Method and system for dynamic optimization of moving targets based on multi-source behavior fusion

Abstract

The application discloses a dynamic optimization method and system of a moving target based on multi-source behavior fusion, and belongs to the technical field of sports health management. The method steps comprise user terminal and wearable device access; multi-source behavior data collection; data preprocessing and feature fusion; user state feature vector construction; working mode determination and target initial generation; psychological constraint adjustment correction; sports target setting and game reward setting; user sports execution and data recovery; continuous behavior evaluation; model parameter adaptive update evaluation, forming a closed loop; and the system is used for executing the above method. The application can improve the reliability of multi-source data processing and the accuracy of target decision-making, realize dynamic, stable and individualized adjustment of the sports target, reduce the risk of target overload and sports interruption through psychological constraint and continuous behavior feedback, and thus improve the operation efficiency, decision stability and long-term use effect of the adaptive sports behavior guidance system.

Description

Technical Field

[0001] This invention belongs to the field of sports and health management technology, specifically relating to a method and system for dynamic optimization of sports goals based on multi-source behavior fusion. Background Technology

[0002] With the widespread adoption of wearable devices such as smartphones, smartwatches, and fitness trackers, sports and health apps are now able to collect data such as users' steps, exercise time, exercise distance, heart rate, and energy consumption, and encourage users to exercise through daily goals, points, badges, leaderboards, and message reminders.

[0003] In daily exercise management, step count is an easy-to-collect, easy-to-understand, and widely applicable metric. Therefore, many fitness and health apps set daily step count goals for users and provide feedback based on whether the user achieves the goal. Common goal-setting methods include setting fixed goals, user-defined goals, or simple adjustments based on recent average step counts.

[0004] However, a user's daily exercise capacity is not constant and is influenced by various factors such as historical exercise habits, study or work schedules, physical fatigue, emotional state, and recent goal achievement. A uniform or fixed exercise goal may be too high for some users, easily leading to frustration; or it may be too low for others, failing to provide effective motivation.

[0005] Some existing systems have begun to adjust exercise goals based on users' historical steps and goal completion rates, but they usually only focus on exercise data and do not adequately consider users' psychological feelings, emotional changes, and in-app interactions. At the same time, many systems only use two evaluation methods, "completed" or "not completed," which makes it difficult to accurately reflect different situations such as users being close to completion, making significant progress, or exceeding the goal.

[0006] Furthermore, existing fitness apps often use gamification methods such as points, levels, virtual rewards, and consecutive check-ins to increase user engagement. However, reward rules are mostly fixed in advance and difficult to dynamically adjust based on different users' actual responses. Data generated by users during use, such as page dwell time, number of task views, reward claim status, consecutive logins, and emotional records, are also usually not fully utilized for subsequent goal optimization.

[0007] Since step count data, goal completion data, sentiment data, and interaction behavior data may come from different functional modules, it is necessary to organize, analyze, and integrate these data in a unified manner in order to more comprehensively determine the user's current state.

[0008] Existing technologies already include sports and health management systems that generate personalized exercise plans based on user health data and exercise test results. These systems typically collect data such as basic health information, fitness test results, exercise goals, and exercise records to assess the user's physical condition and exercise ability, then generate corresponding exercise prescriptions or suggestions. During the user's exercise, data such as heart rate, movement quality, and exercise completion are collected, and finally, the exercise plan is adjusted based on the phased exercise results.

[0009] However, the existing technologies mentioned above have the following shortcomings: First, their core focus is usually on "exercise prescription generation" and "exercise risk management," mainly addressing questions such as what exercises users should do and what the exercise intensity should be, but they do not pay enough attention to users' long-term motivation to persist in exercising; Second, adjustments to exercise plans rely heavily on phased exercise results, resulting in long adjustment cycles and a single feedback dimension, making it difficult to dynamically optimize based on fine-grained behavioral data such as users' daily completion status, continuous login status, and task viewing frequency; Third, goal setting methods are too static or rule-based, lacking a dynamic decision-making mechanism oriented towards daily exercise goals; Fourth, an adaptive closed loop of "goal generation—user interaction—behavioral feedback—model optimization" has not yet been formed; Fifth, there is a lack of gamified interaction mechanisms deeply tied to exercise goals; Sixth, there is insufficient consideration for users' psychological state and behavioral feelings, and a lack of psychological constraint adjustment mechanisms.

[0010] Therefore, although existing technologies can achieve basic personalized exercise program recommendations and exercise feedback management, they still struggle to address issues such as insufficient dynamic adaptation to user exercise goals, inadequate behavioral incentives, untimely feedback optimization, insufficient consideration of psychological constraints, and poor long-term adherence results. Summary of the Invention

[0011] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for dynamic optimization of motion targets based on multi-source behavior fusion. It constructs an adaptive motion behavior guidance system based on multi-source behavior data fusion, dynamic target decision-making, psychological constraint adjustment, gamified interaction, and continuous behavioral feedback. This system uses user terminals or wearable devices as data entry points, user state characteristics as the decision-making basis, and dynamic motion targets as the core intervention means. Through continuous behavior evaluation and model parameter updates, a closed-loop optimization mechanism is formed, enabling the motion targets to be dynamically adjusted according to the user's actual exercise ability, historical performance, psychological state, and behavioral changes.

[0012] To achieve the above objectives, the present invention adopts the following technical solution:

[0013] The dynamic optimization method for moving targets based on multi-source behavior fusion includes the following steps:

[0014] S1, User terminal and wearable device access;

[0015] S2, Multi-source behavioral data acquisition;

[0016] S3, Data Preprocessing and Feature Fusion;

[0017] S4. Constructing user state feature vectors;

[0018] S5. Working Mode Determination and Initial Target Generation:

[0019] S6. Psychological constraint adjustment and correction;

[0020] S7, setting exercise goals and game reward settings;

[0021] S8, User-executed motion and data collection;

[0022] S9. Continuous behavioral evaluation;

[0023] S10, Adaptive updating of model parameters.

[0024] Furthermore, the multi-source behavioral data in step S2 includes, but is not limited to, user basic attribute data, motion behavior data, physiological state data, system interaction data, and user subjective psychological feedback data. All types of data can be collected at the time granularity of minutes, hours, or days, and associated with user identifiers and timestamps.

[0025] Furthermore, step S3 includes:

[0026] S3.1 Perform integrity checks and outlier identification on the collected multi-source behavioral data, and perform adaptive missing value completion on missing data;

[0027] S3.2 Perform quantile truncation filtering and normalization on the completed data;

[0028] S3.3 Extract statistical features from continuous data using a preset time window;

[0029] S3.4. Integrate the processed basic attribute features, motion behavior features, physiological state features, psychological feedback features, and system interaction features to form a unified set of user behavior features.

