Intelligent monitoring and anomaly recognition method and system for strength training and storage medium

Abstract

The invention belongs to the technical field of training data processing, and particularly provides a strength training intelligent monitoring and anomaly recognition method and system and a storage medium, and the method mainly comprises the steps: obtaining a target and real-time training parameter, calculating the training density and an intermittent and strength coupling factor, and obtaining a training result; and generating a dynamic compensation coefficient and an initial intensity correction gradient through a random forest, carrying out constraint, decomposing a deviation rate into a trend term and a mutation term by adopting an attention network, distributing a weight to enhance the trend term, and outputting an efficiency index and a risk level according to stability and a peak value. The problems that in the prior art, strength training mostly adopts a fixed parameter mode and is mainly executed according to a preset plan, and it is difficult to accurately evaluate the training efficiency and recognize abnormal risks can be effectively solved, the training efficiency can be accurately evaluated, the abnormal risks can be effectively recognized, the training scientificity is remarkably improved, the training effect is optimized, and the training process safety is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of training data processing, and in particular relates to a strength training intelligent monitoring and abnormality identification method, system and storage medium. Background Art

[0002] Strength training enhances muscle strength and endurance, increases basal metabolic rate, promotes bone health, improves body posture, enhances athletic ability and quality of life. Monitoring strength training can scientifically evaluate training effects, accurately adjust training plans, effectively prevent sports injuries, and ensure safe and efficient training.

[0003] Currently, strength training mostly adopts a fixed parameter model, which is mainly executed according to preset plans. It is difficult to capture dynamic information such as changes in intervals between groups and fluctuations in training intensity in real time, making it difficult to dynamically adjust training parameters according to actual conditions. This makes it impossible to conduct in-depth analysis based on accurate data, and it is difficult to accurately evaluate training effectiveness and identify abnormal risks. Summary of the invention

[0004] The present disclosure provides a method, system and storage medium for intelligent monitoring and abnormality identification of strength training, which effectively solves the problem that strength training in the prior art mostly adopts a fixed parameter mode and is mainly executed according to a preset plan, making it difficult to accurately evaluate training effectiveness and identify abnormal risks. The method can accurately evaluate training effectiveness, effectively identify abnormal risks, significantly improve the scientific nature of training, optimize training effects, and ensure the safety of the training process.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present disclosure provides a method for intelligent monitoring and abnormality identification of strength training, including: obtaining target training parameters and real-time training parameters; calculating training density and interval and intensity coupling factors according to the target training parameters, using a random forest model to process the training density and interval and intensity coupling factors, outputting a dynamic compensation coefficient and an intensity correction gradient, and performing nonlinear constraints on the intensity correction gradient; using an attention gated network to process the dynamic compensation coefficient, real-time training parameters and intensity correction gradient, including: the real-time training parameters include the actual inter-group interval deviation rate, using the STL decomposition method to perform time series decomposition on the actual inter-group interval deviation rate, and generating trend terms and mutation terms; allocating weights of attention heads through dynamic compensation coefficients and intensity correction gradients, using a multi-head attention mechanism to perform feature enhancement on trend terms, and generating trend enhancement terms; outputting training effectiveness indexes and abnormal risk levels according to the stability of the trend enhancement terms and the peak values ​​of the mutation terms.

[0006] Furthermore, the target training parameters include training volume, training intensity, target interval time between groups, and target interval time within a group; the training density and the interval-intensity coupling factor are calculated based on the target training parameters, including: dividing the training volume by the product of the target interval time between groups and the target interval time within a group to obtain the training density; multiplying the difference between the target interval time between groups and the target interval time within a group by the training intensity to obtain the interval-intensity coupling factor.

[0007] Furthermore, a random forest model is used to process the training density and the intermittent and intensity coupling factors, and output dynamic compensation coefficients and intensity correction gradients, including: normalizing the training density and the intermittent and intensity coupling factors as the dynamic feature weights of each decision tree; based on the dynamic feature weights, the CART algorithm is used to split nodes to generate dynamic compensation coefficients and intensity correction gradients.

[0008] Furthermore, the real-time training parameters also include load weight, and the load fluctuation rate is calculated based on the load weight; nonlinear constraints are applied to the intensity correction gradient, including: setting a basic range, a first threshold, and a minimum upper limit value of the intensity correction gradient, and when the load fluctuation rate exceeds the first threshold, reducing the upper limit value of the basic range to the minimum upper limit value; performing nonlinear compression after scaling the intensity correction gradient with the actual inter-group interval deviation rate; and using a gradient clipping algorithm to perform boundary constraints on the compressed intensity correction gradient based on the basic range.

