A heuristic algorithm-based dance performer training mode optimization method

Abstract

The present application relates to the technical field of education intelligence, and more particularly to a dance performer training mode optimization method based on heuristic algorithm. The method first constructs a multi-dimensional action library containing memory decay, physiological load and kinematic characteristics; calculates the action review urgency based on the forgetting curve model and generates a priority queue; uses an improved genetic algorithm with a segmented constraint mechanism to generate a schedule, which forcibly limits the action intensity of the warm-up and relaxation phases; constructs a multi-objective fitness function that combines review coverage and physiological load variance to guide the population to evolve in the direction of memory reinforcement and physical balance; finally, cognitive pruning is performed by calculating the cosine similarity of the kinematic characteristics of adjacent actions to eliminate confusing schemes. The present application can automatically generate personalized training schedules that conform to human physiological rhythms and cognitive laws, while improving action memory efficiency and effectively reducing the risk of sports injuries.

Description

Technical Field

[0001] This invention relates to the field of intelligent education technology, and in particular to a method for optimizing training modes for dance performers based on heuristic algorithms. Background Technology

[0002] Dance performances require performers to not only master the precise movements but also possess extremely high muscle endurance and a high rate of retaining motor memory. Current dance training schedules mainly rely on the instructor's subjective experience or fixed teaching outlines, typically employing a linear model of new instruction-review-reinforcement. This model focuses on consolidating motor memory through high-frequency repetitive training and relies on manual judgment to adjust training intensity and rest intervals.

[0003] However, existing technologies lack data-driven quantitative analysis methods, making it difficult to maximize training benefits and minimize injury risks. On the one hand, traditional manual scheduling struggles to accurately capture the dynamic decay points of motor memory, often leading to delayed review or ineffective repetitions. It also ignores cognitive interference caused by movements with highly similar spatial trajectories or force exertion rhythms occurring in adjacent sequences, easily causing memory confusion. On the other hand, relying solely on experience makes it difficult to accurately calculate the real-time cumulative load and recovery cycle of various body parts, making it highly susceptible to local muscle strain or joint damage due to unreasonable scheduling structures. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a method for optimizing the training mode of dance performers based on heuristic algorithms. It aims to achieve personalized and scientific scheduling for dance performers through digital modeling, multi-objective optimization, and cognitive pruning.

[0005] This invention provides the following technical solution: a method for optimizing training modes for dance performers based on heuristic algorithms, comprising: S1. Construct a dance movement feature library and configure memory decay feature parameters, physiological load distribution vector, and kinematic feature vector for each movement in the library; S2. Obtain the trainer's historical training data, use the forgetting curve model to calculate the current memory retention rate of each action in the action set to be scheduled, and generate a review urgency priority queue based on the memory retention rate. S3. Construct a scheduling optimization model based on an improved genetic algorithm, mapping the discrete time slots of the training cycle to gene loci, and initializing the action combination sequence as chromosomes; S4. Establish a multi-objective fitness function, which includes a positive index that maximizes the coverage of the review urgency priority queue and a negative penalty term that minimizes the variance of cumulative fatigue of the same muscle group within a specific time window. S5. Perform population iterative evolution calculation, and generate a candidate scheduling scheme set containing several non-dominated solutions based on the convergence of the multi-objective fitness function; S6. Calculate the cosine similarity of the kinematic feature vectors of actions in adjacent time periods in the scheme, prune the scheme using the cosine similarity, and output the final dance training target schedule from the remaining set.

[0006] Preferably, in step S1, the step of constructing the dance movement feature library includes: Configure the memory decay characteristic parameters, which include the base decay rate and the movement difficulty coefficient; Configure the physiological load distribution vector, which includes normalized stress values ​​of different movements on six body parts: neck, shoulder, waist, hip, knee and ankle. Configure the kinematic feature vector, which includes a spatial coordinate sequence representing the movement of the human body's center of mass during the execution of the action, as well as the angular velocity data of the force-generating joints.

[0007] Preferably, in step S2, the step of generating a review urgency priority queue based on the memory retention rate includes: Based on the forgetting curve model, the review urgency score is calculated by combining the memory decay characteristic parameters. Based on the review urgency score, a queue sorting operation is performed to map the set of actions to be scheduled to the review urgency priority queue.