[0030] Furthermore, step S4 constructs a user state feature vector based on the user's current exercise ability, exercise trend, goal completion status, psychological state, and interactive behavior. This unifies the user's multi-source behavioral characteristics, recent exercise ability, exercise change trend, historical goal completion status, psychological state, fatigue risk, and interactive activity into the same state feature space. This enables the dynamic exercise goal decision model to adaptively select candidate exercise goals that match the user's current exercise ability and psychological endurance based on the user's state changes in different decision-making cycles.

[0031] Furthermore, step S5 determines the cold start mode or the dynamic target decision mode based on the number of valid historical periods for the user and the data integrity rate.

[0032] In the cold start mode, the basic activity level is calculated based on user basic information and prior weighted statistics of similar user groups. The initial exercise goal within the restricted range is generated by combining the exercise ability, health risk and psychological fear of difficulty score.

[0033] In the case of a dynamic goal decision-making mode, four types of candidate goals are generated based on the user's reference movement level: conservative, maintain, progressively improve, and challenge. A contextual feature is constructed that integrates user state, goal difficulty, and state-difficulty interaction items. Based on the contextual multi-armed slot machine model, the comprehensive score of each candidate goal is calculated, and the candidate with the highest score is selected as the initial dynamic goal.

[0034] Furthermore, step S6 includes:

[0035] S6.1 Calculate the probability of psychological risk based on the psychological risk feature vector;

[0036] S6.2. Based on the probability of psychological risk, revise the initial dynamic target;

[0037] S6.3 Calculate the range of target variation;

[0038] S6.4. Issue the final motion target based on the range of target changes.

[0039] Furthermore, step S7 includes:

[0040] S7.1 Break down a single sports objective into multiple phased tasks and set tiered rewards;

[0041] S7.2. Associate corresponding rewards based on the degree of goal completion.

[0042] S7.3 Record user clicks, claims, and dwell behaviors for different task formats and reward types as multi-source input data for the next cycle; transmit user interaction behaviors for tasks and rewards back to the continuous behavior evaluation module.

[0043] Furthermore, in step S8, the user carries out actual exercise based on the exercise task released by the system. During the exercise, the user's wearable device continuously collects multi-source data and uploads it to the system in real time or periodically. The collected data updates the task completion progress and provides appropriate prompts to the user. After the user completes the exercise, the system records the actual completion value, completion time, target completion rate, fluctuations in the exercise process, and the user's subjective feedback to form the exercise result data for the current decision cycle.

[0044] Furthermore, step S9 performs a non-binary, continuous evaluation based on the motion result data of multiple decision cycles collected in S8.

[0045] This application also discloses a dynamic optimization system for moving targets based on multi-source behavior fusion, used to perform the above method, specifically including:

[0046] The system includes a multi-source behavioral data acquisition module, a data preprocessing and feature fusion module, a user state feature construction module, a cold start target generation module, a dynamic target decision-making module, a psychological constraint adjustment module, a motion task and gamified interaction module, a continuous behavior evaluation module, a model parameter update module, and a user terminal and wearable acquisition module.

[0047] The output of the multi-source behavioral data acquisition module is connected to the data preprocessing and feature fusion module; the data preprocessing and feature fusion module is connected to the user state feature construction module; the user state feature construction module is connected to the cold start target generation module and the dynamic target decision module respectively; both the cold start target generation module and the dynamic target decision module output to the psychological constraint adjustment module; the psychological constraint adjustment module is connected to the motion task and gamification interaction module; the motion task and gamification interaction module is connected to the user terminal and the wearable acquisition subsystem; the user terminal and the wearable acquisition subsystem send back data to the multi-source behavioral data acquisition module; the user executes motion data input to the continuous behavior evaluation module; the continuous behavior evaluation module outputs to the model parameter update module; the model parameter update module feeds back parameters to the dynamic target decision module, forming a complete closed-loop link.

[0048] Compared with the prior art, the beneficial effects of this application are as follows:

[0049] 1. This application comprehensively utilizes multi-source data such as user basic attributes, historical exercise behavior, physiological state, psychological feedback and system interaction behavior to construct user state characteristics, thereby more comprehensively and accurately reflecting the user's current exercise ability, behavioral stability and goal tolerance, reducing goal deviation caused by a single data source or fixed rules, and improving the matching degree between exercise goals and the user's actual state.

[0050] 2. To address the issue of new users lacking historical activity data and making it difficult to make personalized decisions directly, this application sets up a cold start target generation module. Based on the user's basic attributes, initial activity level, and prior statistical information of similar user groups, it generates an initial target. This avoids the system randomly generating targets or directly using a uniform target value when there is insufficient data, thereby reducing the probability of the initial target being too high or too low and improving the stability and availability of the system during the cold start phase.

[0051] 3. This application constructs the motion goal generation process as a continuous decision-making process. Based on the user's state changes, goal completion status and behavioral feedback in different decision-making cycles, candidate goals are dynamically selected and updated. Compared with static motion goals that remain unchanged for a long time, this application can adjust the goals in a timely manner according to the user's improved exercise ability, state fluctuations, fatigue accumulation or changes in participation willingness, making the goal adjustment process more continuous, stable and adaptive.

[0052] 4. Based on the dynamic goal decision-making results, this application introduces a psychological constraint adjustment mechanism and sets constraints such as the target increase rate, target decrease rate, minimum safe target, and continuous failure response. When the system detects that the user has failed to meet the target for a long time, subjective fatigue has increased, exercise willingness has decreased, or negative feedback has accumulated, the system can lower the target, maintain the target, or postpone the target increase. This reduces the problem of target overload, user frustration, and exercise interruption caused by continuously increasing the target based solely on exercise data. At the same time, it reduces the execution instability caused by excessive changes in the target between adjacent periods.

[0053] 5. This application adopts a continuous behavior evaluation mechanism, which incorporates indicators such as target completion rate, behavior stability, continuous participation, interaction response and fatigue risk into the evaluation process. Through continuous evaluation, different levels of completion and different trends of change can be distinguished, and subtle differences in user behavior can be preserved. This provides more accurate and continuous feedback signals for the target decision model, thereby improving the accuracy of subsequent parameter updates and target selection.

[0054] 6. This application incrementally updates the benefit estimation, feature weights, exploration parameters, reward function weights, and psychological constraint thresholds in the target decision-making model through continuous behavior evaluation results. This enables the model to gradually learn its movement patterns and target acceptance range during continuous user use. Since the system does not need to retrain the complete model every time the target is adjusted, but instead updates the parameters online or periodically based on new behavior data, it can reduce redundant calculations, improve the model update speed and system operating efficiency. At the same time, by jointly judging historical data and the current state, it can reduce the probability of drastic changes in decision-making due to a single abnormal data, and improve the stability of the target generation process.