[0009] Furthermore, the load fluctuation rate is obtained by subtracting the training intensity from the load weight and dividing it by the training intensity.

[0010] Furthermore, the real-time training parameters also include the actual intra-group intermittent deviation rate; the STL decomposition method is used to perform time series decomposition on the actual intermittent deviation rate between groups to generate trend terms and mutation terms, including: filtering outliers and filling missing values ​​on the actual intermittent deviation rate between groups to generate time series data; determining the decomposition window length according to the length of the time series data, and performing smoothing on the data in each window to output the trend term; taking the difference between the time series data and the trend term as the mutation term, and adjusting the amplitude of the mutation term based on the dynamic compensation coefficient.

[0011] Furthermore, the weights of the attention heads are allocated through the dynamic compensation coefficient and the intensity correction gradient, and the multi-head attention mechanism is used to perform feature enhancement on the trend item to generate a trend enhancement item, including: weighted summing the dynamic compensation coefficient and the intensity correction gradient to generate the allocation weight of each attention head in the multi-head attention mechanism; the multi-head attention mechanism is used to perform feature enhancement on the trend item, and the output results of multiple attention heads are normalized and weighted fused to generate a trend enhancement item.

[0012] Furthermore, a training effectiveness index and an abnormal risk level are output according to the stability of the trend enhancement item and the peak value of the mutation item, including: dividing the trend enhancement item into windows, calculating the variance of all trend enhancement items contained in each window as a stability index; extracting the maximum absolute value of the mutation item in the time series as a peak value; assigning weights to the stability index and the peak value, and linearly mapping the weighted sum of the stability index and the peak amplitude to a specified interval to obtain a training effectiveness index; and outputting the abnormal risk level according to the peak value and the dynamic compensation coefficient.

[0013] In a second aspect, the present disclosure provides a strength training intelligent monitoring and abnormality identification system, which includes: a parameter acquisition module, a factor calculation and model derivation module, and a real-time data processing and evaluation output module.

[0014] The parameter acquisition module is used to obtain target training parameters and real-time training parameters.

[0015] The factor calculation and model derivation module is used to calculate the training density and the intermittent and intensity coupling factors according to the target training parameters, use the random forest model to process the training density and the intermittent and intensity coupling factors, output the dynamic compensation coefficient and the initial intensity correction gradient, and perform nonlinear constraints on the intensity correction gradient; The real-time data processing and evaluation output module is used to process dynamic compensation coefficients, real-time training parameters and intensity correction gradients using an attention gated network, including: the real-time training parameters include the actual inter-group interval deviation rate, and the STL decomposition method is used to perform time series decomposition on the actual inter-group interval deviation rate to generate trend terms and mutation terms; the weights of the attention heads are allocated through the dynamic compensation coefficient and the intensity correction gradient, and the multi-head attention mechanism is used to enhance the features of the trend terms to generate trend enhancement terms; the training efficiency index and abnormal risk level are output according to the stability of the trend enhancement term and the peak value of the mutation term.

[0016] In a third aspect, the present disclosure provides a strength training intelligent monitoring and abnormality identification device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of the strength training intelligent monitoring and abnormality identification method as described in the first aspect when executing the computer program.

[0017] In a fourth aspect, the present disclosure provides a storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, the steps of the strength training intelligent monitoring and abnormality identification method as described in the first aspect are executed.

[0018] Beneficial effects of the present invention: The present invention solves the problem that in the prior art, strength training mostly adopts a fixed-parameter mode and mainly executes according to a pre-designed plan, making it difficult to accurately evaluate training efficacy and identify abnormal risks. By obtaining the target and real-time training parameters, calculating the training density and the coupling factor of rest intervals and intensity, generating and constraining the dynamic compensation coefficient and the initial intensity correction gradient through a random forest, decomposing the deviation rate into a trend term and a mutation term using an attention network, allocating weights to enhance the trend term, and outputting the efficacy index and risk level according to stability and peak value, the present invention can accurately evaluate training efficacy, effectively identify abnormal risks, significantly improve the scientific nature of training, optimize the training effect, and ensure the safety of the training process.

