Interactive data processing method based on multi-dimensional data analysis

Abstract

The invention discloses an interactive data processing method based on multi-dimensional data analysis, belongs to the technical field of data interactive processing, and solves the problems that in an existing method, when a multi-dimensional data model processes interactive data, multi-source heterogeneous multi-dimensional data is low in quality, multi-dimensional data noise cannot be filtered out, and data processing efficiency is low. The method comprises the steps that multi-dimensional time sequence data related to user behaviors is collected in real time, a multi-dimensional data analysis model analyzes and processes a time sequence data set based on edge calculation, and visual interaction processing is carried out on a data analysis result in combination with a particle swarm optimization algorithm and an ant colony optimization algorithm. The multi-dimensional data analysis model not only can perform filtering and feature fusion processing on multi-source heterogeneous multi-dimensional time sequence data, but also can perform visual interaction processing on a data analysis result by combining a particle swarm optimization algorithm and an ant colony optimization algorithm, and performs short-term and long-term behavior preference fusion on the user to output an accurate user behavior prediction result.

Description

Technical Field

[0001] The present invention belongs to the technical field of data interaction processing, and in particular relates to an interactive data processing method based on multidimensional data analysis. Background Art

[0002] In the era of big data, with the rapid development of information technology, the scale of data production has grown exponentially. How to organize, understand and utilize massive data has become a problem faced by all walks of life. Real-time data is generally multi-dimensional structured data, semi-structured data and unstructured data, and there is uncertainty, including erroneous, duplicate or incomplete data. Processing this data requires the ability to effectively filter and correct errors to ensure the accuracy and reliability of data interaction.

[0003] Chinese patent CN115658775A discloses a data analysis method, device, electronic device and computer-readable storage medium, the method comprising: collecting IoT customer data based on IoT devices, wherein the IoT customer data includes one or any combination of face data, diagnostic data and signal data; performing statistical analysis on the collected IoT customer data to obtain a multidimensional data model; performing trend forecasting analysis based on the multidimensional data model, but in the existing method, when the multidimensional data model processes interactive data, the multidimensional data of multi-source heterogeneity has low quality and cannot filter out the multidimensional data noise, resulting in incomplete capture of effective information of the multidimensional data and poor fusion effect. In view of the above problems, we propose an interactive data processing method based on multidimensional data analysis. Summary of the invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and provide an interactive data processing method based on multidimensional data analysis, which solves the problems in the prior art that when the multidimensional data model processes interactive data, the quality of multi-source heterogeneous multidimensional data is low and the multidimensional data noise cannot be filtered out, resulting in incomplete capture of effective information in the multidimensional data and poor fusion effect.

[0005] In the existing methods, when multidimensional data models process interactive data, the quality of multi-source heterogeneous multidimensional data is low and the noise of multidimensional data cannot be filtered out, resulting in incomplete capture of effective information of multidimensional data and poor fusion effect. To address the above problems, we propose an interactive data processing method based on multidimensional data analysis. In short, when the method is implemented, it first collects multidimensional time series data related to user behavior in real time based on the Internet of Things gateway, integrates and cleans the multidimensional time series data, and pre-builds a multidimensional data analysis model based on deep learning combined with the improved whale algorithm to traverse the interactive database. Then, the multidimensional data analysis model analyzes and processes the time series data set based on edge computing, outputs data analysis results and data flow recommendation links, triggers data interaction instructions, and interactively uploads the data analysis results to the visualization interaction platform based on the flow recommendation link. Finally, the particle swarm algorithm and ant colony optimization algorithm are combined to visualize and interactively process the data analysis results, and output interactive evaluation results. In an embodiment of the present invention, a multidimensional data analysis model is constructed and trained based on deep learning combined with an improved whale algorithm. The multidimensional data analysis model can not only filter and fuse features of multi-source heterogeneous multidimensional time series data, but also visualize and interactively process data analysis results by combining a particle swarm algorithm and an ant colony optimization algorithm, and fuse users' short-term and long-term behavior preferences to output accurate user behavior prediction results.

[0006] The present invention is implemented as follows: an interactive data processing method based on multidimensional data analysis, the interactive data processing method based on multidimensional data analysis comprising:

[0007] Based on the IoT gateway, multi-dimensional time series data related to user behavior is collected in real time, the multi-dimensional time series data is integrated, cleaned and processed, and the time series data set is output, and the time series data set is uploaded to the interactive database;

[0008] Pre-build a multidimensional data analysis model based on deep learning combined with the improved whale algorithm, traverse the interactive database, load model training samples from the interactive database, train the multidimensional data analysis model through the model training samples, and output a converged multidimensional data analysis model;

[0009] Real-time loading of time-series data sets. The multidimensional data analysis model analyzes and processes the time-series data sets based on edge computing, outputs data analysis results and data flow recommendation links, triggers data interaction instructions, and interactively uploads data analysis results to the visualization interaction platform based on the flow recommendation links.

[0010] In response to data interaction instructions, the visualization interaction platform triggers the visualization interaction logic based on the behavior analysis requirements, combines the particle swarm algorithm and the ant colony optimization algorithm to visualize and interact with the data analysis results, and outputs the interaction evaluation results.

[0011] Preferably, the method of training a multidimensional data analysis model using model training samples specifically includes:

[0012] Construct a multidimensional data analysis model, using a convolutional neural network as the initial model of the multidimensional data analysis model. The initial model includes an input layer, a convolution module, a pooling layer, and a fully connected layer. Define the loss function of the initial model as the SIou loss function, and set the initial model iterative training rounds and hyperparameters.

[0013] Obtain model training samples, divide the model training samples into a training set and a validation set, input the training set, and perform filtering and denoising on the training set by the initial model to obtain a sample denoised set;

[0014] Load the sample denoising set, and use the sample denoising set to iteratively train the initial model until convergence;

[0015] Obtain a validation set, use the validation set as input, execute the initial model, output the validation result, and judge whether the error between the validation result and the actual result meets the preset accuracy threshold;

[0016] If the preset accuracy threshold is met, a converged multidimensional data analysis model is output;

[0017] If the preset accuracy threshold is not met, the learning rate of the initial model is fine-tuned using the validation set combined with the fine-tuning mechanism, and the initial model is iteratively trained until convergence.

