Multi-dimensional data fusion method and system based on adaptive weight
By adopting data fusion method with adaptive weights and Harris Eagle optimization algorithm in multidimensional data classification, the problems of data integrity and feature fusion in multidimensional data environment are solved, and efficient and dynamic classification discrimination ability and model adaptability are achieved.
Patent Information
- Application Number
- CN202411644110.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-03-07
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing classification methods are difficult to meet the integrity, continuity and consistency of data simultaneously in a multi-dimensional data environment, resulting in the impact of model stability and prediction accuracy. Traditional feature processing is difficult to effectively integrate multi-source data, which limits the model's representation ability and generalization ability.
The multi-dimensional data fusion method based on adaptive weights is adopted, and the hyperparameters of the support vector machine are optimized through data preprocessing, feature extraction and comprehensive feature vector construction, combined with the Harris Eagle optimization algorithm of multi-strategy collaboration, and a feedback learning mechanism is introduced to dynamically adjust the model parameters.
It significantly improves the accuracy of multi-dimensional data classification and the adaptability of the model, enhances the efficient and dynamic classification and discrimination ability of complex data, and improves the adaptability and robustness of the model in a dynamic data environment.
Smart Images
Figure CN119577678A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data analysis and machine learning optimization, and in particular to a multidimensional data fusion method and system based on adaptive weights. Background Art
[0002] With the development of big data technology, data classification and discrimination methods are increasingly used in industries, environmental monitoring, medical care, and finance. Classification and discrimination models usually need to process complex multidimensional data, so the accuracy and robustness of the model are crucial in practical applications. However, existing classification methods often face the following technical difficulties in multidimensional data environments:
[0003] Since data in practical applications are diverse and noisy, they usually contain missing values, outliers, and inconsistent feature scales. These factors will affect the stability and prediction accuracy of the model. Existing preprocessing methods often find it difficult to simultaneously meet the integrity, continuity, and consistency of the data, which in turn affects the classification effect.
[0004] Multidimensional data may contain features of different categories (such as physiological features, physical features, and environmental features). Traditional feature processing methods are difficult to effectively fuse multi-source data, which limits the model's ability to represent complex data and reduces classification accuracy and generalization ability.
[0005] The classification performance of models such as support vector machines, which are widely used in classification and discrimination, depends largely on the selection of hyperparameters. Traditional hyperparameter optimization methods such as grid search and random search are computationally intensive and inefficient when seeking the best parameter combination, and are often prone to falling into local optimality, which limits the overall performance of the model.
[0006] Existing classification models are usually static models, that is, the model parameters are no longer updated after training, and the parameters cannot be automatically adjusted according to the feedback of classification results to adapt to data changes. Therefore, in dynamic environments or when data features change, the classification accuracy of the model will gradually decrease.
[0007] In order to solve the above problems, the present invention proposes a multidimensional data fusion method based on adaptive weights. The method not only improves data preprocessing and feature extraction, but also realizes the hyperparameter optimization of support vector machines through the multi-strategy collaborative Harris Hawk Optimization Algorithm (MHHO). In addition, by introducing a feedback learning mechanism, the present invention can dynamically adjust model parameters, improve the adaptability and robustness of the classification model, and thus maintain efficient classification discrimination capabilities in a dynamic data environment. Summary of the invention
[0008] In view of this, the purpose of the present invention is to provide a multidimensional data fusion method and system based on adaptive weights, which realizes efficient classification of multidimensional data through feature extraction, support vector machine algorithm parameter optimization and feedback learning mechanism, and is suitable for real-time classification and intelligent decision support in complex data environments, such as environmental monitoring, industrial process control and multivariate relationship analysis. The present invention can significantly improve the accuracy of multidimensional data classification and the adaptability of the model, and realize efficient and dynamic classification and discrimination of complex data.