[0019] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structure pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0021] Figure 1 FIG. shows a schematic flow chart of the intelligent monitoring and abnormal identification method for strength training according to the present invention; Figure 2 FIG. shows a schematic block diagram of the intelligent monitoring and abnormal identification system for strength training according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to solve the problem that in the prior art, strength training mostly adopts a fixed-parameter mode and mainly executes according to a pre-designed plan, making it difficult to accurately evaluate training efficacy and identify abnormal risks, the present disclosure obtains the target and real-time training parameters, calculates the training density and the coupling factor of rest intervals and intensity, generates and constrains the dynamic compensation coefficient and the initial intensity correction gradient through a random forest, decomposes the deviation rate into a trend term and a mutation term using an attention network, allocates weights to enhance the trend term, and outputs the efficacy index and risk level according to stability and peak value, so as to accurately evaluate training efficacy, effectively identify abnormal risks, significantly improve the scientific nature of training, optimize the training effect, and ensure the safety of the training process.

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In some embodiments, Figure 1 As shown, the present disclosure provides a strength training intelligent monitoring and abnormality identification method, comprising: S100. Obtain target training parameters and real-time training parameters, where the real-time training parameters include actual inter-group interval deviation rate.

[0025] S200. Calculate the training density and the intermittent-intensity coupling factor according to the target training parameters, use the random forest model to process the training density and the intermittent-intensity coupling factor, output the dynamic compensation coefficient and the intensity correction gradient, and perform nonlinear constraints on the intensity correction gradient.

[0026] S300. Uses attention gated network to process dynamic compensation coefficients, real-time training parameters and intensity correction gradients, including: S310. Use the STL decomposition method to perform time series decomposition on the actual inter-group intermittent deviation rate to generate trend terms and mutation terms.

[0027] S320. The weights of the attention heads are allocated through the dynamic compensation coefficient and the intensity correction gradient, and the multi-head attention mechanism is used to enhance the features of the trend items to generate trend enhancement items.

[0028] S330. Output training effectiveness index and abnormal risk level according to the stability of trend enhancement items and the peak value of mutation items.

[0029] In some embodiments, the target training parameters in S100 include training volume, training intensity, target interval time between groups, and target interval time within a group.

[0030] ; Where V represents the training volume, S represents the number of sets, R represents the number of times per set, and W represents the single load weight. For example, if the user completes 3 sets of squats, 10 times per set, and 100 kg per load, the training volume is: .

[0031] Training intensity refers to the percentage of the load weight of a single action to the user's maximum strength. For example, if the user's squat 1RM is 150kg, if 120kg is used in this training, the training intensity is .

[0032] The target interval time between sets is the preset rest time between two sets of training, which is used to control the training density and recovery rhythm.

[0033] The target interval time within a group is the preset interval time between two consecutive actions in the same group, which is used to maintain the rhythm of the action.

[0034] In some embodiments, calculating the training density and the interval-intensity coupling factor according to the target training parameters in S200 includes: Divide the training volume by the product of the target interval time between groups and the target interval time within groups to get the training density.

[0035] Multiply the difference between the target interval time between groups and the target interval time within a group by the training intensity to obtain the interval-intensity coupling factor.

[0036] ; Where D represents training density and V represents training volume. The target rest time between reps. Represents the target time for the interval within the set.

[0037] ; Among them, C represents the interval and intensity coupling factor, and I represents the training intensity.

[0038] For example, the user performs squat training with a training volume of The target interval time between groups is 120 seconds, and the target interval time within the group is 60 seconds. The training density is: .

[0039] Training density reflects the completion of training volume per unit time and can assist in adjusting the training rhythm.

[0040] The rest time between sets is relatively long, the purpose is to allow the body to recover to a greater extent from the fatigue of a set of training, so that the next set of training can be carried out in a better state, while the rest time within a set is relatively short, mainly for briefly adjusting breathing and preparing for the next movement, and the body's recovery is limited.

[0041] By subtracting the target interval time within a group from the target interval time between groups, the difference in recovery degree can be reflected. The larger the difference, the more time there is between groups to restore physical strength and function, and prepare for the next set of high-intensity training.

[0042] Training intensity directly determines the load borne by the muscles, and the interval and intensity coupling factor can predict the muscle fatigue rate to a certain extent.

[0043] In some embodiments, the random forest model is used in S200 to process the training density and the intermittent and intensity coupling factors, and output the dynamic compensation coefficient and the intensity correction gradient, including: S210. Normalize the training density and the interval-intensity coupling factor as the dynamic feature weight of each decision tree.

