New energy unit icing shutdown prediction method and system based on multi-feature interaction threshold

Abstract

The invention discloses a new energy unit icing shutdown prediction method and system based on a multi-feature interaction threshold. The method comprises the steps of collecting and preprocessing multi-dimensional data of a new energy unit; based on meteorological and geographic features, generating a corresponding clustering label for each new energy unit by using a clustering algorithm; constructing a full-connection deep neural network shutdown prediction model, taking the preprocessed data and the clustering labels as input features, and training the icing shutdown probability of a model output unit; for a single feature, a feature fixing strategy is adopted, and a single feature threshold interval is determined; key features are selected for double-feature interaction analysis, a three-dimensional decision boundary is constructed, an interaction effect is quantified, and a multi-feature interaction rule is extracted based on a decision tree algorithm; and constructing a comprehensive discrimination rule, setting risk preference parameters, carrying out adaptive threshold updating, and finally outputting a shutdown prediction result. The method can significantly improve the shutdown prediction precision, and is suitable for different types of new energy equipment such as wind power and photovoltaic equipment.

Description

Technical Field

[0001] This invention belongs to the field of new energy power generation technology, and relates to a method and system for predicting icing shutdown of new energy units, and more particularly to a method and system for predicting icing shutdown of new energy units based on multi-feature interactive thresholds. Background Technology

[0002] As the global energy structure accelerates its transition to a low-carbon model, the installed capacity of new energy power generation is experiencing explosive growth. However, the operational stability of new energy units, especially wind turbines and photovoltaic power generation systems, faces severe challenges under extreme weather conditions. In cold and humid weather such as cold waves, wind turbine blades are highly susceptible to icing, triggering automatic shutdowns of protection systems; simultaneously, snow accumulation on photovoltaic modules can prevent them from receiving effective solar radiation and cause them to stop working. Large-scale unplanned outages not only result in significant power generation losses but also pose a serious threat to the safe and stable operation of the power grid. Therefore, accurately predicting the risk of icing-induced outages of new energy units is crucial for improving the absorption capacity of new energy, ensuring grid security, and optimizing operation and maintenance scheduling.

[0003] Currently, mainstream icing shutdown prediction technologies have many significant shortcomings and cannot meet practical needs:

[0004] 1. Limited forecasting methods: Existing technologies mostly use single meteorological features such as humidity or temperature to determine thresholds, without fully considering the complex interactions between multiple meteorological factors, resulting in limited forecasting accuracy.

[0005] 2. Threshold setting lacks scientific basis: The shutdown threshold in traditional methods is mostly based on experience and lacks theoretical analysis and data-driven scientific basis, which is prone to false alarms or missed alarms.

[0006] 3. Ignoring nonlinear relationships: There is a complex nonlinear relationship between weather conditions and the probability of shutdown, which traditional linear models cannot accurately capture.

[0007] 4. Poor environmental adaptability: Existing methods do not consider the differentiated impact of different geographical locations and environmental conditions on downtime risk, and the model has insufficient generalization ability.

[0008] 5. Conflict between real-time performance and accuracy: Existing systems often sacrifice prediction accuracy in pursuit of real-time performance, making it difficult to balance the two.

[0009] Therefore, there is an urgent need for a new energy unit icing shutdown prediction method that can accurately identify multi-feature interaction relationships, scientifically set prediction thresholds, and has strong adaptability. Summary of the Invention

[0010] This invention provides a method and system for predicting icing shutdown of new energy generating units based on multi-feature interactive thresholds, which solves the technical problems of low prediction accuracy, unscientific threshold setting, and poor environmental adaptability in the prior art, and realizes accurate prediction and scientific judgment of the risk of icing shutdown of new energy generating units.

[0011] The technical solution adopted in this invention is as follows:

[0012] A method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds includes the following steps:

[0013] S1. Collect meteorological characteristic data, operational data, and geographical location information of new energy units, and preprocess them to form a standardized dataset;

[0014] S2. Based on meteorological and geographical characteristics, clustering algorithms are used to divide new energy generating units into different environmental type groups, and corresponding clustering labels are generated for each new energy generating unit;

[0015] S3. Construct a fully connected deep neural network shutdown prediction model, using data and cluster labels in a standardized dataset as input features, and train the model to output the probability of unit shutdown due to icing.