[0008] Preferably, in step S3, the step of constructing a scheduling optimization model based on an improved genetic algorithm includes: Establish a segmented constraint mechanism to divide the action combination sequence, which is a chromosome, into a warm-up segment, a core training segment, and a relaxation segment in time. Perform constraint initialization operations to restrict the gene loci of the warm-up segment and the cool-down segment to be mapped only to a preset low physiological load action subset, while the gene loci of the core training segment are mapped to a high physiological load action subset.

[0009] Preferably, in step S4, the step of establishing the multi-objective fitness function includes: The percentage of the coverage of high-priority actions in the review urgency priority queue by the statistical action combination sequence is taken as the coverage rate of the review urgency priority queue. For the aforementioned action sequence, the cumulative load of each muscle group within a specific time window is calculated, and the variance of the cumulative load values ​​among the muscle groups is calculated. The review urgency priority queue coverage is given a positive weight, and the variance is given a negative weight. The final fitness function value is obtained by linear weighted summation.

[0010] Preferably, in step S5, the step of performing population iterative evolution calculation includes: The individuals in the population are sorted according to the fitness function value, and a preset proportion of elite individuals are directly retained, while parent selection is performed on the remaining individuals. A segmented structure preservation strategy is adopted to restrict gene exchange between parent chromosomes to occur only within corresponding time segments, in order to maintain the temporal structure of warm-up, core training and relaxation segments in offspring chromosomes. A subset constraint mutation strategy is adopted. When performing mutation on a certain gene locus, the replacement action is selected only from the preset action subset corresponding to the time segment to which the gene locus belongs.

[0011] Preferably, in step S6, the step of calculating the cosine similarity of the kinematic feature vectors of actions in adjacent time periods in the scheme includes: Perform vector decoupling operation to decompose the kinematic feature vector into spatial dimension components that represent spatial trajectory and dynamic dimension components that represent force application rhythm; Calculate the first cosine similarity of two actions in adjacent time periods in the spatial dimension component and the second cosine similarity in the dynamic dimension component. A weighted summation is performed on the first cosine similarity and the second cosine similarity to obtain the final kinematic feature vector cosine similarity.

[0012] Preferably, in step S6, the step of pruning the scheme using the cosine similarity includes: Set a cognitive interference threshold and remove any scheme in the candidate scheduling scheme set that contains adjacent action pairs with a cosine similarity higher than the cognitive interference threshold.

[0013] Preferably, the optimization method further includes: Obtain objective fatigue feedback values ​​after the trainee executes the schedule; Calculate the absolute value of the deviation between the objective fatigue feedback value and the cumulative fatigue value; If the absolute value of the deviation exceeds the preset calibration threshold, the physiological load distribution vector of the corresponding action is corrected.

[0014] The present invention has the following beneficial effects: 1. This invention accurately identifies memory weaknesses through a dynamic forgetting model, ensuring that high-priority actions are reviewed in a timely manner. Simultaneously, by utilizing the cosine similarity calculation of kinematic feature vectors, it eliminates easily confused similar actions before output, effectively solving the problem of rhythm disorder and action distortion caused by consecutive similar actions in traditional scheduling.

[0015] 2. This invention ensures a scientific transition in training intensity through a segmented constraint mechanism, avoiding high-intensity exercise during cold starts or periods of fatigue. Its core fitness function introduces a variance penalty term for the load on multiple parts of the body, which can automatically identify and break down movement combinations that may lead to local strain, allowing physiological load to be rationally distributed across all joints of the body. From a data-driven decision-making perspective, this significantly reduces the risk of sports injuries to common areas such as the knees and ankles in dance training. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a method for optimizing training patterns for dance performers based on heuristic algorithms, provided in an embodiment of the present invention; Figure 2 A flowchart for establishing a multi-objective fitness function is provided in an embodiment of the present invention; Figure 3 A flowchart for calculating the cosine similarity of kinematic feature vectors of actions in adjacent time periods is provided in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the implementation of a method for optimizing training modes for dance performers based on heuristic algorithms, as provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In a first embodiment of the present invention, a method for optimizing training patterns for dance performers based on heuristic algorithms is provided, such as... Figure 1 As shown, it includes the following steps: S1. Construct a dance movement feature library and configure memory decay feature parameters, physiological load distribution vector, and kinematic feature vector for each movement in the library; Preferably, in step S1, the step of constructing the dance movement feature library includes: Configure the memory decay characteristic parameters, which include the base decay rate and the movement difficulty coefficient; Configure the physiological load distribution vector, which includes normalized stress values ​​of different movements on six body parts: neck, shoulder, waist, hip, knee and ankle. Configure the kinematic feature vector, which includes a spatial coordinate sequence representing the movement of the human body's center of mass during the execution of the action, as well as the angular velocity data of the force-generating joints.

