Intelligent terminal application intelligent adaptation configuration method

Abstract

The application discloses a kind of intelligent terminal application intelligent adaptation configuration method, it is related to data acquisition control technical field, the specific steps of this method are as follows: terminal feature acquisition, adaptation benchmark construction, adaptation strategy generation, dynamic adaptation execution and adaptation optimization iteration;The benchmark library constructed by the pre-training adaptation configuration prediction model in combination with historical adaptation data in the application generates initial adaptation parameters, then the application interface is atomized and split, based on the function attributes of atomic component and the hard constraint conditions of intelligent terminal, a multi-objective adaptation evaluation function with layout and interaction conflict penalty term is constructed, the weights of different optimization objectives are dynamically adjusted in combination with scene attributes, fine adaptation of each component unit of application interface is realized, while the integrity of information display and the convenience of user operation are guaranteed, avoiding adaptation conflict problems such as component overlapping and shielding, input mode mismatching.

Description

Technical Field

[0001] This invention relates to the field of data acquisition and control technology, specifically to a smart terminal application intelligent adaptation and configuration method. Background Technology

[0002] With the rapid development of mobile internet and smart hardware technologies, smart terminals have become indispensable tools in people's daily lives and work. The application ecosystem has also experienced explosive growth, with smart terminal products of different brands and positioning constantly emerging, covering multiple categories such as smartphones, tablets, and smart wearable devices, meeting the diverse needs of different user groups. At the same time, users' requirements for application operation experience are also constantly improving. The stability of application operation, interface display effect, and operation smoothness on different terminals directly affect the user experience and the market competitiveness of the application. As a key link to ensure the normal operation of applications on different terminals, the quality and efficiency of application adaptation have become core issues that need to be focused on in the application development process.

[0003] However, existing application adaptation technologies mainly rely on developers manually testing, configuring, and packaging applications for mainstream terminal models in advance, or on the terminal system passively adapting the application. This approach has limited coverage and cannot cover all terminal models on the market, which can easily lead to application malfunctions on some niche terminals. The adaptation process often adopts a holistic approach, which cannot finely adjust the smallest components of the application interface, resulting in issues such as abnormal component display and inconvenient operation. At the same time, existing technologies cannot simultaneously accommodate multiple adaptation targets, nor can they dynamically adjust adaptation priorities according to the needs of different usage scenarios, making it difficult to meet the diverse needs of various usage scenarios. Furthermore, existing adaptation effect evaluations are mostly based on instantaneous running data, which cannot accurately reflect the continuous running status of the application, and lack an effective closed-loop optimization mechanism, so the adaptation capability cannot be continuously improved as the application is used. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent adaptation configuration method for smart terminal applications. This invention generates initial adaptation parameters by combining a pre-trained adaptation configuration prediction model with a benchmark library built from historical adaptation data. Then, the application interface is atomically decomposed. Based on the functional attributes of atomic components and the hard constraints of smart terminals, a multi-objective adaptation evaluation function with layout and interaction conflict penalty terms is constructed. The weights of different optimization objectives are dynamically adjusted in combination with scene attributes to achieve fine-grained adaptation of each component of the application interface. At the same time, it ensures the integrity of information display and the convenience of user operation, and avoids adaptation conflict problems such as component overlap and occlusion, and input method mismatch.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a smart terminal application intelligent adaptation configuration method, the specific steps of which are as follows: Terminal feature acquisition: Collect multi-dimensional features of smart terminals, and generate terminal feature vectors after normalization processing; Adaptation benchmark construction: retrieve the historical adaptation dataset of the target application, clean and filter it, build an adaptation configuration benchmark library, and extract the corresponding adaptation configuration parameter set; Adaptation strategy generation: Input the terminal feature vector into the pre-trained adaptation configuration prediction model to output the initial adaptation parameters. Combine the functional attributes of the atomic components of the application interface with the hard constraints of the smart terminal to construct a multi-objective adaptation evaluation function. Iteratively optimize the candidate configuration and dynamically adjust the weights in combination with the scene attributes to generate the optimal target adaptation strategy. Dynamic adaptation execution: The target adaptation strategy is sent to the smart terminal to drive the target application to complete the adaptation execution without being noticed, and the adaptation operation data is continuously collected; Adaptation optimization iteration: Calculate multi-dimensional adaptation effect evaluation values ​​based on continuously collected adaptation operation data, trigger abnormal parameter adjustments, and synchronously update the adaptation configuration benchmark library and incrementally train the pre-trained adaptation configuration prediction model with valid data.