[0016] S4. For a single feature, by fixing other features and changing the target feature alone, draw the response curve and determine the threshold range of the single feature;

[0017] S5. Select key feature pairs for dual-feature interaction analysis, construct a three-dimensional decision boundary, quantify the interaction effect, and extract multi-feature interaction rules based on the decision tree algorithm;

[0018] S6. Construct a comprehensive discrimination rule by combining the shutdown probability, single feature threshold range, and multi-feature interaction rules, set risk preference parameters and perform adaptive threshold updates, and finally output the shutdown prediction result.

[0019] Furthermore, the meteorological characteristic data includes the station's highest temperature, lowest temperature, average humidity, average wind speed, precipitation type, and precipitation amount; the operational data includes actual power, theoretical power, and operational status; and the geographical location information includes longitude, latitude, and altitude.

[0020] Furthermore, the clustering algorithm is the K-means algorithm, which uses Euclidean distance as the similarity metric for K-means clustering, and the number of clusters k is determined by the elbow rule.

[0021] Furthermore, the fully connected deep neural network shutdown prediction model includes an input layer, 3-5 hidden layers, and an output layer; the input features of the input layer include basic meteorological features, cluster labels, power difference features, and time features; the hidden layers use the ReLU activation function; and the output layer uses the Sigmoid activation function to output the shutdown probability.

[0022] Furthermore, step S4 specifically includes:

[0023] For the target feature to be analyzed, all other features are fixed to their corresponding mean values. The target feature is then assigned values ​​at equal intervals within its range. Each feature value is substituted into the trained fully connected deep neural network shutdown prediction model to obtain the corresponding shutdown probability prediction value. A feature-probability response curve is plotted, and the feature value interval where the shutdown probability exceeds a preset threshold is taken as the single feature threshold interval.

[0024] Furthermore, step S5 specifically includes:

[0025] For each selected feature pair (F1, F2), points are taken at equal intervals within the value range of feature F1 and the value range of feature F2, and the value points of the two features are combined to form a grid. The shutdown probability corresponding to each grid point is calculated, and a three-dimensional decision boundary between the feature pair and the shutdown probability is constructed.

[0026] The impact of the interaction effect between two features is quantified by the shutdown probability when feature pair (F1, F2) works together, the shutdown probability when feature F1 works alone, the shutdown probability when feature F2 works alone, and the shutdown probability when both features are means.

[0027] Threshold rules for multi-feature interactions are extracted based on the decision tree algorithm, including: when conditions C1 and C2 are met simultaneously, the probability of shutdown exceeds the threshold. When conditions C3 or C4 are met, the probability of shutdown exceeds the threshold. Where C1, C2, C3, and C4 are conditions based on eigenvalues. , This represents the corresponding downtime probability threshold.

[0028] Furthermore, the comprehensive discrimination rule is constructed by combining the overall shutdown probability, single feature threshold range, and multi-feature interaction rules, specifically as follows:

[0029] ,

[0030] in, This indicates the comprehensive discrimination rule. This represents the single-feature threshold discrimination rule. This represents the multi-feature interaction threshold discrimination rule. This represents the shutdown probability output by the fully connected deep neural network shutdown prediction model. Let be the weighting coefficient, satisfying .

[0031] Furthermore, the setting of risk preference parameters and adaptive threshold updates specifically includes:

[0032] (1) Setting risk preference parameters , The sensitivity of the threshold is adjusted by combining the base threshold and the standard deviation of the threshold to adapt to different risk requirements. The adjustment formula is as follows:

[0033] ,

[0034] in, The adjusted threshold, Based on the threshold, The threshold standard deviation;

[0035] (2) Set the target accuracy of the model prediction and introduce the learning rate. The threshold is dynamically updated based on the difference between the real-time prediction accuracy and the target accuracy, using the following formula:

[0036] ,

[0037] in, For the updated threshold, The current threshold, For the target accuracy, To improve real-time prediction accuracy.

