New energy power generation equipment intelligent regulation and control method and system based on big data

Abstract

The invention discloses a new energy power generation equipment intelligent regulation and control method and system based on big data. The method comprises the steps of obtaining operation data of new energy power generation equipment, performing data dimension reduction processing to obtain structured data, and performing feature extraction and clustering to obtain an influence factor candidate set; according to the influence factor candidate set, performing spatial-temporal feature fusion analysis to obtain a multi-dimensional feature vector, and performing continuous sampling and quantization to obtain a quantization sequence; on the basis of the quantization sequence, potential risks are predicted through a long-short term memory network, risk marking is carried out, risk distribution map dynamic adjustment control parameters are obtained, and an optimization strategy set is obtained; based on the optimization strategy set, equipment operation parameters are selected and adjusted through a real-time feedback mechanism, a system stability index is obtained, multi-dimensional evaluation verification is carried out, and if an expected threshold value is not reached, regulation and control parameters are optimized through a genetic algorithm, and a final control instruction is output. According to the invention, real-time prediction of risks and intelligent regulation and control of equipment can be realized.

Description

Technical Field

[0001] The present invention relates to the field of new energy power generation technology, and in particular to a method and system for intelligent control of new energy power generation equipment based on big data. Background Art

[0002] With the large-scale deployment of new energy power generation equipment, the real-time collection and analysis of massive amounts of operational data has become a core requirement for improving system stability and energy efficiency. New energy power generation equipment (such as wind and photovoltaic units) operates in complex and changing natural environments. It is necessary to accurately identify key factors affecting stability from high-dimensional, heterogeneous data, build multi-dimensional assessment models to comprehensively reflect equipment status, and design adaptive control strategies to cope with dynamically changing operating conditions. Furthermore, real-time prediction and closed-loop control of potential risks are required to ensure reliable power supply and efficient equipment operation.

[0003] Currently, traditional approaches primarily rely on centralized databases to store data, combined with static models to assess system stability, and employing rule bases or simple feedback mechanisms for control. For example, some technologies use principal component analysis (PCA) to reduce dimensionality, support vector machines (SVMs) or linear regression algorithms to predict risk, and trigger control instructions based on preset thresholds.

[0004] In summary, existing technologies are unable to cope with the scale and real-time requirements of high-dimensional heterogeneous data, and are unable to achieve real-time prediction and closed-loop control of risks. Summary of the Invention

[0005] The present invention provides a method for intelligent control of new energy power generation equipment based on big data to achieve real-time risk prediction and intelligent control of equipment.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a method for intelligent control of new energy power generation equipment based on big data, comprising:

[0007] Obtain the operating data of new energy power generation equipment, perform data dimensionality reduction processing, and obtain structured data;

[0008] Based on the structured data, a deep learning network is used to perform feature extraction and clustering to obtain a candidate set of influencing factors;

[0009] Performing spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector;

[0010] According to the multidimensional feature vector, continuous sampling and quantization are performed using a sliding window technique to obtain a quantized sequence;

[0011] Based on the quantified sequence, potential risks are predicted and risk labels are performed using a long short-term memory network to obtain a risk distribution map;

[0012] According to the risk distribution map, the control parameters are dynamically adjusted using a reinforcement learning framework to obtain an optimized strategy set;

[0013] Based on the optimization strategy set, the device operating parameters are selected and adjusted through a real-time feedback mechanism to obtain a system stability index;

[0014] Based on the system stability indicators, multi-dimensional evaluation and verification are carried out. If the expected threshold is not reached, the control parameters are optimized through genetic algorithms and the final control instructions are output to achieve intelligent control of new energy power generation equipment.

[0015] As an optional implementation manner, the acquisition of the operating data of the new energy power generation equipment and the data dimensionality reduction processing to obtain structured data include:

[0016] Acquire multi-dimensional heterogeneous data on voltage, current, power output, and environmental sensors of new energy power generation equipment, perform data cleaning operations, remove outliers and missing values, and obtain preprocessed data;

[0017] Performing normalization processing on the preprocessed data to eliminate dimensional differences and obtain a standardized data matrix;

[0018] According to the standardized data matrix, feature importance evaluation is performed to obtain a feature subset;

[0019] According to the feature subset, orthogonal transformation is performed using principal component analysis to obtain a dimension-reduced feature vector;

[0020] A spatiotemporal alignment operation is performed based on the dimensionally reduced feature vector to obtain structured data.

[0021] As an optional implementation, the method of performing feature extraction and clustering using a deep learning network based on the structured data to obtain a candidate set of influencing factors includes:

[0022] Extracting spatiotemporal features based on the structured data using a convolutional neural network and a long short-term memory network to obtain a spatiotemporal feature vector;

[0023] According to the spatiotemporal feature vector, a self-attention mechanism is used to perform feature weighting processing to obtain a weighted feature matrix;

[0024] Based on the weighted feature matrix, cluster analysis is performed using the K-means algorithm to obtain a candidate set of influencing factors.

[0025] As an optional implementation, performing spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector includes:

[0026] Extracting periodic characteristic components based on the candidate set of influencing factors to obtain a time series characteristic vector;

[0027] Based on the time series feature vector and the spatial topological structure of the device, a graph neural network is used to extract spatial features to obtain spatial correlation features;

[0028] Perform weighted splicing and fusion based on the temporal feature vector and the spatial correlation feature to obtain preliminary fusion features;

[0029] Based on the preliminary fusion features, principal component analysis and dimensionality reduction are performed to obtain the final multidimensional feature vector.