[0006] Furthermore, in the terminal feature acquisition, the acquired multi-dimensional features include hardware features, system features, operating status features, and current scene features. Among them, hardware features include the number of processor cores, processor clock speed, memory capacity, storage capacity, screen resolution, and screen refresh rate; system features include operating system version, system kernel version, and installed operating environment version; operating status features include processor utilization rate, memory utilization rate, remaining storage capacity, and network bandwidth; and current scene features include current usage scenario, ambient brightness, network type, and scene urgency index.

[0007] Furthermore, in the construction of the adaptation benchmark, the historical adaptation dataset includes terminal feature vectors of the target application on different smart terminals, historical adaptation configuration parameters, adaptation operation data, adaptation effect feedback data, and comprehensive adaptation score. During the data filtering process, duplicate adaptation data is deleted, and invalid adaptation data with a comprehensive adaptation score lower than 0.6 is also deleted. After filtering, the data is classified and stored according to two dimensions: hardware configuration level and system version. The adaptation data under each category is sorted from high to low according to the comprehensive adaptation score, thus constructing a structured adaptation configuration benchmark library. When selecting the corresponding adaptation configuration parameter set from the structured adaptation configuration benchmark library, the similarity is calculated, and the adaptation configuration parameters corresponding to the terminal feature vectors in the top 10 similarity benchmarks are selected to form the adaptation configuration parameter set.

[0008] Furthermore, in the adaptation strategy generation, the pre-trained adaptation configuration prediction model adopts an improved lightweight Transformer encoder structure, including an input layer, encoder, hidden layer, and output layer. The pre-training dataset for the adaptation configuration prediction model is all valid historical adaptation data in the adaptation configuration benchmark library. The pre-training process uses the mean squared error loss function, with the terminal feature vector as input and the corresponding verified valid historical adaptation configuration parameters as labels to complete the training. After training, the underlying feature extraction network is frozen. The terminal feature vector is input into the pre-trained adaptation configuration prediction model, and the output includes initial adaptation configuration parameters including interface layout parameters, resource loading parameters, performance scheduling parameters, and functional compatibility parameters.

[0009] Furthermore, in the adaptation strategy generation, after outputting the initial adaptation configuration parameters, the interface of the target application is atomically decomposed, and all atomic components that make up the interface are identified. Atomic components include text components, icon components, button components, input box components, and multimedia components. The functional attributes, minimum displayable size, input method dependency, and information priority label of each atomic component are obtained one by one. The functional attributes are divided into display type, interaction type, and input type, and the information priority label is divided into level one to level five. At the same time, the visible area, input capability, and interaction modality of the current smart terminal are determined as hard constraints.

[0010] Furthermore, in the adaptation strategy generation, based on the functional attributes and hard constraints of atomic components, a multi-objective adaptation evaluation function with layout and interaction conflict penalties is constructed. The expression of the multi-objective adaptation evaluation function is as follows: ,in, The overall adaptation score for candidate configurations, with a value ranging from 0 to 1. Weighting coefficients for maintaining the goal of information integrity. The weighting coefficients for the operational efficiency objective are, and , Adjust dynamically based on the current urgency index of the scenario. To maintain a score for information integrity, the score is determined by the ratio of the number of first- and second-priority atomic components that are displayed normally under candidate configurations to the total number of first- and second-priority atomic components. The operational efficiency score is calculated by taking the normalized reciprocal of the average click path length of all interactive atomic components under the candidate configurations. The conflict penalty factor has a value range of 0-1. It is obtained by weighting and normalizing the number of atomic components with overlapping occlusion and the number of interactive atomic components with mismatched input methods under the candidate configuration. The value is 0 when there is no conflict.

[0011] Furthermore, in the adaptation strategy generation, the output initial adaptation configuration parameters and the adaptation configuration parameter set are used together as a candidate configuration set. Under hard constraints, the layout and interaction mapping of atomic components of each candidate configuration are iteratively adjusted. In each iteration, the size and position of the fourth and fifth priority atomic components are adjusted first, followed by the third priority atomic components. The first and second priority atomic components are only adjusted when the constraints cannot be met. After each adjustment, the comprehensive adaptation score is calculated through a multi-objective adaptation evaluation function. The weight coefficients are dynamically adjusted in combination with the current scene urgency index. After the iteration is completed, the configuration with the highest comprehensive adaptation score is selected from all the adjusted candidate configurations as the final target adaptation strategy.