[0019] Specifically, a digital index of dance movements is established, and for each standard dance movement in the library, parametric modeling is performed from the dimensions of cognitive psychology, biomechanics, and kinematics.

[0020] In terms of configuring memory decay characteristic parameters, the system assigns a base decay rate and a motion difficulty coefficient to each action. The base decay rate is set based on the general law of the Ebbinghaus forgetting curve, which represents the natural decay rate of the action in the brain without review intervention. The motion difficulty coefficient is set based on the limb coordination complexity and sequence length of the action, and the value range is set between 1.0 and 2.0.

[0021] In configuring the physiological load distribution vector, the system establishes a body part model encompassing six dimensions: neck, shoulders, waist, hips, knees, and ankles. Through biomechanical simulation software or by collecting wearable sensor data from professional dancers, the system obtains the peak impact force or torque generated on these six body parts during standard execution of the movement. To eliminate differences in the physiological tolerance of different body parts, the system normalizes the collected physical quantities, ultimately constructing the physiological load distribution vector. Each element in this vector is a dimensionless value between 0 and 1, which directly quantifies the degree of wear and tear on a specific joint caused by the movement.

[0022] In configuring kinematic feature vectors, the system employs a discretized time sampling method to record the three-dimensional spatial coordinate changes of the human body's center of mass and the angular velocity changes of the main force-generating joints during the action execution cycle. The system extracts the center of mass coordinates during the action at a preset sampling frequency, forming a spatial coordinate sequence that characterizes the spatial displacement trajectory of the action. Simultaneously, it extracts the instantaneous angular velocities of the force-generating joints at each sampling point, forming an angular velocity data sequence that characterizes the force exertion rhythm and speed characteristics of the action.

[0023] Through the above implementation methods, abstract dance movements are transformed into computable multidimensional digital models, providing a precise data foundation for subsequent implementation of scheduling optimization based on memory patterns, injury prevention based on physiological load, and cognitive interference elimination based on similarity.

[0024] S2. Obtain the trainer's historical training data, use the forgetting curve model to calculate the current memory retention rate of each action in the action set to be scheduled, and generate a review urgency priority queue based on the memory retention rate. Preferably, in step S2, the step of generating a review urgency priority queue based on the memory retention rate includes: Based on the forgetting curve model, the review urgency score is calculated by combining the memory decay characteristic parameters. Based on the review urgency score, a queue sorting operation is performed to map the set of actions to be scheduled to the review urgency priority queue.

[0025] Specifically, the system reads the trainee's historical training logs, retrieves the timestamp of the last effective training for each action in the action set to be scheduled, and calculates the time interval between the current time and that timestamp. The system then calls the feature library constructed in step S1, substitutes the time interval as an independent variable, and, combined with the action's base decay rate and difficulty coefficient, calculates the real-time memory retention rate for that action. The calculation model for the real-time memory retention rate is defined by the following formula: ; in, Representing the action after a time interval The probability of memory retention after that, This is the base rate of decline for this action. This represents the difficulty coefficient of the action. The formula indicates that the higher the difficulty coefficient, the faster the exponential decay rate, resulting in a lower memory retention rate.

[0026] Secondly, based on the calculated real-time memory retention rate, the system further quantifies the review requirement for each action. The system defines the memory loss rate as the difference between a unit value of 1 and the real-time memory retention rate, and introduces an action weighting coefficient to distinguish between core and secondary actions. One way to implement the review urgency score is as follows: ; in, To review the urgency score, This is a preset action weighting coefficient. Through this calculation, actions with higher forgetting rates and greater importance will receive higher scores. Finally, a queue sorting operation is performed, arranging all actions to be scheduled in descending order based on the calculated review urgency scores. Actions at the head of the queue represent the content most in need of review in the current training cycle, while actions at the tail of the queue represent content with good memory retention or lower importance.

[0027] By combining the natural forgetting patterns over time with the inherent characteristics of the actions themselves, the above implementation method enables a quantitative assessment of the priority of training content, ensuring that the generated training plan accurately targets the trainee's weak points in memory.

