55 results about "Maximal information coefficient" patented technology

The invention discloses a method for detecting network traffic anomaly based on feature selection and density peak clustering. The method comprises the following stages: a stage of acquiring the traffic: monitoring a network through a network analysis tool, and acquiring monitored data packets in the local; a stage of extracting features: extracting the data packets belonging to the same stream from the data packets, performing feature extraction of the data packets, and normalizing the extracted features; a stage of selecting the features: evaluating the importance of each feature on classification decision by utilizing a maximal information coefficient, simply clustering the features according to the redundancy among the features, selecting one feature having the highest importance, and adding the feature having the highest importance into a feature sub-set; and a stage of clustering and analyzing: clustering the features by adopting an improved clustering method based on a density peak so as to obtain clusters in a plurality of traffic types, performing little sampling of the cluster in each traffic type, performing class detection, and covering the traffic types of the clusters in the whole traffic types by utilizing the modal classified traffic types in a sampled sample, such that the anomaly traffic can be detected.

The invention relates to a feature selection and classification method based on a maximum information coefficient and a feature selection and classification device based on the maximum information coefficient. The feature selection method comprises the following steps: S1, sequencing primitive features based on mutual information criterions, deleting features with correlation degree lower than a threshold value, and forming a primary feature subset by use of features with correlation degree higher than the threshold value; S2, calculating the maximum information coefficient among the features in the primary feature subset; S3, deleting redundancy features in the primitive feature subset according to the maximum information coefficient, thereby obtaining a low-dimension feature subset. According to the feature selection method, feature selection is performed by virtue of a mode of using mutual information and the maximum information coefficient, so that the redundancy features are removed, and the data dimension is reduced.

The invention discloses an electricity stealing detection method and device of using an integrated entropy method and a density clustering method, wherein the detection method comprises the followingsteps of: acquiring user intelligent meter data and total energy consumption data in a period of time to be detected in a target area to obtain a user power consumption curve set and non-technical loss data; processing the user power consumption curve set and non-technical loss data according to a maximum information coefficient calculation method, and obtaining a power consumption correlation measurement index; obtaining a shape anomaly index of each curve; according to the measurement index and the shape anomaly index, dividing the power consumption curve of each user into an abnormal curveand a normal curve through the k-Mean value clustering, and obtaining an abnormal index of power quantity and an abnormal index of shape; and then obtaining the power quantity anomaly rank and shape anomaly rank of the users to obtain the comprehensive anomaly rank. The method can improve the accuracy of detection in various power stealing modes, realize the unsupervised detection of the power stealing behavior under large data, and enlarge the application scope.

The invention relates to a multi-model integrated load prediction method based on wavelet transform, which is divided into four stages: 1, on the basis of considering multiple influence factors, historical load related data is subjected to a maximum information coefficient feature selection technology to obtain a feature candidate set with high correlation; 2, in order to obtain a stable load sequence and improve the prediction precision, multiple wavelet transforms are integrated into a multi-prediction model; and 3, each load correlation sequence after wavelet function decomposition is trained by an intelligent prediction sub-model, and the sub-models provide different predictions in the same hour; 4, combining the optimal prediction of each time period and providing a final prediction result by adopting online secondary learning in an integrated decision process. According to the method, the load prediction precision can be further improved on the basis of various single predictionmodels. The method is high in generalization ability, can adapt to various environments, has high applicability, and is beneficial to reducing the operation cost of a power system.

The invention belongs to the technical field of wind speed prediction, and discloses a wind speed prediction method and system based on a neighborhood gate long-short-term memory network. The Pearsoncorrelation coefficient and the maximum information coefficient are respectively adopted to explore the linear and nonlinear correlation among variables to screen the wind speed correlation factors. On the basis of correlation analysis, the Granger causality test is used to explore the statistical causality of wind speed and wind speed factors. The structure of causality is divided into five types, and all types of causality are unified into an equivalent tree causality structure by the method of 'decomposition-dummy variable-pruning'. Aiming at the causality structure of equivalent tree, a long-term and short-term memory network model based on neighborhood gates is proposed to predict wind speed. The prediction method (NLSTM) of the invention accurately considers the causality between thewind speed and the wind speed factors, effectively improves the prediction accuracy of the wind speed, and plays a vital role in the application of the wind power and the dispatching of the power grid.