[0055] 7. This application further transforms dynamic exercise goals into phased, actionable exercise tasks, and provides feedback to users by combining progress display, tiered rewards, and gamified interaction mechanisms. Even if users do not fully complete the goal, they can still receive corresponding feedback based on their actual progress, avoiding the weakening of user motivation caused by traditional binary reward mechanisms. This reduces the perceived pressure on users regarding exercise goals, improves task comprehension and execution willingness, reduces exercise interruptions caused by consecutive failures, and enhances the stability of users' long-term participation in exercise tasks.

[0056] 8. This application forms a closed-loop technical process of "data collection - feature fusion - state construction - goal generation - psychological adjustment - task execution - behavior evaluation - parameter update". After each decision cycle is completed by the user, the system can revise the subsequent goal strategy based on the new data, so that the goal generation gradually transitions from the initial group prior to individualized decision-making. As the usage time increases, the system can continuously identify the user's changes in exercise ability, behavior fluctuation patterns, psychological tolerance range and reward preferences, so that the subsequent exercise goals are more in line with the user's actual situation, avoiding the problem of traditional system goal strategies being fixed for a long time and difficult to adapt to user changes;

[0057] 9. This application reduces the reliance on manual intervention in the process of setting and adjusting exercise goals by automating data collection, user status analysis, motion goal generation, and model parameter updates. As the number of users increases, the system can simultaneously process data and goal decision-making tasks for multiple users based on a unified modular process, reducing the costs of manual evaluation, scheme modification, and long-term tracking, and improving the system's deployment efficiency and scalability in universities, communities, and health management platforms. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the steps of the dynamic optimization method for moving targets based on multi-source behavior fusion according to the present invention;

[0059] Figure 2 This is a schematic diagram illustrating the specific process of the dynamic optimization method for moving targets based on multi-source behavior fusion according to the present invention;

[0060] Figure 3 This is a schematic diagram of the module connection structure of the dynamic optimization system for moving targets based on multi-source behavior fusion according to the present invention. Detailed Implementation

[0061] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0062] like Figure 1 As shown, the dynamic optimization method for moving targets based on multi-source behavior fusion includes the following steps:

[0063] S1. User terminal and wearable device access:

[0064] Users access the system via smartphones, tablets, smartwatches, smart bracelets, or other terminal devices with motion data collection capabilities. User terminals are used for user registration, basic information entry, viewing of exercise goals, task execution, gamified interaction, psychological state feedback, and display of exercise results; wearable devices are used to collect users' physiological and exercise behavior data during daily activities and exercise.

[0065] When a user uses the system for the first time, they enter basic information such as age, gender, height, weight, occupation type, daily activity level, exercise history, exercise preferences, health risk information, and periodic exercise needs. The system also requests necessary data access authorization from the user and establishes an individual exercise behavior profile corresponding to the user's unique identifier.

[0066] S2. Multi-source behavioral data acquisition:

[0067] The multi-source behavioral data acquisition module acquires multi-dimensional data related to user movement behavior from user terminals, wearable devices, and system interaction records according to a preset data acquisition cycle.

[0068] The collected data mainly includes:

[0069] I. Basic attribute data, including age, body mass index, occupation type, exercise experience, exercise preferences, and health risk level;

[0070] II. Exercise behavior data, including daily steps, exercise duration, exercise frequency, exercise distance, activity intensity, sedentary time, goal completion rate, and number of consecutive exercise days;

[0071] III. Physiological status data, including heart rate, resting heart rate, sleep duration, sleep quality, energy consumption, and fatigue level;

[0072] IV. System interaction data, including the number of times tasks are viewed, page dwell time, task response latency, reward collection frequency, virtual resource redemption frequency, consecutive login interval, and gamification function usage frequency, etc.

[0073] V. User-initiated feedback data, including subjective fatigue, perceived exercise difficulty, willingness to exercise, emotional state, goal acceptance, and task satisfaction.

[0074] Various types of data can be collected at the time granularity of minutes, hours, or days, and associated with user identifiers and timestamps to provide basic data for subsequent data fusion and state modeling.

[0075] The set of available data for a user before the start of the t-th decision period is represented as follows: ;

[0076] In the formula, Indicates user In the The set of historical data that has been generated before the start of each decision cycle.

[0077] S3. Data Preprocessing and Feature Fusion: This involves cleaning, validating, standardizing, and fusing the collected multi-source heterogeneous data. The steps include:

[0078] S3.1 Perform integrity checks and outlier identification on the collected multi-source behavioral data. For missing data caused by device disconnection, sensor error or user not wearing the device, the data can be processed by interpolation of nearby time data, historical mean completion, sliding window estimation or missing state marking. For data that obviously exceeds the reasonable physiological range or movement range, the data can be removed or corrected according to the preset threshold.

[0079] S3.1.1 Data Completion and Validity Check:

[0080] For users In the The first decision cycle The original features, with missing value imputation results, are represented as follows:

[0081] ,

[0082] in:

[0083] ;

[0084] In the formula:

[0085] This represents the feature value after missing values ​​have been filled in; Represents the original eigenvalues; This represents a data validity indicator variable; its value is [value] when the original data is valid. When missing, the value is... ; This represents the corresponding feature value of the user's previous effective decision-making cycle; Indicates to users Among historical user groups with similar characteristics, the first Statistical values ​​of each feature; represents the individual historical information credibility weight determined based on the user's valid historical data volume, which is used to adjust the relative contribution of the user's own historical features and the statistical features of the similar user group in the missing value imputation process; represents the number of valid historical cycles of the user before the th decision cycle; represents the equivalent sample size of the group prior.

[0086] When the user's historical data is scarce, is relatively small, and the imputation result is mainly determined by the statistical value of the similar user group; as the user's historical data increases, gradually increases, and the imputation result relies more on the user's own historical data. The parameter is determined according to the verification of offline historical data. In one embodiment, can be set as the minimum number of valid cycles required in the cold start phase.

[0087] S3.2. Perform quantile truncation filtering and normalization processing on the imputed data:

[0088] ,

[0089] where:

[0090] ;

[0091] In the formula:

[0092] represents the normalized feature value; and respectively represent the lower quantile threshold and the upper quantile threshold of the th feature; represents a preset small positive number for preventing the denominator from being zero. represents a truncation function, which is used to limit the input value x between the lower limit L and the upper limit U; when x<L, the value is L; when x>U, the value is U; when x is within the interval [L, U], the original value remains unchanged. This function is used to reduce the influence of outliers or extreme values on the subsequent feature normalization results, and improve the stability and robustness of the data preprocessing process.

[0093] The quantile threshold is obtained by statistics from historical valid data, and is determined in groups according to the user's age, occupational activity level or health risk level.

