Teenager long-distance running strategy method integrating deep learning and improved swarm intelligence optimization algorithm

Abstract

The invention discloses a teenager long-distance running strategy method fusing deep learning and an improved swarm intelligence optimization algorithm, and the method comprises the following steps: collecting multi-dimensional time sequence features in the historical training and competition process of athletes, and constructing and training a CNN-LSTM-Attention prediction model based on the multi-dimensional time sequence features; establishing an integral form long-distance running strategy optimization model framework with the goal of minimizing the game completion time; dividing the long-distance running process into a plurality of continuous time sequence windows, calling a CNN-LSTM-Attention model at the starting moment of each window to generate a state prediction result as a dynamic constraint, and constructing a state transition equation; and based on the state transition equation, solving the optimal speed under the current time sequence window through an improved dung beetle optimization algorithm, rolling to the next window after execution, and carrying out iterative execution until the optimal speeds under all time sequence windows are output.

Description

Technical Field

[0001] This invention relates to the field of sports intelligent planning technology, and in particular to a long-distance running strategy method for teenagers that integrates deep learning and improved swarm intelligence optimization algorithms. Background Technology

[0002] With the rapid development of competitive sports, the scientific optimization of long-distance running strategies for young athletes has become a core issue in improving athlete performance, preventing sports injuries, and optimizing training plans. Traditional methods often rely on coaching experience or simplified segmented models, which have the following limitations: Strong reliance on experience: Subjective experience makes it difficult to accurately quantify the dynamic coupling relationship between energy consumption, heart rate changes, and speed distribution. Insufficient global optimization capability: Rule-based or gradient descent-based methods are prone to getting trapped in local optima and cannot find the global optimum under complex constraints (such as energy limits and heart rate safety thresholds). Poor real-time performance: Algorithms such as dynamic programming have high computational complexity and are difficult to adapt to the real-time strategy adjustment needs during competitions.

[0003] In recent years, swarm intelligence optimization algorithms (such as particle swarm optimization and genetic algorithms) have been increasingly applied in the field of sports science due to their parallel search and global optimization capabilities. For example, particle swarm optimization (PSO) achieves multi-objective optimization by simulating the foraging behavior of bird flocks, but its convergence speed is affected by parameter sensitivity; while genetic algorithms (GA) possess strong robustness, they are prone to premature convergence in high-dimensional problems. Furthermore, existing research often neglects the dynamic coupling effect of physiological constraints (such as the nonlinear relationship between heart rate and speed), leading to a disconnect between optimization results and actual physiological limits.

[0004] Meanwhile, deep learning technology has made significant progress in the field of time series prediction. In particular, the CNN-LSTM-Attention hybrid model, which extracts local features through convolutional neural networks, captures long-term dependencies through long short-term memory networks, and focuses key information through attention mechanisms, can accurately predict the changing trends of physiological parameters such as athletes' heart rate and energy consumption. However, current research has not yet deeply integrated these accurate predictions with the strategy optimization process.

[0005] To address the aforementioned issues, some scholars have proposed dung beetle optimization algorithms based on the habits of dung beetles. However, in long-running strategy optimization, balancing global exploration and local development, and efficiently handling multiple constraint couplings remain technical challenges. This is particularly true in real-time strategy adjustment under dynamic environments, where effective integration with accurate prediction models is lacking. This invention proposes an improved dung beetle optimization algorithm (HDBO), which integrates SPM chaotic initialization, Lévy flight perturbation, and differential evolution strategies, and combines the output of a CNN-LSTM-Attention prediction model as dynamic constraints, thus overcoming the shortcomings of traditional methods. This hybrid approach fully leverages the accurate prediction capabilities of deep learning and the global optimization advantages of swarm intelligence algorithms, providing more precise theoretical support and technical assurance for real-time dynamic optimization of long-running strategies. Summary of the Invention

[0006] In view of the above-mentioned prior art, the present invention provides a method and system for developing long-distance running strategies for teenagers that integrates deep learning and improved swarm intelligence optimization algorithms, mainly to solve the technical problems existing in the background art.