[0044] S220. Based on the dynamic feature weight, the CART algorithm is used to split the nodes and generate dynamic compensation coefficients and intensity correction gradients.

[0045] The Min-Max normalization method can be used to process the training density and the intermittent and intensity coupling factor, and the normalized training density and intermittent and intensity coupling factor are used as the dynamic feature weights of each decision tree.

[0046] According to the dynamic feature weight, an optimal feature and an optimal splitting point on the feature are selected from the training density and the intermittent and intensity coupling factors. The optimal splitting point can be found with the goal of minimizing the mean square error.

[0047] In each decision tree, nodes represent tests of training density or intermittent-intensity coupling factors, branches represent the results of the tests, and leaf nodes represent dynamic compensation coefficients and intensity correction gradients.

[0048] When building a decision tree, each node is evaluated.

[0049] When evaluating a node, the mean square error between the generated result and the actual target value is calculated when different values ​​of the training density and the intermittent and intensity coupling factor are used as split points. For example, if the mean square error is smaller when the training density is 0.6 as the split point, then 0.6 is the optimal split point for the node based on the training density feature, and the data set is divided accordingly to form the left and right child nodes of the decision tree.

[0050] For the generated left and right child nodes, the operations of feature selection and node splitting are repeated until the pre-set stopping condition is met.

[0051] The stopping condition can be that the depth of the tree reaches a preset value, such as setting the maximum depth to 5 layers; it can also be that the number of samples in the node is less than a certain threshold, such as stopping the split when the number of node samples is less than 10.

[0052] Through continuous recursive construction, a complete decision tree structure is formed. Many decision trees together form a random forest. Finally, the output results of multiple decision trees are integrated, such as taking the average, to generate a dynamic compensation coefficient and intensity correction gradient.

[0053] The intensity-corrected gradient is used in subsequent attention mechanisms and anomaly detection.

[0054] The intensity correction gradient is used to dynamically adjust the training intensity according to real-time training data (such as load fluctuations, inter-group interval deviations, etc.), helping the trainer to automatically reduce the intensity adjustment range when the load fluctuates greatly to ensure training safety; in a stable state, a larger range of intensity optimization is allowed to improve the effect.

[0055] In some embodiments, the real-time training parameter in S100 further includes the load weight, and the load volatility can be calculated according to the load weight.

[0056] The non-linear constraint on the intensity correction gradient in S200 includes: S230. Set the basic range, the first threshold, and the lowest upper limit value of the intensity correction gradient. When the load volatility exceeds the first threshold, reduce the upper limit value of the basic range to the lowest upper limit value.

[0057] S240. Perform non-linear compression after scaling the intensity correction gradient with the actual inter-group interval deviation rate.

[0058] S250. Use the gradient clipping algorithm to perform boundary constraint on the compressed intensity correction gradient based on the basic range.

[0059] The basic range can be , the first threshold can be 15%, and the lowest upper limit value can be 8%.

[0060] When the load volatility is , the upper limit value of the basic range drops from to

[0061] , which can effectively avoid excessive intensification under high fluctuations, thereby reducing the risk of injury. ; where represents the intensity correction gradient after scaling, represents the intensity correction gradient before scaling, represents the actual inter-group interval deviation rate.

[0062] The non-linear compression formula can be: ; where represents the intensity correction gradient after non-linear compression by the tanh function, represents the upper limit value of the basic range.

[0063] For example, if the intensity correction gradient is and the actual inter-group interval deviation rate is 20%, then the intensity correction gradient after scaling is: .

[0064] Use the gradient clipping algorithm to force the compressed gradient to be within the dynamically adjusted basic range, and the rules are as follows: If the compressed strength correction gradient is greater than the upper limit of the base range, the compressed strength correction gradient takes the upper limit of the base range; if the compressed strength correction gradient is less than the lower limit of the base range, the compressed strength correction gradient takes the lower limit of the base range.

[0065] Gradient clipping can effectively ensure that the correction gradient is strictly within the safe range.

[0066] In some embodiments, the difference between the load weight and the training intensity is divided by the training intensity to obtain the load volatility. The reference formula is: ; where F represents the load volatility, W represents the actual load weight, and I represents the training intensity.

[0067] In some embodiments, the real-time training parameters in S100 further include the actual intra-group interval deviation rate.