[0094] S3.3. Statistical feature extraction is performed on continuous data within a preset time window to form feature indicators reflecting short-term and long-term user behavior. These feature indicators may include the average number of steps taken over several days, step count trends, average target completion rate, completion rate fluctuations, number of consecutive days meeting the target, probability of exercise interruption, average response time, and trends in psychological state.

[0095] The data completeness rate for the user's current decision-making cycle is represented as:

[0096] ;

[0097] In the formula:

[0098] This indicates the number of valid features among the features required to construct the current state;

[0099] This indicates the total number of features required.

[0100] when Below the preset integrity rate threshold When this happens, the system will not improve the motion target based on the current incomplete data, and will suspend the use of the data from that period to update the model.

[0101] S3.4. Integrate the processed basic attribute features, motion behavior features, physiological state features, psychological feedback features, and system interaction features to form a unified set of user behavior features.

[0102] S4. User State Feature Vector Construction: After completing the preprocessing and feature fusion of multi-source behavioral data, a user state feature vector is constructed based on the user's current exercise ability, exercise trend, goal completion status, psychological state, and interactive behavior. The user's multi-source behavioral characteristics, recent exercise ability, exercise change trend, historical goal completion status, psychological state, fatigue risk, and interactive activity are uniformly mapped to the same state feature space, so that the dynamic exercise goal decision model can adaptively select candidate exercise goals that match the user's current exercise ability and psychological endurance based on the user's state changes in different decision-making cycles.

[0103] Among them, the estimation of athletic ability:

[0104] The system based on the user's recent Estimating current athletic ability based on actual exercise volume over one effective cycle:

[0105] ;

[0106] In the formula:

[0107] This represents an estimate of the user's current physical ability. Indicates the user's position in the first month. The actual amount of exercise within each effective cycle; Indicates the length of the historical time window used for motion capability estimation; This represents the median function. Using the median can reduce the impact of abnormal daily activity levels on capability estimation results.

[0108] When there is insufficient valid historical data For each cycle, all existing valid cycle data are used; when no valid historical data exists, the basic activity level determined by the cold start module is used.

[0109] Based on athletic ability,

[0110] User at The reference motion level within each decision cycle is represented as follows:

[0111] ;

[0112] In the formula:

[0113] , This indicates the level of user activity determined during the cold start phase.

[0114] Trends in motion

[0115] The recent trend of a user's exercise capacity can be represented by the linear slope of historical exercise volume:

[0116] .

[0117] In the formula:

[0118] This represents the recent trend of change in athletic ability; i represents the period number within the historical time window, with values ​​ranging from 1, 2, ..., L. The larger the number, the closer the corresponding historical period is to the current decision period; L represents the length of the historical time window used to calculate the trend of change in athletic ability, i.e., the number of historical decision periods included in the calculation. This represents the average of historical period numbers; This represents the average actual amount of exercise within the historical time window. When, it indicates that the user's exercise level is on the rise; when When this occurs, it indicates that the user's activity level is declining.

[0119] State feature vector

[0120] The user state feature vector is represented as:

[0121] .

[0122] In the formula: , Indicates the recent trend in athletic ability;

[0123] Represents the user's state feature vector within the current decision-making cycle; constant Indicates the bias term; This represents a preprocessed and normalized multi-source behavioral feature vector; This indicates the target completion rate of the previous decision-making cycle; This indicates the degree of fluctuation in the target completion rate over the past few periods; This indicates the currently available emotional state or willingness to exercise. This indicates the current level of fatigue or a risk score related to exercise. This indicates the recent activity level of the application. This indicates the data integrity rate.

[0124] The emotional state, fatigue level, and interaction activity were all measured using the [missing information]. Data collected before the start of each decision cycle.

[0125] S5. Working mode determination and initial target generation: Determine the cold start mode or dynamic target decision mode based on the number of valid historical periods of the user and the data integrity rate.

[0126] Specifically, the system determines the target generation mode based on the number of valid historical periods for the user and the data integrity rate:

[0127] ;

[0128] In the formula:

[0129] This indicates that a cold start target generation mode is used; This indicates the adoption of a dynamic goal-oriented decision-making model; This represents the number of valid historical periods accumulated by user u before the t-th decision period, where a valid historical period refers to a historical decision period that meets the preset data integrity and validity requirements. This represents the minimum number of valid historical periods required to enter dynamic target decision-making; This represents the minimum data integrity threshold.

[0130] In cold start mode, when a user uses the system for the first time or when historical behavior data is insufficient, the cold start target generation module generates an initial movement target based on the user's basic information, initial movement ability, and prior group data.

[0131] The cold start phase does not directly use a fixed step threshold. Instead, it makes a comprehensive judgment based on the user's age, body mass index, occupational activity level, past exercise habits, health risks, and self-reported goals. The system can extract statistical features from historical user groups with similar characteristics to the current user and combine them with robust means, medians, quantiles, or confidence intervals to determine the initial target range.

[0132] For example, the system can first determine the user's basic activity level, and then set a predetermined percentage increment based on that level to obtain the step count or exercise duration target for the first decision cycle. For users with weaker exercise foundations or higher psychological pressure, the initial target increment should be relatively low; for users with stronger exercise abilities and a clear need for improvement, the initial target can be appropriately increased.

[0133] Cold start target generation:

[0134] Basic activity level calculation,

[0135] User activity levels are represented as follows:

[0136] ,

[0137] in:

[0138] ;

[0139] In the formula:

[0140] Indicates the user's basic activity level; This represents the median activity level of a historical user group with similar user characteristics. This indicates an individual's activity level based on initial user testing, short-term device data, or user-reported data. This represents the number of valid historical samples of individuals that user u has accumulated before the t-th decision period; Represents the group prior weights. This represents the prior strength parameter.

[0141] When a user has no personal historical data near It primarily uses statistical results from similar user groups; as personal data increases, Gradually decreasing, the level of basic activity gradually shifts towards users' own data.

[0142] Cold start target calculation;

[0143] The cold start objective is represented as:

[0144] ;

[0145] In the formula:

[0146] This indicates a range truncation function, used to restrict the calculated base target to a preset reasonable target range, in order to avoid individual adjustments causing the target value to exceed the reasonable activity level range of similar user groups. Indicates the increment of the basic target; This indicates the user's initial athletic ability score; This indicates the user's goal of improving their needs score; Indicates health risk score; This indicates a score reflecting psychological stress or difficulty with exercise. , , and This indicates the corresponding adjustment weight; and These represent the lower and upper limits of the initial goals for similar user groups, respectively.

[0147] The adjustment weights are determined using historical cold-start user samples. Specifically, the completion rate, psychological acceptance, and fatigue feedback after the first execution of the objective are used as training results, and the weights for each item are determined through ridge regression or grid search with non-negative constraints.