[0018] Preferably, the convolution module consists of a first convolution layer, a second convolution layer, and a third convolution layer. The first convolution layer is frozen and replaced by an iterative filter. The number of channels of the second convolution layer is 256, and the convolution kernel size is 3*3. The third convolution layer is a SimAM module, and the number of channels of the third convolution layer is 512, and the convolution kernel size is 3*3. The activation function of the convolution module is a ReLU function. The pooling layer introduces the SimAM attention mechanism, and the pooling layer adopts maximum pooling. The local area of ​​the features filtered out by the convolution is taken as the maximum value to extract the most important feature information. The link discrimination network architecture is introduced into the initial model, and the improved whale algorithm is incorporated into the link discrimination network architecture. The link discrimination network architecture recommends links based on the feature weights of the training samples and the link response ratio discrimination data flow.

[0019] Preferably, the method for analyzing and processing a time series data set based on edge computing by the multidimensional data analysis model specifically includes:

[0020] Obtain a time series data set, perform dimensionality reduction processing on the multi-dimensional time series data in the time series data set, and generate one-dimensional sequence data;

[0021] Among them, the one-dimensional sequence data is expressed as:

[0022]

[0023] Among them, X represents the input representation of multidimensional time series data, i is the dimension of the data, and θ t Represents the weight parameter of the data. The weight parameter of the data is defined based on the Huber weight function. x(t) is the output representation of the one-dimensional sequence data.

[0024]

[0025] Among them, θ t represents the weight parameter of the data, m is the number of data types, ε is the preset weight threshold in Huber regression, σ(X) is the residual of multidimensional time series data, Represents the mean of multidimensional time series data;

[0026] Load the one-dimensional sequence data, perform Fourier transform on the one-dimensional sequence data based on the Blackman window, and obtain a unified two-dimensional time-frequency spectrum;

[0027] An iterative filter is used to filter the two-dimensional time-frequency spectrum and output a filtered data set. The third convolutional layer fuses the filtered data set to obtain a feature fusion set.

[0028] Load the feature fusion set. The pooling layer simplifies the features in the feature fusion set based on the SimAM attention mechanism. The sensitivity analysis algorithm is used to consider the influence of the associated data associated with the filtered data. The filtered data in the feature fusion set is mapped to a low-dimensional latent attention space, and the attention vector is collected from the latent attention space.

[0029] Use the attention vector to predict three sets of fusion properties of the filter data: weight inertia coefficient, sensitivity coefficient, and priority coefficient, and output the weight inertia coefficient, sensitivity coefficient, and priority coefficient of the filter data;

[0030] Loading filtered data, where the filtered data includes an attention vector, a weight inertia coefficient, a sensitivity coefficient, and a priority coefficient. The link discrimination network architecture calculates the link responsiveness of the filtered data based on an improved whale algorithm, and sets at least one set of link responsiveness of the filtered data as a cluster center.

[0031] Based on the improved whale algorithm, the link response ratio of the flow link is calculated, and the link response ratio is clustered using the fuzzy clustering algorithm. The flow links that meet the cluster center are screened, at least one group of flow links are integrated, and the flow links are set as data flow recommended links.

[0032] Output data analysis results and recommended data flow links, and trigger data interaction instructions.

[0033] Preferably, when Fourier transform processing is performed on one-dimensional sequence data based on the Blackman window:

[0034]

[0035] S t =|Stft t (t,f s ) 2 (4)

[0036] Among them, S t is the output representation of the two-dimensional time-frequency spectrum, t is the sampling time, f s is the sampling frequency, x(t) represents one-dimensional sequence data, is the mean of the one-dimensional sequence data, B(x) represents the Blackman window function, Stft t (t,f s ) is a two-dimensional function of one-dimensional sequence data;

[0037]

[0038] Among them, x(t) represents one-dimensional sequence data, is the mean of the one-dimensional sequence data, l and α represent the length and shape parameters of the Blackman window respectively;

[0039] The iterative filter filters the two-dimensional time-frequency spectrum and obtains the filtered data set through the following formula:

[0040]

[0041] Among them, S L t is the output representation of the filtered data set, L is the number of filtering times of the iterative filter, G(w) represents the amplitude response of the iterative filter, arg(·) represents the argument, w(D) represents the iterative filter, and SL is the sector loss of the iterative filter;

[0042] The weight inertia coefficient, sensitivity coefficient, and priority coefficient are determined by the following formula:

[0043]

[0044]

[0045]

[0046] Among them, T R is the attention vector input representation, θ t ,q t , γ represents the weight inertia coefficient, priority coefficient, and sensitivity coefficient, β is the attention space mapping coefficient, η is the sensitivity constant, S1 and S2 are S L tThe filtered data set is based on the first-order mapping coefficients and the second-order mapping coefficients of the sensitivity analysis algorithm.

[0047] Preferably, the method for interactively processing the data analysis results by combining the particle swarm algorithm and the ant colony optimization algorithm specifically includes:

[0048] Obtain data analysis results, initialize pheromones of the ant colony optimization algorithm, and generate the initial positions of ants;

[0049] The ants are sorted by integrating the pheromone and path length of the ant colony optimization algorithm to obtain the ranking of at least one group of the ants, and the pheromone weighting is updated based on the ranking combined with the particle swarm optimization algorithm;

[0050] The weighted update of pheromones based on ranking combined with particle swarm optimization algorithm is expressed as:

[0051] ω i (t+1)=(1-ψ)ω i (t)+c1m1(L best ×Q+Δω i )+c2m2(L best ×Q-Δω i )(10)

[0052] Among them, ω i (t+1) represents the weighted update result of pheromone after t+1 weighting, ω i (t) is the current pheromone input, ψ is the ranking weight, Q is the number of ants, Δω i Indicates the update of pheromone, L best represents the optimal path length, c1, c2 represent the particle swarm optimization acceleration factors, m1, m2 are random heuristic factors;

[0053] The ant positions are updated in a spiral manner. When the ant positions are updated, the fitness function constraint is used to calculate the fitness value of the ant positions, and the ant positions corresponding to the maximum fitness value are recorded until the maximum number of ant iterations is reached.