[0028] S3. Construct a scheduling optimization model based on an improved genetic algorithm, mapping the discrete time slots of the training cycle to gene loci, and initializing the action combination sequence as chromosomes; Preferably, in step S3, the step of constructing a scheduling optimization model based on an improved genetic algorithm includes: Establish a segmented constraint mechanism to divide the action combination sequence, which is a chromosome, into a warm-up segment, a core training segment, and a relaxation segment in time. Perform constraint initialization operations to restrict the gene loci of the warm-up segment and the cool-down segment to be mapped only to a preset low physiological load action subset, while the gene loci of the core training segment are mapped to a high physiological load action subset.

[0029] Specifically, the coding rules and temporal structure of the chromosome are established. The system discretizes the preset training cycle into several consecutive time slots, defining the chromosome. For a length of An integer vector, where each gene position For each time slot, the value on the gene locus represents the specific action ID scheduled within that time slot. The system establishes a segmented constraint mechanism, dividing chromosomes into three consecutive intervals on the temporal index based on the fundamental principles of exercise physiology: the warm-up interval... Core training interval and relaxation interval .in and This is a dynamically set dividing point based on the total training duration.

[0030] Secondly, the exercise library is preprocessed and stratified based on physiological load intensity. The system reads the physiological load distribution vector configured in step S1 and calculates the comprehensive load index for each exercise. A physiological load threshold is then set. The action library is divided into two mutually exclusive subsets, one of which is the low physiological load action subset. This includes a comprehensive load index less than or equal to Actions; and a subset of actions with high physiological load. This includes a comprehensive load index greater than The action.

[0031] Finally, constraint initialization is performed. When generating the initial population or performing gene mutations, the system enforces constraints on the value range of each gene locus. For any gene locus on a chromosome... Its value space The specific constraint logic is determined by the time segment in which it occurs, and is defined by the following piecewise function: ; During the initialization phase, the system iterates through each chromosome in the population and, according to the aforementioned function, randomly selects an action ID from the corresponding action subset for each gene position and assigns it a value. If subsequent operations involve gene segment crossover or single-point mutation, the system also verifies this constraint to ensure that the solution generated at any time conforms to the temporal intensity structure.

[0032] Through the above implementation method, the temporal structure constraints of exercise physiology are directly embedded at the coding level of the algorithm, which effectively reduces the search space of the genetic algorithm and eliminates the unsafe scheduling scheme of high-intensity movements occurring in the warm-up or cool-down phase from the root.

[0033] S4. Establish a multi-objective fitness function, which includes a positive index that maximizes the coverage of the review urgency priority queue and a negative penalty term that minimizes the variance of cumulative fatigue of the same muscle group within a specific time window. Preferably, in step S4, the step of establishing the multi-objective fitness function is as follows: Figure 2 As shown, it includes: The percentage of the coverage of high-priority actions in the review urgency priority queue by the statistical action combination sequence is taken as the coverage rate of the review urgency priority queue. For the aforementioned action sequence, the cumulative load of each muscle group within a specific time window is calculated, and the variance of the cumulative load values ​​among the muscle groups is calculated. The review urgency priority queue coverage is given a positive weight, and the variance is given a negative weight. The final fitness function value is obtained by linear weighted summation.

[0034] Specifically, the system reads the review urgency priority queue generated in step S2 and selects the top section of the queue. An action or prior The actions constitute a high-priority review set, denoted as Subsequently, the system iterates through the current sequence of actions that constitute a chromosome, extracting all non-repeating actions contained in the sequence to form the current scheduling set, denoted as . The system calculates the ratio of the number of elements in the intersection of two sets to the number of elements in the higher-priority review set, which is used as the review urgency priority queue coverage. ,in, This represents the total number of actions in the high-priority review set. This indicates the number of high-priority actions successfully covered in the current scheduling plan. The closer this value is to 1, the better the scheduling plan is at reviewing areas where memory is weak.

[0035] For the current sequence of movements, the system accumulates the normalized stress values ​​of six body parts—neck, shoulder, waist, hip, knee, and ankle—within a set specific time window to obtain the cumulative load vector for each muscle group. The system then calculates the variance of the six values ​​in the vector. This is used to characterize the degree of dispersion of force on different parts of the body.