[0148] When there are insufficient historical samples, the initial weights can be set by experts in the field of sports and health, and then recalibrated using historical data after a sufficient number of samples have been accumulated.

[0149] The cold start target generation module can avoid setting overly high or low motion targets due to a lack of historical data, thus reducing the risk of frustration for users during their first participation.

[0150] During the cold start phase, the system continues to record the actual release goals, user execution results, and continuous behavior evaluation values, and updates the historical matrix of the dynamic decision-making model. When the dynamic mode switching conditions are met, the system directly uses the accumulated data to enter the dynamic goal decision-making process, rather than re-initializing the model, thereby achieving a continuous transition from group priors to individual models.

[0151] In the dynamic goal decision-making mode, after obtaining certain historical behavioral data, the system enters the dynamic goal decision-making stage. The dynamic goal decision-making module selects a suitable movement goal for the current user from the candidate goal set based on the user's current state characteristics, historical goal completion results, and the expected benefits of different goal schemes.

[0152] The system can build multiple candidate goals based on the user's recent average exercise level, such as conservative goals, maintenance goals, progressive improvement goals, and challenging goals.

[0153] The system utilizes a contextual multi-armed slot machine model to estimate the expected returns of different candidate targets based on user state characteristics, and balances between "utilizing the currently known best target" and "exploring potential better targets".

[0154] The benefits considered in goal-oriented decision-making include not only whether the user completes the goal, but also the degree of completion, the stability of completion, the exercise burden, psychological acceptance, willingness to participate continuously, and the level of gamified interaction. The system ultimately selects candidate goals with higher expected overall benefits to form the initial dynamic exercise goals.

[0155] During the target adjustment process, the system also sets the maximum growth rate, the maximum decline rate, and the minimum safe target to avoid drastic changes in the target within adjacent decision-making cycles.

[0156] Dynamic candidate target generation;

[0157] when At this point, the system enters the dynamic goal decision-making stage.

[0158] The system constructs a set of candidate targets based on the user's reference motion level:

[0159] ,

[0160] in: .

[0161] The corresponding target adjustment ratio can be expressed as: ;

[0162] In the formula:

[0163] Indicates candidate actions Corresponding candidate motion targets; This indicates that the integer part is rounded according to the preset target granularity. Indicates user The reference exercise level in the t-th decision period is used as the benchmark for generating candidate exercise goals. It can be determined based on the user's recent actual exercise volume, estimated exercise capacity, and historical goal completion status. Indicates a conservative target; Indicates a maintenance objective; This indicates a progressively improving goal; This indicates a challenging goal. The goal adjustment ratio is not arbitrarily set, but is obtained statistically from historical data of similar users. Represents the candidate target space, the The median increase in the target success rate of similar users under conditions of no significant fatigue, consecutive failures, or decline in participation can be used to determine the target. The higher quantile of the increase in the successful target can be used to determine it; The median percentage decrease in the target that allows participants to resume normal participation after a series of unfulfilled targets can be used to determine the target.

[0164] User candidate target context construction;

[0165] The standardized difficulty of the candidate target relative to the user's current motor ability is represented as:

[0166] ,

[0167] The user's state features, the difficulty of the candidate target, and the interaction between the two are combined into a contextual feature vector:

[0168] ;

[0169] In the formula:

[0170] Indicates user In the Within each decision cycle, for candidate targets Constructed context feature vector; Indicates the user's current state; This indicates the standardization difficulty of the candidate target relative to the user's current capabilities; This represents the interaction between the difficulty of the candidate target and the user's state; the superscript T indicates the transpose of a vector or matrix. Represents user state feature vector The transpose of is used to concatenate with the candidate target difficulty and interaction features; the outermost transpose symbol T is used to convert the concatenated row vector into a column-form context feature vector. .

[0171] By introducing interactive terms, the same level of difficulty can yield different benefit estimates under different fatigue states, emotional states, movement trends, and historical completion conditions.

[0172] Initialize the group prior model;

[0173] The system first uses historical user samples to calculate the population model parameters:

[0174] ;

[0175] In the formula:

[0176] This represents the prior parameters of the group obtained based on historical user data; Represents the vector of the model to be optimized; This indicates the search for model parameters that minimize the objective function. Indicates the sequence number of the historical sample; This indicates that the calculation starts from the first historical sample; This represents the total number of historical samples used for training the ensemble model; This represents the actual continuous behavior evaluation value corresponding to the nth historical sample; Indicates the first Context feature vectors of historical samples; Indicates the first Context feature vector of a historical sample transpose; Represents the vector of the model to be optimized; This represents the continuous behavioral evaluation prediction value obtained by the model based on the features of the nth historical sample; Represents the regularization coefficient; This represents the squared L2 norm of the model parameter vector, used to limit the magnitude of the model parameters and reduce the risk of overfitting.

[0177] The user model is initialized as follows:

[0178] ,

[0179] ,

[0180] ;

[0181] In the formula:

[0182] Represents the initial feature correlation matrix for user u; This represents the identity matrix corresponding to the context feature dimension; The regularization coefficient indicates that it is used to control the constraint strength of the group's prior information during the model initialization phase, and to ensure that the matrix... reversible; This represents the group prior parameter vector obtained by pre-training based on historical user samples; This represents the parameter estimation vector for user u during the model initialization phase; This represents the cumulative vector of revenue-related factors during the initialization of the user model.

[0183] Therefore, when individual user data is insufficient, the model parameters are close to the group prior; as personal data accumulates, the model gradually shifts towards individualized parameters.

[0184] Exploration coefficient;

[0185] The exploration coefficient for the user's current decision-making cycle is expressed as:

[0186] ;

[0187] In the formula:

[0188] This represents the exploration coefficient used by the user in the t-th decision-making cycle; Indicates the initial exploration coefficient; Indicates the lower limit of the exploration coefficient; This indicates the number of valid feedback cycles that a user has accumulated.

[0189] When there is limited user historical data, the exploration coefficient is relatively high; as the amount of valid historical data increases, the exploration coefficient gradually decreases, but remains no lower than [a certain value]. .

[0190] and Through offline playback experiments using historical data, we determined the optimal parameter combination that yielded the best results among cumulative behavioral evaluation values, target completion rate, and fatigue risk.

[0191] Candidate target selection;

[0192] The system calculates the overall decision score for each candidate objective:

[0193] ,

[0194] And select the candidate target with the highest score:

[0195] ,

[0196] The initial dynamic objectives are:

[0197] ;

[0198] in:

[0199] This represents the expected overall benefit of the candidate target under the current user state; This represents compensation for the uncertainty of the candidate target; This represents a feature relevance matrix formed based on historical context features; Indicates the user's current model parameters; This indicates the search for model parameters that maximize the objective function; Indicates the current set of candidate targets In the process, we seek to improve the overall decision-making score. ; This indicates the optimal candidate target selected accordingly; This represents the initial dynamic moving target selected by the dynamic target decision model.