[0009] In order to achieve the above objectives, the present invention is implemented through the following technical solutions:
[0010] A multidimensional data fusion method based on adaptive weights comprises the following steps:
[0011] S1. Preprocess the collected data, including data cleaning, missing value filling, outlier detection and processing, and feature standardization to ensure data quality and consistency;
[0012] S2, feature extraction and comprehensive feature vector construction, by extracting the main features and auxiliary features from different data sources, generating a comprehensive feature vector for representing multidimensional data;
[0013] S3, Harris Hawk optimization algorithm based on multi-strategy collaboration optimizes the parameters of support vector machine, uses mixed mapping to initialize population distribution, optimal solution retention strategy, uses cosine decreasing escape energy algorithm to update population, uses dual strategy mutation to perturb the calculated optimal solution and adaptive weight factor to update prey, and finally obtains the optimization algorithm MHHO with good performance;
[0014] S4, using the multi-strategy coordinated Harris Eagle optimization algorithm obtained in step S3 to optimize the hyperparameters of the SVM model (such as the penalty coefficient C and the kernel function parameter γ) to obtain an optimized MHHO-SVM classification model;
[0015] S5. Use the MHHO-SVM model optimized in step S4 to classify and distinguish the multidimensional data, and evaluate and improve the prediction results of the model through a feedback learning mechanism, so as to dynamically adjust the model parameters and further improve the classification performance.
[0016] Further, the step S2 comprises the following steps:
[0017] 2.1 Select the main features and auxiliary features in other specific fields;
[0018] 2.2 Generate comprehensive feature vector F composite , to represent the input data by combining multi-dimensional features:
[0019] F composite=[F1,F2,...,F m ,T] (7)
[0020] Among them, F1, F2, ..., F m represents the multidimensional main features, and T represents the auxiliary features.
[0021] Further, the step S3 comprises the following steps:
[0022] 3.1 Design a hybrid mapping initialization population distribution method so that the population initialization is in a relatively random distribution state, and design a mapping factor to control the concentration of the random distribution, which helps the global development stage and the local exploration stage to fully explore the solution space. The specific formula is:
[0023]
[0024] Among them, X i (t) represents the position of the i-th individual, X bottom and X top They represent the lower and upper bounds of the individual distribution, r(t) is the mapping factor, and t is the current number of iterations;
[0025] The mapping factor controls the random distribution concentration of the population initialization. The larger r is, the more dispersed the population distribution is. The maximum value is usually set below 10 according to the algorithm trend and the range of SVM hyperparameter values. Therefore, in order to effectively balance the global development stage and the local exploration stage, a hybrid attenuation strategy is designed to control the change of the mapping factor, so that the mapping factor is larger in the global search stage and smaller in the local search stage. The specific formula is:
[0026]
[0027] Among them, r max and r min are the initial and final values respectively, β is the shape parameter controlling the decay speed of the function;
[0028] 3.2 Using the optimal solution retention strategy to retain the optimal solution X new_exlite =X old_exlite ;
[0029] 3.3 The transition between the two stages of global search for the optimal solution and local exploration of the optimal solution is determined by the escape energy of the prey. In order to balance the ratio of the number of iterations between the two stages and enable the algorithm to better find the optimal solution, the escape energy update formula is designed as the attenuation part of the cosine function. The specific formula is as follows:
[0030]
[0031] Among them, E maxis the maximum escape energy, t is the current iteration number, and T is the maximum iteration number;
[0032] 3.4 In the process of finding the optimal solution for the MHHO algorithm, the eagle group will concentrate on a certain optimal solution too early, resulting in a phenomenon of falling into a local optimal solution. In order to solve this problem, two different perturbation functions are designed:
[0033] In the global search stage, large-scale mutations are used to perturb the position of the optimal solution combination. The characteristic of this function is that the mutation span is large, which has great advantages for large-scale searches in the early stage of the search, and can significantly enhance the global search ability of traditional HHO. In the local exploration stage, small-scale mutations that obey the normal distribution are used to perform individual small-scale refined mutations. The specific formula is:
[0034]
[0035] Where ΔX i (t) is the variation of the i-th individual in the current iteration; γ is the scale factor of variation, which controls the magnitude of variation; Γ(1+δ) is the Gamma function; δ is the degree of variation control, which is generally taken as (0,2]; r i is a random number; σ is usually taken from the standard normal distribution to control the amplitude of variation; N(0,1) is a random number generated by the standard normal distribution, indicating variation under the Gaussian distribution; thread is determined by the escape energy;
[0036] 3.5 In the later stage of the optimization algorithm, that is, the local development stage, four hunting methods are used to find the optimal solution. In this stage, a weight update method is added to perturb the prey position and enhance the algorithm to jump out of the local optimal solution. Specifically, an exponential decay function is used. The specific formula is as follows:
[0037]
[0038] Among them, W min and W max are the minimum and maximum weight values respectively, and the selected range is based on the SVM hyperparameter selection range.