[0068] The actual intra-group interval deviation rate represents the deviation percentage of the actual interval time between two adjacent actions in the same group of training from the preset intra-group interval target time, and is calculated after obtaining the actual intra-group interval time.

[0069] The calculation formula is: , where represents the actual inter-group interval deviation rate, represents the actual inter-group interval time, the actual intra-group interval time represents the actual interval time between two consecutive actions within the same group, represents the inter-group interval target time.

[0070] The training status of the trainer changes continuously over time, and the actual inter-group interval deviation rate reflects the rhythm changes during the training process.

[0071] In some embodiments, the attention gating network in S300 can capture the trends and mutation information during the training process by deeply analyzing this deviation rate, combining the dynamic compensation coefficient and the strength correction gradient, so as to provide more targeted training suggestions for coaches and trainers.

[0072] For example, when the trend term shows that the training intensity gradually and steadily increases, and the mutation term shows an abnormal peak, it may indicate that the trainer is at risk of fatigue or injury. At this time, the system can output the abnormal risk level in a timely manner and adjust the training plan.

[0073] The input of the attention gating network is the dynamic compensation coefficient, real-time training parameters, and strength correction gradient, and the output is the training efficiency index and the abnormal risk level.

[0074] The training effectiveness index can comprehensively reflect the effect and efficiency of training and provide quantitative indicators for evaluating training quality. The abnormal risk level can classify possible risks according to abnormal situations in the training process, helping coaches and trainers to take timely response measures.

[0075] The attention gating network consists of the following core modules: time series decomposition module, weight distribution module, multi-head attention layer and evaluation module.

[0076] The time series decomposition module is used to receive the actual inter-group interval deviation rate in the real-time training parameters, and output the trend term and mutation term using the STL decomposition method.

[0077] The weight assignment module is used to receive the dynamic compensation coefficient and the intensity correction gradient and calculate the weights of the multi-head attention heads.

[0078] The multi-head attention layer is used to perform feature enhancement on trend items and generate trend enhancement items.

[0079] The evaluation module is used to output the training effectiveness index and abnormal risk level according to the stability of the trend enhancement item and the peak value of the mutation item.

[0080] The connection relationship of the core modules is as follows: The actual inter-group intermittent deviation rate is input into the time series decomposition module, the decomposed trend term is input into the multi-head attention layer, and the mutation term is input into the evaluation module.

[0081] The dynamic compensation coefficient and intensity correction gradient are input into the weight distribution module, and the attention head weights are generated and passed to the multi-head attention layer.

[0082] The trend enhancement term and mutation term output by the multi-head attention layer are input into the evaluation module together to generate the final result.

[0083] In some embodiments, in S310, the STL decomposition method is used to perform time series decomposition on the actual inter-group intermittent deviation rate to generate trend terms and mutation terms, including: S311. Perform outlier filtering and missing value filling on the actual inter-group interval deviation rate to generate time series data.

[0084] The actual inter-group intermittent deviation rate data may contain outliers, such as extreme values ​​and missing values ​​caused by measurement errors. Outliers can be filtered out by statistical methods, such as outlier detection based on standard deviation, and missing values ​​can be filled using interpolation to generate continuous time series data. , ,in, represents the Tth actual inter-group interval deviation rate, and T represents the number of actual inter-group interval deviation rates.

[0085] S312. Determine the decomposed window length according to the length of the time series data, perform smoothing on the data in each window, and output a trend item.

[0086] Determine the appropriate window length based on the length of the time series data. For example, the LOWESS algorithm of STL decomposition can be used to set the period window length L, L = After determining the window length, smooth the data in the window and set the smoothing parameters , through the calculation process of iterating 3 times, the trend item is finally output .

[0087] S313. The difference between the time series data and the trend term is used as the mutation term, and the amplitude of the mutation term is adjusted based on the dynamic compensation coefficient.

[0088] Mutation Item , through the dynamic compensation coefficient Adjustment amplitude, the mutation term after adjustment amplitude is .

[0089] The dynamic compensation coefficient will amplify or reduce the value of the mutation term. For example, when the compensation coefficient is high, abnormal fluctuations will be marked more significantly.

[0090] In some embodiments, in S320, the weights of the attention heads are allocated by the dynamic compensation coefficient and the intensity correction gradient, and the multi-head attention mechanism is used to enhance the features of the trend items to generate the trend enhancement items, including: S321. Perform weighted summation of the dynamic compensation coefficient and the intensity correction gradient to generate the allocation weight of each attention head in the multi-head attention mechanism.