[0200] S6. Psychological Constraint Adjustment and Correction: The target output by the dynamic target decision module is further input into the psychological constraint adjustment module. This module is used to combine the purely data-driven target decision results with the user's psychological state, subjective experience, and behavioral tolerance, avoiding the psychological pressure caused by the system continuously raising the target based solely on motion data.

[0201] The psychological constraint adjustment module focuses on analyzing the user's recent subjective fatigue level, target difficulty evaluation, number of consecutive failures, emotional changes, task response delay, and risk of exercise interruption. When the system detects that the user has repeatedly failed to meet the target, experienced a significant increase in subjective fatigue, a decrease in exercise willingness, or accumulated negative emotions, it can lower, maintain, or temporarily postpone the increase of the target.

[0202] For example, if a user fails to complete the goal for several consecutive periods, the system will not continue to increase the goal, but will appropriately reduce the goal difficulty; if the user does not fully achieve the goal, but the completion rate continues to improve and the psychological feedback is good, the system can keep the goal unchanged; if the user completes the goal continuously and the subjective difficulty is low, the system allows the goal to be gradually increased within a preset range.

[0203] After psychological adjustment, the final exercise goal is obtained and presented to the user. This goal balances the effectiveness of exercise promotion with the user's psychological acceptability.

[0204] S6.1 Calculate the probability of psychological risk based on the psychological risk feature vector;

[0205] The system constructs psychological risk characteristics based on the user's current fatigue level, emotional state, consecutive unfulfilled goals, task response latency, and recent interaction activity. Specifically, fatigue level reflects the user's current physiological and subjective workload; emotional state reflects the user's psychological acceptance of the current exercise task, converting positive emotional levels into corresponding negative risk representations; consecutive unfulfilled goals reflect the user's recent level of frustration, normalized according to a preset maximum number of consecutive failure cycles; task response latency reflects the user's sluggishness in viewing, confirming, or starting the task after receiving the exercise goal, and is also normalized; and recent interaction activity reflects the user's continued engagement with the system and the exercise task, converting interaction activity into corresponding interaction decline risk representations.

[0206] Through the above processing, the various psychological risk characteristics are aligned with a consistent risk direction; that is, the larger the characteristic value, the higher the potential risk of the user exhibiting goal resistance, activity interruption, or negative feedback. The system uses all of these psychological risk characteristics as input to the psychological risk probability model.

[0207] S6.2. Based on the probability of psychological risk, revise the initial dynamic target;

[0208] The system performs graded corrections to the initial dynamic goals generated by the dynamic goal decision-making module based on the user's current psychological risk probability.

[0209] When the user's psychological risk probability is lower than the medium psychological risk threshold, the user's current psychological state is considered relatively stable, and the system retains the preliminary dynamic goal generated by the dynamic goal decision module without making any additional adjustments.

[0210] When a user's psychological risk probability reaches a moderate level but has not yet reached a high psychological risk threshold, the system restricts the target from increasing further. If the initial dynamic target is lower than or equal to the final target actually released in the previous decision-making cycle, the initial dynamic target is retained; if the initial dynamic target is higher than the final target of the previous decision-making cycle, the target for this cycle is restricted to the final target of the previous decision-making cycle, thereby avoiding further increasing exercise requirements when the user has a certain psychological burden.

[0211] When a user's psychological risk probability reaches or exceeds the high psychological risk threshold, the system proactively lowers the target level. Based on the final target actually released in the previous decision-making cycle, the system calculates the upper limit of the allowable target under high psychological risk conditions according to a preset target reduction ratio. The system then compares the initial dynamic target with this upper limit and selects the lower target as the psychological risk-corrected target to reduce the user's current exercise burden.

[0212] Among them, the medium psychological risk threshold is used to distinguish between low-risk and medium-risk states, the high psychological risk threshold is used to distinguish between medium-risk and high-risk states, and the high psychological risk threshold is greater than the medium psychological risk threshold; the target descent ratio is used to control the magnitude of the reduction of the moving target under the high psychological risk state.

[0213] Through the aforementioned graded correction mechanism, the system maintains the original dynamic decision-making results when the psychological risk is low, restricts the continued growth of the target when the psychological risk is at a moderate level, and actively lowers the target when the psychological risk is high, thereby reducing the risk of target resistance and movement interruption caused by continuous failures, fatigue accumulation, or negative emotions.

[0214] S6.3 Calculate the range of target variation;

[0215] To prevent the movement target from changing too much in adjacent decision-making cycles, which could lead to excessive exercise burden or pressure on the user to complete the target, the system sets upper and lower limit constraints on the movement target for the current cycle based on the user's current status, historical target execution, and risk level.

[0216] Specifically, the system first determines the basic range of change based on the final exercise goal actually released in the previous decision-making cycle, and sets a minimum effective exercise goal lower limit. When the goal calculated by the system is lower than this lower limit, the current goal is restricted to the minimum effective exercise level to avoid interruption of exercise behavior or decline in long-term exercise effect due to the goal being too low.

[0217] Meanwhile, the system limits the target's descent based on a preset maximum descent percentage to prevent excessive drops in the user's exercise target within a single cycle. For example, when a user experiences short-term fluctuations in their condition but there is no significant risk, the system only allows the target to decrease within a reasonable range, ensuring the continuity of the exercise plan.

[0218] Regarding the direction of target growth, the system sets the maximum growth rate based on the user's current exercise ability, risk level, and historical performance. When the user's condition is stable, the risk is low, and the exercise target is achieved consecutively, the target is allowed to be appropriately increased in adjacent cycles. When the user experiences fatigue accumulation, consecutive failures, or increased psychological risk, the system automatically reduces the target growth rate to avoid causing frustration due to excessively rapid target increases.

[0219] Among them, the minimum effective exercise target represents the minimum target level required for a user to maintain basic exercise behavior; the maximum decrease ratio represents the maximum allowable decrease in the target within adjacent decision periods; the maximum increase ratio represents the maximum allowable increase in the target within adjacent decision periods; and the upper limit of the user's safety target represents the maximum acceptable exercise target calculated based on the user's current exercise ability and risk level.

[0220] The upper limit of a user's safety target is determined based on the user's current estimated athletic ability and the target increase coefficient corresponding to the risk level. The higher the risk level, the smaller the allowable increase in the target; when the risk level is low and the user's performance is good, the system allows the target to be increased appropriately.

[0221] By using the aforementioned target change range constraint mechanism, the process of adjusting the motion target is made gradual and stable. While ensuring personalized dynamic optimization, it avoids fatigue accumulation, increased psychological pressure, and decreased user participation caused by excessively rapid target changes.