[0030] As an optional implementation manner, the method of continuously sampling and quantizing the multidimensional feature vector using a sliding window technique to obtain a quantized sequence includes:

[0031] According to the multidimensional feature vector, a sliding window with a fixed time span is set to perform data slicing processing to obtain window feature fragments;

[0032] Calculating statistical features of features in each dimension based on the window feature fragments to obtain statistical feature values;

[0033] Performing feature quantization classification based on the statistical feature values to obtain a quantization level sequence;

[0034] According to the quantization level sequence, the quantization levels are arranged and combined in time order to generate a final quantization sequence.

[0035] As an optional implementation, based on the quantization sequence, predicting potential risks and marking risks through a long short-term memory network to obtain a risk distribution map includes:

[0036] According to the quantified sequence, time series features are extracted through a bidirectional LSTM network to obtain a risk feature vector;

[0037] According to the risk feature vector, a fully connected layer is used to predict the risk probability to obtain a risk prediction value;

[0038] Based on the risk prediction value, the risk level is divided by comparing with the preset threshold to obtain the risk marking result;

[0039] Based on the risk labeling results, combined with the equipment spatial distribution information, visual mapping is performed to generate a risk distribution map.

[0040] As an optional implementation, the control parameters are dynamically adjusted using a reinforcement learning framework based on the risk distribution map to obtain an optimized strategy set, including:

[0041] According to the risk distribution map, a deep Q network is used to evaluate the strategy and obtain the initial control strategy;

[0042] Performing policy gradient iterative optimization according to the initial control strategy to generate a candidate optimization strategy set;

[0043] According to the preset stability index and efficiency index, the candidate optimization strategy set is evaluated and screened to obtain an optimization strategy set.

[0044] As an optional implementation, the selecting and adjusting of device operating parameters based on the optimization strategy set through a real-time feedback mechanism to obtain a system stability index includes:

[0045] Performing strategy optimization according to the optimization strategy set to obtain the optimal control strategy;

[0046] Generate a device control instruction set based on the optimal control strategy, dynamically adjust the parameters of the generator set, and obtain real-time operating data;

[0047] Performing state stability calculation based on the real-time operating data to obtain a preliminary stability index;

[0048] Based on the preliminary stability index, an index quantification process is performed to obtain a final system stability index.

[0049] As an optional implementation, the multi-dimensional evaluation and verification is performed based on the system stability index. If the expected threshold is not reached, the control parameters are optimized by genetic algorithm and the final control instructions are output to realize the intelligent control of the new energy power generation equipment, including:

[0050] Based on the system stability indicators, evaluation and verification are performed through preset indicator dimensions to obtain a comprehensive evaluation result;

[0051] When the comprehensive evaluation result does not reach the preset threshold, an initial population is constructed according to the current control parameters, and an optimized parameter set is obtained by iterative optimization through a genetic algorithm;

[0052] Recalculate the system stability index based on the optimized parameter set and verify it, and generate the final control instruction if it meets the requirements;

[0053] According to the final control instruction, the operating parameters of the generator set are adjusted through the device control interface to complete the intelligent regulation of the new energy power generation equipment.

[0054] In a second aspect, the present invention provides an intelligent control system for new energy power generation equipment based on big data, comprising:

[0055] The data dimension reduction module is used to obtain the operating data of new energy power generation equipment, perform data dimension reduction processing, and obtain structured data;

[0056] A feature extraction module is used to extract and cluster features based on the structured data using a deep learning network to obtain a candidate set of influencing factors;

[0057] A feature fusion module is used to perform spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector;

[0058] A feature quantization module is used to perform continuous sampling and quantization based on the multidimensional feature vector using a sliding window technique to obtain a quantized sequence;

[0059] A risk marking module is used to predict potential risks and mark risks based on the quantified sequence through a long short-term memory network to obtain a risk distribution map;

[0060] A strategy optimization module is used to dynamically adjust control parameters based on the risk distribution map using a reinforcement learning framework to obtain an optimized strategy set;

[0061] A parameter optimization module is used to select and adjust equipment operating parameters based on the optimization strategy set through a real-time feedback mechanism to obtain a system stability index;

[0062] The intelligent control module is used to perform multi-dimensional evaluation and verification based on the system stability indicators. If the expected threshold is not reached, the control parameters are optimized through genetic algorithms and the final control instructions are output to achieve intelligent control of new energy power generation equipment.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] (1) The multi-dimensional stability evaluation model constructed by the present invention based on deep learning network and spatiotemporal fusion technology can comprehensively capture the spatiotemporal correlation and dynamic evolution law of equipment operation, overcoming the limitations of traditional single-dimensional or static models.

[0065] (2) The present invention predicts the potential risks of quantitative sequences through long short-term memory networks, generates high-precision risk distribution maps, realizes early warning and precise positioning of abnormal working conditions, and improves the comprehensiveness and predictive ability of system status monitoring.