A method to identify a relationship between at least pairs of variables in a dataset generates a novel measure of dependence, referred to as a Maximal Information Coefficient (MIC), that captures a wide range of interesting associations (not limited to specific function types such as linear, exponential or periodic, or to all functional relationships), and that gives similar scores to equally noisy relationships of different types. MIC is based on the idea that if a relationship exists between two variables, then a grid can be drawn on a scatterplot of the two variables that partitions the data to encapsulate that relationship.

The invention belongs to the technical field of runoff prediction, and discloses a runoff probability prediction method and system based on deep learning, and the method comprises the steps: employinga maximum information coefficient to analyze the linear and nonlinear correlation between variables, so as to screen a runoff correlation factor; building an extreme gradient boosting tree model on the basis of correlation analysis, and inputting runoff correlation factors into a trained XGB model to complete runoff point prediction; inputting a point prediction result obtained by the XGB model into a GPR model, and performing secondary prediction to obtain a runoff probability prediction result; selecting confidence and acquiring a runoff interval prediction result under the corresponding confidence through Gaussian distribution; and optimizing hyper-parameters in the XGB model and the GPR model by adopting a Bayesian optimization algorithm. A high-precision runoff point prediction result, an appropriate runoff prediction interval and reliable runoff probability prediction distribution can be obtained, and the prediction method plays a crucial role in utilization of water resourcesand reservoir scheduling.

The invention discloses a protein subnuclear localization method for feature extraction and fusion based on an improved PSSM, and relates to the field of biological and information technologies. The method comprises the steps: firstly, denormalizing a position-specific scoring matrix for the evolution information of a protein sequence through using a Z-SoftMax function; secondly, respectively performing the feature extraction of the position-specific scoring matrix in different directions at different hopping intervals through using a proposed SC-PSSM-C and SC-PSSM-R, and fixing the length ofthe PSSM; thirdly selecting the features of the fused features through an improved maximum information coefficient algorithm; finally, performing the final classification prediction through a W-SVM classifier with optimized parameters. The method has the function of overcoming limitations and unicity of the conventional feature extraction, and improves the capability of protein subnuclear localization.

The invention discloses a wind generation set fault detection method and device based on a quick gradient elevator. Target state features are selected from a state feature set based on a maximum information coefficient correlation analysis method, and a fault detection model is obtained according to a cost sensitive quick gradient elevator, so that in the wind generation set fault detection process, the calculation amount is reduced, consumed calculation time is shortened, and the fault detection efficiency and real-time performance of a wind generation set are improved. The method comprises the steps that the state feature set of the wind generation set is obtained, wherein the state feature set comprises at least one state feature; target state features are selected from the state feature set based on the maximum information coefficient correlation analysis method; the fault detection model is obtained according to the cost sensitive quick gradient elevator; and according to the target state features and the fault detection model, a fault detection result of the wind generation set is obtained through prediction.

The invention relates to the field of intelligent water affairs, and discloses a residential water consumption prediction method based on an MIC-XGBoost algorithm, and the method comprises the steps: obtaining influence factor values and water consumption of each month, constructing a water consumption-influence factor value corresponding table of water consumption influence factors, and employing a maximum information coefficient algorithm to obtain a water consumption-influence factor value corresponding table of the water consumption influence factors; calculating a target influence factor which has the maximum influence on the water consumption, training the original XGBoost algorithm model according to the water consumption and the influence factor numerical value of each month to obtain a target XGBoost algorithm model, and predicting the water consumption by utilizing the target XGBoost algorithm model according to the influence factor numerical value of the current month to obtain the predicted water consumption of the current month. The invention further provides a resident water consumption prediction device based on the MIC-XGBoost algorithm, electronic equipment and a computer readable storage medium. According to the method, the problem that the prediction accuracy is low when the water consumption of residents is predicted by only depending on the XGBoost algorithm can be solved.