[0222] S6.4. Issue the final motion target based on the range of target changes.

[0223] After completing the dynamic target generation, psychological risk correction, and target change range constraint, the system obtains the final motion target executed by the user in the current period.

[0224] The system first compares the psychologically risk-adjusted target with the range of target variation allowed in the current cycle, and then limits the exercise target to a preset reasonable range through target clipping to avoid the target exceeding the user's current acceptable ability or falling below the basic exercise requirements.

[0225] The constrained target is further discretized to convert it into an integer target value that conforms to the actual motion scenario, and finally the motion target is published to the user.

[0226] Among them, the final published target represents the exercise target that the system actually displays to the user and serves as the basis for the user's execution; the psychological risk-corrected target represents the intermediate result after psychological risk constraints; and the target variation range represents the allowable target range calculated based on the user's historical execution, current exercise ability, and risk status.

[0227] It should be noted that the motion targets used in subsequent model update phases are not the initial targets generated by the dynamic target decision module, but rather the final published targets that the user actually receives and executes. That is, the system calculates behavioral feedback based on the actual completion status corresponding to the final published targets, and uses this feedback for subsequent model parameter updates.

[0228] When a user is in the initial stage of use or has not yet formed effective historical motion data, the system does not have the final goal of the previous decision cycle. At this time, the preset minimum effective motion goal is used as the initial reference goal to construct the range of change of the initial goal.

[0229] Through the above processing, the system achieves a complete closed loop from candidate target generation, psychological risk adjustment, variation range constraint to final target release, so that the moving target maintains dynamic optimization capability and meets the stability and security requirements of users in the actual execution process.

[0230] S7, setting exercise goals and game reward settings;

[0231] S7.1 Break down a single sports objective into multiple phased tasks and set tiered rewards;

[0232] S7.2. Associate corresponding rewards based on the degree of goal completion.

[0233] S7.3 Record user clicks, claims, and dwell behaviors for different task formats and reward types as multi-source input data for the next cycle; transmit user interaction behaviors for tasks and rewards back to the continuous behavior evaluation module.

[0234] The exercise task and gamification interaction module transforms the final exercise goal into specific exercise tasks that users can understand and execute, and enhances users' willingness to participate through gamification mechanisms.

[0235] The system can break down a single exercise goal into multiple phased tasks, such as morning activity tasks, daily cumulative tasks, and evening supplementary tasks, and display the task progress, estimated completion time, and current completion percentage to the user.

[0236] Meanwhile, the system links users' activity with virtual resources, growth levels, scene unlocks, and reward feedback. For example, after completing a certain number of steps, users can obtain virtual seeds, energy, items, or growth points, which can be used for plant cultivation, garden construction, virtual asset accumulation, or reward redemption. The system can also set tiered rewards based on the degree of goal completion, allowing users to receive some positive feedback even if they do not fully complete the goal, avoiding the frustration caused by traditional binary evaluation.

[0237] The gamified interaction module not only provides feedback on results, but also records users' responses to different reward methods, task formats, and interactive content, and feeds the relevant interaction data back to the subsequent continuous behavior evaluation module.

[0238] S8, User-executed motion and data collection;

[0239] Users engage in actual exercise based on the exercise tasks issued by the system. During the exercise, the user's terminal or wearable device continuously collects data such as steps, exercise duration, heart rate, exercise intensity, and energy consumption, and uploads it to the system in real time or periodically.

[0240] The system updates the task completion progress based on real-time data collection and provides appropriate prompts to the user. When the user has been inactive for an extended period, a low-stress reminder can be sent; when the user is approaching the goal, positive encouragement can be provided; and when the user has reached the goal, high-intensity reminders can be stopped to prevent overexertion.

[0241] After a user completes an exercise, the system records the actual completion value, completion time, target completion rate, fluctuations during the exercise process, and the user's subjective feedback, forming the exercise result data for the current decision-making cycle.

[0242] S9. Continuous behavioral evaluation;

[0243] The continuous behavior evaluation module provides a non-binary, continuous evaluation of the user's exercise results. Unlike judging user performance solely based on "completed" or "not completed," this technical solution generates continuous behavior evaluation values ​​based on goal completion rate, completion stability, sustained participation, psychological feedback, and interactive behavior.

[0244] For example, a user's evaluation should be higher when they complete 90% of their goal than when they only complete 50%; a user can also receive a positive evaluation if they haven't fully met the goal but have shown significant improvement compared to the previous period. This allows for the retention of information on gradual changes in user behavior, providing more detailed feedback signals for model learning.

[0245] S10, Adaptive updating of model parameters;

[0246] The model parameter update module updates the key parameters in the dynamic target decision-making model based on the continuous behavior evaluation results.

[0247] The updated parameters may include the estimated returns of candidate objectives, the weights of different state characteristics, the exploration coefficient, the weight of the reward function, the objective adjustment range, and the psychological constraint threshold. When an objective consistently achieves high evaluations under a specific user state, the system increases the probability of that objective being selected under similar states; conversely, when an objective leads to consecutive failures, increased fatigue, or decreased user engagement, its probability of being selected decreases.

[0248] Model updates can be performed either through online incremental learning, updating parameters at the end of each decision cycle, or in batches according to a preset time window. The updated model parameters are then fed back to the dynamic target decision module for target generation in the next decision cycle.

[0249] Through the above steps, the system forms a closed-loop operation mechanism of "data acquisition - state construction - target generation - target adjustment - task execution - behavior evaluation - model update".

[0250] In the initial stage of the system, the cold start goal generation module provides users with safe and robust initial goals based on limited information; after accumulating a certain amount of historical data, the dynamic goal decision module gradually takes over the goal generation process; the psychological constraint adjustment module corrects the data-driven results at the behavioral and psychological level; after the user performs the exercise task, the continuous behavior evaluation module extracts feedback signals; and the model parameter update module continuously optimizes the goal decision strategy based on the feedback results.

[0251] As users' usage time increases, the system can gradually identify patterns in changes in users' athletic ability, target acceptance range, psychological fluctuation characteristics, and reward preferences, thereby transitioning from group-based prior knowledge to individualized decision-making. Ultimately, this allows exercise goals to be dynamically adjusted according to changes in user status, improving user participation, goal completion rate, and long-term exercise persistence while ensuring exercise safety and psychological acceptability.

[0252] like Figure 3 As shown, this application also discloses a dynamic optimization system for moving targets based on multi-source behavior fusion, including:

[0253] The system includes a multi-source behavioral data acquisition module, a data preprocessing and feature fusion module, a user state feature construction module, a cold start target generation module, a dynamic target decision-making module, a psychological constraint adjustment module, a motion task and gamified interaction module, a continuous behavior evaluation module, a model parameter update module, and a user terminal and wearable acquisition module.