[0039] Further, in step S4, the hyperparameter optimization of the support vector machine model includes the following steps:
[0040] 4.1 Generate the initial population, each individual is a hyperparameter combination (C, γ);
[0041] 4.2 The optimal individual is obtained by cross-validating each set of hyperparameters through the fitness function. The fitness function is defined as error f(C,γ)=1-Accuracy(C,γ);
[0042] 4.3 After multiple iterations of the optimization algorithm, the optimal hyperparameter combination (C*, γ*) is finally obtained.
[0043] Further, in step S5, the feedback learning mechanism includes the following steps:
[0044] 5.1 Compare the classification results with the actual values and calculate the classification error rate;
[0045] 5.2 Dynamically adjust the hyperparameters of the SVM model based on the feedback results to optimize the model performance.
[0046] The present invention also provides a multidimensional data fusion system based on adaptive weights, which includes a data preprocessing module, a feature extraction module, an SVM classification and parameter optimization module, a feedback learning module, and a result interpretation and display module;
[0047] The data preprocessing module is used to preprocess the collected data to ensure the quality and consistency of the data; the data preprocessing module includes data cleaning, missing value filling, outlier detection and processing, and feature standardization operations;
[0048] The feature extraction module is used to extract and fuse the main features from different data sources from the multidimensional data to form a comprehensive feature vector; the comprehensive feature vector contains multiple features, etc., which are used to enhance the feature expression ability of the classification model;
[0049] The SVM classification and parameter optimization module is based on the support vector machine classification model, and optimizes the hyperparameters of the SVM model (such as the penalty coefficient C and the kernel function parameter γ) through the multi-strategy coordinated Harris Eagle optimization algorithm, thereby improving the classification accuracy and generalization ability;
[0050] The feedback learning module is used to evaluate the classification and discrimination results, and dynamically adjust the model parameter combination according to the classification error feedback information so that the model can adapt to data changes;
[0051] The result interpretation and display module is used to provide model interpretability and user interface display, including feature contribution analysis and time series display of the discrimination results, so as to enhance the user's understanding of the classification results.