The invention provides construction of a loss prediction model combining an echo state network and a support vector machine. Aiming at a non-ideal single model prediction effect, a classifier construction method is explored and improved, and the construction of the classifier can be divided into three aspects of a training set construction strategy, a classifier selection strategy and a predictionvalue determination strategy, comprehensively evaluating features and categories by fusing the Mahalanobis distance and the maximum information coefficient, and jointly measuring the redundancy between the features and the correlation degree between the features and the categories. The advantages of an echo state network and a support vector machine are integrated, and the problem of secondary optimization is solved through linear constraint, so that the performance of a telecom customer loss system is improved, and the prediction effect of a customer loss prediction model is improved.

A method to identify a relationship between at least pairs of variables in a dataset generates a novel measure of dependence, referred to as a Maximal Information Coefficient (MIC), that captures a wide range of interesting associations (not limited to specific function types such as linear, exponential or periodic, or to all functional relationships), and that gives similar scores to equally noisy relationships of different types. MIC is based on the idea that if a relationship exists between two variables, then a grid can be drawn on a scatterplot of the two variables that partitions the data to encapsulate that relationship.

The embodiment of the invention provides a hydroelectric generating set vibration trend prediction method and system. The method comprises the steps: collecting an original signal feature set of a hydroelectric generating set; analyzing the correlation between each intrinsic mode component and the environment variable based on a maximum information coefficient (MIC), and extracting the environment variable of which the correlation is greater than a preset correlation threshold as the environment feature of each intrinsic mode component; forming a feature input sequence by each intrinsic mode component and the corresponding environment feature, taking the feature input sequence as the input of a neural network according to a time sequence, and performing neural network training to obtain a vibration trend prediction model; and enabling each intrinsic mode component of the current hydroelectric generating set to be predicted and the corresponding environmental characteristics to form a characteristic input vector, inputting the characteristic input vector into the vibration trend prediction model for prediction, and fusing all prediction results to obtain a future trend prediction value of the hydroelectric generating set, so that the precision and stability of the vibration trend prediction result are greatly improved.

The invention discloses a high-dimensional data feature selection method based on a filtering method and a genetic algorithm. Traditional feature selection methods are not suitable for high-dimensional and small-sample data because they have the limitation of high probability of falling into local optimization and deleting useful features. According to the invention, firstly, the maximum information coefficient is adopted to calculate the correlation between features of input data and class marks; then, the features are sorted in descending order according to values of the correlation, a threshold value is set, and features with weak correlation are deleted; and finally, the remaining features with strong correlation are subjected to random searching optimization by the genetic algorithm to obtain an optimal feature subset. The invention can effectively carry out feature selection on high-dimensional data and realize dimension reduction. The result of feature selection has important significance for sample class judgment, and when the method is applied to gene expression profile data, the selected features also have important biological significance.

The invention relates to a short-term photovoltaic output probability prediction method based on a simplest gated neural network, and the method comprises the following steps: 1), carrying out the normalization of original data containing a plurality of to-be-selected weather variables, and carrying out the reduction of the dimension of the original data through employing a maximum information coefficient MIC; 2) dividing the reduced feature data set into a training data set and a test data set, and respectively dividing the training data set and the test data set into four weather type data of sunny days, cloudy days, cloudy days and rainy days by adopting a K-means algorithm; 3) constructing a neural network quantile regression model and performing training by adopting the training dataset; and 4) performing prediction by adopting the trained neural network quantile regression model to obtain quantiles under various conditions, and obtaining an approximately complete probability density function through kernel density estimation. Compared with the prior art, the invention has the advantages that the prediction reliability and precision are improved, the prediction interval is narrower, the coverage rate is higher, and the method is simple and rapid.