[0012] Furthermore, in the dynamic adaptation execution, the target adaptation strategy is delivered using a differential delivery method. The target adaptation strategy is compared with the adaptation configuration parameters currently in effect on the smart terminal, and only the parameter content that differs is transmitted. After receiving the transmitted parameter content, the smart terminal completes the parameter replacement during the cold start or warm start phase of the target application. The parameter replacement process does not interrupt the operation of the target application. The continuous collection of adaptation running data is executed in the background using a low-priority thread, which does not occupy the main thread resources of the target application and does not affect the normal operation of the target application.

[0013] Furthermore, in the adaptation optimization iteration, a sliding evaluation time window is set to cover continuous adaptation operation data. The sliding evaluation time window slides forward over time. Based on the continuous adaptation operation data within the sliding evaluation time window, an adaptation effect evaluation function with time decay and fluctuation penalty is used to calculate the adaptation effect evaluation values ​​in four dimensions: interaction smoothness, visual comfort, power consumption, and thermal impact. The expression of the adaptation effect evaluation function with time decay and fluctuation penalty is as follows: ,in, This represents the adaptation effect evaluation value for the k-th dimension, ranging from 0 to 1. k is the dimension index, corresponding to the four dimensions: interaction smoothness, visual comfort, power consumption, and thermal impact. T is the total number of sampling points within the sliding evaluation time window, fixed at 30. t is the time index of the sampling point; t=1 represents the earliest sampling point within the sliding evaluation time window, and t=T represents the latest sampling point within the sliding evaluation time window. The time decay coefficient, fixed at 0.95, was obtained through regression analysis of historical adaptation operation data. The actual adaptation operation data for the k-th dimension collected at the t-th sampling point. The worst-case fit threshold corresponding to the k-th dimension. The optimal adaptation threshold corresponds to the k-th dimension. The data were determined by the 5th percentile and 95th percentile of the corresponding dimension of the adaptation and operation data of the same type of smart terminal in the adaptation configuration benchmark library. The fluctuation penalty coefficient is fixed at 0.3 and is determined by statistically analyzing the correlation between the fluctuation amplitude of historical adaptation operation data and the corresponding dimension adaptation effect evaluation value. The standard deviation of all adapted running data in the k-th dimension within the sliding evaluation time window.

[0014] Furthermore, in the adaptation optimization iteration, based on the adaptation effect evaluation values ​​of four dimensions, when the adaptation effect evaluation value of any dimension is lower than the corresponding preset qualified threshold, the abnormal adaptation parameters affecting that dimension are located through the correlation mapping relationship between the dimension and the adaptation parameters, and then the abnormal adaptation parameters are adjusted by gradient. When the adaptation effect evaluation values ​​of all dimensions are higher than the corresponding preset qualified threshold, the terminal feature vector, adaptation configuration parameters, adaptation running data, and comprehensive adaptation score corresponding to this adaptation are packaged into a valid sample and stored in the adaptation configuration benchmark library. At the same time, the valid sample is added to the incremental training sample set of the pre-trained adaptation configuration prediction model. During the training process, only the output layer parameters of the pre-trained adaptation configuration prediction model are updated.

[0015] Compared with existing technologies, this intelligent terminal application intelligent adaptation configuration method has the following beneficial effects: I. This invention generates initial adaptation parameters by combining a pre-trained adaptation configuration prediction model with a benchmark library built from historical adaptation data. Then, the application interface is atomically decomposed. Based on the functional attributes of atomic components and the hard constraints of smart terminals, a multi-objective adaptation evaluation function with layout and interaction conflict penalty terms is constructed. The weights of different optimization objectives are dynamically adjusted in combination with scene attributes to achieve fine-grained adaptation of each component of the application interface. At the same time, it ensures the integrity of information display and the convenience of user operation, and avoids adaptation conflict problems such as component overlap and occlusion, and input method mismatch.