[0052] Furthermore, the data preprocessing module includes data cleaning, missing value filling, outlier detection and processing, and feature standardization operations; the data cleaning ensures the integrity and correctness of the data by removing invalid items, null values, and duplicate items in the data; for missing value filling, linear interpolation is used to fill the missing values, and the specific formula is:
[0053]
[0054] Among them, xi is the missing value to be filled, x i-1 and x i+1 are the previous and next known values of the missing value, respectively, t i is the corresponding time point;
[0055] The outlier detection and processing uses the Z-score algorithm to detect abnormal data. The specific formula is:
[0056]
[0057] According to a standard normal distribution, about 99.7% of the data will fall within plus or minus 3 standard deviations of the mean. i When |>3, the data is regarded as an outlier, and the detected outlier is smoothed by the moving average method. The smoothing calculation formula is:
[0058]
[0059] Among them, x new is the smoothed value, n is the window size;
[0060] The feature normalization step scales each feature value to a uniform scale to reduce the impact of scale differences between features. The normalization formula is:
[0061]
[0062] Among them, X is the original eigenvalue, μ is the characteristic mean, and σ is the characteristic standard deviation;
[0063] Furthermore, the features extracted by the feature extraction module include physical features (such as temperature, humidity, etc.) and field-specific features (such as pollutant concentration, industrial indicators, etc.) from different data sources, and a comprehensive feature vector is generated by combining them;
[0064] The comprehensive feature vector is expressed as:
[0065] F composite =[F1,F2,...,F m ,T] (17)
[0066] Furthermore, the SVM classification and parameter optimization module optimizes the hyperparameters of SVM through the multi-strategy coordinated Harris Hawk Optimization Algorithm (MHHO). The algorithm first initializes the population distribution through mixed mapping, and the specific formula is:
[0067]
[0068] Furthermore, the mapping factor controls the random distribution concentration of the population initialization. The larger r is, the more dispersed the population distribution is. Therefore, in order to effectively balance the global development stage and the local exploration stage, a hybrid attenuation strategy is designed to control the change of the mapping factor, so that the mapping factor is larger in the global search stage and smaller in the local search stage. The specific formula is:
[0069]
[0070] The MHHO algorithm evaluates each individual hyperparameter combination by fitness, and the fitness function is defined as:
[0071] f(C,γ)=1-Accuracy(C,γ) (20)
[0072] Among them, Accuracy(C,γ) is the classification accuracy of cross validation;
[0073] The escape energy update function in the MHHO algorithm uses a cosine decreasing function combined with a random perturbation factor to improve the early convergence speed. The specific formula is:
[0074]
[0075] The individual variation includes two stages, global search and local development, which execute different functions, corresponding to the individual energy E being higher or lower than a threshold value;
[0076] Furthermore, in the process of finding the optimal solution, the MHHO algorithm uses large-scale mutations to perturb the position of the optimal solution combination in the global search stage, and performs individual small-scale refined mutations by small-scale mutations that obey the normal distribution in the local exploration stage. The specific formula is:
[0077]
[0078] The adaptive weight factor update uses an exponential decay function to dynamically calculate the weight factor to dynamically adjust the position of the prey. The formula is:
[0079]
[0080] Furthermore, the optimal solution individual obtained by the optimal solution retention strategy in each iteration is denoted as X elite ;
[0081] Furthermore, the feedback learning module is used to perform feedback analysis on the errors of the classification results and introduce the feedback information into the model optimization process to improve the adaptability and classification accuracy of the model. This module calculates the classification error rate and feeds the error back to the SVM hyperparameter optimization process to achieve adaptive adjustment of the parameter combination.
[0082] Furthermore, the result interpretation and display module uses the Shapley value to calculate the contribution of each feature to the classification result. The specific formula is:
[0083]
[0084] Among them, N is the feature set, S is the feature subset, and v(S) is the contribution value of subset S. Furthermore, the time series dynamic display module displays the trend of feature changes and classification results through time series, provides a visual display of historical data, real-time data and forecast data, and provides users with intuitive classification analysis support.
[0085] Compared with the prior art, the advantages and positive effects of the present invention are:
[0086] The present invention provides a multidimensional data fusion method and system based on adaptive weights. On the one hand, the data preprocessing module cleans, processes outliers, fills in missing values, and standardizes multi-source data to ensure the integrity and consistency of the data; on the other hand, the feature extraction module fuses multidimensional features from different sources to generate a comprehensive feature vector, thereby enhancing the data feature expression capability and providing rich input features for the classification model. In addition, the Harris Hawk Optimization Algorithm (MHHO) based on multi-strategy collaboration optimizes the hyperparameters of the support vector machine (SVM), so that the classification model has stronger classification accuracy and generalization ability; in the classification process, the present invention also introduces a feedback learning mechanism to dynamically optimize the model parameters through classification error feedback, so that the model can maintain efficient classification performance when the data features change.
