameters of the wind farm equipment are dynamically adjusted, and real-time monitoring is optimized.

2. A wind farm monitoring method based on real-time database according to claim 1, characterized in that: The method of analyzing the real-time operation characteristic data in the real-time database based on the fault diagnosis algorithm to identify the operation fault data of the current wind farm includes the following steps: S21, acquiring real-time operation feature data in the real-time database, and dividing the real-time operation feature data into a training set and a data set; S22, initializing fault diagnosis algorithm parameters, and building an initial fault detection model based on the fault diagnosis algorithm; S23, randomly generating an initial fault detection model parameter set, each set of parameters corresponding to a different fault detection strategy; S24, inputting the training set and the initial fault detection model parameters into the initial fault detection model, optimizing the initial fault detection model using a model optimization algorithm, and outputting the optimal initial fault detection model parameters; S25. Based on the optimal initial fault detection model parameters, a final fault detection model is constructed, and the data set is input into the final fault detection model for detection to identify the operating fault data of the current wind farm.

3. A wind farm monitoring method based on real-time database according to claim 2, characterized in that: Initializing the fault diagnosis algorithm parameters and constructing an initial fault detection model based on the fault diagnosis algorithm includes the following steps: S221, initializing fault diagnosis algorithm parameters and establishing an initial diagnosis matrix based on real-time operation characteristic data of the wind farm; S222, randomly generate an initial fault detection model parameter combination, each set of parameters corresponding to a different fault detection strategy; S223, sorting the initial fault detection model evaluation results, selecting the parameter combination with the highest recognition accuracy, and if the selected parameter combination is consistent with the current initial fault detection model parameters, then keep it unchanged; otherwise, adjust the initial fault detection model parameters to the parameter combination with the highest recognition accuracy; S224, evaluate the unoptimized parameter combinations one by one, and insert new parameter combinations for testing as needed. After completing the evaluation of each set of parameters, return to step S222 to continue optimizing and adjusting the initial fault detection model parameters; S225. Update the initial fault detection model structure according to the final evaluation result, calculate the diagnostic accuracy of the overall initial fault detection model, save the optimized initial fault detection model parameters and structure, and complete the construction of the initial fault detection model.

4. A wind farm monitoring method based on real-time database according to claim 3, characterized in that: The randomly generated initial fault detection model parameter combination, each set of parameters corresponding to a different fault detection strategy, comprises the following steps: S2221, determining a parameter set of an initial fault detection model, and presetting a variation range for each parameter; S2222. Randomly generate each group of fault detection model parameters, where each group of parameters represents a fault detection strategy, and ensure that each generated parameter combination is not repeated in structure; S2223. Compare the fault detection strategy of each parameter combination with known fault characteristics to evaluate the adaptability of the detection strategy.

5. A wind farm monitoring method based on real-time database according to claim 4, characterized in that: The step of inputting the training set and the initial fault detection model parameters into the initial fault detection model, optimizing the initial fault detection model using a model optimization algorithm, and outputting the optimal initial fault detection model parameters comprises the following steps: S241, setting the population size and maximum number of iterations of the optimization process, and initializing the initial fault detection model parameter set and the objective function of the initial fault detection model; S242, randomly initializing a parameter set in the initial fault detection model parameter space, defining the initial fault detection model structure and calculating the objective function; S243, using the training set to evaluate the initial fault detection model, calculating the performance index of the initial fault detection model corresponding to the current parameter combination, and recording the performance results; S244, based on the model optimization algorithm update rule, the current parameter combination is updated, if the new parameter combination has the best performance, the current initial fault detection model parameters are updated to the new parameter combination; if the new parameter combination performance is higher than the current optimal parameter combination performance, the global optimal parameters are updated to the new parameter combination, and the initial fault detection model structure is expanded to adapt to the new parameters; S245. If the current optimization fails to find the optimal solution, the parameter search range is expanded again, the search strategy is dynamically adjusted, and the parameter combination with the lowest performance is eliminated; S246, determine whether the maximum number of iterations has been reached, if so, jump to step S247; otherwise, return to S244 to step S245 to continue optimization; S247, terminate the optimization process, and output the global optimal initial fault detection model parameters.