[0091] Through grid search and cross-validation, the model performance of different weight combinations can be compared on the historical training data set to obtain the dynamic compensation coefficient and intensity correction gradient The allocation weight of .

[0092] For example, if the dynamic compensation coefficient and intensity correction gradient The weight coefficients are 0.6 and 0.4 respectively, so the allocation weight of each attention head is Dynamic compensation coefficient and intensity correction gradient The weighted sum is normalized by Softmax, refer to the formula: .

[0093] S322. Use a multi-head attention mechanism to enhance the features of trend items, and perform normalized weighted fusion on the output results of multiple attention heads to generate trend enhancement items.

[0094] Each attention head maps the trend item into a query matrix, a key matrix, and a value matrix through a linear transformation. The query matrix is ​​multiplied by the transpose of the key matrix, and after scaling, the attention weight is generated through the Softmax function. The attention weight is used to weight the sum of the value matrix and output the single-head attention result.

[0095] Using the assigned weight of each attention head, the outputs of multiple attention heads are weighted and summed according to the weights to finally generate a trend enhancement term.

[0096] The dynamic compensation coefficient is combined with the intensity correction gradient to determine the degree of focus of the model on the trend term. For example, when the compensation coefficient is high, the model pays more attention to the long-term stable trend and weakens the short-term noise.

[0097] In some embodiments, outputting the training effectiveness index and the abnormal risk level according to the stability of the trend enhancement item and the peak value of the mutation item in S330 includes: S331. Divide the trend enhancement items into windows, and calculate the variance of all trend enhancement items contained in each window as a stability indicator.

[0098] S332. Extract the maximum absolute value of the mutation item in the time series as the peak value.

[0099] S333. Assign weights to the stability index and the peak value, linearly map the weighted sum of the stability index and the peak amplitude to the specified interval, and obtain the training effectiveness index; output the abnormal risk level according to the peak value and the dynamic compensation coefficient.

[0100] According to the variance calculation formula, the variance of all trend enhancement items contained in the window is calculated, and the variance in each window is normalized to obtain the stability index. The formula can be: S= , where S represents the stability index, Represents the variance of all trend enhancement items contained in the current window, Represents the maximum value of all window variances.

[0101] If the mutation term is sequence M, the peak value is the maximum value in sequence M.

[0102] The stability index and the peak value can be weighted and summed at a weight of 50% each, and then linearly mapped to a specified interval, such as the 0-100 interval, to obtain the training effectiveness index. The formula can be:

[0103] Among them, E represents the training efficiency index, S represents the stability index, and P represents the peak value. Indicates the preset peak upper limit.

[0104] Exemplarily, the risk levels may include high risk, medium risk, and low risk.

[0105] When the peak And the dynamic compensation coefficient , it is judged as high risk.

[0106] when or , it is judged as medium risk.

[0107] When the high risk and medium risk conditions are not met, it is judged as low risk.

[0108] When training the attention gated network, the dynamic compensation coefficient, real-time training parameters and intensity correction gradient are taken as input, and the trend enhancement term, training effectiveness index and abnormal risk level are taken as output.

[0109] The mean square error can be used for regression tasks of training performance index, the cross entropy loss can be used for classification tasks of abnormal risk level, etc., and the Adam optimization algorithm can be used to minimize the loss function.

[0110] During the training process, the parameters of the attention gating network are continuously adjusted to make the output of the network as close to the true value as possible. The gradient of the loss function with respect to the network parameters is calculated through the back-propagation algorithm, and the parameters are updated according to the optimization algorithm.

[0111] Through multiple iterations, the parameters of the network are gradually adjusted until the loss function converges to a smaller value or reaches the preset number of training rounds.

[0112] The training effectiveness index can reflect the overall effect and efficiency of the current training. If the index is low, it means that the training effect may not be ideal. You can consider adjusting the training plan, such as changing parameters such as training intensity, interval time or training volume, to improve training effectiveness.

[0113] For example, if it is found that the training density is low due to the long interval between groups, which in turn affects the training efficiency index, the interval time between groups can be appropriately shortened.

[0114] The risk level can directly indicate the potential risks in the training process. If you are in a high-risk or medium-risk state, you need to take timely measures to adjust your training to avoid injury or overtraining.