[0066] (3) The present invention adopts a reinforcement learning framework to dynamically adjust control parameters and combines it with a genetic algorithm optimization strategy, so that the control algorithm can adaptively generate the optimal control solution based on real-time operation data, breaking through the rigid limitations of traditional fixed thresholds or rule bases. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1This is a flow chart of a method for intelligently controlling new energy power generation equipment based on big data provided by an embodiment of the present invention;

[0068] Figure 2 This is a structural diagram of an intelligent control system for new energy power generation equipment based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0070] Reference Figure 1 The first embodiment of the present invention provides a method for intelligent control of new energy power generation equipment based on big data, comprising the following steps:

[0071] S11, obtaining the operating data of the new energy power generation equipment, performing data dimensionality reduction processing, and obtaining structured data;

[0072] S12, performing feature extraction and clustering using a deep learning network based on the structured data to obtain a candidate set of influencing factors;

[0073] S13, performing spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector;

[0074] S14, performing continuous sampling and quantization using a sliding window technique based on the multidimensional feature vector to obtain a quantized sequence;

[0075] S15, based on the quantified sequence, predicting potential risks and marking risks through a long short-term memory network to obtain a risk distribution map;

[0076] S16, dynamically adjusting control parameters using a reinforcement learning framework according to the risk distribution map to obtain an optimized strategy set;

[0077] S17, based on the optimization strategy set, selecting and adjusting device operating parameters through a real-time feedback mechanism to obtain a system stability index;

[0078] S18, performing multi-dimensional evaluation and verification based on the system stability index. If the expected threshold is not reached, optimizing the control parameters through genetic algorithm and outputting the final control instructions to realize intelligent control of the new energy power generation equipment.

[0079] In step S11, the operation data of the new energy power generation equipment is obtained, and data dimension reduction processing is performed to obtain structured data, including:

[0080] Acquire multi-dimensional heterogeneous data on voltage, current, power output, and environmental sensors of new energy power generation equipment, perform data cleaning operations, remove outliers and missing values, and obtain preprocessed data;

[0081] Performing normalization processing on the preprocessed data to eliminate dimensional differences and obtain a standardized data matrix;

[0082] According to the standardized data matrix, feature importance evaluation is performed to obtain a feature subset;

[0083] According to the feature subset, orthogonal transformation is performed using principal component analysis to obtain a dimension-reduced feature vector;

[0084] A spatiotemporal alignment operation is performed based on the dimensionally reduced feature vector to obtain structured data.

[0085] It should be noted that the dimensionality reduction processing of the operating data of new energy power generation equipment refers to the process of converting the original heterogeneous operating data into structured data through steps such as data cleaning, normalization, feature importance assessment, principal component analysis dimensionality reduction and spatiotemporal data alignment. This operation can provide high-quality data input for subsequent feature analysis, risk prediction and optimization and regulation. In an embodiment of the present invention, the dimensionality reduction processing of the operating data is based on the real-time monitoring data of the equipment, and is achieved through steps such as automated data cleaning, intelligent feature screening and spatiotemporal fusion. In the management of new energy power stations, the use of this dimensionality reduction processing method can significantly improve data processing efficiency and feature effectiveness, and ensure the accuracy of subsequent intelligent analysis algorithms.

[0086] Among them, data acquisition and cleaning are carried out by collecting equipment operation data at a frequency of once per second through industrial standard communication protocols, including parameters such as voltage (normal range 0-1500 volts) and current (0-1000 amperes). Sliding window technology is used to process abnormal data: the median is calculated by taking 2 data points before and after, totaling 5 points, to replace the abnormal value; linear interpolation is used for missing data. When there are no more than 3 consecutive missing points, interpolation is performed, and when there are more than 3 points, the entire segment is eliminated. The specific implementation uses Python's Pandas library, calling rolling(5).median() and interpolate() methods to complete. Data normalization processing sets normalization schemes according to the characteristics of different types of data: voltage values are divided by the upper limit of 1500 volts; current values are divided by the larger value of the rated current and the historical maximum current; ambient temperature values are first added by 20 degrees and then divided by 80 degrees. Special parameter processing includes: solar irradiance is normalized to the range of 0-1200 watts / square meter, and wind speed data is first logarithmically transformed and then normalized by adding 1. All normalization operations are implemented through the MinMaxScaler component of the scikit-learn library to ensure that the output values are between 0 and 1.

[0087] It is worth noting that the feature importance assessment uses an improved random forest model for feature screening, setting 100 decision trees, with a maximum depth of 10 layers per tree and a minimum number of node splits of 5. Power-related features are given a weight of 1.2 times, and temperature-related features are given a weight of 1.1 times. The feature selection threshold is dynamically adjusted based on the total number of features, with a base value of 0.015 and increasing logarithmically with the number of features. Key features with importance above the threshold are selected for subsequent analysis. Principal component analysis dimensionality reduction first calculates the covariance matrix between each feature: Where X represents the normalized data matrix, ∑ represents the covariance matrix, n-1 represents the number of samples minus 1, used for unbiased estimation, and T represents the transpose operation. Matrix eigendecomposition is then performed: λ, v = eig(∑), where λ represents the eigenvalue vector (arranged in descending order), v represents the eigenvector matrix (each column is the eigenvector of the corresponding eigenvalue), eig represents the matrix eigendecomposition function, and ∑ represents the covariance matrix. Principal components with cumulative contributions exceeding 85% and individual contributions exceeding 5% are automatically selected. The calculation automatically stops when the contribution of a newly added component falls below 1% to improve efficiency. In practical applications, the original features are typically reduced to 5 to 8 dimensions, significantly reducing data dimensionality while maintaining information content. Spatiotemporal data alignment uses spline interpolation to normalize the data to 1-minute intervals in the temporal dimension. A four-level coding system (equipment type - plant number - region code - equipment serial number) is established in the spatial dimension to identify each device. The final output is structured data containing timestamps, device codes, geographic coordinates, eigenvectors, and topological relationships. This data is stored in JSON format and can be directly integrated into various monitoring systems.