[0007] To achieve the above objectives, the technical solution of this invention is implemented as follows: a method for developing long-distance running strategies for teenagers that integrates deep learning and improved swarm intelligence optimization algorithms, the method comprising the following steps: Collect multidimensional temporal features from athletes’ historical training and competition processes, and construct and train a CNN-LSTM-Attention prediction model based on the multidimensional temporal features. Establish a framework for a points-based long-distance running strategy optimization model with the objective of minimizing the finishing time; The long-distance running process is divided into multiple consecutive time windows. At the beginning of each window, the CNN-LSTM-Attention model is called to generate state prediction results as dynamic constraints to construct state transition equations. Based on the state transition equation, the optimal speed under the current time window is solved by improving the dung beetle optimization algorithm. After execution, it scrolls to the next window and iterates until the optimal speed under all time windows is output.

[0008] Optionally, the multidimensional time-series features include instantaneous speed sequence, heart rate sequence, energy consumption rate, distance run, environmental data, and individual characteristics. The environmental data includes temperature data, humidity data, and altitude data, and the individual characteristics include maximum oxygen uptake and fatigue index.

[0009] Optionally, the CNN-LSTM-Attention prediction model is composed of a CNN convolutional neural network, an LSTM long short-term memory network, and an attention mechanism unit connected in series. The CNN convolutional neural network is used for local feature extraction to capture the local dependencies between multiple features within a short time period. The LSTM (Long Short-Term Memory) network is used to receive the high-level feature sequences extracted by the CNN (Convolutional Neural Network) and capture long-term dependencies within a long time period. The attention mechanism unit is used to assign different weights to the hidden states of the LSTM long short-term memory network at all time steps and output a sequence of state prediction results for the next τ time steps.

[0010] Optionally, a framework for optimizing long-distance running strategies in integral form, with the objective of minimizing the finishing time, is constructed, specifically including: Define a race time objective function with the total length of the track as the integration interval and the goal of minimizing the race time; An energy consumption constraint based on exercise metabolism is introduced, and the remaining energy prediction result in the CNN-LSTM-Attention prediction model is used as a dynamic upper limit condition. Add velocity smoothness constraints and construct heart rate-velocity nonlinear coupling constraints, and set a safety threshold margin by combining the heart rate prediction results in the CNN-LSTM-Attention prediction model.

[0011] Optionally, the expression for the defined completion objective function is:

[0012] Where v(x) represents the speed at a distance x, and L represents the total length of the track.

[0013] Optionally, the long-distance running process can be divided into n consecutive time windows of equal length.

[0014] Optionally, the athlete's real-time state data can be input into the CNN-LSTM-Attention model to obtain a sequence of state prediction results consisting of multiple state prediction results, including remaining energy prediction results and heart rate prediction results.

[0015] Optionally, construct the state transition equations, specifically including: Construct the state transition equation for the remaining energy:

[0016] in, This represents the athlete's remaining energy at the start of the (i+1)th time window. Let represent the athlete's remaining energy at the start of the i-th time window, and let a, b, and c represent different energy consumption coefficients. This represents the instantaneous velocity used within the i-th time window. Indicates the length of the time window. For integration operations; Construct the transition equation for the distance already traveled:

[0017] in, This represents the cumulative distance the athlete has run at the start of the (i+1)th time window. This represents the cumulative distance the athlete has run at the start of the i-th time window; Constructing a real-time heart rate equation:

[0018] in, For the fitting parameters, This represents the athlete's real-time heart rate at the start of the (i+1)th time window. This indicates the resting heart rate.

[0019] Optionally, based on the state transition equation, the optimal speed under the current time window is solved by an improved dung beetle optimization algorithm, specifically including: The initial population is generated using a sinusoidal product mapping. Update the individual's position by performing a rolling ball action; Introduce the Levy flight disturbance for local search; Implement differential evolution strategies to induce population mutation and crossover; Perform constraint processing and fitness calculation to determine the optimal speed for the current window.