[0087] The present invention forms a complete discrimination process from data preprocessing, multi-dimensional feature fusion, SVM classification and optimization to feedback learning, which significantly improves the adaptability, classification accuracy and dynamic adjustment ability of the classification model in a complex multi-dimensional data environment. The multi-strategy collaborative Harris Eagle optimization algorithm is used to improve global search and local development capabilities to ensure that good classification results can be achieved on different data sets. By introducing an adaptive feedback mechanism, the present invention effectively improves the adaptability of the model to real-time data changes, and provides reliable support for real-time classification discrimination in complex data environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0088] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:
[0089] Figure 1The system structure block diagram of the multi-dimensional data fusion method based on adaptive weights of the present invention. The system includes a data preprocessing module, a feature extraction module, an SVM classification and parameter optimization module (Harris Eagle optimization algorithm based on multi-strategy collaboration), a feedback learning module, and a result interpretation and display module. The figure shows the logical relationship between the modules and the data flow;
[0090] Figure 2 This is a detailed flowchart of the data preprocessing module. The figure shows the four steps of the data preprocessing module: data cleaning, missing value filling, outlier detection and processing, and feature standardization. Each step describes in detail the execution process of each data processing method to ensure the consistency and integrity of the input data;
[0091] Figure 3 The figure is a schematic diagram of the structure of the feature extraction module. The figure shows the feature extraction process of different data sources, including the extraction of main factors and auxiliary factors, and the construction process of comprehensive feature vectors. The figure shows the feature fusion process of multiple data sources to generate a comprehensive feature vector for classification;
[0092] Figure 4 This is a flow chart for the parameter optimization of the support vector machine based on the multi-strategy coordinated Harris Hawk optimization algorithm. The figure shows the overall optimization process of the multi-strategy coordinated Harris Hawk optimization algorithm, including the initial population generation, fitness calculation, energy update, exploration and development behavior selection, position update and other steps. The figure also shows the weight factor update method based on Logistic mapping and exponential decay, as well as the switching conditions between global search and local development;
[0093] Figure 5 The workflow diagram of the feedback learning module shows that the feedback learning module compares the error between the classification result and the actual result, analyzes the classification error and feeds it back to the SVM hyperparameter optimization process to achieve dynamic adjustment and optimization of the model parameters; DETAILED DESCRIPTION
[0094] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.
[0095] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.
[0096] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.
[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.
[0098] See also Figure 1 The present invention is a classification and discrimination system based on multi-dimensional feature fusion and adaptive optimization, and its structure includes multiple functional modules: data preprocessing module, feature extraction module, SVM classification and parameter optimization module, feedback learning module and result interpretation and display module. Among them: the data preprocessing module is used to clean, fill and standardize the input multi-dimensional data to ensure the integrity and consistency of the data; the feature extraction module extracts and fuses the features of different data sources to generate a comprehensive feature vector; the SVM classification and parameter optimization module uses the multi-strategy collaborative Harris Eagle optimization algorithm to optimize the hyperparameters of the support vector machine; the feedback learning module dynamically adjusts the model parameters according to the error of the classification result to enhance the adaptability of the model; the result interpretation and display module explains the discrimination results and provides a user-friendly visual display.
[0099] Step 1: Data preprocessing module, used for data anomaly monitoring, deletion, filling and other processing operations, specifically including the following implementation methods:
[0100] 1.1 Data cleaning and missing value processing: Clean the input multidimensional data, remove invalid data, null values and duplicates to ensure data integrity. For missing values, use interpolation methods to fill them so that the data maintains continuity in subsequent processing. The specific formula is:
[0101]
[0102] 1.2 Outlier detection and processing,The outliers in the data are detected by the Z-score algorithm.,For the detected outliers, the moving average method is used for smoothing,to reduce the negative impact of the abnormal data on the model training process.