[0254] The multi-source behavioral data acquisition module output is connected to the data preprocessing and feature fusion module; the data preprocessing and feature fusion module is connected to the user state feature construction module; the user state feature construction module is connected to the cold start target generation module or the dynamic target decision module; both the cold start target generation module and the dynamic target decision module output to the psychological constraint adjustment module; the psychological constraint adjustment module is connected to the motion task and gamification interaction module; the motion task and gamification interaction module is connected to the user terminal and the wearable acquisition subsystem; the user terminal and the wearable acquisition subsystem send data back to the multi-source behavioral data acquisition module; the user executes motion data and inputs it into the continuous behavior evaluation module; the continuous behavior evaluation module outputs to the model parameter update module; the model parameter update module feeds back parameters to the dynamic target decision module, forming a complete closed-loop link.

[0255] In summary, this invention can improve the reliability of multi-source data processing and the accuracy of target decision-making, while achieving dynamic, stable, and individualized adjustment of moving targets. It also reduces the risk of target overload and motion interruption through psychological constraints and continuous behavioral feedback, thereby improving the operating efficiency, decision-making stability, and long-term effectiveness of the adaptive motion behavior guidance system.

[0256] It should be noted that the accompanying drawings merely illustrate the technical concept of the present invention and should not be used to limit the scope of protection of the present invention. For those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A dynamic optimization method for moving targets based on multi-source behavior fusion, characterized in that: Includes the following steps: S1, User terminal and wearable device access; S2, Multi-source behavioral data acquisition; S3, Data Preprocessing and Feature Fusion; S4. Constructing user state feature vectors; S5. Working Mode Determination and Initial Target Generation: S6. Psychological constraint adjustment and correction; S7, setting exercise goals and game reward settings; S8, User-executed motion and data collection; S9. Continuous behavioral evaluation; S10, Adaptive updating of model parameters.

2. The dynamic optimization method for moving targets according to claim 1, characterized in that: The multi-source behavioral data in step S2 includes, but is not limited to, user basic attribute data, motion behavior data, physiological state data, system interaction data, and user subjective psychological feedback data. All types of data are collected at the time granularity of minutes, hours, or days, and are associated based on user identifiers and timestamps.

3. The dynamic optimization method for moving targets according to claim 1, characterized in that: Step S3 includes: S3.1 Perform integrity checks and outlier identification on the collected multi-source behavioral data, and perform adaptive missing value completion on missing data; S3.2 Perform quantile truncation filtering and normalization on the completed data; S3.3 Extract statistical features from continuous data using a preset time window; S3.

4. Integrate the processed basic attribute features, motion behavior features, physiological state features, psychological feedback features, and system interaction features to form a unified set of user behavior features.

4. The dynamic optimization method for moving targets according to claim 1, characterized in that: Step S4 constructs a user state feature vector based on the user's current exercise ability, exercise trend, goal completion status, psychological state, and interactive behavior. This vector maps the user's multi-source behavioral characteristics, recent exercise ability, exercise change trend, historical goal completion status, psychological state, fatigue risk, and interactive activity to the same state feature space. This enables the dynamic exercise goal decision model to adaptively select candidate exercise goals that match the user's current exercise ability and psychological endurance based on the user's state changes in different decision-making cycles.

5. The dynamic optimization method for moving targets according to claim 1, characterized in that: Step S5 determines the cold start mode or the dynamic target decision mode based on the number of valid historical periods of the user and the data integrity rate. In the cold start mode, the basic activity level is calculated based on user basic information and prior weighted statistics of similar user groups. The initial exercise goal within the restricted range is generated by combining the exercise ability, health risk and psychological fear of difficulty score. If it is a dynamic goal decision-making mode, four types of candidate goals are generated based on the user's reference exercise level: conservative, maintain, progressively improve, and challenge. A contextual feature is constructed that integrates user status, goal difficulty, and status-difficulty interaction items. The comprehensive score of each candidate goal is calculated, and the candidate with the highest score is selected as the initial dynamic goal.

6. The dynamic optimization method for moving targets according to claim 1, characterized in that: Step S6 includes: S6.1 Calculate the probability of psychological risk based on the psychological risk feature vector; S6.

2. Based on the probability of psychological risk, revise the initial dynamic target; S6.3 Calculate the range of target variation; S6.

4. Issue the final motion target based on the range of target changes.

7. The dynamic optimization method for moving targets according to claim 1, characterized in that: Step S7 includes: S7.1 Break down a single sports objective into multiple phased tasks and set tiered rewards; S7.

2. Associate corresponding rewards based on the degree of goal completion. S7.3 Record user clicks, claims, and dwell behaviors for different task formats and reward types as multi-source input data for the next cycle; transmit user interaction behaviors for tasks and rewards back to the continuous behavior evaluation module.

8. The dynamic optimization method for moving targets according to claim 1, characterized in that: In step S8, the user carries out actual exercise according to the exercise task issued by the system. During the exercise, the user's wearable device continuously collects multi-source data and uploads it to the system in real time or periodically. The collected data updates the task completion progress and provides appropriate prompts to the user. After the user completes the exercise, the system records the actual completion value, completion time, target completion rate, fluctuations in the exercise process, and the user's subjective feedback to form the exercise result data for the current decision cycle.

9. The dynamic optimization method for moving targets according to claim 1, characterized in that: Step S9 performs a non-binary, continuous evaluation based on the motion result data of multiple decision cycles collected in S8.

10. A dynamic optimization system for moving targets based on multi-source behavior fusion, characterized in that: For performing the method according to any one of claims 1-9, comprising: The system includes a multi-source behavioral data acquisition module, a data preprocessing and feature fusion module, a user state feature construction module, a cold start target generation module, a dynamic target decision-making module, a psychological constraint adjustment module, a motion task and gamified interaction module, a continuous behavior evaluation module, a model parameter update module, and a user terminal and wearable acquisition module. The output of the multi-source behavioral data acquisition module is connected to the data preprocessing and feature fusion module; the data preprocessing and feature fusion module is connected to the user state feature construction module; the user state feature construction module is connected to the cold start target generation module and the dynamic target decision module respectively; both the cold start target generation module and the dynamic target decision module output to the psychological constraint adjustment module; the psychological constraint adjustment module is connected to the motion task and gamification interaction module; the motion task and gamification interaction module is connected to the user terminal and the wearable acquisition subsystem; the user terminal and the wearable acquisition subsystem send back data to the multi-source behavioral data acquisition module; the user executes motion data input to the continuous behavior evaluation module; the continuous behavior evaluation module outputs to the model parameter update module; the model parameter update module feeds back parameters to the dynamic target decision module, forming a complete closed-loop link.

Images