[0054] Integrate at least one set of ant positions, output user behavior prediction results based on the ant positions, use visualization tools to process the user behavior prediction results, and output interaction evaluation results.

[0055] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0056] In an embodiment of the present invention, a multidimensional data analysis model is constructed and trained based on deep learning combined with an improved whale algorithm. The multidimensional data analysis model can not only filter and fuse features of multi-source heterogeneous multidimensional time series data, but also visualize and interactively process data analysis results by combining a particle swarm algorithm and an ant colony optimization algorithm, and fuse short-term and long-term user behavior preferences to output accurate user behavior prediction results, thereby overcoming the problems in existing methods of multidimensional data models processing interactive data, such as low quality of multi-source heterogeneous multidimensional data and inability to filter out multidimensional data noise, resulting in incomplete capture of effective information in multidimensional data and poor fusion effect.

[0057] In an embodiment of the present invention, a multidimensional data analysis model and a training method thereof are provided. The multidimensional data analysis model is combined with a convolutional neural network, a SimAM attention mechanism, and an iterative filter, so that the multidimensional data analysis model has significant advantages in user behavior analysis, which can not only improve feature extraction and representation capabilities, improve model performance and generalization capabilities, but also can be combined with an improved whale algorithm and a link discrimination network architecture to achieve optimal allocation of data flow links. The link discrimination network architecture uses a deep learning model to perform real-time discrimination and optimization of data flow links. This architecture can make full use of the data features of the attention vector, weight inertia coefficient, sensitivity coefficient, and priority coefficient to achieve accurate prediction and adjustment of the link state. The link discrimination network architecture combined with the improved whale algorithm can find the optimal link allocation solution in a broader search space. For the optimal allocation of data flow links, it means that the optimal link can be found faster and the data transmission efficiency can be improved.

[0058] In an embodiment of the present invention, an attention vector is used to comprehensively evaluate user behavior, and the attention vector integrates the weight inertia coefficient, sensitivity coefficient, and priority coefficient of the filtered data. Through weighted fusion of the weight inertia coefficient, sensitivity coefficient, and priority coefficient, on the one hand, it can assist the particle swarm algorithm and the ant colony optimization algorithm in visualizing and interactively processing the data analysis results, and accurately realize the user behavior prediction results. On the other hand, it can also assist in screening the flow links that meet the clustering center, thereby providing data flow efficiency and processing speed through edge computing, and reducing the overall load of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the implementation flow of an interactive data processing method based on multidimensional data analysis provided by the present invention.

[0060] Figure 2 The present invention shows a schematic diagram of the implementation process of a method for training a multidimensional data analysis model through model training samples.

[0061] Figure 3The figure shows a schematic diagram of the implementation process of a multidimensional data analysis model for analyzing and processing time-series data sets based on edge computing.

[0062] Figure 4 The figure shows a schematic diagram of the implementation process of the method for interactively visualizing data analysis results by combining the particle swarm algorithm and the ant colony optimization algorithm.

[0063] Figure 5 It is a structural schematic diagram of an interactive data processing system based on multidimensional data analysis provided by the present invention. DETAILED DESCRIPTION

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0065] In the existing methods, when multidimensional data models process interactive data, the quality of multi-source heterogeneous multidimensional data is low and the noise of multidimensional data cannot be filtered out, resulting in incomplete capture of effective information of multidimensional data and poor fusion effect. To address the above problems, we propose an interactive data processing method based on multidimensional data analysis. In short, when the method is implemented, it first collects multidimensional time series data related to user behavior in real time based on the Internet of Things gateway, integrates and cleans the multidimensional time series data, and pre-builds a multidimensional data analysis model based on deep learning combined with the improved whale algorithm to traverse the interactive database. Then, the multidimensional data analysis model analyzes and processes the time series data set based on edge computing, outputs data analysis results and data flow recommendation links, triggers data interaction instructions, and interactively uploads the data analysis results to the visualization interaction platform based on the flow recommendation link. Finally, the particle swarm algorithm and ant colony optimization algorithm are combined to visualize and interactively process the data analysis results, and output interactive evaluation results. In an embodiment of the present invention, a multidimensional data analysis model is constructed and trained based on deep learning combined with an improved whale algorithm. The multidimensional data analysis model can not only filter and fuse features of multi-source heterogeneous multidimensional time series data, but also visualize and interactively process data analysis results by combining a particle swarm algorithm and an ant colony optimization algorithm, and fuse short-term and long-term user behavior preferences to output accurate user behavior prediction results, thereby overcoming the problems in existing methods of multidimensional data models processing interactive data, such as low quality of multi-source heterogeneous multidimensional data and inability to filter out multidimensional data noise, resulting in incomplete capture of effective information in multidimensional data and poor fusion effect.

[0066] The embodiment of the present invention provides an interactive data processing method based on multidimensional data analysis. Figure 1 The following is a schematic diagram of the implementation process of the interactive data processing method based on multidimensional data analysis. The interactive data processing method based on multidimensional data analysis specifically includes:

[0067] Step S10, based on the Internet of Things gateway, real-time collection of multi-dimensional time series data related to user behavior, integration and cleaning of the multi-dimensional time series data, output of the time series data set, and uploading of the time series data set to the interactive database;

[0068] It should be noted that the IoT gateway is connected to multiple groups of sensors, including but not limited to edge gateways, industrial gateways, wireless gateways, multi-protocol gateways, fog gateways, and multi-protocol gateways that support multiple communication protocols and standards, allowing devices using different protocols to connect and communicate seamlessly. These gateways are widely used and can handle different IoT device ecosystems, thereby simplifying integration and interoperability challenges, while sensors include but are not limited to pressure sensors, vibration sensors, photoelectric sensors, angular displacement sensors, position sensors, biosensors, temperature sensors, and gas sensors. Multidimensional time series data can be structured, unstructured, or semi-structured interactive behavior images, videos, and logs.