The invention relates to an electric power CPS information attack identification method based on a stack type self-encoding network model, which is characterized by comprising the following steps: introducing a maximum information coefficient to select data characteristics according to properties such as CPS data non-function dependence and non-linear correlation, and determining an optimal attackcharacteristic set; constructing an information attack identification model based on a stack type self-encoding network, and setting an unsupervised pre-training encoder and a supervised fine tuningclassifier to perform network parameter training updating; model initial parameter optimization based on the adaptive cuckoo algorithm is realized. The problems of complex data characteristics, relatively low identification precision and the like in the power CPS information attack identification process are solved, and the method has the advantages of being scientific, reasonable, high in applicability, good in effect and the like.

The invention discloses a power load prediction method based on an improved long-term and short-term memory network. The method comprises steps of the characteristic of historical load being preliminarily screened by adopting the maximum information coefficient; the influence caused by load related factors being considered; and further screening the historical load by adopting a maximum correlation minimum redundancy algorithm, taking the screened feature set as the input of the model, carrying out power load prediction by adopting an improved long-short memory network, and verifying the obtained prediction result and the actual power grid load to prove the practicability of the model. According to the forecasting method (H-ILSTM) provided by the invention, the power load and related factors influencing the load are accurately considered, precision of power load forecasting is effectively improved, and safety and economy of power grid operation are improved to a certain extent.

The invention provides an electric power emergency repair light source evaluation method based on a grid acquisition method and a maximum information coefficient method. The evaluation method includes following steps: establishing an index system for comprehensive evaluation by employing the grid acquisition method and dividing the system into index hierarchies; calculating each index value, performing standardization processing on the index values, and obtaining dimensionless index data; simplifying the index system by employing the maximum information coefficient method, and screening important index items; for the screened important index items, determining index weights by employing a fuzzy Borda number analytic method; and calculating comprehensive performance evaluation scores of illuminating lamps according to the weights, and determining the to-be-repaired light source according to the comprehensive performance evaluation score. The evaluation method is applicable to the performance evaluation of the illuminating lamps employed in occasions including substations, emergency repair vehicles, and high-voltage circuit towers etc. in an emergency repair system, and the most appropriate illuminating equipment can be selected according to the special requirements of the illuminating equipment in the emergency repair system.

The invention discloses an ultra-short-term load prediction method and system considering data missing and feature redundancy, and the method comprises the steps of carrying out the processing of a data set with a data missing problem through a missing data completion algorithm based on an improved KNN; obtaining an optimal feature set of ultra-short-term load prediction through a package type feature selection method based on a maximum information coefficient MIC, and reducing the feature redundancy of multivariable time series data; adopting an S2SGRU ultra-short-term load prediction model,performing load prediction from sequence to sequence, improving the processing capacity of time sequence information, and therefore the precision of ultra-short-term load prediction is improved. According to the invention, the morphological similarity of the load data is considered, the optimal feature set can be effectively screened out, and the precision of ultra-short-term load prediction is improved. An S2SGRU ultra-short-term load prediction model is adopted to perform load prediction through sequences, so that the prediction capability of the algorithm on a long-time sequence model is further improved, and the limitation of the lengths of an input sequence and an output sequence in a traditional load prediction task is avoided.

The invention discloses a reservoir downstream water level prediction method based on a deep learning model. The method comprises the following steps: screening downstream water level related factorsby adopting a maximum information coefficient; optimizing by adopting a genetic algorithm on the basis of correlation analysis to obtain an optimal feature combination among single correlation factors; constructing a deep learning model (CNNLSTM) based on a convolutional neural network and a long-term and short-term memory network by taking the optimal feature combination of the downstream water level related factors as input; and training a CNNLSTM model weight variable by adopting an Adam gradient optimization algorithm, and taking the trained CNNLSTM as a reservoir downstream water level prediction model. According to the prediction method, the correlation factors of the downstream water level are finely considered, the feature combination of the correlation factors is optimized, the deep learning prediction model is adopted, the prediction precision of the downstream water level is effectively improved, and the prediction method plays a crucial role in accurately calculating the generated output in reservoir dispatching.