[0036] The formula for calculating variance is as follows: ; in, For the first The cumulative load on each body part The variance is calculated as the arithmetic mean of the cumulative load on all body parts. A smaller variance indicates more even stress distribution across body parts during training, and a lower risk of localized strain. Finally, a multi-objective linear weighted fitness function is constructed. The system pre-defines positive weight coefficients. With negative weighting coefficient These correspond to the relative importance placed on the benefits of memory review and the risk of physical injury, respectively. The two indicators are then integrated into a single fitness function value using a linear weighted summation method. During the iterative process of the genetic algorithm, the system strives to find a way to make the genetic algorithm more efficient and efficient. The solution that maximizes the value.

[0037] Through the above implementation methods, abstract teaching objectives are transformed into quantifiable mathematical optimization problems, which not only ensures the frequency of review of key movements, but also enforces the balanced distribution of physical load through the variance penalty mechanism, thereby achieving dual optimization of training effect and physical protection.

[0038] S5. Perform population iterative evolution calculation, and generate a candidate scheduling scheme set containing several non-dominated solutions based on the convergence of the multi-objective fitness function; Preferably, in step S5, the step of performing population iterative evolution calculation includes: The individuals in the population are sorted according to the fitness function value, and a preset proportion of elite individuals are directly retained, while parent selection is performed on the remaining individuals. A segmented structure preservation strategy is adopted to restrict gene exchange between parent chromosomes to occur only within corresponding time segments, in order to maintain the temporal structure of warm-up, core training and relaxation segments in offspring chromosomes. A subset constraint mutation strategy is adopted. When performing mutation on a certain gene locus, the replacement action is selected only from the preset action subset corresponding to the time segment to which the gene locus belongs.

[0039] Specifically, setting population size parameters (For example ), generate containing The initial population of individuals The system calculates the current population. Each individual fitness function value The population is then sorted in descending order based on this value. An elite retention ratio parameter is set. The system directly sorts the top-ranked items. Each individual is marked as an elite individual and is replicated seamlessly into the next generation of the population. In the middle. For the remaining For each individual position, the system uses either the roulette wheel selection algorithm or the tournament selection algorithm to select the parent from the current population.

[0040] To prevent temporal structure disorder, the system employs a "segmented independent crossover" strategy. Assume the chromosomes are divided into warm-up segments. Core segment and relaxation phase When performing a crossover, the system independently selects the crossover point within the corresponding segments of parent A and parent B, respectively. For example, for the warm-up segment... The system randomly selects the intersection point. Swap parent A and parent B in Duan Zhong Subsequent genes; similarly, for and Select the intersection points respectively and The operation is performed. This operation ensures that the exchanging offspring chromosomes logically maintain a strict temporal structure and do not violate the length definitions of each stage.

[0041] Finally, the subset constraint mutation operation is performed. The system uses a preset mutation probability. Traverse every gene locus on the offspring chromosome. When determining the first... When a gene locus needs to be mutated, the system first identifies that gene locus. Time segment Subsequently, the system invokes the constraint mapping relationship defined in step S3 to obtain the subset of allowed actions corresponding to that time segment. The system only uses this subset. A new action ID is randomly selected to replace the original gene value. For example, if the gene position... If the exercise is in the warm-up phase, the new movement generated by the variation must belong to a subset of low physiological load movements, and it is strictly forbidden to vary it into high-load core training movements.

[0042] Through the above implementation methods, strong topological constraints are introduced into the iterative process of the evolutionary algorithm, which not only ensures that the population converges towards higher fitness, but also eliminates physiological and logical errors caused by randomness at the algorithm operator level, ensuring that each generation of scheduling schemes is executable and secure.

[0043] S6. Calculate the cosine similarity of the kinematic feature vectors of actions in adjacent time periods in the scheme, prune the scheme using the cosine similarity, and output the final dance training target schedule from the remaining set.

[0044] Preferably, in step S6, the step of calculating the cosine similarity of the kinematic feature vectors of actions in adjacent time periods in the scheme is as follows: Figure 3As shown, it includes: Perform vector decoupling operation to decompose the kinematic feature vector into spatial dimension components that represent spatial trajectory and dynamic dimension components that represent force application rhythm; Calculate the first cosine similarity of two actions in adjacent time periods in the spatial dimension component and the second cosine similarity in the dynamic dimension component. A weighted summation is performed on the first cosine similarity and the second cosine similarity to obtain the final kinematic feature vector cosine similarity.