[0069] For example, multi-dimensional time series data can be web page interaction data such as user click times, page dwell time, scroll depth, etc. It can also be in-application operation records, such as button clicks, menu selections, function usage frequency, etc. The interaction database consists of storage tables, indexes, views, and triggers, and the interaction database can be a NoSQL database.

[0070] In this embodiment, the method of integrating and cleaning the multi-dimensional time series data may be to delete outliers and process missing values.

[0071] Step S20, pre-constructing a multidimensional data analysis model based on deep learning combined with an improved whale algorithm, traversing the interactive database, loading model training samples from the interactive database, training the multidimensional data analysis model through the model training samples, and outputting a converged multidimensional data analysis model;

[0072] In this embodiment, when loading model training samples from the interactive database, the batch export tool mongoexport can be used to export the samples in batches. The data of the model training samples come from a wide range of sources and can be collected from public websites, software (Gaussian, VASP, etc.) calculations, and literature.

[0073] Step S30, real-time loading of time series data sets, multi-dimensional data analysis model analyzes and processes the time series data sets based on edge computing, outputs data analysis results and data flow recommendation links, triggers data interaction instructions, and interactively uploads data analysis results to the visualization interaction platform based on the flow recommendation links.

[0074] Step S40, in response to the data interaction instruction, the visualization interaction platform triggers the visualization interaction logic based on the behavior analysis requirements, combines the particle swarm algorithm and the ant colony optimization algorithm to visualize and interact with the data analysis results, and outputs the interaction evaluation results.

[0075] In this embodiment, when the visualization interaction platform triggers the visualization interaction logic based on the behavior analysis requirements, the behavior analysis requirements can be user preference requirements, user interest analysis requirements, user threat detection requirements analysis, and user demand expectation analysis. Through in-depth analysis of user behavior, problems and deficiencies in the product can be discovered, providing a basis for product iteration. On the other hand, through the analysis of user behavior data, the user's interests, preferences, and purchasing behaviors can be understood, thereby achieving precision marketing, improving marketing effects, and achieving precision marketing.

[0076] In an embodiment of the present invention, a multidimensional data analysis model is constructed and trained based on deep learning combined with an improved whale algorithm. The multidimensional data analysis model can not only filter and fuse features of multi-source heterogeneous multidimensional time series data, but also visualize and interactively process data analysis results by combining a particle swarm algorithm and an ant colony optimization algorithm, and fuse short-term and long-term user behavior preferences to output accurate user behavior prediction results, thereby overcoming the problems in existing methods of multidimensional data models processing interactive data, such as low quality of multi-source heterogeneous multidimensional data and inability to filter out multidimensional data noise, resulting in incomplete capture of effective information in multidimensional data and poor fusion effect.

[0077] The embodiment of the present invention provides a method for training a multidimensional data analysis model through model training samples. Figure 2 The schematic diagram of the implementation process of the method for training a multidimensional data analysis model through model training samples is shown. The method for training a multidimensional data analysis model through model training samples specifically includes:

[0078] Step S101, constructing a multidimensional data analysis model, using a convolutional neural network as the initial model of the multidimensional data analysis model, the initial model includes an input layer, a convolution module, a pooling layer and a fully connected layer, defining the loss function of the initial model as a SIou loss function, and setting the initial model iterative training rounds and hyperparameters;

[0079] It should be noted that the training rounds of the initial model are 200-500 times, the hyperparameter learning rate is 0.002, and the Adam optimizer is used to adaptively adjust the hyperparameters.

[0080] In this embodiment, the convolution module consists of a first convolution layer, a second convolution layer, and a third convolution layer. The first convolution layer is frozen and replaced by an iterative filter. The number of channels of the second convolution layer is 256, and the convolution kernel size is 3*3. The third convolution layer is a SimAM module, the number of channels of the third convolution layer is 512, and the convolution kernel size is 3*3. The activation function of the convolution module is a ReLU function. The pooling layer introduces the SimAM attention mechanism, and the pooling layer adopts maximum pooling. The local area of ​​the feature filtered by the convolution is taken as the maximum value to extract the most important feature information. The link discrimination network architecture is introduced into the initial model, and the improved whale algorithm is incorporated into the link discrimination network architecture. The link discrimination network architecture recommends links based on the feature weights of the training samples and the link response ratio discrimination data flow.

[0081] It should be noted that the input layer is connected to the convolution module, the convolution module is connected to the pooling layer, a link discrimination network architecture is set between the pooling layer and the fully connected layer, the link discrimination network architecture is connected to the fully connected layer, and the activation function of the fully connected layer can be a Sigmoid function, which is more sensitive to outliers (i.e., data points where the error between the predicted value and the actual value is much larger than other errors), thereby ensuring that the model will be more stable and accurate at the end of training.

[0082] Step S102, obtain model training samples, divide the model training samples into a training set and a validation set, input the training set, and perform filtering and denoising on the training set by the initial model to obtain a sample denoised set; in this embodiment, the ratio of the training set to the validation set can be 9:1.

[0083] Step S103, loading a sample denoising set, and iteratively training the initial model using the sample denoising set until convergence;

[0084] Step S104, obtaining a verification set, taking the verification set as input, executing the initial model, and outputting a verification result;

[0085] Step S105, judging whether the result error between the verification result and the actual result meets a preset accuracy threshold;

[0086] Step S106, if it meets the preset accuracy threshold, output a converged multidimensional data analysis model;

[0087] Step S107, if it does not meet the preset accuracy threshold, use the verification set combined with the fine-tuning mechanism to fine-tune the learning rate of the initial model, and return to step S103 to continue iterative training of the initial model until convergence.