[0020] The beneficial effects of this invention are as follows: 1. By introducing a CNN-LSTM-Attention deep learning prediction model, it can accurately predict key physiological indicators such as heart rate and energy consumption in multiple future time windows. This makes the optimization strategy no longer a passive reaction based on the current state, but a proactive planning based on the prediction results, greatly improving the scientific nature of the strategy and enabling athletes to adjust their pace in advance to avoid risks such as exhaustion or heart rate overload; 2. By dividing the long-distance running process into multiple time windows, within each window, the system re-solves and executes a short-term optimization problem based on the latest real-time data and prediction information. This mechanism enables the system to respond in real time to uncertainties such as fluctuations in athlete condition and changes in the track environment, achieving dynamic adaptive adjustment and ensuring the real-time performance and effectiveness of the strategy; 3. An improved dung beetle optimization algorithm (HDBO) is adopted, which integrates SPM chaotic initialization, Lévy flight perturbation, and differential evolution strategy. This combination effectively balances the algorithm's global exploration and local exploitation capabilities, enabling it to efficiently escape local optima and, under multiple complex constraints such as energy, heart rate, and speed smoothness, search for globally optimal or near-optimal solutions, thereby formulating strategies with better overall performance. These improvements allow the algorithm to achieve both faster convergence speed and higher solution quality when solving complex optimization problems with high dimensions and multiple constraints. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the adolescent long-distance running strategy method that integrates deep learning and improved swarm intelligence optimization algorithms in the embodiments of this application.

[0022] Figure 2 This is a flowchart illustrating the solution obtained using the improved dung beetle optimization algorithm in the embodiments of this application. Detailed Implementation

[0023] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. In the following description, the expression "some embodiments" refers to a subset of all possible embodiments; however, it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0024] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0025] It should be understood that the present invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Furthermore, the terminology used herein is intended only to describe particular embodiments and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “compose” and / or “comprising,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0026] It should also be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "inner," "outer," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0027] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.

[0028] Please refer to the attached document. Figure 1 This application provides a method for developing long-distance running strategies for teenagers that integrates deep learning and improved swarm intelligence optimization algorithms. The method includes the following steps: S1. Collect multi-dimensional temporal features from the athlete's historical training and competition process, and construct and train a CNN-LSTM-Attention prediction model based on the multi-dimensional temporal features; Specifically, the multidimensional time-series features include instantaneous speed sequence, heart rate sequence, energy consumption rate, distance run, environmental data, and individual characteristics. The environmental data includes temperature data, humidity data, and altitude data, while the individual characteristics include maximum oxygen uptake and fatigue index.

[0029] Furthermore, the multidimensional temporal features are standardized, denoised, and sequence aligned to construct a supervised learning dataset.

[0030] Furthermore, the CNN-LSTM-Attention prediction model is composed of a CNN convolutional neural network, an LSTM long short-term memory network, and an attention mechanism unit connected in series. The CNN convolutional neural network is used for local feature extraction, capturing the local dependencies between multiple features within a short time period. It uses one-dimensional convolutional layers (Conv1D) and pooling layers to process the input time-series data.

[0031] When using a one-dimensional convolutional layer to extract local temporal features, the output of the l-th convolutional kernel at time t is:

[0032] Where k is the kernel size, and represents the trainable weights and biases; X represents the input data over a period of time; ReLU is the activation function; the outputs of multiple convolutional kernels are concatenated and downsampled through pooling layers to obtain a high-level feature map.

[0033] The LSTM (Long Short-Term Memory) network is used to receive high-level feature sequences extracted by the CNN (Convolutional Neural Network) and capture long-term dependencies over long time periods, such as heart rate drift after continuous high-speed running and energy decay trends.

[0034] The attention mechanism unit is used to assign different weights to the hidden states of the LSTM (Long Short-Term Memory) network at all time steps. This allows the model to focus on the most critical historical moments for predicting future states, such as the sprint phase or the moment of the extreme point, enhancing the model's interpretability and prediction accuracy, and outputting a sequence of state predictions for the next τ time steps. The attention weights are:

[0035]

[0036] in, Let be the hidden state at time t; is a trainable weight matrix used to perform linear transformations on the hidden state; is a trainable bias vector; This is the new representation after the hidden state has been transformed.

[0037] By summing all weights with their corresponding LSTM hidden states, we obtain a context vector c that encapsulates all the key information.