[0088] In step S12, the structured data is subjected to feature extraction and clustering using a deep learning network to obtain a candidate set of influencing factors, including:

[0089] Extracting spatiotemporal features based on the structured data using a convolutional neural network and a long short-term memory network to obtain a spatiotemporal feature vector;

[0090] According to the spatiotemporal feature vector, a self-attention mechanism is used to perform feature weighting processing to obtain a weighted feature matrix;

[0091] Based on the weighted feature matrix, cluster analysis is performed using the K-means algorithm to obtain a candidate set of influencing factors.

[0092] It is worth noting that the spatiotemporal feature extraction uses a CNN-LSTM hybrid network architecture to process structured data: Convolutional Neural Network (CNN) part: uses 3 convolutional layers (the number of filters is 32, 64, and 128 respectively), with ReLU activation function and MaxPooling layer to extract spatial local features; Long Short-Term Memory Network (LSTM) part: sets 128 hidden units to process time series dependencies; Feature Fusion: splices the spatial features output by CNN with the temporal features output by LSTM to form a spatiotemporal feature vector containing 256-dimensional features; Use the Keras framework to build, and the input data needs to be reshaped into the format of (number of samples, time steps, number of features). Feature weighting processing adopts a feature optimization implementation scheme based on the self-attention mechanism: (1) Construct a multi-head attention mechanism (head_num=3), and the attention dimension d of each head is 1. k =d v =64; (2) Scaled dot product attention is used to calculate the query-key attention weight, and the formula is Where Q represents the set of feature vectors to be queried, K represents the key matrix, V represents the actual set of feature vectors, and d kRepresents the dimension of the key vector, softmax represents the normalization function, and T represents the matrix transpose operation; (3) Dynamically weight the 256-dimensional spatiotemporal feature vector, and keep the output dimension unchanged through residual connection and layer normalization; (4) Output the weighted feature matrix, and the key feature weight is increased by 30%-50%. Python implementation uses TensorFlow's MultiHeadAttention layer. Cluster analysis uses improved K-means clustering, and the specific implementation process is as follows: (1) Automatically determine the number of clusters k: draw the SSE curve with k = 1 to 10 through the elbow method, select the inflection point (the point where the slope decreases significantly and slows down), and select the k value that maximizes the silhouette coefficient (usually 3-5 categories); (2) Use the k-means++ algorithm to initialize the cluster center, and set the upper limit of the number of iterations to 300 times; (3) Use cosine similarity as the distance metric, and the calculation formula is Where distance(x, y) represents the cosine distance between vector x and vector y, x and y represent input vectors and feature vectors of the same dimension, and || || represents the L2 norm of the vector. (4) An empty cluster detection mechanism is introduced, which triggers reinitialization when the number of samples in any cluster is less than 1% of the total number of samples. (5) The center points of each cluster are output as candidate sets of influencing factors.

[0093] In step S13, the spatiotemporal feature fusion analysis is performed based on the candidate set of influencing factors to obtain a multi-dimensional feature vector, including:

[0094] Extracting periodic characteristic components based on the candidate set of influencing factors to obtain a time series characteristic vector;

[0095] Based on the time series feature vector and the spatial topological structure of the device, a graph neural network is used to extract spatial features to obtain spatial correlation features;

[0096] Perform weighted splicing and fusion based on the temporal feature vector and the spatial correlation feature to obtain preliminary fusion features;

[0097] Based on the preliminary fusion features, principal component analysis and dimensionality reduction are performed to obtain the final multidimensional feature vector.

[0098] It should be noted that the extraction of periodic characteristic components uses a time series decomposition algorithm to extract key periodic features. The STL (Seasonal-Trend decomposition using Loess) decomposition method is used, and the sliding window size is set to 24 hours (the daily cycle of photovoltaic power generation). The trend component (reflecting the long-term performance changes of the equipment), the period component (capturing the daily power generation pattern) and the residual component (identifying abnormal fluctuations) are extracted. The decomposition formula of the STL method is as follows: t =T t +St +R t , where Y t Represents the original time series data, T t represents the trend component; S t Expressed as periodic component; R t is the residual term; t is the current time series value; this is implemented using the STL class from the statsmodels library, with the period parameter set to 1440 minutes. Spatial feature extraction is based on graph neural network processing of device topology, constructing a graph structure (nodes represent devices, edges represent electrical connections). The GAT (Graph Attention Network) model is used, with a two-layer network configuration (four attention heads per layer, 64 node feature dimensions, and attention leakage ReLU activation). The attention coefficient is calculated as:

[0099]

[0100] Among them, α ij represents the attention coefficient of node i to j; W is the trainable weight matrix; h i 、h j is the original feature vector of nodes i and j; || represents the concatenation operation; a represents the trainable parameter vector; N i represents the neighbor set of node i (usually including itself, i.e. self-loop connection); k represents the neighbor node; h k Represents the feature vector of the neighbor node; LeakyReLU represents the rectified linear unit activation function with leakage; exp represents the exponential function, which maps the attention score to the positive real number domain for normalization; parameter settings: learning rate 0.001, training 50 epochs, batch_size = 32.