[0115] For example, when it is judged to be high risk, you may need to stop the current training immediately and conduct a comprehensive assessment and adjustment of the training plan; if it is medium risk, you can appropriately reduce the training intensity or increase the interval time to reduce the risk.

[0116] Different trainees have different physical conditions and training goals. Through the training effectiveness index and risk level, a personalized training plan can be developed for each trainee. According to the training effectiveness index, we can understand each trainee's adaptation to different training parameters, so as to optimize the training plan to make it more in line with individual characteristics.

[0117] For example, for trainees with a high training effectiveness index but also a high risk level, the risk can be reduced by adjusting the training parameters while maintaining the training effect; and for trainees with a low training effectiveness index, the training method can be adjusted in a targeted manner to improve the training effect.

[0118] Continuously recording the training effectiveness index and risk level can track and evaluate the long-term training effect and physical condition of the trainees.

[0119] Observe the changing trend of the training effectiveness index to understand whether the training plan is effective in the long term and whether the trainee's body is adapted to this training method. At the same time, according to the changes in the risk level, timely discover possible problems in the training process, such as overtraining and accumulation of physical fatigue, so as to take preventive measures in advance to ensure the trainee's physical health and the continuity of training.

[0120] In some embodiments, Figure 2 As shown, the present disclosure provides a strength training intelligent monitoring and abnormality identification system, which includes: a parameter acquisition module, a factor calculation and model derivation module, and a real-time data processing and evaluation output module.

[0121] The parameter acquisition module is used to obtain target training parameters and real-time training parameters.

[0122] The factor calculation and model derivation module is used to calculate the training density and the intermittent and intensity coupling factors according to the target training parameters, use the random forest model to process the training density and the intermittent and intensity coupling factors, output the dynamic compensation coefficient and the initial intensity correction gradient, and perform nonlinear constraints on the intensity correction gradient.

[0123] The real-time data processing and evaluation output module is used to process dynamic compensation coefficients, real-time training parameters and intensity correction gradients using an attention gated network, including: the real-time training parameters include the actual inter-group interval deviation rate, and the STL decomposition method is used to perform time series decomposition on the actual inter-group interval deviation rate to generate trend terms and mutation terms; the weights of the attention heads are allocated through the dynamic compensation coefficient and the intensity correction gradient, and the multi-head attention mechanism is used to enhance the features of the trend terms to generate trend enhancement terms; the training efficiency index and abnormal risk level are output according to the stability of the trend enhancement term and the peak value of the mutation term.

[0124] This embodiment has all the advantages of the strength training intelligent monitoring and abnormality identification method, and can automatically execute the steps of the strength training intelligent monitoring and abnormality identification method.

[0125] In some embodiments, the present disclosure provides a strength training intelligent monitoring and abnormality identification device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of the strength training intelligent monitoring and abnormality identification method when executing the computer program.

[0126] In some embodiments, the present disclosure provides a storage medium, in which computer program instructions are stored. When the computer program instructions are read and executed by a processor, the steps of the strength training intelligent monitoring and abnormality identification method are executed.

[0127] Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0128] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0129] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent monitoring and abnormality identification of strength training, characterized in that: include: Obtain target training parameters and real-time training parameters; The training density and the coupling factor between interval and intensity are calculated according to the target training parameters, and the training density and the coupling factor between interval and intensity are processed by the random forest model. The dynamic compensation coefficient and the intensity correction gradient are output, and the intensity correction gradient is nonlinearly constrained. An attention gated network is used to process dynamic compensation coefficients, real-time training parameters, and intensity correction gradients, including: The real-time training parameters include the actual inter-group interval deviation rate, which is decomposed into time series using the STL decomposition method to generate trend terms and mutation terms. The weights of the attention heads are allocated through dynamic compensation coefficients and intensity correction gradients, and the multi-head attention mechanism is used to enhance the features of trend items and generate trend enhancement items. The training effectiveness index and abnormal risk level are calculated based on the stability of the trend enhancement item and the peak output of the mutation item.

2. The strength training intelligent monitoring and abnormality identification method according to claim 1 is characterized in that: Target training parameters include training volume, training intensity, target interval time between groups, and target interval time within a group. Calculate training density and interval-intensity coupling factors based on target training parameters, including: Divide the training volume by the product of the target interval time between groups and the target interval time within a group to get the training density. Multiply the difference between the target interval time between groups and the target interval time within a group by the training intensity to obtain the interval-intensity coupling factor.