[0103] The specific detection formula is:
[0104]
[0105] The specific smoothing formula is:
[0106]
[0107] 1.3 Feature standardization: standardize the data and scale each feature value to a uniform scale to ensure that the weight of each feature is consistent in subsequent analysis and classification, and avoid unnecessary influence of different feature scales on the classification results. The specific formula is:
[0108]
[0109] Step 2: Comprehensive feature vector construction, extract multiple features from different data sources and fuse these features into a comprehensive feature vector. The comprehensive feature vector can better express the characteristics of multidimensional data and provide effective input for the classification model.
[0110] Step 3: SVM classification and parameter optimization module, in order to build a high-precision MHHO-SVM algorithm model and realize data fusion, the specific implementation method is as follows:
[0111] 3.1 Initial population generation (C, γ), using the multi-strategy coordinated Harris Eagle optimization algorithm to initialize the hyperparameter population of the support vector machine model, including the penalty coefficient C and the kernel parameter γ. The initial population is generated by mixed mapping to increase the diversity of the initial individuals. The mixed mapping formula is:
[0112]
[0113] Where r(t) is the mapping factor and t is the current iteration number;
[0114] The mapping factor controls the random distribution concentration of the population initialization. The larger r is, the more dispersed the population distribution is. Therefore, in order to effectively balance the global development stage and the local exploration stage, a hybrid attenuation strategy is designed to control the change of the mapping factor, so that the mapping factor is larger in the global search stage and smaller in the local search stage. The specific formula is:
[0115]
[0116] 3.2 Fitness Evaluation,The fitness of each individual in the initial population is evaluated, and the classification error is calculated using the cross-validation method to measure the pros and cons of each set of (C,γ) hyperparameter combinations.
[0117] 3.3MHHO algorithm iterative optimization, dynamically adjust the position and parameter combination of individuals according to the current fitness information and the escape energy state of individuals. MHHO algorithm searches for the optimal hyperparameter combination in multidimensional space by combining global search and local development.
[0118] The specific escape energy formula is:
[0119]
[0120] The specific dual-strategy individual variation formula is:
[0121]
[0122] 3.4SVM model training, the optimal hyperparameter combination (C*, γ*) obtained by the MHHO optimization algorithm is used to train the support vector machine model and build the final classification model.
[0123] Step 4: Feedback learning: Feedback the error to the hyperparameter optimization process of the SVM model, and dynamically adjust the hyperparameter combination to further optimize the adaptability and classification performance of the model. Figure 5 Workflow diagram for the feedback learning module.
[0124] Step 5: Result display: display the classification results, error analysis and optimized model parameters on the user interface, provide intuitive classification discrimination results and data trend analysis support to help users make decisions.
[0125] See also Figure 2 ,This figure shows the specific process of the data preprocessing module, which includes the following steps:
[0126] Data cleaning: removing invalid items, null values and duplicate items in the data to ensure data integrity;
[0127] Missing value filling: linear interpolation method is used to fill missing values;
[0128] Outlier detection and processing, using the Z-score algorithm to detect outliers and smoothing them using the moving average method;
[0129] Feature standardization scales each feature to a uniform scale, eliminating the dimensional differences between different features to facilitate subsequent analysis.
[0130] See also Figure 3 , which is a schematic diagram of the structure of the feature extraction module. The specific implementation method includes: first, extracting multidimensional features from different data sources, such as main factors and auxiliary factors. Then, combining each feature into a comprehensive feature vector for input into the classification model. The comprehensive feature vector is represented by F composite =[F1,F2,...,F m ,T], where F1,F2,...,F m represents the multidimensional main feature, and T represents the auxiliary feature.
[0131] See also Figure 4 , which is a flowchart of the Harris Hawk optimization algorithm for multi-strategy collaboration in the present invention. The specific implementation method is as follows:
[0132] Initial population generation, using mixed mapping to generate the position of the initial population. The population distribution generated by mixed mapping has good chaos, which helps the diversity of initial individuals and enhances the global search ability of the algorithm.
[0133] Fitness calculation, calculate the fitness of each individual in the population to measure the quality of the current solution. The fitness function is usually based on the classification error of the support vector machine model.