[0045] Specifically, the system extracts adjacent time periods from the scheduling scheme. and The two actions are denoted as actions. With action Read the kinematic feature vector constructed in step S1. It is a high-dimensional composite vector. The system, based on a predefined data structure definition, converts the vector... Decomposed into spatial dimension components Components of the dynamic dimension .in, It contains a normalized coordinate sequence of the human body's center of mass in three-dimensional space, representing the spatial trajectory of the movement; It includes angular velocity sequences of key joints such as the hip, knee, and ankle, representing the force exertion rhythm of the movement.

[0046] The system employs a cosine similarity algorithm to measure the directional similarity between two vectors. For the spatial dimension component, the system calculates the action... With action First cosine similarity For the dynamic dimension component, the system calculates the action. With action Second cosine similarity Finally, a weighted fusion of multidimensional similarity is performed. The system presets spatial weight coefficients. With dynamic weighting coefficient And satisfy This allows for the adjustment of the degree to which different features influence cognitive confusion. For example, for beginner learners, similar spatial trajectories are more likely to cause confusion, so a larger adjustment can be applied. The final kinematic feature vector cosine similarity .

[0047] The above implementation method solves the technical defect that traditional single-dimensional similarity calculation cannot distinguish similar actions, realizes a refined measurement of the risk of cognitive interference between actions, and provides a high-precision judgment basis for subsequent scheduling and pruning.

[0048] Preferably, in step S6, the step of pruning the scheme using the cosine similarity includes: Set a cognitive interference threshold and remove any scheme in the candidate scheduling scheme set that contains adjacent action pairs with a cosine similarity higher than the cognitive interference threshold.

[0049] Specifically, the system sets a cognitive interference threshold to define the pruning boundary of the schemes. It iterates through the candidate scheduling scheme set output in step S5, checking the cosine similarity of the kinematic feature vectors of all adjacent action pairs in each scheme. Once it detects that the similarity value of any pair of adjacent actions in a scheme exceeds the cognitive interference threshold, the system determines that the scheme does not meet the cognitive safety standard and removes it directly from the candidate set. Finally, the system outputs the remaining scheduling schemes after the above filtering steps as the final dance training target scheduling table.

[0050] Through the above implementation method, a threshold mechanism is used to quickly filter out training plans that are prone to causing memory confusion, thereby achieving quality control of the cognitive dimension of scheduling schemes with low computational cost.

[0051] Preferably, the optimization method further includes: Obtain objective fatigue feedback values ​​after the trainee executes the schedule; Calculate the absolute value of the deviation between the objective fatigue feedback value and the cumulative fatigue value; If the absolute value of the deviation exceeds the preset calibration threshold, the physiological load distribution vector of the corresponding action is corrected.

[0052] Specifically, after the trainee completes the system-generated schedule, the system records the trainee's physiological state in response to the training intensity, preferentially using parameters such as heart rate variability or lactate accumulation as fatigue feedback values, and records the objective fatigue feedback value. .

[0053] Secondly, read the theoretical cumulative fatigue value calculated for this scheduling scheme in step S4. To achieve dimensional uniformity, the system employs a max-min normalization method to map the theoretical cumulative fatigue value to the same numerical range as the objective feedback value. Subsequently, the system calculates the absolute value of the deviation between the two values.

[0054] ; The system sets a preset calibration threshold. If the calculated absolute value of the deviation If the value exceeds this threshold, it is determined that there is a significant error between the physiological load parameters in the current database and the trainee's actual physical condition, triggering the parameter correction mechanism.

[0055] Finally, the physiological load distribution vector is corrected. The system identifies all action IDs included in this scheduling and, for each action, corrects the physiological load distribution vector stored in step S1. Update using preset correction coefficients. Adjust vector values: ; like This indicates that the actual action is more strenuous, and the system automatically increases the load value of the relevant action; conversely, it decreases it. (Corrected vector) The original parameters are stored in the feature library and used as the basis for the next scheduling calculation.

[0056] Through the above implementation methods, a closed-loop feedback system with self-learning capabilities is constructed, enabling the dance movement feature database to be dynamically calibrated according to changes in the trainee's physical fitness and individual differences, thereby continuously improving the physiological accuracy of the scheduling plan.