6. A wind farm monitoring method based on real-time database according to claim 1, characterized in that: The method of establishing a fault prediction model based on the identified current wind farm operation fault data, and using the fault prediction model to predict the operation faults of the wind farm at a future moment, and obtaining the key factors affecting the operation faults of the wind farm includes the following steps: S31, quantifying the identified operation fault data of the current wind farm, and dividing the quantified operation fault data set into a training set and a test set; S32, setting the maximum number of iterations and population size of the meta-heuristic algorithm, initializing the population position using convex lens imaging reverse learning, and setting the number of training times, number of hidden layer neurons, and learning rate of the fault prediction model; S33, updating the nonlinear escape energy according to the change mode of the logic curve characteristics, and dynamically adjusting the objective function weight of the fault prediction model in combination with the current operating fault data; S34, evaluating the fitness value of the fault prediction model under the current fault prediction model parameter combination according to the objective function, and ranking the fitness values ​​of all fault prediction model parameter combinations; S35, judging by nonlinear escape energy, if the escape energy is less than the threshold, adopting a local optimization strategy to update the fault prediction model parameters; otherwise, adopting a global optimization strategy to update the fault prediction model parameters, and synchronously adjusting the auxiliary parameter combination according to the update direction of the dominant parameter combination; S36, comparing the optimization result of the current fault prediction model with the current optimal parameter combination, and recording the fault prediction model parameters with the best performance; S37, introducing a disturbance mechanism, randomly disturbing the current fault prediction model parameters, optimizing the disturbed parameters using a local optimization algorithm, obtaining the local optimal fault prediction model parameters in the disturbance area, and using a greedy strategy mechanism for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, proceed to step S38, otherwise, return to step S32 for re-optimization; S38. If the current number of iterations reaches the maximum number, the optimal parameters and fault prediction results of the fault prediction model are output, and the key factors affecting the operation failures of the wind farm are extracted from the fault prediction model; otherwise, the fault prediction model results are subjected to dimension-by-dimensional Gaussian mutation, and the process returns to step S31 for re-optimization.

7. A wind farm monitoring method based on real-time database according to claim 6, characterized in that: The formula for updating the nonlinear escape energy according to the change mode of the logistic curve characteristics is: Where T represents the nonlinear escape energy of the current fault prediction model parameter combination; T0 represents the nonlinear escape energy of the initial fault prediction model parameter combination; T max and T min They represent the maximum nonlinear escape energy and the minimum nonlinear escape energy of the fault prediction model parameter combination; f s (·) represents the Sigmoid function; x represents the value range of nonlinear energy; λ represents the adjustment coefficient; M represents the maximum number of iterations; k represents the number of iterations of the current optimization process; and e represents the base of the natural logarithm.

8. A wind farm monitoring method based on real-time database according to claim 7, characterized in that: The disturbance mechanism is introduced to randomly disturb the current fault prediction model parameters, and the disturbed parameters are optimized by using a local optimization algorithm to obtain the local optimal fault prediction model parameters in the disturbance area, and the greedy strategy mechanism is used for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, step S38 is entered, otherwise, step S32 is returned to re-optimization, which includes the following steps: S371, setting parameters of the local optimization algorithm, initializing the wind farm operation fault data, quantifying the fault data characteristics, generating a parameter combination of the initial fault prediction model, and performing random perturbations; S372, based on the change trend of the wind farm operation fault data, calculate the initial fitness value of each parameter combination, and evaluate the performance of the fault prediction model under the current parameter combination; S373, using the behavior rules of the local optimization algorithm, updating the position of each fault prediction model parameter combination, and dynamically adjusting the parameters of the fault prediction model; S374, performing local search optimization on the currently found fault prediction model parameters through a local sequence floating backward selection mechanism; S375. When the local optimization algorithm reaches the maximum number of iterations, the local optimal fault prediction model parameters of the disturbance area are output, and the greedy strategy mechanism is used for judgment. If the local optimal fault prediction model parameters reach the optimal parameter combination, step S38 is entered; otherwise, step S32 is returned to re-optimize.

9. A wind farm monitoring method based on real-time database according to claim 8, characterized in that: The local search optimization of the currently found fault prediction model parameters by the local sequence floating backward selection mechanism comprises the following steps: S3741, initializing a floating backward selection mechanism, and randomly adjusting each fault prediction model parameter according to the floating mechanism to generate a current candidate fault prediction model parameter combination; S3742, for the current candidate fault prediction model parameter combinations, eliminate the parameter combinations with the lowest fitness one by one, and re-evaluate the fitness of the remaining parameter combinations after each elimination until a preset stop condition is reached; S3743. Calculate the fitness value of each retained candidate parameter combination, evaluate its influence on the fault prediction model, compare the evaluation result with the current optimal fault prediction model parameter combination, and record the parameter combination with the highest fitness; S3744. If the parameter combination after the current local search optimization is better than the current optimal parameter combination, update it to the optimal parameter combination, otherwise retain the original optimal parameter combination; S3745. If the maximum number of iterations of the local search is reached, the local search is terminated; otherwise, return to step S3741 to perform a new round of floating perturbation and backward selection.

10. A wind farm monitoring system based on a real-time database, used to implement the wind farm monitoring method based on a real-time database as claimed in any one of claims 1 to 9, characterized in that: The wind farm monitoring system includes: A data acquisition module is used to acquire real-time operation data of the wind farm, pre-process the real-time operation data to obtain real-time operation characteristic data, and store the obtained real-time operation characteristic data in a real-time database; A fault diagnosis module is used to analyze the real-time operation characteristic data in the real-time database based on the fault diagnosis algorithm to identify the operation fault data of the current wind farm; A fault prediction module is used to establish a fault prediction model based on the identified operating fault data of the current wind farm, and use the fault prediction model to predict the operating faults of the wind farm at a future time to obtain key factors affecting the operating faults of the wind farm; The early warning monitoring module is used to issue fault early warnings based on the key factors affecting the operation failure of the wind farm, dynamically adjust the operation parameters of the wind farm equipment and optimize real-time monitoring.

Images