[0088] In this embodiment, the preset accuracy threshold can be 0.9-0.95. After the model is trained to convergence, statistical analysis software can be used to perform leave-one-out cross-validation calculations on the model, and a cross-validation correlation coefficient of 0.876 is calculated, ensuring that the cross-validation correlation coefficient value is greater than 0.5, proving that the model has good robustness.

[0089] In an embodiment of the present invention, a multidimensional data analysis model and a training method thereof are provided. The multidimensional data analysis model is combined with a convolutional neural network, a SimAM attention mechanism, and an iterative filter, so that the multidimensional data analysis model has significant advantages in user behavior analysis, which can not only improve feature extraction and representation capabilities, improve model performance and generalization capabilities, but also can be combined with an improved whale algorithm and a link discrimination network architecture to achieve optimal allocation of data flow links. The link discrimination network architecture uses a deep learning model to perform real-time discrimination and optimization of data flow links. This architecture can make full use of the data features of the attention vector, weight inertia coefficient, sensitivity coefficient, and priority coefficient to achieve accurate prediction and adjustment of the link state. The link discrimination network architecture combined with the improved whale algorithm can find the optimal link allocation solution in a broader search space. For the optimal allocation of data flow links, it means that the optimal link can be found faster and the data transmission efficiency can be improved.

[0090] The embodiment of the present invention provides a method for analyzing and processing a time series data set based on edge computing using a multidimensional data analysis model. Figure 3 The schematic diagram of the implementation process of the multidimensional data analysis model based on edge computing for analyzing and processing a time series data set is shown. The method of analyzing and processing a time series data set based on edge computing by the multidimensional data analysis model specifically includes:

[0091] Step S201, obtaining a time series data set, performing dimensionality reduction processing on multi-dimensional time series data in the time series data set, and generating one-dimensional sequence data;

[0092] Among them, the one-dimensional sequence data is expressed as:

[0093]

[0094] Where X represents the input representation of multidimensional time series data, i is the dimension of the data, which can be 1-10, θ t Represents the weight parameter of the data. The weight parameter of the data is defined based on the Huber weight function. x(t) is the output representation of the one-dimensional sequence data.

[0095]

[0096] Among them, θ trepresents the weight parameter of the data, m is the number of data types, in this embodiment, the number of data types is 4-15, ε is the weight threshold preset in Huber regression, the weight threshold is set to 0.1-0.6, σ(X) is the residual of the multidimensional time series data, and X represents the mean of the multidimensional time series data;

[0097] In the embodiment of the present invention, by reducing the dimension of multi-dimensional time series data to generate one-dimensional sequence data, the problems of high computational complexity and low efficiency that may be faced by direct processing can be overcome. Through dimensionality reduction processing, data can be mapped from high-dimensional space to low-dimensional space, thereby reducing the dimension of the data and improving the efficiency of data processing and analysis. In the process of dimensionality reduction, the Huber weight function can be combined to ensure that the reduced-dimensional data can still reflect the main trends and characteristics of the original data.

[0098] Step S202, loading the one-dimensional sequence data, and performing Fourier transform processing on the one-dimensional sequence data based on the Blackman window to obtain a unified two-dimensional time-frequency spectrum diagram;

[0099] It should be noted that the Blackman window has certain filtering characteristics and can realize signal denoising while performing Fourier transform, which is very useful for signal analysis containing noise. It is suitable for a variety of application scenarios and can also be applied to multiple fields such as audio processing and image processing. Its wide application prospects make this method have high practical value.

[0100] Step S203, using an iterative filter to filter the two-dimensional time-frequency spectrum, outputting a filtered data set, and fusing the filtered data set with the third convolution layer to obtain a feature fusion set;

[0101] In this embodiment, when Fourier transform processing is performed on one-dimensional sequence data based on the Blackman window:

[0102]

[0103] S t =|Stft t (t,f s ) 2 (4)

[0104] Among them, S t is the output representation of the two-dimensional time-frequency spectrum, t is the sampling time, f s is the sampling frequency, x(t) represents one-dimensional sequence data, is the mean of the one-dimensional sequence data, B(x) represents the Blackman window function, Stft t (t,f s ) is a two-dimensional function of one-dimensional sequence data;

[0105]

[0106] Among them, x(t) represents one-dimensional sequence data, is the mean of the one-dimensional sequence data, l and α represent the length and shape parameters of the Blackman window respectively.

[0107] Step S204, loading the feature fusion set, the pooling layer simplifies the features in the feature fusion set based on the SimAM attention mechanism, uses the sensitivity analysis algorithm to consider the influence of the associated data associated with the filtered data, maps the filtered data in the feature fusion set to a low-dimensional potential attention space, and collects the attention vector from the potential attention space;

[0108] In this embodiment, the sensitivity analysis algorithm can be a global sensitivity analysis and a local sensitivity analysis algorithm. When the sensitivity analysis algorithm is used to consider the influence of the associated data associated with the filtered data, the global sensitivity analysis focuses on the influence of the associated data associated with the filtered data on the entire input space, while the local sensitivity analysis focuses on the influence of the associated data associated with the filtered data in a specific area. When the feature fusion concentrates the filtered data and maps it to a low-dimensional potential attention space, a deep learning model (such as an autoencoder, a variational autoencoder, etc.) can be used to map the fused features to a low-dimensional potential attention space. This potential space can capture the inherent structure and pattern in the data and provide support for subsequent attention mechanisms. In the mapping process, the model parameters can be optimized by minimizing the reconstruction error or maximizing the mutual information and other objective functions.

[0109] Step S205, use the attention vector to predict three sets of fusion properties of the filter data: weighted inertia coefficient, sensitivity coefficient, and priority coefficient, and output the weighted inertia coefficient, sensitivity coefficient, and priority coefficient of the filter data.

[0110] In an embodiment of the present invention, an attention vector is used to comprehensively evaluate user behavior, and the attention vector integrates the weight inertia coefficient, sensitivity coefficient, and priority coefficient of the filtered data. Through weighted fusion of the weight inertia coefficient, sensitivity coefficient, and priority coefficient, on the one hand, it can assist the particle swarm algorithm and the ant colony optimization algorithm in visualizing and interactively processing the data analysis results, and accurately realize the user behavior prediction results. On the other hand, it can also assist in screening the flow links that meet the clustering center, thereby providing data flow efficiency and processing speed through edge computing, and reducing the overall load of the system.