[0016] Second, this invention sets a sliding evaluation time window and combines an adaptation effect evaluation function with time decay and fluctuation penalty to calculate multi-dimensional adaptation effect evaluation values ​​based on continuous adaptation operation data, accurately reflecting the continuous operation status of the application. Then, it locates abnormal parameters and performs gradient adjustment through the correlation mapping relationship between dimensions and adaptation parameters. At the same time, it synchronizes the verified and effective adaptation operation data to the adaptation configuration benchmark library and the pre-trained adaptation configuration prediction model, forming a complete closed-loop optimization mechanism, so that the adaptation capability can continuously evolve with the accumulation of data.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 A flowchart of a smart terminal application intelligent adaptation configuration method; Figure 2 A framework diagram for generating adaptation strategies in a smart terminal application intelligent adaptation configuration method; Figure 3 This is a framework diagram of the adaptation optimization iteration in an intelligent adaptation configuration method for smart terminal applications. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example: In the scenario of intelligent adaptation and configuration of social applications on Android smartphones, the entire process of intelligent adaptation and configuration of interface layout, interaction mapping, resource loading and performance scheduling of social applications is carried out for Android smartphones with different hardware configurations and system versions. This solves the problems of interface misalignment, inconvenient operation, running lag and excessive power consumption of social applications on heterogeneous Android terminals. This embodiment takes Android smartphones as the target smart terminal and mainstream social applications as the target applications.

[0022] Terminal Feature Acquisition: First, the terminal feature acquisition process is initiated. Multi-dimensional features of the target smart terminal are collected according to a preset execution frequency. These features include hardware features, system features, operating status features, and current scene features. Hardware features include the number of processor cores, processor clock speed, memory capacity, storage capacity, screen resolution, and screen refresh rate. System features include the operating system version, system kernel version, and operating environment version. Operating status features include processor utilization, memory utilization, remaining storage capacity, and network bandwidth. Current scene features include the current usage scenario, ambient brightness, network type, and scene urgency index. After acquisition, a min-max normalization method is used to map all feature values ​​to the 0-1 range. These features are then concatenated in a fixed order according to hardware features, system features, operating status features, and current scene features to generate a 28-dimensional terminal feature vector. Figure 1 As shown, the vector dimension and order remain fixed throughout the process, providing a unified feature input basis for subsequent processes.

[0023] Adaptation Benchmark Construction: Historical adaptation datasets since the target social application's launch are retrieved, containing five categories of data: terminal feature vectors for different Android devices, historical adaptation configuration parameters, adaptation runtime data, adaptation effect feedback data, and comprehensive adaptation scores. First, the historical adaptation dataset is cleaned and filtered, removing redundant data with completely identical terminal feature vectors, and then invalid adaptation data with a comprehensive adaptation score below 0.6. After filtering, the dataset is categorized and stored according to two dimensions: hardware configuration level and system version. Each category of data is sorted from highest to lowest comprehensive adaptation score, constructing a structured adaptation configuration benchmark library. Then, the similarity between the current terminal feature vector and the vectors in the adaptation configuration benchmark library is calculated based on the terminal feature similarity calculation formula with feature dimension weights. The terminal feature similarity calculation formula with feature dimension weights is as follows: ,in, The current terminal feature vector The similarity between the terminal feature vector Y and the configuration benchmark library is measured in terms of value, ranging from 0 to 1. The dimension of the terminal feature vector. Let i be the eigenvalue of the i-th dimension of the current terminal feature vector. To adapt to the eigenvalues ​​of the i-th dimension of the terminal feature vector in the configuration benchmark library, The weight coefficient of the i-th feature is 0-1, and the sum of the weight coefficients of all dimensions is 28. The weight coefficient is calculated by the random forest feature importance algorithm of historical adaptation data. The adaptation configuration parameters corresponding to the terminal feature vectors of the top 10 adaptation configuration benchmark library with the highest similarity are selected to form an adaptation configuration parameter set, which provides candidate parameter basis for the generation of adaptation strategy.