[0038] Furthermore, the output shutdown prediction results include: the probability value of shutdown due to icing of new energy units, the shutdown risk level, the key meteorological or operational characteristics affecting the shutdown due to icing of units, and the response recommendations for different risk levels.

[0039] A new energy unit icing shutdown prediction system based on multi-feature interaction thresholds, used to implement the above method, includes:

[0040] Data processing module: Used to collect meteorological characteristic data, operational data and geographical location information of new energy units, and to preprocess them to form a standardized dataset;

[0041] Feature clustering module: Based on meteorological and geographical features, clustering algorithms are used to divide new energy generating units into different environmental type groups, and generate corresponding clustering labels for each new energy generating unit;

[0042] Model training module: used to build a fully connected deep neural network shutdown prediction model, using data and cluster labels in the standardized dataset as input features, and train the model to output the probability of unit shutdown due to icing.

[0043] Single Feature Analysis Module: Used to plot response curves and determine single feature threshold ranges by fixing other features and changing the target feature alone for a single feature;

[0044] Multi-feature analysis module: used to select key feature pairs for dual-feature interaction analysis, construct a three-dimensional decision boundary, quantify interaction effects, and extract multi-feature interaction rules based on decision tree algorithm;

[0045] The comprehensive prediction module constructs a comprehensive discrimination rule by combining the downtime probability, single feature threshold range, and multi-feature interaction rules, sets risk preference parameters and performs adaptive threshold updates, and finally outputs the downtime prediction result.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] Compared with the prior art, the present invention has the following significant advantages:

[0048] 1) Significantly improved prediction accuracy: Through multi-feature interaction analysis and deep neural network modeling, more reliable decision-making basis is provided for power grid dispatch.

[0049] 2) Scientific threshold setting: The critical threshold is determined based on response curve analysis and data-driven methods, which overcomes the subjectivity and inaccuracy of traditional experience-based setting and improves the scientificity and reliability of risk identification.

[0050] 3) Strong ability to handle complex relationships: It can accurately identify and process the nonlinear interaction relationships between multiple meteorological features, especially the complex coupling effects under extreme weather conditions, and the prediction reliability is significantly improved from 79% of the traditional method to more than 85%.

[0051] 4) Excellent environmental adaptability: By grouping different environmental conditions through cluster analysis, the model can adapt to different geographical locations and climatic conditions, with strong generalization ability and wide applicability.

[0052] 5) Wide range of applications: It is applicable to different types of new energy equipment such as wind power and photovoltaics, and can be promoted to new energy power plants with different climate regions and installed capacity. Attached Figure Description

[0053] Figure 1 This is a visualization of sample similarity during the clustering process in an embodiment of the present invention.

[0054] Figure 2 This is a clustering feature distribution diagram in an embodiment of the present invention.

[0055] Figure 3 This is a structural diagram of a fully connected deep neural network shutdown prediction model in an embodiment of the present invention.

[0056] Figure 4 This is a visualization of the single-feature response curve in an embodiment of the present invention.

[0057] Figure 5 This is a visualization of the impact of 3D feature interaction on the downtime probability in an embodiment of the present invention.

[0058] Figure 6 This is a visualization of the multi-feature interaction effect in an embodiment of the present invention. Detailed Implementation

[0059] The technical solution of this invention will be further described clearly and in detail below with reference to the accompanying drawings and specific examples. This invention provides a method and system for predicting icing shutdown of new energy power units based on multi-feature interactive thresholds. It abandons single-threshold judgment and comprehensively integrates multi-dimensional meteorological factors such as temperature, humidity, wind speed, air pressure, and precipitation patterns, as well as environmental elements. It deeply analyzes and quantifies the complex nonlinear interaction mechanisms between various features, constructing a prediction model with high information capacity. Relying on massive historical icing event data and corresponding meteorological parameters, and based on a fully connected deep neural network model, it performs adaptive, multi-level dynamic threshold optimization settings to ensure that the thresholds have a solid statistical basis and theoretical interpretability, significantly reducing the false alarm rate and false negative rate. While ensuring high accuracy of the prediction model, it can meet the low latency and high concurrency processing requirements of practical engineering applications, effectively solving the contradiction between real-time performance and accuracy. This invention is applicable to icing shutdown prediction of wind turbine units and photovoltaic equipment, focusing on blade icing for wind turbine units and snow accumulation on photovoltaic modules.