[0038]

[0039] Passing the context vector c through a fully connected layer outputs predicted values ​​for the next τ time steps, such as heart rate and energy:

[0040]

[0041] During training, the dataset is divided into training, validation, and test sets. Supervised learning is performed using mean squared error (MSE) as the loss function. For the predicted value, its loss function is:

[0042] S2. Establish a framework for a long-distance running strategy optimization model based on points, with the goal of minimizing the finishing time. Specifically, a framework for optimizing long-distance running strategies using an integral form, with the objective of minimizing the finishing time, is constructed. First, a finishing time objective function is defined, with the total course length as the integral interval and the goal of minimizing the finishing time. The expression for the defined finishing time objective function is as follows:

[0043] Where v(x) represents the speed at a distance x, and L represents the total length of the track.

[0044] An energy consumption constraint based on exercise metabolism is introduced, and the remaining energy prediction result in the CNN-LSTM-Attention prediction model is used as a dynamic upper limit condition. Add velocity smoothness constraints and construct heart rate-velocity nonlinear coupling constraints, and set a safety threshold margin by combining the heart rate prediction results in the CNN-LSTM-Attention prediction model.

[0045] To ensure athletes do not exceed their energy reserves during competition and guarantee the sustainability of the competition, an energy expenditure constraint based on an exercise metabolism model is introduced. The prediction results from the previous step are used as dynamic constraints. The expression for the energy expenditure constraint is as follows:

[0046] in, These are energy consumption coefficients (J / m), which are calibrated experimentally and reflect the energy consumption patterns at different speeds. It is the remaining energy prediction result output by the CNN-LSTM-Attention prediction model.

[0047] To avoid abrupt changes in athlete speed, reduce unnecessary energy expenditure, and help athletes maintain a stable rhythm, a speed smoothness constraint is added to prevent drastic fluctuations. The expression for the speed smoothness constraint is:

[0048] in, It is the velocity change rate threshold ((m / s) / m), used to limit acceleration per unit distance.

[0049] In long-distance running, it is necessary to monitor the athlete's heart rate in real time to ensure it remains within a safe range. The relationship between an athlete's heart rate and speed is not a simple linear one, but rather a complex nonlinear one, described by the following expression:

[0050] in Represents real-time heart rate (bpm). It is the instantaneous velocity (m / s) at a distance x. For the fitting parameters, It is the resting heart rate (bpm).

[0051] The specific expression for the heart rate coupling constraint is:

[0052]

[0053] It will change with speed It changes with the change. When the athlete's speed is 0, that is, when he is stationary, ,at this time This aligns with reality, meaning that the resting heart rate is equal to the heart rate at rest. It is the future heart rate output by the CNN-LSTM-Attention prediction model, which is a safety margin. This constraint ensures that the speed strategy not only meets the instantaneous safety requirements, but also anticipates the future heart rate entering the danger zone.

[0054] in, This primarily reflects the maximum range of heart rate change with speed. This value will vary depending on the athlete's physical condition. For example, athletes with better physical fitness will experience a relatively smaller increase in heart rate at the same speed change. The value may be smaller; while for athletes with poorer physical fitness, the increase in heart rate is greater. The value is relatively large. It is used to adjust the shape and rate of change of the speed-heart rate curve, affecting how quickly the heart rate increases with speed.

[0055] Exponential function The introduction of this factor makes the relationship between heart rate and speed non-linear. As speed... Gradually increase The value gradually approaches 0. The value gradually approaches This means that the heart rate will not increase indefinitely with increasing speed, but will approach a limit value, which is in line with the physiological laws of the human body.

[0056] S3. Divide the long-distance running process into multiple consecutive time windows. At the beginning of each window, call the CNN-LSTM-Attention model to generate state prediction results as dynamic constraints and construct state transition equations. Specifically, based on the total duration of the long-distance race and actual needs, the entire race is divided into multiple fixed-length time windows. The length of each time window needs to comprehensively consider various factors, such as the frequency of changes in athlete condition, computational resources, and accuracy requirements. If the time window is too long, it may not be able to respond promptly to rapid changes in athlete condition and track environment; if it is too short, it will increase the computational load and may lead to overly frequent strategy adjustments, affecting the athlete's rhythm. The total duration of the long-distance race is set as... Divide it into n time windows of equal length, each time window having a length of... Then we have:

[0057] At the beginning of each time window, the athlete's real-time state is constructed, and state transition equations are constructed, including the state transition equation for remaining energy, the state transition equation for distance already run, and the real-time heart rate equation.