[0111] The iterative filter filters the two-dimensional time-frequency spectrum and obtains the filtered data set through the following formula:

[0112]

[0113] Among them, S L tis the output representation of the filtered data set, L is the number of filtering times of the iterative filter, G(w) represents the amplitude response of the iterative filter, arg(·) represents the argument, w(D) represents the iterative filter, and SL is the sector loss of the iterative filter;

[0114] The weight inertia coefficient, sensitivity coefficient, and priority coefficient are determined by the following formula:

[0115]

[0116]

[0117]

[0118] Among them, T R is the attention vector input representation, θ t ,q t , γ represent weight inertia coefficient, priority coefficient, and sensitivity coefficient respectively, β is the attention space mapping coefficient. In this embodiment, the attention space mapping coefficient can be 0.1-0.5, η is the sensitivity constant. In this embodiment, the sensitivity constant is 1-5, S1 and S2 are S L t The filtered data set is based on the first-order mapping coefficients and the second-order mapping coefficients of the sensitivity analysis algorithm.

[0119] In this embodiment, the first-order mapping coefficient and the second-order mapping coefficient are the first-order effect index and the second-order effect index based on the sensitivity analysis algorithm. The first-order mapping coefficient measures the contribution of a single parameter to the variance of the output variable. It reflects the proportion of the output variance change within a range of parameter changes. The second-order mapping coefficient is used to measure the contribution of two input variables and their combination to the variance of the output variable.

[0120] Step S206, loading the filter data, wherein the filter data includes an attention vector, a weight inertia coefficient, a sensitivity coefficient, and a priority coefficient, and the link discrimination network architecture calculates the link responsiveness of the filter data based on the improved whale algorithm, and sets at least one group of link responsiveness of the filter data as a cluster center;

[0121] It should be noted that the flow link usually consists of the previous hop node where the data comes from, the current node where the data is located, and the next hop node where the data is sent. During the data transmission process, these nodes together constitute a complete flow link, which is the entire transmission path from the source to the destination of the data. Its clarity is directly related to the transmission efficiency and security of the data.

[0122] Step S207, calculating the link response ratio of the flow link based on the improved whale algorithm, clustering the link response ratio using the fuzzy clustering algorithm, screening the flow link that meets the cluster center, integrating at least one group of flow links, and setting the flow link as the data flow recommended link;

[0123] The fuzzy clustering algorithm can be a fuzzy C-means (FCM) algorithm. When the link response ratio data is clustered by the fuzzy clustering algorithm, parameters such as the number of clusters, fuzzy factor, iterative error allowed, and maximum number of iterations are set. Through iterative calculation, the algorithm will output the membership matrix and cluster center of each data point (i.e., each link), and then determine the degree to which each link belongs to which cluster center based on the membership matrix. The higher the membership, the closer the link is to the corresponding cluster center. A threshold value (such as 0.8) is set, and links with a membership higher than the threshold value are considered to belong to the cluster center. This threshold value can be adjusted according to actual needs. Thus, the flow links that meet the specific cluster center are screened out.

[0124] Step S208: output the data analysis results and the recommended data flow link, and trigger the data interaction instruction.

[0125] The embodiment of the present invention provides a method for interactively processing data analysis results by combining a particle swarm algorithm and an ant colony optimization algorithm. Figure 4 The schematic diagram of the implementation process of the method for visualizing and interactively processing data analysis results by combining the particle swarm algorithm and the ant colony optimization algorithm is shown. The method for visualizing and interactively processing data analysis results by combining the particle swarm algorithm and the ant colony optimization algorithm specifically includes:

[0126] Step S301, obtaining data analysis results, initializing pheromones of the ant colony optimization algorithm, and generating initial positions of ants;

[0127] Step S302, integrating the pheromone and path length of the ant colony optimization algorithm to sort the ants, obtain the ranking of at least one group of the ants, and update the pheromone weights based on the ranking combined with the particle swarm optimization algorithm;

[0128] The weighted update of pheromones based on ranking combined with particle swarm optimization algorithm is expressed as:

[0129] ω i (t+1)=(1-ψ)ω i (t)+c1m1(L best ×Q+Δω i )+c2m2(L best ×Q-Δω i )(10)

[0130] Among them, ω i(t+1) represents the weighted update result of pheromone after t+1 weighting, ω i (t) represents the current pheromone input, ψ represents the ranking weight value. In this embodiment, the ranking weight value can be calculated by subjective weighting, objective weighting, and combined weighting. Q is the number of ants, Δω i Indicates the update of pheromone, L best represents the optimal path length, c1, c2 represent the particle swarm optimization acceleration factors, m1, m2 are random heuristic factors, and the value range of random heuristic factors is 0-1;

[0131] Step S303, using a spiral method to update the ant position. When the ant position is updated, the fitness function constraint is used to calculate the fitness value of the ant position, and the ant position corresponding to the maximum fitness value is recorded until the maximum number of ant iterations is reached;

[0132] Step S304: integrate at least one group of ant positions, output user behavior prediction results based on the ant positions, process the user behavior prediction results using a visualization tool, and output an interaction evaluation result.

[0133] It should be noted that the visualization tools can be Tableau, Power-BI, and Python libraries. Tableau provides a rich variety of chart types and interactive functions, allowing users to customize and explore data. Through Tableau, companies can create intuitive dashboards and reports that reflect user shopping behavior trends, key indicators, and potential problems.

[0134] In this embodiment, the particle swarm algorithm and the ant colony optimization algorithm are combined to perform visual interactive processing on the data analysis results, and the user behavior prediction results are output, which has the advantages of strong optimization ability, balance between global search and local development, high adaptability and flexibility, intuitive visual interactive processing, high prediction accuracy, real-time and dynamic update, etc. These advantages make this method have broad application prospects in many fields.