[0101] Feature fusion uses a weighted splicing method for spatiotemporal features. First, the temporal feature vector and the spatial feature matrix are aligned, and an adaptive weight allocation mechanism is designed: time weight = period component variance / total variance, spatial weight = 1 - time weight. Preliminary fused features are then spliced together according to the weights, and dynamic weight adjustment is implemented to ensure a balance between spatiotemporal and temporal features. The optimized PCA dimensionality reduction process calculates the covariance matrix of the fused features: Where ∑ represents the covariance matrix, X represents the standardized fused feature matrix, T represents the transpose operation, and n represents the number of samples. Principal components with a cumulative contribution rate ≥ 90% are retained, and a minimum dimension of 5 is set to ensure that key information is not lost. The final 5- to 8-dimensional feature vector is output.

[0102] In step S14, the sliding window technique is used to continuously sample and quantize the multidimensional feature vector to obtain a quantized sequence, including:

[0103] According to the multidimensional feature vector, a sliding window with a fixed time span is set to perform data slicing processing to obtain window feature fragments;

[0104] Calculating statistical features of features in each dimension based on the window feature fragments to obtain statistical feature values;

[0105] Performing feature quantization classification based on the statistical feature values to obtain a quantization level sequence;

[0106] According to the quantization level sequence, the quantization levels are arranged and combined in time order to generate a final quantization sequence.

[0107] It should be noted that this step converts continuous features into discrete quantized sequences by performing sliding window analysis and quantization on the multidimensional feature vector. The sliding window processing sets a dynamically adjusted sliding window mechanism with a basic window size of 10 minutes (corresponding to 60 sampling points and a sampling frequency of 0.1Hz), an adaptive window step size (range 1-5 minutes), and a 30% overlap rate design. The core window division formula is: W t ={x t-k , x t-k+1 ,...,x t}, where W t represents the window data at time t, k is the window size parameter (default value is 60), x t Represents the sampled value at historical time t. This is implemented using the rolling(window=60, min_periods=30) method of Pandas. Statistical feature calculation: Calculates four-dimensional statistics for each window, including mean μ, standard deviation σ, kurtosis κ, and skewness γ, and outputs a standardized statistical feature matrix.

[0108] Among them, feature quantization classification: using the improved equal frequency binning method, automatically determine the number of bins k = 5, and define the quantization level 1 (very low) to 5 (very high). Binning formula: Among them, Q j is the quantization level of the jth feature, s i is the statistical eigenvalue, b qis the bin boundary point, and q is the quantization level from 1 (very low) to 5 (very high). Sequence generation: Sort the quantization results by timestamp, add the device location code prefix, and generate a standard JSON format sequence. Example structure: {"device":"WT-01-B-05","timestamp":"2023-05-01T14:30:00Z","quant_seq":[3, 2, 4, 1, 5],"stats":[0.32, 1.05, 0.78, 0.56, 1.12]}, where device represents the device code, timestamp represents the timestamp, quant_seq represents the quantization sequence, and stats represents the status.

[0109] In step S15, based on the quantization sequence, potential risks are predicted and risk labels are performed using a long short-term memory network to obtain a risk distribution map, including:

[0110] According to the quantified sequence, time series features are extracted through a bidirectional LSTM network to obtain a risk feature vector;

[0111] According to the risk feature vector, a fully connected layer is used to predict the risk probability to obtain a risk prediction value;

[0112] Based on the risk prediction value, the risk level is divided by comparing with the preset threshold to obtain the risk marking result;

[0113] Based on the risk labeling results, combined with the equipment spatial distribution information, visual mapping is performed to generate a risk distribution map.

[0114] It should be noted that the time series feature extraction uses a bidirectional LSTM network to process the quantized sequence. The network structure consists of two layers of LSTM (128 hidden units per layer, dropout rate 0.3), and the forward and reverse outputs are fused through the splicing layer. The calculation formula is: and in, and Represents the forward / reverse LSTM output, Q 1：t and Q t：1 The input sequence is the final output of the 256-dimensional risk feature vector. Risk probability prediction: A fully connected neural network (FCN) is constructed for risk prediction. The network structure contains 3 hidden layers (dimensions are 128, 64, and 32, respectively), using LeakyReLU activation (α = 0.1), and the output layer uses Sigmoid activation. Key calculation formula: Among them, p is the risk probability, h is the input feature, w T is a trainable parameter, b is a bias vector, and the output risk prediction value range is [0, 1].