[0134] Escape energy update, apply the cosine decreasing function to dynamically update the energy E of each individual. This energy value is used to guide the algorithm to select appropriate search strategies at different stages.
[0135] Determine the energy state and select a global search or local search strategy based on the current energy E.
[0136] Global search, when the energy value is greater than the threshold, a global search is performed to expand the search space. In order to increase the diversity of the population, a large-scale mutation strategy is adopted to randomly perturb the current optimal individual and increase the search space.
[0137] Local search: When the energy value is less than the threshold, a local search is performed to refine the search near the optimal solution. This stage introduces an adaptive weight update mechanism to update the prey position, and adopts a multi-strategy local development mode. A small-amplitude mutation strategy is used to perturb individuals to escape the local optimal trap.
[0138] The optimal solution individual is updated after each round of search to ensure that the current optimal solution is preserved. The optimal solution individual will be used as a candidate for the global optimal solution, and will continue to improve and approach the optimal solution.
[0139] Iteration termination condition, judge whether the number of iterations exceeds the preset value or meets the accuracy requirement. If the termination condition is met, the optimal solution individual is output as the final global optimal solution; otherwise, enter the next round of iteration.
[0140] The above-described embodiments only express the implementation methods of the present application, and those skilled in the art will easily think of other embodiments of the present invention after considering the specification and practicing the invention invented here. This application is intended to cover any modification, use or adaptive change of the present invention, which follows the general principles of the present invention and includes common knowledge or customary technical means in the art that the present invention has not invented. It should be understood that the present invention is not limited to the structural flow described above and shown in the drawings, and can be modified or changed without departing from its scope.
Claims
1. A multidimensional data fusion method based on adaptive weights, characterized in that: The following steps are involved: S1. Preprocess the collected data, including data cleaning, missing value filling, outlier detection and processing, and feature standardization to ensure data quality and consistency; S2, feature extraction and comprehensive feature vector construction, by extracting the main features and auxiliary features from different data sources, generating a comprehensive feature vector for representing multidimensional data; S3, Harris Hawk optimization algorithm based on multi-strategy collaboration optimizes the parameters of support vector machine, uses mixed mapping to initialize population distribution, optimal solution retention strategy, uses cosine decreasing escape energy algorithm to update population, uses dual strategy mutation to perturb the calculated optimal solution and adaptive weight factor to update prey, and finally obtains the optimization algorithm MHHO with good performance; S4, using the multi-strategy coordinated Harris Eagle optimization algorithm obtained in step S3 to optimize the hyperparameters of the SVM model to obtain an optimized MHHO-SVM classification model; S5. Use the MHHO-SVM model optimized in step S4 to classify and distinguish the multidimensional data, and evaluate and improve the prediction results of the model through a feedback learning mechanism, so as to dynamically adjust the model parameters and further improve the classification performance.
2. The multidimensional data fusion method based on adaptive weights according to claim 1, characterized in that: The step S2 comprises the following steps: 2.1 Select the main features and auxiliary features in other specific fields; 2.2 Generate comprehensive feature vector F composite , to represent the input data by combining multi-dimensional features: F composite =[F1,F2,...,F m ,T] (1) Among them, F1, F2, ..., F m represents the multidimensional main features, and T represents the auxiliary features.