[0060] The method of the present invention includes the following steps:

[0061] S1. Collect meteorological characteristic data, operational data, and geographic location information of the new energy generating units, and preprocess them to form a standardized dataset. Among them, the meteorological characteristic data includes the station's highest temperature, lowest temperature, average humidity, average wind speed, precipitation type, and precipitation amount; the operational data includes actual power, theoretical power, and operating status; and the geographic location information includes longitude, latitude, and altitude.

[0062] S2. Based on meteorological and geographical characteristics, a clustering algorithm is used to divide the new energy generating units into different environmental type groups, and a corresponding cluster label is generated for each new energy generating unit. The clustering algorithm can be the K-means algorithm, using Euclidean distance as the similarity metric for K-means clustering, and the number of clusters k is determined by the elbow rule.

[0063] S3. Construct a fully connected deep neural network outage prediction model, using data and cluster labels from a standardized dataset as input features, and train the model to output the outage probability of the unit due to icing. The fully connected deep neural network outage prediction model includes an input layer, 3-5 hidden layers, and an output layer; the input features of the input layer include basic meteorological features, cluster labels, power difference features, and time features; the hidden layers use the ReLU activation function; the output layer uses the Sigmoid activation function and outputs the outage probability.

[0064] S4. For a single feature, by fixing other features and changing only the target feature, plot the response curve and determine the threshold range for that single feature. Specifically, this includes:

[0065] For the target feature to be analyzed, all other features are fixed to their corresponding mean values. The target feature is then assigned values ​​at equal intervals within its range. Each feature value is substituted into the trained fully connected deep neural network shutdown prediction model to obtain the corresponding shutdown probability prediction value. A feature-probability response curve is plotted, and the feature value interval where the shutdown probability exceeds the preset shutdown probability threshold (≥0.5) is taken as the single feature threshold interval.

[0066] S5. Select key feature pairs for dual-feature interaction analysis, construct a three-dimensional decision boundary, quantify interaction effects, and extract multi-feature interaction rules based on the decision tree algorithm. Specifically, this includes:

[0067] For each selected feature pair (F1, F2), points are taken at equal intervals within the value range of feature F1 and the value range of feature F2, and the value points of the two features are combined to form a grid. The shutdown probability corresponding to each grid point is calculated, and a three-dimensional decision boundary between the feature pair and the shutdown probability is constructed.

[0068] The impact of the interaction effect between two features is quantified by the shutdown probability when feature pair (F1, F2) works together, the shutdown probability when feature F1 works alone, the shutdown probability when feature F2 works alone, and the shutdown probability when both features are means.

[0069] Threshold rules for multi-feature interactions are extracted based on the decision tree algorithm, including: when conditions C1 and C2 are met simultaneously, the probability of shutdown exceeds the threshold. When conditions C3 or C4 are met, the probability of shutdown exceeds the threshold. Where C1, C2, C3, and C4 are conditions based on eigenvalues. , This represents the corresponding downtime probability threshold.

[0070] S6. Construct a comprehensive discrimination rule by combining the shutdown probability, single feature threshold range, and multi-feature interaction rules, set risk preference parameters and perform adaptive threshold updates, and finally output the shutdown prediction result.

[0071] Construct comprehensive discrimination rules, specifically as follows:

[0072] ,

[0073] in, This indicates the comprehensive discrimination rule. This represents the single-feature threshold discrimination rule. This represents the multi-feature interaction threshold discrimination rule. This represents the shutdown probability output by the fully connected deep neural network shutdown prediction model. Let be the weighting coefficient, satisfying .

[0074] Setting risk preference parameters and performing adaptive threshold updates specifically includes:

[0075] (1) Setting risk preference parameters , The sensitivity of the threshold is adjusted by combining the base threshold and the standard deviation of the threshold to adapt to different risk requirements. The adjustment formula is as follows:

[0076] ,

[0077] in, The adjusted threshold, Based on the threshold, The threshold standard deviation;

[0078] (2) Set the target accuracy of the model prediction and introduce the learning rate. The threshold is dynamically updated based on the difference between the real-time prediction accuracy and the target accuracy, using the following formula:

[0079] ,

[0080] in, For the updated threshold, The current threshold, For the target accuracy, To improve real-time prediction accuracy.