[0024] Adaptation strategy generation: The generated 28-dimensional terminal feature vector is input into the pre-trained adaptation configuration prediction model. The pre-trained adaptation configuration prediction model adopts an improved lightweight Transformer encoder structure, which consists of an input layer, encoder, hidden layer, and output layer connected sequentially. The dimension of the input layer perfectly matches the dimension of the terminal feature vector. The encoder sets a 3-layer multi-head attention mechanism to extract the correlation between features. The dimension of the hidden layer is set to 128 for feature transformation. The dimension of the output layer corresponds one-to-one with the total number of adaptation configuration parameters of the target social application. During the pre-training stage of the pre-trained adaptation configuration prediction model, all valid historical adaptation data in the adaptation configuration benchmark library are used as the training set, and the mean squared error loss function is used to calculate... The error between the predicted and actual values ​​is calculated, and the parameters of the pre-trained adaptive configuration prediction model are updated using the backpropagation algorithm. After training, the underlying feature extraction network of the pre-trained adaptive configuration prediction model is frozen, retaining only the updatable parameters of the output layer. After inputting the terminal feature vector into the pre-trained adaptive configuration prediction model, the model directly outputs initial adaptation configuration parameters, including four categories: interface layout parameters, resource loading parameters, performance scheduling parameters, and functional compatibility parameters. Interface layout parameters include component spacing, margins, and font size; resource loading parameters include image resolution and video bitrate; performance scheduling parameters include thread count and CPU frequency limits; and functional compatibility parameters... The parameters include special function switches, compatibility modes, etc. After obtaining the initial adaptation configuration parameters, all interfaces of the target social application are atomically decomposed. The decomposition process is based on the application's UI component tree structure, identifying each indivisible interface element as an atomic component, ultimately resulting in five types of atomic components: text components, icon components, button components, input box components, and multimedia components. Each identified atomic component is labeled with attributes, including functional attributes, minimum displayable size, input method dependency, and information priority. The functional attributes are labeled as display type, interaction type, or input type. The minimum displayable size is labeled as the minimum pixel size required for the component to display its content correctly. The input method dependency is labeled... For touch input, voice input, or peripheral input, information priority is labeled from level one to five, with level one being the highest priority and corresponding to components related to the core functions of the application, and level five being the lowest priority and corresponding to auxiliary information components. Simultaneously, the screen viewing area, supported input capabilities, and interaction modalities of the current smart terminal are extracted as hard constraints. These hard constraints are insurmountable adaptation boundaries, and all subsequent adaptation adjustments must meet these requirements. Based on the labeled atomic component attributes and the determined hard constraints, a multi-objective adaptation evaluation function with layout and interaction conflict penalties is constructed. This multi-objective adaptation evaluation function simultaneously optimizes information integrity and operational efficiency. The expression for the multi-objective adaptation evaluation function is as follows: ,in, The overall adaptation score for candidate configurations, with a value ranging from 0 to 1. Weighting coefficients for maintaining the goal of information integrity. The weighting coefficients for the operational efficiency objective are, and , Adjust dynamically based on the current urgency index of the scenario. To maintain a score for information integrity, the score is determined by the ratio of the number of first- and second-priority atomic components that are displayed normally under candidate configurations to the total number of first- and second-priority atomic components. The operational efficiency score is calculated by taking the normalized reciprocal of the average click path length of all interactive atomic components under the candidate configurations. The conflict penalty factor, ranging from 0 to 1, is obtained by weighting and normalizing the number of atomic components with overlapping occlusion and the number of interactive atomic components with mismatched input methods under the candidate configuration. It has a value of 0 when there are no conflicts. The initial adaptation configuration parameters and all parameters in the adaptation configuration parameter set are used as the candidate configuration set. Each candidate configuration undergoes a hard constraint check, and invalid candidate configurations that do not meet the hard constraints are eliminated. The remaining valid candidate configurations undergo iterative adjustments to the atomic component layout and interaction mapping. The iterative adjustment process follows a principle of low to high priority, with each iteration prioritizing the adjustment of the size of level 4 and 5 priority atomic components. After adjusting the position of the atomic components, the third-level priority atomic components are adjusted. First- and second-level priority atomic components are only adjusted if the hard constraints are still not met after all lower-priority components have been adjusted. The maximum number of iterations is set to 50. After each iteration, the comprehensive adaptation score of the candidate configuration is calculated using a multi-objective adaptation evaluation function. During the calculation, the weight coefficients of the two optimization objectives are dynamically adjusted based on the current scenario urgency index. The scenario urgency index is calculated by weighting the values ​​of the current scenario type, user operation frequency, and network status, with a value range of 0-1. When the scenario urgency index is higher than 0.7, the operation efficiency objective is prioritized. The value range is 0.3-0.4. The value range is 0.6-0.7. When the urgency index is below 0.3, the information integrity objective is prioritized. The value range is 0.6-0.7. The value range is 0.3-0.4. When the scenario urgency index is in the range of 0.3-0.7, the weight coefficients of both optimization objectives are set to 0.5. After the weight coefficients are adjusted, the comprehensive adaptation score of all candidate configurations is recalculated. After all candidate configurations are iteratively adjusted, the configuration with the highest comprehensive adaptation score is selected from all adjusted candidate configurations as the final target adaptation strategy, such as... Figure 2 As shown.