[0135] The embodiment of the present invention provides an interactive data processing system based on multidimensional data analysis. Figure 5 The structure diagram of the interactive data processing system based on multidimensional data analysis is shown, and the interactive data processing system based on multidimensional data analysis specifically includes:

[0136] The data collection module 100 collects multi-dimensional time series data related to user behavior in real time based on the Internet of Things gateway, integrates and cleans the multi-dimensional time series data, outputs a time series data set, and uploads the time series data set to an interactive database;

[0137] Model building module 200, the model building module 200 pre-builds a multidimensional data analysis model based on deep learning combined with an improved whale algorithm, traverses the interactive database, loads model training samples from the interactive database, trains the multidimensional data analysis model through the model training samples, and outputs a converged multidimensional data analysis model;

[0138] The multidimensional analysis module 300 is used to load time-series data sets in real time. The multidimensional data analysis model analyzes and processes the time-series data sets based on edge computing, outputs data analysis results and data flow recommendation links, triggers data interaction instructions, and interactively uploads data analysis results to the visualization interaction platform based on the flow recommendation links.

[0139] The interactive processing module 400 responds to the data interactive instruction. The visual interactive platform triggers the visual interactive logic based on the behavior analysis requirements, combines the particle swarm algorithm and the ant colony optimization algorithm to visually interactively process the data analysis results, and outputs the interactive evaluation results.

[0140] In this embodiment, the interactive transmission of data is realized by DTU or 5G communication between the data acquisition module 100, the model building module 200, the multidimensional analysis module 300, and the interactive processing module 400. It is understandable that the interactive data processing system based on multidimensional data analysis provided in the embodiment of the present invention corresponds to the interactive data processing method based on multidimensional data analysis, and the explanation, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the interactive data processing method based on multidimensional data analysis, which will not be repeated here.

[0141] On the other hand, an embodiment of the present invention further provides a computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and the computer program implementing any one of the above-mentioned methods when executed by the processor.

[0142] As a non-volatile computer-readable storage medium, the memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the program instructions / modules corresponding to the interactive data processing method based on multidimensional data analysis in the embodiment of the present application. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of the interactive data processing method based on multidimensional data analysis, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the local module via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0143] The processor executes various functional applications and data processing of the server by running non-volatile software programs, instructions and modules stored in the memory, that is, implements the interactive data processing method based on multidimensional data analysis of the above method embodiment.

[0144] In summary, the present invention provides an interactive data processing method based on multidimensional data analysis. In an embodiment of the present invention, a multidimensional data analysis model is constructed and trained based on deep learning combined with an improved whale algorithm. The multidimensional data analysis model can not only filter and fuse features of multi-source heterogeneous multidimensional time series data, but also visualize and interactively process data analysis results by combining a particle swarm algorithm and an ant colony optimization algorithm, and fuse user short-term and long-term behavior preferences to output accurate user behavior prediction results, thereby overcoming the problems in existing methods of multidimensional data models processing interactive data, such as low quality of multi-source heterogeneous multidimensional data and inability to filter out multidimensional data noise, resulting in incomplete capture of effective information in multidimensional data and poor fusion effect.

[0145] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0146] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. An interactive data processing method based on multidimensional data analysis, characterized in that: The interactive data processing method based on multidimensional data analysis includes: Based on the IoT gateway, multi-dimensional time series data related to user behavior is collected in real time, the multi-dimensional time series data is integrated, cleaned and processed, and the time series data set is output, and the time series data set is uploaded to the interactive database; Pre-build a multidimensional data analysis model based on deep learning combined with the improved whale algorithm, traverse the interactive database, load model training samples from the interactive database, train the multidimensional data analysis model through the model training samples, and output a converged multidimensional data analysis model; Real-time loading of time-series data sets. The multidimensional data analysis model analyzes and processes the time-series data sets based on edge computing, outputs data analysis results and data flow recommendation links, triggers data interaction instructions, and interactively uploads data analysis results to the visualization interaction platform based on the flow recommendation links.

2. The interactive data processing method based on multidimensional data analysis according to claim 1, characterized in that: The interactive data processing method based on multidimensional data analysis also includes: In response to data interaction instructions, the visualization interaction platform triggers the visualization interaction logic based on the behavior analysis requirements, combines the particle swarm algorithm and the ant colony optimization algorithm to visualize and interact with the data analysis results, and outputs the interaction evaluation results.

3. The interactive data processing method based on multidimensional data analysis according to claim 1, characterized in that: The method for training a multidimensional data analysis model through model training samples specifically includes: Construct a multidimensional data analysis model, using a convolutional neural network as the initial model of the multidimensional data analysis model. The initial model includes an input layer, a convolution module, a pooling layer, and a fully connected layer. Define the loss function of the initial model as the SIou loss function, and set the initial model iterative training rounds and hyperparameters. Obtain model training samples, divide the model training samples into a training set and a validation set, input the training set, and perform filtering and denoising on the training set by the initial model to obtain a sample denoised set; Load the sample denoising set, and use the sample denoising set to iteratively train the initial model until convergence; Obtain a validation set, use the validation set as input, execute the initial model, output the validation result, and judge whether the error between the validation result and the actual result meets the preset accuracy threshold; If the preset accuracy threshold is met, a converged multidimensional data analysis model is output; If the preset accuracy threshold is not met, the learning rate of the initial model is fine-tuned using the validation set combined with the fine-tuning mechanism, and the initial model is iteratively trained until convergence.

4. The interactive data processing method based on multidimensional data analysis according to claim 3, characterized in that: The convolution module consists of a first convolution layer, a second convolution layer, and a third convolution layer. The first convolution layer is frozen and replaced by an iterative filter. The number of channels of the second convolution layer is 256, and the convolution kernel size is 3*3. The third convolution layer is a SimAM module. The number of channels of the third convolution layer is 512, and the convolution kernel size is 3*3. The activation function of the convolution module is the ReLU function. The pooling layer introduces the SimAM attention mechanism. The pooling layer adopts maximum pooling, takes the maximum value of the local area of ​​the features filtered out by the convolution, and extracts the most important feature information. The link discrimination network architecture is introduced into the initial model, and the improved whale algorithm is incorporated into the link discrimination network architecture. The link discrimination network architecture recommends links based on the feature weights of the training samples and the link response ratio discrimination data flow.