[0057] In one embodiment, taking the pre-competition training scenario of professional dancers as an example, the complete execution flow of the training mode optimization method of the present invention is illustrated, and the execution flow is as follows: Figure 2 As shown. The training subject is a professional dancer preparing for the National Championships. The dancer's memory of the recently choreographed competition routine is still unstable, and there is an old injury to the right knee joint requiring strict control of continuous load.

[0058] To address the above scenario, this system generates and optimizes the schedule according to the following steps: The system first performs attribute labeling and parameter configuration on the competition movements in the movement library. For newly choreographed competition routines, the system sets the difficulty coefficient of the movement in its memory decay feature parameters to a high level, meaning that these movements will be marked as forgotten in the model more quickly. For all movements involving jumping and sudden stops, the system significantly increases the normalized pressure value of the knee joint dimension in its physiological load distribution vector to identify key targets for protection. At the same time, for rotational movements, the system records their kinematic feature vectors containing rotation axis and angular velocity data to facilitate subsequent identification of easily confused rotational movements.

[0059] The system scans dancers' historical training logs and discovers that several key connecting movements in the new competition routine have not been performed in the past three days. It then activates a priority generation mechanism. Based on a forgetting curve model, the system determines that the current memory retention rate of these movements is below a warning threshold. Combining this with a high-weighting coefficient from pre-competition preparation, the system calculates an extremely high review urgency score. Based on this score, the system forcibly maps these movements to the front of the review urgency priority queue, ensuring they appear in subsequent scheduling, thus achieving precise coverage of memory weaknesses.

[0060] The system begins generating an initial scheduling plan, establishing a strict segmented constraint mechanism in the process. The system divides the schedule temporally into a warm-up activation phase, a core competition phase, and a recovery and relaxation phase, and applies subset mapping restrictions to the gene loci for each phase. Specifically, during the initialization of the warm-up activation phase, the system locks in a subset of low-physiological-load movements, automatically filtering out all high-impact jumping movements and selecting only joint lubrication and muscle activation movements. This logically avoids the risk of knee injury recurrence due to cold starts, ensuring the physiological safety of the initial solution.

[0061] The system generates multiple candidate scheduling schemes and evaluates their merits based on coverage and physiological load variance. Schemes that, while covering competition movements frequently, are too densely packed, leading to a sharp increase in cumulative load on the knee joint and excessively large force variance across different body parts, are deemed to have an extremely high risk of injury and given a low fitness score. Conversely, schemes that intelligently incorporate upper limb performance training, allowing the knee joint to rest intermittently and with a relatively balanced distribution of physiological load throughout the body, are given a high fitness score.

[0062] The system performs genetic evolution on the optimal scheme, strictly maintaining the temporal structure of the schedule. When performing crossover, the system adopts a segmented structure preservation strategy to ensure that the core competition segment of the parent scheme is not mistakenly swapped with the relaxation segment of another scheme, preventing high-intensity movements from occurring during the fatigue period at the end of training. When performing mutation, if a movement within the core competition segment is detected to need replacement, the system restricts the selection of replacements to a subset of high-intensity movements, ensuring that the training intensity always meets the target.

[0063] The system outputs several evolved high-scoring schemes and performs a final cognitive safety check. Considering the dancers are under high pressure before the competition, the system automatically lowers the cognitive interference threshold and iterates through the kinematic feature vectors of all adjacent movements in the schemes. Once a scheme is detected to contain two movements with highly similar spatial trajectories, such as a left gyro spin and a reverse spiral spin, the system determines that it is highly likely to cause confusion in terms of direction and performs a pruning operation to directly remove the scheme. Finally, the system outputs a schedule that inserts linear movement movements as buffers between easily confused movements.

[0064] After the dancer completed the rehearsal, physiological indicators showed "good heart rate, but core muscle fatigue did not meet the standard." The system's comparative calculations revealed a significant discrepancy between the estimated lower back and abdominal load and the objective feedback. This triggered a correction mechanism, automatically increasing the load parameters for the lower back dimension in the relevant movements in the movement library. In the next rehearsal, the system will automatically increase the proportion of core training to approximate the dancer's true physical limits.