[0115] The risk level classification is based on the establishment of a dynamic threshold mechanism, which divides the risk prediction value into 5 levels (1-5, with 5 being the highest risk). The threshold setting formula is: Among them, τ i is the i-th level threshold, β is the adjustment coefficient (default is 0.2), σ p = is the standard deviation of historical risk probability, i represents the i-th level. Output the risk labeling results with confidence. Map generation: Develop a WebGL-based visualization system, associate risk labels with device GIS coordinates, and use heat map gradient coloring (blue → yellow → red corresponds to risk level 1 → 5). Introduce 3D terrain mapping technology, and the risk intensity calculation formula is: Where I is the risk intensity at location l, D is the device set, r d is the risk level of device d, l d is the device coordinate, l is the 3D coordinate of the target device. Output an interactive risk distribution map.

[0116] In step S16, the control parameters are dynamically adjusted using a reinforcement learning framework according to the risk distribution map to obtain an optimized strategy set, including:

[0117] According to the risk distribution map, a deep Q network is used to evaluate the strategy and obtain the initial control strategy;

[0118] Performing policy gradient iterative optimization according to the initial control strategy to generate a candidate optimization strategy set;

[0119] According to the preset stability index and efficiency index, the candidate optimization strategy set is evaluated and screened to obtain an optimization strategy set.

[0120] It should be noted that

[0121] in,

[0122] In step S17, based on the optimization strategy set, the device operating parameters are selected and adjusted through a real-time feedback mechanism to obtain a system stability index, including:

[0123] Performing strategy optimization according to the optimization strategy set to obtain the optimal control strategy;

[0124] Generate a device control instruction set based on the optimal control strategy, dynamically adjust the parameters of the generator set, and obtain real-time operating data;

[0125] Performing state stability calculation based on the real-time operating data to obtain a preliminary stability index;

[0126] Based on the preliminary stability index, an index quantification process is performed to obtain a final system stability index.

[0127] It is worth noting that the strategy evaluation is performed by constructing a deep Q network (DQN). The state space is a 256-dimensional risk map feature vector, and the action space is a 6-dimensional continuous control parameter adjustment. The reward function is designed as: r = 0.6R s +0.3R e -0.1R r , where r is the available reward, R s is the stability bonus (based on the standard deviation of power fluctuation), R e is the efficiency bonus (based on power generation), R r is the risk penalty (based on the risk level). The network adopts a 3-layer fully connected structure (256-128-64 nodes) and uses the Adam optimizer (1r=0.001) to output the initial policy π. The policy optimization uses the PPO algorithm for iterative optimization, and the policy network uses a Gaussian distribution output (mean μ, variance σ). The advantage estimation formula is: The discount factor γ=0.99, the trade-off factor λ=0.95, represents the advantage estimate, l is the time step offset, which represents the number of steps from the current moment to the past, and δ t+l The TD residual quantifies the difference between the current prediction and the subsequent actual reward. Each iteration generates 50 sets of candidate strategies, and the update amplitude is controlled by KL divergence (threshold 0.01). Strategy screening: Establish a multi-objective evaluation system and define a comprehensive scoring function: Among them, ΔP is the performance fluctuation after the strategy is executed, E is the power generation, R is the risk level, and P rate is the system baseline performance threshold, E max is the theoretical maximum power generation effect, and score is the comprehensive score. The non-dominated solutions on the Pareto front are retained, and 3-5 optimal strategies are ultimately output to form the optimization strategy set.

[0128] In step S18, a multi-dimensional evaluation and verification is performed based on the system stability index. If the expected threshold is not reached, the control parameters are optimized by a genetic algorithm and a final control instruction is output to realize intelligent control of the new energy power generation equipment, including:

[0129] Based on the system stability indicators, evaluation and verification are performed through preset indicator dimensions to obtain a comprehensive evaluation result;

[0130] When the comprehensive evaluation result does not reach the preset threshold, an initial population is constructed according to the current control parameters, and an optimized parameter set is obtained by iterative optimization through a genetic algorithm;

[0131] Recalculate the system stability index based on the optimized parameter set and verify it, and generate the final control instruction if it meets the requirements;

[0132] According to the final control instruction, the operating parameters of the generator set are adjusted through the device control interface to complete the intelligent regulation of the new energy power generation equipment.

[0133] It should be noted that the multi-dimensional evaluation verification establishes an evaluation system with three dimensions: voltage stability: measures the bus voltage fluctuation rate, with a requirement of <2.5%; frequency stability: detects system frequency deviation, with a threshold of ±0.2Hz; power balance: evaluates power generation / load matching, with a target value of >95%. The weighted summation is used to calculate the comprehensive evaluation score: comprehensive score = 0.5×voltage stability score + 0.3×frequency stability score + 0.2×power balance score. The preset qualified threshold is 85 points (out of 100), and the optimization process is triggered when the score is lower than the threshold. Genetic algorithm optimization: The coding scheme uses real number coding, and each chromosome contains 6 control parameters (including excitation voltage, pitch angle, active power setting value, reactive power compensation, voltage regulator gain, and damping controller parameters); the initial population is based on the current parameters, and 200 individuals are randomly generated within the range of ±15%; fitness function design: Where ΔV represents the voltage deviation (the maximum deviation from the rated value), V max Indicates the maximum allowable voltage deviation, P output Indicates the actual output active power of the system, P max Indicates the maximum output active power of the system, T settle Indicates the dynamic response adjustment time, T refDenotes the reference adjustment time threshold, Fitness denotes the fitness value, and w1, w2, and w3 denote the weights for voltage deviation, active power, and adjustment time, respectively. A tournament selection (size = 5) was performed, randomly selecting five individuals from the population as a group (repeated sampling was allowed). The fitness value (Fitness) of each individual within the group was calculated. The individual with the highest fitness within the group was selected as the winner and entered the next generation. This process was repeated until the new population size reached a preset value (200 individuals). The top 10% of individuals in the current generation were retained to advance directly to the next generation to prevent loss of high-quality genes. A binary crossover (probability 0.8) was simulated, pairing selected parent individuals. For each pair of parents, a crossover was performed with an 80% probability; otherwise, the parent was simply copied. Gaussian mutation (probability 0.05, σ = 0.1) was performed, mutating each parameter of each individual with a 5% probability. Gaussian noise was added to the selected parameters, and if the mutated parameters exceeded the parameter range, they were forcibly truncated to the boundary value. The termination criteria are a maximum of 50 iterations or improvement of less than 1% for 10 consecutive generations. The control instruction generation and optimization verification process includes three levels of verification: primary verification to ensure stability indicators meet standards; intermediate verification to check equipment safety limits; and advanced verification to assess economic efficiency (change in electricity cost less than 3%). Parameter sets that pass the verification are converted into standard control instructions.