[0025] Dynamic Adaptation Execution: The target adaptation strategy is transmitted using a differential delivery method. First, the target adaptation strategy is compared with the currently effective adaptation configuration parameters of the smart terminal. The parameters that differ are extracted, encrypted, and compressed before transmission. Transmitting only the differing parameters can significantly reduce the amount of data transmitted and reduce network bandwidth usage. After receiving the differing parameters, the smart terminal completes the parameter replacement during the cold or warm start phase of the target social application. The parameter replacement process is executed in the application's background thread, without interrupting the normal operation of the application or requiring a restart of the smart terminal, achieving seamless adaptation execution. After the adaptation execution is completed, a continuous adaptation data collection process is initiated. The collection process is executed in a low-priority background thread, collecting adaptation operation data such as interface rendering frame rate, interface layout offset, function response time, processor utilization, memory utilization, overall power consumption, and device temperature. The collection thread does not occupy the application's main thread resources and does not affect the normal use of the application's social functions.

[0026] Adaptation optimization iteration: A sliding evaluation time window is set, covering the most recent 30 seconds of continuous adaptation operation data. The sliding evaluation time window slides forward over time, always retaining the latest operation data and discarding outdated historical data to avoid interference from transient abnormal data in the evaluation results. Then, based on the adaptation effect evaluation function with time decay and fluctuation penalty, the adaptation effect evaluation values ​​of four dimensions are calculated: interaction smoothness, visual comfort, power consumption, and thermal impact. The expression of the adaptation effect evaluation function with time decay and fluctuation penalty is: ,in, This represents the adaptation effect evaluation value for the k-th dimension, ranging from 0 to 1. k is the dimension index, corresponding to the four dimensions: interaction smoothness, visual comfort, power consumption, and thermal impact. T is the total number of sampling points within the sliding evaluation time window, fixed at 30. t is the time index of the sampling point; t=1 represents the earliest sampling point within the sliding evaluation time window, and t=T represents the latest sampling point within the sliding evaluation time window. The time decay coefficient, fixed at 0.95, was obtained through regression analysis of historical adaptation operation data. The actual adaptation operation data for the k-th dimension collected at the t-th sampling point. The worst-case fit threshold corresponding to the k-th dimension. The optimal adaptation threshold corresponds to the k-th dimension. The data were determined by the 5th percentile and 95th percentile of the corresponding dimension of the adaptation and operation data of the same type of smart terminal in the adaptation configuration benchmark library. The fluctuation penalty coefficient is fixed at 0.3 and is determined by statistically analyzing the correlation between the fluctuation amplitude of historical adaptation operation data and the corresponding dimension adaptation effect evaluation value. The standard deviation of all adaptation data for the k-th dimension within the sliding evaluation time window is defined as follows: The preset pass thresholds for each dimension are taken from the 25th percentile of the corresponding dimension scores of historical adaptation data with a comprehensive adaptation score higher than 0.8 in the adaptation configuration benchmark library. When the adaptation effect evaluation value of any dimension is lower than the corresponding preset pass threshold, the abnormal adaptation parameters affecting that dimension are located through the correlation mapping relationship between the dimension and the adaptation parameters. These abnormal adaptation parameters are then adjusted in a gradient manner, with each adjustment ranging from 5% to 15% of the current parameter value. The adjustment direction is determined based on the trend of the dimension evaluation value after the previous adjustment. After the adjustment is completed, the adaptation execution and effect evaluation are re-executed until the adaptation effect evaluation values ​​of all dimensions are higher than the corresponding preset pass thresholds. When the adaptation effect evaluation values ​​of all dimensions are higher than the corresponding preset qualified thresholds, the terminal feature vector, adaptation configuration parameters, adaptation running data, and comprehensive adaptation score corresponding to this adaptation are packaged into a valid sample and stored in the adaptation configuration benchmark library. At the same time, the valid sample is added to the incremental training sample set of the pre-trained adaptation configuration prediction model. Incremental training is performed once every certain number of newly added valid samples. The incremental training process only updates the output layer parameters of the pre-trained adaptation configuration prediction model and does not change the underlying feature extraction network of the pre-trained adaptation configuration prediction model. After training, the old output layer parameters of the pre-trained adaptation configuration prediction model are automatically replaced, so as to continuously improve the prediction accuracy of the pre-trained adaptation configuration prediction model. Figure 3 As shown.