[0065] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing training patterns for dance performers based on heuristic algorithms, characterized in that, include: S1. Construct a dance movement feature library and configure memory decay feature parameters, physiological load distribution vector, and kinematic feature vector for each movement in the library; S2. Obtain the trainer's historical training data, use the forgetting curve model to calculate the current memory retention rate of each action in the action set to be scheduled, and generate a review urgency priority queue based on the memory retention rate. S3. Construct a scheduling optimization model based on an improved genetic algorithm, mapping the discrete time slots of the training cycle to gene loci, and initializing the action combination sequence as chromosomes; S4. Establish a multi-objective fitness function, which includes a positive index that maximizes the coverage of the review urgency priority queue and a negative penalty term that minimizes the variance of cumulative fatigue of the same muscle group within a specific time window. S5. Perform population iterative evolution calculation, and generate a candidate scheduling scheme set containing several non-dominated solutions based on the convergence of the multi-objective fitness function; S6. Calculate the cosine similarity of the kinematic feature vectors of actions in adjacent time periods in the scheme, prune the scheme using the cosine similarity, and output the final dance training target schedule from the remaining set.

2. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, In step S1, the steps for constructing the dance movement feature library include: Configure the memory decay characteristic parameters, which include the base decay rate and the movement difficulty coefficient; Configure the physiological load distribution vector, which includes normalized stress values ​​of different movements on six body parts: neck, shoulder, waist, hip, knee and ankle. Configure the kinematic feature vector, which includes a spatial coordinate sequence representing the movement of the human body's center of mass during the execution of the action, as well as the angular velocity data of the force-generating joints.

3. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, In step S2, the step of generating a review urgency priority queue based on the memory retention rate includes: Based on the forgetting curve model, the review urgency score is calculated by combining the memory decay characteristic parameters. Based on the review urgency score, a queue sorting operation is performed to map the set of actions to be scheduled to the review urgency priority queue.

4. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, In step S3, the steps for constructing a scheduling optimization model based on an improved genetic algorithm include: Establish a segmented constraint mechanism to divide the action combination sequence, which is a chromosome, into a warm-up segment, a core training segment, and a relaxation segment in time. Perform constraint initialization operations to restrict the gene loci of the warm-up segment and the cool-down segment to be mapped only to a preset low physiological load action subset, while the gene loci of the core training segment are mapped to a high physiological load action subset.

5. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, In step S4, the steps for establishing the multi-objective fitness function include: The percentage of the coverage of high-priority actions in the review urgency priority queue by the statistical action combination sequence is taken as the coverage rate of the review urgency priority queue. For the aforementioned action sequence, the cumulative load of each muscle group within a specific time window is calculated, and the variance of the cumulative load values ​​among the muscle groups is calculated. The review urgency priority queue coverage is given a positive weight, and the variance is given a negative weight. The final fitness function value is obtained by linear weighted summation.

6. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, In step S5, the steps for performing population iterative evolution calculations include: The individuals in the population are sorted according to the fitness function value, and a preset proportion of elite individuals are directly retained, while parent selection is performed on the remaining individuals. A segmented structure preservation strategy is adopted to restrict gene exchange between parent chromosomes to occur only within corresponding time segments, in order to maintain the temporal structure of warm-up, core training and relaxation segments in offspring chromosomes. A subset constraint mutation strategy is adopted. When performing mutation on a certain gene locus, the replacement action is selected only from the preset action subset corresponding to the time segment to which the gene locus belongs.

7. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, In step S6, the step of calculating the cosine similarity of the kinematic feature vectors of actions in adjacent time periods in the scheme includes: Perform vector decoupling operation to decompose the kinematic feature vector into spatial dimension components that represent spatial trajectory and dynamic dimension components that represent force application rhythm; Calculate the first cosine similarity of two actions in adjacent time periods in the spatial dimension component and the second cosine similarity in the dynamic dimension component. A weighted summation is performed on the first cosine similarity and the second cosine similarity to obtain the final kinematic feature vector cosine similarity.

8. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, In step S6, the step of pruning the scheme using the cosine similarity includes: Set a cognitive interference threshold and remove any scheme in the candidate scheduling scheme set that contains adjacent action pairs with a cosine similarity higher than the cognitive interference threshold.

9. The method for optimizing the training mode of dance performers based on heuristic algorithms according to claim 1, characterized in that, The optimization method further includes: Obtain objective fatigue feedback values ​​after the trainee executes the schedule; Calculate the absolute value of the deviation between the objective fatigue feedback value and the cumulative fatigue value; If the absolute value of the deviation exceeds the preset calibration threshold, the physiological load distribution vector of the corresponding action is corrected.

Images