The invention relates to a mobile user exit feature selection method based on an Fisher score and an approximate Markov blanket, and belongs to the field of data mining. Firstly, a Fisher criterion isutilized to reserve features with high classification capability, and irrelevant features and weak relevant features are eliminated; secondly, two measurement methods of a maximum information coefficient MIC and a symmetric uncertainty SU are fused, a correlation measurement standard MSCC is designed, and the MSCC standard is used for further removing irrelevant features; and finally, in combination with an MSCC measurement standard, redundant features in the Fisher candidate feature set are eliminated by utilizing an approximate Markov-Blanket judgment condition, and finally, an optimal feature subset with a relatively small dimension scale is obtained. According to the method, the exit features of the mobile users can be effectively selected, and the classification accuracy of the modelis improved.

The invention discloses a thermal power plant boiler NOx emission modeling variable selection method. Whether the accuracy and the generalization ability of a model are affected or not is judged through variable selection. Aiming at the problem of variable selection, the maximum information coefficient variable selection method is provided by combining with the maximum information coefficient on the basis of the mutual information variable selection method, the maximum information coefficient is introduced to improve the original variable selection method, and the effectiveness of the method is verified by using the experimental data set of the nominal model.

The invention discloses a power load characteristic extraction method and system, and the method comprises the following steps of 1, obtaining a power load curve, and processing the power load curve through the empirical mode decomposition, and obtaining a new data sequence h1 (t); step 2, acquiring a maximum information coefficient of the load characteristic X and each empirical mode decomposition component Y of the load time sequence; and step 3, carrying out correlation analysis on the load characteristic X and a target imf component according to the maximum correlation coefficient, step 4,obtaining different imf component final load feature subsets according to the new data sequence, superposing the different final load feature subsets, and obtaining a load feature set TBEST. According to the method and the system, the original power load time sequence is decomposed by using the empirical mode decomposition, and then the power load characteristics are screened by using the maximumcorrelation coefficient of the load characteristics and the load components, so that compared with the prior art, the precision of the power load characteristic prediction is greatly improved, and the calculated amount is effectively reduced.

The invention provides a control tower control effectiveness evaluation index association relationship analysis method, which comprises the following steps of: firstly, cleaning original data, removing abnormal values and repeated values, and filling missing values; selecting a part of calculable representative indexes of the air traffic control tower based on a process method, establishing a maximum interaction information method non-parametric model between the index pairs, and solving a maximum information coefficient (MIC) to determine the correlation strength between the indexes; and identifying the associated direction in an assisted manner according to a Spearman correlation coefficient method, and finally, obtaining the air traffic control tower index association relationship. By using the Spearman correlation coefficient method, the problem that the maximum nonparametric model cannot determine the correlation direction between the indexes is solved, the limitation and accuracyof the Spearman correlation coefficient method for judging the correlation degree between the nonlinear indexes are improved through the maximum nonparametric model, and a foundation is laid for improving the control efficiency of a control tower.

The invention discloses a fan output power prediction method, which comprises the following steps of: firstly, establishing a multivariate polynomial regression model based on a maximum information coefficient and multivariate polynomial regression; and then, determining a test set, and predicting the output power of the fan on the test set by using the multivariate polynomial regression model. In addition, the invention further provides a fan output power prediction system. According to the method and system, the maximum information coefficient is used for selecting the important variables influencing the output power of the fan in the process of establishing the multivariate polynomial regression model, multiple important variables are comprehensively considered, the problem that analysis is not comprehensive enough due to the fact that influence factors of the output power of the fan are analyzed manually is solved, and further, the relation between multivariable and the fan output power is modeled through a polynomial method, so that prediction of the fan output power has high accuracy and low model complexity.

The invention relates to the technical field of industry user load prediction, in particular to an industry user electric quantity prediction method based on mode extraction and error adjustment, and the method comprises the following operation steps: S1, employing a K-means clustering algorithm to extract a typical industry annual load curve; s2, analyzing the correlation degree between external macroscopic factors such as holidays and festivals, weather and the like and the electric quantity of the user to be predicted by using a maximum information coefficient method; s3, according to a typical industry load curve and external factors such as holidays and festivals, weather and the like, electric quantity prediction is carried out by utilizing a sparrow to search the optimized BP neural network; s4, a non-parameter estimation method is utilized, probability distribution of residual errors is obtained, a prediction result is adjusted, and a final electric quantity predicted value is obtained.According to the method, the maximum information coefficient method is adopted, the influence degree of external factors on the electric quantity is analyzed, universality and fairness are achieved, and the influence degree of different factors on the electric quantity can be described more accurately.