[0027] In summary, in the context of social applications on Android smartphones, this approach achieves refined, automated, and continuously optimized adaptation configuration for target applications on heterogeneous terminals through five key steps: precise collection of terminal features, construction of a structured adaptation configuration benchmark library, generation of atomic component-level adaptation strategies, seamless dynamic adaptation execution, and multi-dimensional closed-loop iterative optimization. This completely solves the problems of coarse granularity, poor scenario adaptation, and lack of iterative capabilities in traditional adaptation methods, effectively improving the operational stability and user experience of applications on different terminals.

[0028] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent adaptation and configuration of smart terminal applications, characterized in that, The specific steps of this method are as follows: Terminal feature acquisition: Collect multi-dimensional features of smart terminals, and generate terminal feature vectors after normalization processing; Adaptation benchmark construction: retrieve the historical adaptation dataset of the target application, clean and filter it, build an adaptation configuration benchmark library, and extract the corresponding adaptation configuration parameter set; Adaptation strategy generation: Input the terminal feature vector into the pre-trained adaptation configuration prediction model to output the initial adaptation parameters. Combine the functional attributes of the atomic components of the application interface with the hard constraints of the smart terminal to construct a multi-objective adaptation evaluation function. Iteratively optimize the candidate configuration and dynamically adjust the weights in combination with the scene attributes to generate the optimal target adaptation strategy. Dynamic adaptation execution: The target adaptation strategy is sent to the smart terminal to drive the target application to complete the adaptation execution without being noticed, and the adaptation operation data is continuously collected; Adaptation optimization iteration: Calculate multi-dimensional adaptation effect evaluation values ​​based on continuously collected adaptation operation data, trigger abnormal parameter adjustments, and synchronously update the adaptation configuration benchmark library and incrementally train the pre-trained adaptation configuration prediction model with valid data.

2. The intelligent terminal application intelligent adaptation configuration method according to claim 1, characterized in that, The terminal feature collection includes multi-dimensional features such as hardware features, system features, operating status features, and current scene features. Hardware features include the number of processor cores, processor clock speed, memory capacity, storage capacity, screen resolution, and screen refresh rate. System features include the operating system version, system kernel version, and installed runtime environment version. Operating status features include processor utilization, memory utilization, remaining storage capacity, and network bandwidth. Current scene features include the current usage scenario, ambient brightness, network type, and scenario urgency index.

3. The intelligent adaptation and configuration method for smart terminal applications according to claim 1, characterized in that, In the construction of the adaptation benchmark, the historical adaptation dataset includes the terminal feature vectors of the target application on different smart terminals, historical adaptation configuration parameters, adaptation running data, adaptation effect feedback data, and comprehensive adaptation score. During the data filtering process, duplicate data is removed, and invalid adaptation data with a comprehensive adaptation score below 0.6 is also removed. After filtering, the data is classified and stored according to two dimensions: hardware configuration level and system version. The adaptation data under each category is sorted from high to low according to the comprehensive adaptation score, thus building a structured adaptation configuration benchmark library. When selecting the corresponding adaptation configuration parameter set from the structured adaptation configuration benchmark library, the similarity is calculated, and the adaptation configuration parameters corresponding to the terminal feature vectors in the adaptation configuration benchmark library with the top 10 similarity are selected to form the adaptation configuration parameter set.

4. The intelligent adaptation and configuration method for smart terminal applications according to claim 1, characterized in that, In the adaptation strategy generation, the pre-trained adaptation configuration prediction model adopts an improved lightweight Transformer encoder structure, including an input layer, encoder, hidden layer, and output layer; the pre-training dataset of the adaptation configuration prediction model is all valid historical adaptation data in the adaptation configuration benchmark library; the terminal feature vector is input into the pre-trained adaptation configuration prediction model, and the output includes initial adaptation configuration parameters including interface layout parameters, resource loading parameters, performance scheduling parameters, and functional compatibility parameters.

5. The intelligent terminal application intelligent adaptation configuration method according to claim 4, characterized in that, In the adaptation strategy generation process, after outputting the initial adaptation configuration parameters, the interface of the target application is atomically decomposed, all atomic components that make up the interface are identified, and the functional attributes, minimum displayable size, input method dependency, and information priority labeling of each atomic component are obtained one by one. The functional attributes are divided into display type, interaction type, and input type, and the information priority labeling is divided into level one to level five. At the same time, the visible area, input capability, and interaction modality of the current smart terminal are determined as hard constraints.