[0081] The output outage prediction results include: the probability value of icing outage of new energy units, the outage risk level, the key meteorological or operational characteristics affecting the icing outage of units, and the response measures recommended for different risk levels.

[0082] Example

[0083] In this embodiment, a method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds includes the following steps:

[0084] Step S1: Preprocessing of multidimensional meteorological feature data

[0085] S1.1 Data Collection: Collect operating data of new energy units, including but not limited to:

[0086] Meteorological characteristic data: maximum temperature T_max, minimum temperature T_min, average humidity H_avg, average wind speed V_avg, precipitation type, precipitation amount, etc.; Unit operation data: actual power P_real, theoretical power P_theory, operating status Status, etc.; Geographical location information: longitude, latitude, altitude, etc.

[0087] S1.2 Data Cleaning: Identify and handle outliers, using 3... The criteria include removing obviously outlier data; handling missing values ​​by using linear interpolation, mean imputation, or imputation based on similar samples; and standardizing the data format by unifying latitude and longitude coordinates into decimal format.

[0088] S1.3 Data Classification and Organization: Establish independent datasets for wind power data and photovoltaic data respectively; index and organize the data according to time series and geographical location.

[0089] Step S2: Clustering and grouping based on similarity of meteorological and geographical conditions

[0090] S2.1 Feature Standardization: Perform Z-score standardization on meteorological features: , where Z is the standardized eigenvalue, X is the original eigenvalue, μ is the eigenmean, and σ is the eigenstandard deviation.

[0091] S2.2 Similarity Measurement: Euclidean distance is used as the similarity measurement standard. ,in, This represents the Euclidean distance between two samples, where A and B are the feature vectors of the two samples, and n is the feature dimension. See similarity visualization below. Figure 1 .

[0092] S2.3 K-means Clustering Analysis: Clustering is performed based on features such as the highest and lowest temperatures, average humidity, and average wind speed at the site; the elbow rule is used to determine the optimal number of clusters k; the K-means algorithm is executed, with the objective function being: ,in, For the i-th cluster, It is the center of the i-th cluster.

[0093] S2.4 Cluster Label Generation: The clustering results are used as cluster labels (Cluster_Label) as auxiliary features for subsequent modeling. The feature distribution is as follows: Figure 2 As shown.

[0094] Step S3: Construct a fully connected deep neural network shutdown prediction model

[0095] S3.1 Network Architecture Design: Construct a multi-layer fully connected neural network, including: Input layer: receives multi-dimensional feature vectors; Hidden layers: 3-5 fully connected layers, with the number of neurons being 64, 128, 256, 128, and 64 respectively; Output layer: 1 neuron, outputting the halting probability. The model structure is as follows: Figure 3 As shown.

[0096] S3.2 Input Feature Construction: The input feature vector includes: basic meteorological features: T_max, T_min, H_avg, V_avg; cluster label: Cluster_Label; power difference features: Time characteristics: hour, date, season, etc.

[0097] S3.3 Activation Function Selection: The hidden layer uses the ReLU activation function. The output layer uses the Sigmoid activation function. .

[0098] S3.4 Model Training: The loss function uses binary cross-entropy. The optimizer uses the Adam algorithm with a learning rate of 0.001; an early stopping mechanism is used to prevent overfitting.

[0099] Step S4: Single Feature Threshold Response Curve Analysis

[0100] S4.1 Feature Fixation Strategy: For target features Fix all other features to their corresponding means: .

[0101] S4.2 Target Feature Variation: The target feature is varied within its value range. Divide the data into equal intervals and take values: Where n is the number of division points, usually taken as 100. For the sampled value of the i-th target feature, These are the minimum and maximum values ​​of the target feature.

[0102] S4.3 Response Curve Plotting:

[0103] Each Substitute the values ​​into the trained fully connected deep neural network shutdown prediction model; obtain the corresponding shutdown probability prediction values. ;draw and The response curve.