3. The multidimensional data fusion method based on adaptive weights according to claim 2, characterized in that: The step S3 comprises the following steps: 3.1 Design a hybrid mapping initialization population distribution method so that the population initialization is in a relatively random distribution state, and design a mapping factor to control the concentration of random distribution. The specific formula is: Among them, X i (t) represents the position of the i-th individual, X bottom and X top They represent the lower and upper bounds of the individual distribution, r(t) is the mapping factor, and t is the current number of iterations; A hybrid attenuation strategy is designed to control the change of the mapping factor, so that the mapping factor is larger in the global search phase and smaller in the local search phase. The specific formula is: Among them, r max and r min are the initial and final values respectively, β is the shape parameter controlling the decay speed of the function; 3.2 Using the optimal solution retention strategy to retain the optimal solution X new_exlite =X old_exlite ; 3.3 For the optimization algorithm, the transition between the global search for the optimal solution and the local exploration of the optimal solution is determined by the escape energy of the prey. The escape energy update formula is designed as the attenuation part of the cosine function. The specific formula is as follows: Among them, E max is the maximum escape energy, t is the current iteration number, and T is the maximum iteration number; 3.4 In the process of finding the optimal solution for the MHHO algorithm, the eagle group will concentrate on a certain optimal solution too early, resulting in a phenomenon of falling into a local optimal solution. In order to solve this problem, two different perturbation functions are designed: In the global search stage, large-scale mutations are used to perturb the position of the optimal solution combination. The characteristic of this function is that the mutation span is large, which has great advantages for large-scale searches in the early stage of the search, and can significantly enhance the global search ability of traditional HHO. In the local exploration stage, small-scale mutations that obey the normal distribution are used to perform individual small-scale refined mutations. The specific formula is: Where ΔX i (t) is the variation of the i-th individual in the current iteration; γ is the scale factor of variation, which controls the magnitude of variation; Γ(1+δ) is the Gamma function; δ is the degree of variation control, which is generally taken as (0,2]; r i is a random number; σ is usually taken from the standard normal distribution to control the amplitude of variation; N(0,1) is a random number generated by the standard normal distribution, indicating variation under the Gaussian distribution; thread is determined by the escape energy; 3.5 In the later stage of the optimization algorithm, that is, the local development stage, four hunting methods are used to find the optimal solution. In this stage, a weight update method is added to perturb the prey position and enhance the algorithm to jump out of the local optimal solution. Specifically, an exponential decay function is used. The specific formula is as follows: Among them, W min and W max are the minimum and maximum weight values respectively, and the selected range is based on the SVM hyperparameter selection range.
4. The multidimensional data fusion method based on adaptive weights according to claim 3 is characterized in that: In step S4, the hyperparameter optimization of the support vector machine model includes the following steps: 4.1 Generate the initial population, each individual is a hyperparameter combination (C, γ); 4.2 The optimal individual is obtained by cross-validating each set of hyperparameters through the fitness function. The fitness function is defined as error f(C,γ)=1-Accuracy(C,γ); 4.3 After multiple iterations of the optimization algorithm, the optimal hyperparameter combination (C*, γ*) is finally obtained.
5. The multidimensional data fusion method based on adaptive weight according to claim 4, characterized in that: In step S5, the feedback learning mechanism includes the following steps: 5.1 Compare the classification results with the actual values and calculate the classification error rate; 5.2 Dynamically adjust the hyperparameters of the SVM model based on the feedback results to optimize the model performance.
6. A multidimensional data fusion system based on adaptive weights, characterized in that: The system includes a data preprocessing module, a feature extraction module, an SVM classification and parameter optimization module, a feedback learning module, and a result interpretation and display module; The data preprocessing module is used to preprocess the collected data to ensure the quality and consistency of the data; the data preprocessing module includes data cleaning, missing value filling, outlier detection and processing, and feature standardization operations; The feature extraction module is used to extract and fuse the main features from different data sources from the multidimensional data to form a comprehensive feature vector; the comprehensive feature vector contains multiple features, etc., which are used to enhance the feature expression ability of the classification model; The SVM classification and parameter optimization module is based on the support vector machine classification model, and optimizes the hyperparameters of the SVM model through the multi-strategy collaborative Harris Eagle optimization algorithm, thereby improving the classification accuracy and generalization ability; The feedback learning module is used to evaluate the classification discrimination results and dynamically adjust the model parameter combination according to the classification error feedback information so that the model can adapt to data changes; The result interpretation and display module is used to provide model interpretability and user interface display, including feature contribution analysis and time series display of the discrimination results, so as to enhance the user's understanding of the classification results.