6. The intelligent adaptation configuration method for smart terminal applications according to claim 5, characterized in that, In the adaptation strategy generation, a multi-objective adaptation evaluation function with layout and interaction conflict penalties is constructed based on the functional attributes and hard constraints of atomic components. The expression of the multi-objective adaptation evaluation function is as follows: ,in, The overall adaptation score for candidate configurations, with a value ranging from 0 to 1. The weighting coefficients for maintaining the goal of information integrity. The weighting coefficients for the operational efficiency objective are, and , To maintain a score for information integrity, Score for operational efficiency. This is the conflict penalty factor, which takes a value of 0 when there are no conflicts.

7. The intelligent adaptation and configuration method for smart terminal applications according to claim 6, characterized in that, In the adaptation strategy generation process, the output initial adaptation configuration parameters and the adaptation configuration parameter set are used together as a candidate configuration set. Under hard constraints, the layout and interaction mapping of atomic components of each candidate configuration are iteratively adjusted. In each iteration, the size and position of the fourth and fifth priority atomic components are adjusted first, followed by the third priority atomic components. The first and second priority atomic components are only adjusted when the constraints cannot be met. After each adjustment, the comprehensive adaptation score is calculated through a multi-objective adaptation evaluation function. The weight coefficients are dynamically adjusted in combination with the current scene urgency index. After the iteration is completed, the configuration with the highest comprehensive adaptation score is selected from all the adjusted candidate configurations as the final target adaptation strategy.

8. The intelligent terminal application intelligent adaptation configuration method according to claim 1, characterized in that, In the dynamic adaptation execution, the target adaptation strategy is delivered using a differential delivery method. The target adaptation strategy is compared with the adaptation configuration parameters currently in effect on the smart terminal, and only the parameter content that differs is transmitted. After receiving the transmitted parameter content, the smart terminal completes the parameter replacement during the cold start or warm start phase of the target application. The parameter replacement process does not interrupt the operation of the target application. The continuous collection of adaptation running data is executed in the background using a low-priority thread, which does not occupy the main thread resources of the target application.

9. The intelligent adaptation configuration method for smart terminal applications according to claim 1, characterized in that, In the adaptation optimization iteration, a sliding evaluation time window is set to cover continuous adaptation operation data. The sliding evaluation time window slides forward over time. Based on the continuous adaptation operation data within the sliding evaluation time window, an adaptation effect evaluation function with time decay and fluctuation penalty is used to calculate the adaptation effect evaluation values ​​in four dimensions: interaction smoothness, visual comfort, power consumption, and thermal impact. The expression of the adaptation effect evaluation function with time decay and fluctuation penalty is as follows: ,in, Let be the adaptation performance evaluation value for the k-th dimension, where k is the dimension index, T is the total number of sampling points within the sliding evaluation time window, t is the time index of the sampling point, t=1 is the earliest sampling point within the sliding evaluation time window, and t=T is the latest sampling point within the sliding evaluation time window. The time decay coefficient, The actual adaptation and operation data for the k-th dimension collected at the t-th sampling point. The worst-case fit threshold corresponding to the k-th dimension. The optimal adaptation threshold corresponds to the k-th dimension. The fluctuation penalty coefficient, The standard deviation of all adapted running data in the k-th dimension within the sliding evaluation time window.

10. The intelligent terminal application intelligent adaptation configuration method according to claim 9, characterized in that, In the adaptation optimization iteration, based on the adaptation effect evaluation values ​​of four dimensions, when the adaptation effect evaluation value of any dimension is lower than the corresponding preset qualified threshold, the abnormal adaptation parameters affecting that dimension are located through the correlation mapping relationship between the dimension and the adaptation parameters, and then the abnormal adaptation parameters are adjusted by gradient. When the adaptation effect evaluation values ​​of all dimensions are higher than the corresponding preset qualified threshold, the terminal feature vector, adaptation configuration parameters, adaptation running data, and comprehensive adaptation score corresponding to this adaptation are packaged into a valid sample and stored in the adaptation configuration benchmark library. At the same time, the valid sample is added to the incremental training sample set of the pre-trained adaptation configuration prediction model. During the training process, only the output layer parameters of the pre-trained adaptation configuration prediction model are updated.

Images