5. The interactive data processing method based on multidimensional data analysis according to claim 4, characterized in that: The multidimensional data analysis model is based on the edge computing method for analyzing and processing time series data sets, specifically including: Obtain a time series data set, perform dimensionality reduction processing on the multi-dimensional time series data in the time series data set, and generate one-dimensional sequence data; Among them, the one-dimensional sequence data is expressed as: Among them, X represents the input representation of multidimensional time series data, i is the dimension of the data, and θ t Represents the weight parameter of the data. The weight parameter of the data is defined based on the Huber weight function. x(t) is the output representation of the one-dimensional sequence data. Among them, θ t represents the weight parameter of the data, m is the number of data types, ε is the preset weight threshold in Huber regression, σ(X) is the residual of multidimensional time series data, Represents the mean of multidimensional time series data; Load the one-dimensional sequence data, perform Fourier transform on the one-dimensional sequence data based on the Blackman window, and obtain a unified two-dimensional time-frequency spectrum; An iterative filter is used to filter the two-dimensional time-frequency spectrum and output a filtered data set. The third convolutional layer fuses the filtered data set to obtain a feature fusion set. Load the feature fusion set. The pooling layer simplifies the features in the feature fusion set based on the SimAM attention mechanism. The sensitivity analysis algorithm is used to consider the influence of the associated data associated with the filtered data. The filtered data in the feature fusion set is mapped to a low-dimensional latent attention space, and the attention vector is collected from the latent attention space. Use the attention vector to predict three sets of fusion properties of the filtered data: weighted inertia coefficient, sensitivity coefficient, and priority coefficient, and output the weighted inertia coefficient, sensitivity coefficient, and priority coefficient of the filtered data.

6. The interactive data processing method based on multidimensional data analysis according to claim 5, characterized in that: The multidimensional data analysis model is based on the edge computing method for analyzing and processing time series data sets, and specifically also includes: Loading filtered data, where the filtered data includes an attention vector, a weight inertia coefficient, a sensitivity coefficient, and a priority coefficient. The link discrimination network architecture calculates the link responsiveness of the filtered data based on an improved whale algorithm, and sets at least one set of link responsiveness of the filtered data as a cluster center. Based on the improved whale algorithm, the link response ratio of the flow link is calculated, and the link response ratio is clustered using the fuzzy clustering algorithm. The flow links that meet the cluster center are screened, at least one group of flow links are integrated, and the flow links are set as data flow recommended links. Output data analysis results and recommended data flow links, and trigger data interaction instructions.

7. The interactive data processing method based on multidimensional data analysis according to claim 5, characterized in that: When Fourier transform is performed on one-dimensional sequence data based on the Blackman window: S t =|Stft t (t,f s ) 2 (4) Among them, S t is the output representation of the two-dimensional time-frequency spectrum, t is the sampling time, f s is the sampling frequency, x(t) represents one-dimensional sequence data, is the mean of the one-dimensional sequence data, B(x) represents the Blackman window function, Stft t (t,f s ) is a two-dimensional function of one-dimensional sequence data; Among them, x(t) represents one-dimensional sequence data, is the mean of the one-dimensional sequence data, l and α represent the length and shape parameters of the Blackman window respectively.

8. The interactive data processing method based on multidimensional data analysis according to claim 7, characterized in that: The iterative filter filters the two-dimensional time-frequency spectrum and obtains the filtered data set through the following formula: Among them, S L t is the output representation of the filtered data set, L is the number of filtering times of the iterative filter, G(w) represents the amplitude response of the iterative filter, arg(·) represents the argument, w(D) represents the iterative filter, and SL is the sector loss of the iterative filter; The weight inertia coefficient, sensitivity coefficient, and priority coefficient are determined by the following formula: Among them, T R is the attention vector input representation, θ t ,q t , γ represents the weight inertia coefficient, priority coefficient, and sensitivity coefficient, β is the attention space mapping coefficient, η is the sensitivity constant, S1 and S2 are S L t The filtered data set is based on the first-order mapping coefficients and the second-order mapping coefficients of the sensitivity analysis algorithm.

9. The interactive data processing method based on multidimensional data analysis according to claim 2, characterized in that: The method for interactively processing the data analysis results by combining the particle swarm algorithm and the ant colony optimization algorithm specifically includes: Obtain data analysis results, initialize pheromones of the ant colony optimization algorithm, and generate the initial positions of ants; The ants are sorted by integrating the pheromone and path length of the ant colony optimization algorithm to obtain the ranking of at least one group of the ants, and the pheromone weighting is updated based on the ranking combined with the particle swarm optimization algorithm; The weighted update of pheromones based on ranking combined with particle swarm optimization algorithm is expressed as: oh i (t+1)=(1-ψ)ω i (t)+c1m1(L best ×Q+Give i )+c2m2(L best ×Q-Do i (10) Among them, ω i (t+1) represents the weighted update result of pheromone after t+1 weighting, ω i (t) is the current pheromone input, ψ is the ranking weight, Q is the number of ants, Δω i Indicates the update of pheromone, L best represents the optimal path length, c1, c2 represent the particle swarm optimization acceleration factors, and m1, m2 are random heuristic factors.

10. The interactive data processing method based on multidimensional data analysis according to claim 9, characterized in that: The method for interactively processing the data analysis results by combining the particle swarm algorithm and the ant colony optimization algorithm specifically includes: The ant positions are updated in a spiral manner. When the ant positions are updated, the fitness function constraint is used to calculate the fitness value of the ant positions, and the ant positions corresponding to the maximum fitness value are recorded until the maximum number of ant iterations is reached. Integrate at least one set of ant positions, output user behavior prediction results based on the ant positions, use visualization tools to process the user behavior prediction results, and output interaction evaluation results.

Images