3. The strength training intelligent monitoring and abnormality identification method according to claim 2 is characterized in that: The random forest model is used to process the training density and the intermittent and intensity coupling factors, and the dynamic compensation coefficient and intensity correction gradient are output, including: Normalize the training density and the intermittent-intensity coupling factor as the dynamic feature weight of each decision tree; Based on the dynamic feature weights, the CART algorithm is used to split nodes and generate dynamic compensation coefficients and intensity correction gradients.

4. The strength training intelligent monitoring and abnormality identification method according to claim 3 is characterized in that: The real-time training parameters also include load weight, and the load fluctuation rate is calculated based on the load weight; Nonlinear constraints are imposed on the intensity correction gradient, including: A basic range, a first threshold value and a minimum upper limit value of the intensity correction gradient are set, and when the load fluctuation rate exceeds the first threshold value, the upper limit value of the basic range is reduced to the minimum upper limit value; Nonlinear compression is performed after scaling the intensity correction gradient with the actual inter-group interval deviation rate; A gradient clipping algorithm is used to perform boundary constraints on the compressed intensity correction gradient based on the base range.

5. The method for intelligent monitoring and abnormality identification of strength training according to claim 4 is characterized in that: The load fluctuation rate is calculated by subtracting the training intensity from the load weight and dividing it by the training intensity.

6. The strength training intelligent monitoring and abnormality identification method according to claim 3 is characterized in that: Real-time training parameters also include actual within-group interval deviation rate; The STL decomposition method is used to perform time series decomposition on the actual inter-group intermittent deviation rate to generate trend terms and mutation terms, including: The actual inter-group intermittent deviation rate was filtered for outliers and filled with missing values ​​to generate time series data; Determine the decomposed window length according to the length of the time series data, perform smoothing on the data in each window, and output the trend item; The difference between the time series data and the trend term is taken as the mutation term, and the amplitude of the mutation term is adjusted based on the dynamic compensation coefficient.

7. The method for intelligent monitoring and abnormality identification of strength training according to claim 6, characterized in that: The weights of the attention heads are allocated through dynamic compensation coefficients and intensity correction gradients, and the multi-head attention mechanism is used to enhance the features of trend items to generate trend enhancement items, including: The dynamic compensation coefficient and the intensity correction gradient are weighted and summed to generate the allocation weight of each attention head in the multi-head attention mechanism; A multi-head attention mechanism is used to enhance the features of trend items, and the output results of multiple attention heads are normalized and weighted fused to generate trend enhancement items.

8. The strength training intelligent monitoring and abnormality identification method according to claim 3 is characterized in that: According to the stability of trend enhancement items and the peak output of mutation items, the training efficiency index and abnormal risk level are as follows: The trend enhancement items are divided into windows, and the variance of all trend enhancement items contained in each window is calculated as a stability indicator; Extract the maximum absolute value of the mutation item in the time series as the peak value; Assign weights to the stability index and the peak value, linearly map the weighted sum of the stability index and the peak amplitude to the specified interval to obtain the training effectiveness index; output the abnormal risk level based on the peak value and the dynamic compensation coefficient.

9. A strength training intelligent monitoring and abnormality identification system, characterized in that: It includes: A parameter acquisition module, which is used to obtain target training parameters and real-time training parameters; The factor calculation and model derivation module is used to calculate the training density and the intermittent and intensity coupling factors according to the target training parameters, process the training density and the intermittent and intensity coupling factors using the random forest model, output the dynamic compensation coefficient and the initial intensity correction gradient, and perform nonlinear constraints on the intensity correction gradient; Real-time data processing and evaluation output module; It is used to process dynamic compensation coefficients, real-time training parameters and intensity correction gradients using an attention gated network, including: The real-time training parameters include the actual inter-group interval deviation rate, which is decomposed into time series using the STL decomposition method to generate trend terms and mutation terms. The weights of the attention heads are allocated through dynamic compensation coefficients and intensity correction gradients, and the multi-head attention mechanism is used to enhance the features of trend items and generate trend enhancement items. The training effectiveness index and abnormal risk level are calculated based on the stability of the trend enhancement item and the peak output of the mutation item.

10. A storage medium, characterized in that: The storage medium stores computer program instructions, and when the computer program instructions are read and executed by a processor, the steps of the strength training intelligent monitoring and abnormality identification method described in any one of claims 1-8 are executed.

Images