[0134] Reference Figure 2 The second embodiment of the present invention provides an intelligent control system for new energy power generation equipment based on big data, including:

[0135] The data dimension reduction module is used to obtain the operating data of new energy power generation equipment, perform data dimension reduction processing, and obtain structured data;

[0136] A feature extraction module is used to extract and cluster features based on the structured data using a deep learning network to obtain a candidate set of influencing factors;

[0137] A feature fusion module is used to perform spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector;

[0138] A feature quantization module is used to perform continuous sampling and quantization based on the multidimensional feature vector using a sliding window technique to obtain a quantized sequence;

[0139] A risk marking module is used to predict potential risks and mark risks based on the quantified sequence through a long short-term memory network to obtain a risk distribution map;

[0140] A strategy optimization module is used to dynamically adjust control parameters based on the risk distribution map using a reinforcement learning framework to obtain an optimized strategy set;

[0141] A parameter optimization module is used to select and adjust equipment operating parameters based on the optimization strategy set through a real-time feedback mechanism to obtain a system stability index;

[0142] The intelligent control module is used to perform multi-dimensional evaluation and verification based on the system stability indicators. If the expected threshold is not reached, the control parameters are optimized through genetic algorithms and the final control instructions are output to achieve intelligent control of new energy power generation equipment.

[0143] It should be noted that the intelligent control system for new energy power generation equipment based on big data provided in an embodiment of the present invention is used to execute all the process steps of the intelligent control method for new energy power generation equipment based on big data in the above embodiment. The working principles and beneficial effects of the two correspond one to one, so they will not be repeated here.

[0144] The embodiment of the present invention further provides a terminal device. The terminal device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a program for intelligent control of new energy power generation equipment based on big data. When the processor executes the computer program, the steps in each of the above-mentioned embodiments of the method for intelligent control of new energy power generation equipment based on big data are implemented, such as Figure 1 Alternatively, the processor implements the functions of the modules / units in the above-mentioned system embodiments when executing the computer program.

[0145] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0146] The terminal device may be a computing device such as a desktop computer, laptop, PDA, or smart tablet. The terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the aforementioned components are merely examples of terminal devices and do not constitute a limitation of the terminal device. The terminal device may include more or fewer components than those described above, or a combination of certain components or different components. For example, the terminal device may also include input / output devices, network access devices, buses, and the like.

[0147] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0148] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store an operating system, at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory can include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0149] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned embodiment method by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0150] It should be noted that the system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the system embodiment provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0151] The specific embodiments described above further illustrate the objectives, technical solutions, 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 intelligent control of new energy power generation equipment based on big data, characterized in that: include: Obtain the operating data of new energy power generation equipment, perform data dimensionality reduction processing, and obtain structured data; Based on the structured data, a deep learning network is used to perform feature extraction and clustering to obtain a candidate set of influencing factors; Performing spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector; According to the multidimensional feature vector, continuous sampling and quantization are performed using a sliding window technique to obtain a quantized sequence; Based on the quantified sequence, potential risks are predicted and risk labels are performed using a long short-term memory network to obtain a risk distribution map; According to the risk distribution map, the control parameters are dynamically adjusted using a reinforcement learning framework to obtain an optimized strategy set; Based on the optimization strategy set, the device operating parameters are selected and adjusted through a real-time feedback mechanism to obtain a system stability index; Based on the system stability indicators, multi-dimensional evaluation and verification are carried out. If the expected threshold is not reached, the control parameters are optimized through genetic algorithms and the final control instructions are output to achieve intelligent control of new energy power generation equipment.

2. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: The step of obtaining the operating data of the new energy power generation equipment and performing data dimensionality reduction processing to obtain structured data includes: Acquire multi-dimensional heterogeneous data on voltage, current, power output, and environmental sensors of new energy power generation equipment, perform data cleaning operations, remove outliers and missing values, and obtain preprocessed data; Performing normalization processing on the preprocessed data to eliminate dimensional differences and obtain a standardized data matrix; According to the standardized data matrix, feature importance evaluation is performed to obtain a feature subset; According to the feature subset, orthogonal transformation is performed using principal component analysis to obtain a dimension-reduced feature vector; A spatiotemporal alignment operation is performed based on the dimensionally reduced feature vector to obtain structured data.

3. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: According to the structured data, feature extraction and clustering are performed using a deep learning network to obtain a candidate set of influencing factors, including: Extracting spatiotemporal features based on the structured data using a convolutional neural network and a long short-term memory network to obtain a spatiotemporal feature vector; According to the spatiotemporal feature vector, a self-attention mechanism is used to perform feature weighting processing to obtain a weighted feature matrix; Based on the weighted feature matrix, cluster analysis is performed using the K-means algorithm to obtain a candidate set of influencing factors.

4. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: The method of performing spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector includes: Extracting periodic characteristic components based on the candidate set of influencing factors to obtain a time series characteristic vector; Based on the time series feature vector and the spatial topological structure of the device, a graph neural network is used to extract spatial features to obtain spatial correlation features; Perform weighted splicing and fusion based on the temporal feature vector and the spatial correlation feature to obtain preliminary fusion features; Based on the preliminary fusion features, principal component analysis and dimensionality reduction are performed to obtain the final multidimensional feature vector.

5. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: The method of continuously sampling and quantizing the multidimensional feature vector using a sliding window technique to obtain a quantized sequence includes: According to the multidimensional feature vector, a sliding window with a fixed time span is set to perform data slicing processing to obtain window feature fragments; Calculating statistical features of features in each dimension based on the window feature fragments to obtain statistical feature values; Performing feature quantization classification based on the statistical feature values to obtain a quantization level sequence; According to the quantization level sequence, the quantization levels are arranged and combined in time order to generate a final quantization sequence.

6. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: The method of predicting potential risks and labeling risks based on the quantified sequence through a long short-term memory network to obtain a risk distribution map includes: According to the quantified sequence, time series features are extracted through a bidirectional LSTM network to obtain a risk feature vector; According to the risk feature vector, a fully connected layer is used to predict the risk probability to obtain a risk prediction value; Based on the risk prediction value, the risk level is divided by comparing with the preset threshold to obtain the risk marking result; Based on the risk labeling results, combined with the equipment spatial distribution information, visual mapping is performed to generate a risk distribution map.

7. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: According to the risk distribution map, the control parameters are dynamically adjusted using a reinforcement learning framework to obtain an optimized strategy set, including: According to the risk distribution map, a deep Q network is used to evaluate the strategy and obtain the initial control strategy; Performing policy gradient iterative optimization according to the initial control strategy to generate a candidate optimization strategy set; According to the preset stability index and efficiency index, the candidate optimization strategy set is evaluated and screened to obtain an optimization strategy set.

8. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: The system stability index is obtained by selecting and adjusting the device operating parameters based on the optimization strategy set through a real-time feedback mechanism, including: Performing strategy optimization according to the optimization strategy set to obtain the optimal control strategy; Generate a device control instruction set based on the optimal control strategy, dynamically adjust the parameters of the generator set, and obtain real-time operating data; Performing state stability calculation based on the real-time operating data to obtain a preliminary stability index; Based on the preliminary stability index, an index quantification process is performed to obtain a final system stability index.

9. The intelligent control method for new energy power generation equipment based on big data according to claim 1 is characterized in that: The multi-dimensional evaluation and verification is performed based on the system stability index. If the expected threshold is not reached, the control parameters are optimized by genetic algorithm and the final control instructions are output to realize the intelligent control of the new energy power generation equipment, including: Based on the system stability indicators, evaluation and verification are performed through preset indicator dimensions to obtain a comprehensive evaluation result; When the comprehensive evaluation result does not reach the preset threshold, an initial population is constructed according to the current control parameters, and an optimized parameter set is obtained by iterative optimization through a genetic algorithm; Recalculate the system stability index based on the optimized parameter set and verify it, and generate the final control instruction if it meets the requirements; According to the final control instruction, the operating parameters of the generator set are adjusted through the device control interface to complete the intelligent regulation of the new energy power generation equipment.

10. An intelligent control system for new energy power generation equipment based on big data, characterized in that: include: The data dimension reduction module is used to obtain the operating data of new energy power generation equipment, perform data dimension reduction processing, and obtain structured data; A feature extraction module is used to extract and cluster features based on the structured data using a deep learning network to obtain a candidate set of influencing factors; A feature fusion module is used to perform spatiotemporal feature fusion analysis based on the candidate set of influencing factors to obtain a multidimensional feature vector; A feature quantization module is used to perform continuous sampling and quantization based on the multidimensional feature vector using a sliding window technique to obtain a quantized sequence; A risk marking module is used to predict potential risks and mark risks based on the quantified sequence through a long short-term memory network to obtain a risk distribution map; A strategy optimization module is used to dynamically adjust control parameters based on the risk distribution map using a reinforcement learning framework to obtain an optimized strategy set; A parameter optimization module is used to select and adjust equipment operating parameters based on the optimization strategy set through a real-time feedback mechanism to obtain a system stability index; The intelligent control module is used to perform multi-dimensional evaluation and verification based on the system stability indicators. If the expected threshold is not reached, the control parameters are optimized through genetic algorithms and the final control instructions are output to achieve intelligent control of new energy power generation equipment.

Images