According to the equipment parameter monitoring evidence network construction method based on independence analysis and testing, different equipment parameters of target equipment are synchronously acquired by utilizing a plurality of sensors, and each equipment parameter is used as a node in the parameter monitoring evidence network of the target equipment. Wherein all the sensors synchronously sample according to the same time interval within the operation time of the target equipment, and monitoring data obtained by all the sensors through sampling within the operation time of the target equipment are used as node input data of all the nodes. And constructing an undirected evidence network structure model based on the maximum information coefficient among the nodes. And finally, on thebasis of the undirected evidence network model, determining the direction of the undirected edge according to the noise adding model, and constructing an equipment parameter monitoring evidence network corresponding to the target equipment. Historical data of multiple different equipment parameters of the target equipment are detected by using the sensor, and the network structure is constructed based on the historical data, so that the objectivity and the accuracy of the evidence network structure are improved.

The embodiment of the invention provides a pumped storage unit generator temperature early warning method and system, which are combined with the maximum information coefficient (MIC), can discover the nonlinear relationship between characteristics while ensuring low operand, and have good universality and fairness. An established Attention-TCN model has a self-adaptive function for the features, the Attention mechanism can dynamically allocate a weight for each sample in the training process, compared with a traditional method for determining the weight for the features, the method is more flexible and practical, the established model is more accurate in combination with the characteristics of few TCN parameters and good accuracy, and the probability of prediction error abnormity can be effectively reduced. Compared with a traditional method of taking an error maximum value, setting an alarm threshold value according to experience and the like, the method is more objective by analyzing skewness and kurtosis early warning of an error sequence, has good interpretability, gives consideration to sensitivity and reliability, is small in calculation amount, and guarantees the real-time performance of early warning.

The invention belongs to the technical field of data mining, and discloses a method and a system for calculating a maximum information coefficient between a single variable and multiple variables in abig data set, which are used for calculating the maximum information coefficient between the single variable Y and m variables (X1, X2,..., Xm). The method: searching optimal grid division of a multi-dimensional (m + 1) space; , fixing division of m variables (X1, X2,..., Xm) by utilizing a maximum information coefficient algorithm of a single variable and m-1 variables; dividing a variable Y, and finding out more suitable division for the variable Y and m variables (X1, X2,..., Xm); and calculating a normalized maximum mutual information value to serve as a maximum information coefficient value. The problem that the existing method for calculating the maximum information coefficient cannot be applied to the calculation of the maximum information coefficient between a single variable anda multivariable can be solved; according to the approximate processing method and system for the maximum information coefficient between the single variable and the multiple variables, the maximum information coefficient value between the single variable and the multiple variables can be calculated.

The invention belongs to the technical field of identification and evaluation of radiation source signals in electronic countermeasures, and discloses a radiation source signal identification efficiency evaluation method and system, an intelligent terminal and application, wherein the method comprises the steps of carrying out the incidence relation analysis of indexes through a maximum information coefficient analysis method, and obtaining an incidence relation matrix; analyzing the incidence relation matrix by adopting a rapid community detection method to obtain an optimal community division index; constructing a network evaluation model with constraint, wherein the network evaluation model adopts an Elman network structure based on sparrow search optimization; and training the network evaluation model by using a sample data set, and performing score prediction on an unknown index set by using the trained network, thereby realizing effective prediction of the evaluation score of the radiation source signal identification method. According to the method, the reverse order problem and the incomprehensive evaluation problem caused by a traditional multi-attribute decision-making method are reasonably and effectively solved, and the all-around requirements of a battlefield evaluation model for accuracy, real-time performance and intelligence can be met.