[0104] S4.4 Single Feature Threshold Determination: Finding the Downtime Probability The eigenvalue range: For visualization effects, see Figure 4 .

[0105] Step S5: Multi-feature interaction threshold determination

[0106] S5.1 Dual Feature Interaction Analysis: Select key feature pairs for interaction analysis, including but not limited to:

[0107] (1) Maximum temperature at the station × Average humidity at the station

[0108] (2) Maximum temperature at the station × Average wind speed at the station

[0109] (3) Minimum temperature at the station × Average wind speed at the station

[0110] (4) Average humidity at the site × Average wind speed at the site

[0111] S5.2 3D Decision Boundary Construction: For feature pairs (F1, F2), construct a mesh:

[0112] ,

[0113] in, These are two features involved in interactive analysis. , These are the minimum and maximum values ​​of feature F1. , These are the minimum and maximum values ​​of feature F2. m and n are the number of grids used to refine the two-dimensional feature space. The impact of three-dimensional feature interactions on the halting probability is as follows: Figure 5 As shown.

[0114] S5.3 Quantification of interaction effects, calculating the impact of characteristic interaction terms:

[0115] .

[0116] in, This refers to the additional effect of the combined effect of two features relative to the independent effect of a single feature. This represents the probability of downtime. Features The mean,

[0117] S5.4 Multi-feature threshold rule extraction: Extracting multi-feature interaction rules based on decision tree algorithm: When condition C1 AND C2, the probability of halting. When condition C3 OR C4, the probability of shutdown is... in, As a characteristic condition, The corresponding threshold. See the visualization of the decision tree analysis matrix for feature interactions. Figure 6 .

[0118] Step S6: Fusion discrimination threshold optimization strategy

[0119] S6.1 Construction of Comprehensive Discrimination Rules: Combining single-feature thresholds and multi-feature interaction thresholds, a comprehensive discrimination rule is established: in, Let be the weighting coefficient, satisfying .

[0120] S6.2 Risk Preference Parameter Setting: Introduce the risk preference parameter λ to adjust threshold sensitivity: in, , The threshold standard deviation.

[0121] S6.3 Adaptive Threshold Update: Dynamically adjusts threshold parameters based on feedback from prediction results. in, This is the learning rate.

[0122] S6.4 Final Prediction Output: Output the downtime prediction results, including: downtime probability value, risk level (high, medium, low), key influencing factors, and recommended measures.

[0123] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0125] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0126] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0127] The above description is merely a preferred embodiment of the present invention. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the technical solutions of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall still fall within the protection scope of the technical solutions of the present invention.

Claims

1. A method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds, characterized in that, Includes the following steps: S1. Collect meteorological characteristic data, operational data, and geographical location information of new energy units, and preprocess them to form a standardized dataset; S2. Based on meteorological and geographical characteristics, clustering algorithms are used to divide new energy generating units into different environmental type groups, and corresponding clustering labels are generated for each new energy generating unit; S3. Construct a fully connected deep neural network shutdown prediction model, using data and cluster labels in a standardized dataset as input features, and train the model to output the probability of unit shutdown due to icing. S4. For a single feature, by fixing other features and changing the target feature alone, draw the response curve and determine the threshold range of the single feature; S5. Select key feature pairs for dual-feature interaction analysis, construct a three-dimensional decision boundary, quantify the interaction effect, and extract multi-feature interaction rules based on the decision tree algorithm; S6. Construct a comprehensive discrimination rule by combining the shutdown probability, single feature threshold range, and multi-feature interaction rules, set risk preference parameters and perform adaptive threshold updates, and finally output the shutdown prediction result.

2. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, The meteorological characteristic data includes the station's highest temperature, lowest temperature, average humidity, average wind speed, precipitation type, and precipitation amount. The operational data includes actual power, theoretical power, and operational status. The geographical location information includes longitude, latitude, and altitude.

3. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, The clustering algorithm described is the K-means algorithm, which uses Euclidean distance as the similarity metric for K-means clustering, and the number of clusters k is determined by the elbow rule.

4. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, The fully connected deep neural network shutdown prediction model includes an input layer, 3-5 hidden layers, and an output layer. The input features of the input layer include basic meteorological features, cluster labels, power difference features, and time features. The hidden layers use the ReLU activation function. The output layer uses the Sigmoid activation function to output the shutdown probability.

5. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, Step S4 specifically includes: For the target feature to be analyzed, all other features are fixed to their corresponding mean values. The target feature is then assigned values ​​at equal intervals within its range. Each feature value is substituted into the trained fully connected deep neural network shutdown prediction model to obtain the corresponding shutdown probability prediction value. A feature-probability response curve is plotted, and the feature value interval where the shutdown probability exceeds a preset threshold is taken as the single feature threshold interval.

6. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, Step S5 specifically includes: For each selected feature pair (F1, F2), points are taken at equal intervals within the value range of feature F1 and the value range of feature F2, and the value points of the two features are combined to form a grid. The shutdown probability corresponding to each grid point is calculated, and a three-dimensional decision boundary between the feature pair and the shutdown probability is constructed. The impact of the interaction effect between two features is quantified by the shutdown probability when feature pair (F1, F2) works together, the shutdown probability when feature F1 works alone, the shutdown probability when feature F2 works alone, and the shutdown probability when both features are means. Threshold rules for multi-feature interactions are extracted based on the decision tree algorithm, including: when conditions C1 and C2 are met simultaneously, the probability of shutdown exceeds the threshold. When conditions C3 or C4 are met, the probability of shutdown exceeds the threshold. Where C1, C2, C3, and C4 are conditions based on eigenvalues. , This represents the corresponding downtime probability threshold.

7. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, The comprehensive discrimination rule is constructed by combining the overall shutdown probability, single feature threshold range, and multi-feature interaction rules, specifically as follows: ， in, This indicates the comprehensive discrimination rule. This represents the single-feature threshold discrimination rule. This represents the multi-feature interaction threshold discrimination rule. This represents the shutdown probability output by the fully connected deep neural network shutdown prediction model. For the weighting coefficients, satisfying .

8. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, The setting of risk preference parameters and adaptive threshold updates specifically includes: (1) Setting risk preference parameters , The sensitivity of the threshold is adjusted by combining the base threshold and the standard deviation of the threshold to adapt to different risk requirements. The adjustment formula is as follows: ， in, The adjusted threshold, Based on the threshold, The threshold standard deviation; (2) Set the target accuracy of the model prediction and introduce the learning rate. The threshold is dynamically updated based on the difference between the real-time prediction accuracy and the target accuracy, using the following formula: ， in, For the updated threshold, The current threshold, For the target accuracy, To improve real-time prediction accuracy.

9. The method for predicting icing shutdown of new energy generating units based on multi-feature interaction thresholds according to claim 1, characterized in that, The output outage prediction results include: the probability value of icing outage of new energy units, the outage risk level, the key meteorological or operational characteristics affecting the icing outage of units, and the response measures recommended for different risk levels.

10. A new energy unit icing shutdown prediction system based on multi-feature interactive thresholds, characterized in that, For implementing the method as described in any one of claims 1-9, comprising: Data processing module: Used to collect meteorological characteristic data, operational data and geographical location information of new energy units, and to preprocess them to form a standardized dataset; Feature clustering module: Based on meteorological and geographical features, clustering algorithms are used to divide new energy generating units into different environmental type groups, and generate corresponding clustering labels for each new energy generating unit; Model training module: used to build a fully connected deep neural network shutdown prediction model, using data and cluster labels in the standardized dataset as input features, and train the model to output the probability of unit shutdown due to icing. Single Feature Analysis Module: Used to plot response curves and determine single feature threshold ranges by fixing other features and changing the target feature alone for a single feature; Multi-feature analysis module: used to select key feature pairs for dual-feature interaction analysis, construct a three-dimensional decision boundary, quantify interaction effects, and extract multi-feature interaction rules based on decision tree algorithm; The comprehensive prediction module constructs a comprehensive discrimination rule by combining the downtime probability, single feature threshold range, and multi-feature interaction rules, sets risk preference parameters and performs adaptive threshold updates, and finally outputs the downtime prediction result.

Images