[0058] When constructing the residual energy state transition equation, define the state variables and assume... Indicates the first The state at the start of each time window For the distance already run, For the remaining energy, This is the real-time heart rate.

[0059] Calculate remaining energy using an energy consumption model. Assuming at time arrive Inside, athletes use speed Exercise, energy expenditure coefficient is Energy consumption for:

[0060] If in time The energy at that time is So in time Remaining energy at time for: Then the residual energy transfer equation can be obtained:

[0061] in, This represents the athlete's remaining energy at the start of the (i+1)th time window. Let represent the athlete's remaining energy at the start of the i-th time window, and let a, b, and c represent different energy consumption coefficients. This represents the instantaneous velocity used within the i-th time window. Indicates the length of the time window. This is an integration operation.

[0062] Furthermore, based on the relationship between speed and time, if within the time window... Internal velocity is The increase in distance already run for:

[0063] If in time The distance already run is So in time Distance already run for:

[0064] Then we can obtain the transition equation for the distance already covered:

[0065] in, This represents the cumulative distance the athlete has run at the start of the (i+1)th time window. This represents the cumulative distance the athlete has run at the start of the i-th time window; Furthermore, it is known that there is a non-linear relationship between heart rate and speed. ,in For the fitting parameters, For resting heart rate, a real-time heart rate equation can be constructed using real-time monitored velocity v:

[0066] in, For the fitting parameters, This represents the athlete's real-time heart rate at the start of the (i+1)th time window. This indicates the resting heart rate.

[0067] S4. Based on the state transition equation, the optimal speed under the current time window is solved by improving the dung beetle optimization algorithm. After execution, it scrolls to the next window and iterates until the optimal speed under all time windows is output.

[0068] See Figure 2 Specifically, when solving for the optimal velocity within the current time window using an improved dung beetle optimization algorithm based on the state transition equation, the initial population is first generated using a sinusoidal product mapping (SPM) to improve distribution uniformity. The SPM chaotic mapping expression is:

[0069] in, A random number between 0 and 1. and At that time, the function is in a chaotic state. After linear mapping to the speed range This provides the algorithm with diverse initial velocity values, increasing the likelihood that the algorithm will find the global optimum.

[0070] The individual position is updated by performing a global search through a rolling ball action, expressed as:

[0071] in, , It is the total time corresponding to the current optimal solution. This represents the fitness of the current optimal solution. The dynamic inertia weight is set so that the algorithm has a strong global exploration capability in the early stage of the search, and gradually strengthens the local exploration capability as the iteration progresses.

[0072] Introducing Levi Flight to enhance local exploration capabilities:

[0073] in It is the random step size of the Lévy distribution. ; The perturbation intensity coefficient is used to improve the quality of the solution by using the Levy flight perturbation to search for the optimal solution more precisely in the local area.

[0074] Population mutation and crossover are achieved by implementing a differential evolution strategy:

[0075]

[0076] in Scaling factor The crossover probability, differential evolution strategy helps increase the diversity of the population and avoids the algorithm getting trapped in local optima too early.

[0077] Constraint processing and fitness calculation are used to ensure that the optimization results satisfy energy, heart rate, and speed smoothness constraints. The fitness function is:

[0078] In summary, during the solution process, a sinusoidal product mapping is first used to initialize the algorithm population, generating a uniformly distributed and diverse initial solution set, effectively avoiding the local convergence problem caused by traditional random initialization. Then, a rolling ball behavior update mechanism is executed, simulating the dung beetle's rolling ball navigation characteristics to perform a global search for the positions of individuals in the population. By introducing the exploration and development capabilities of a dynamic inertial weight balancing algorithm, the algorithm ensures efficient location of potential optimal regions in complex non-convex search spaces. To further enhance local search accuracy, a Lévy flight perturbation mechanism is introduced, utilizing its long-tail distribution characteristics to deeply explore the neighborhood of the current optimal solution, effectively improving the algorithm's ability to escape local optima. Simultaneously, a differential evolution strategy is integrated, maintaining algorithm diversity through mutation and crossover operations on the population, preventing premature convergence, and improving global search efficiency.

[0079] In each iteration, the energy consumption, heart rate response and speed smoothness under each candidate speed strategy are calculated based on the state transition equation. The fitness is evaluated in combination with the dynamic constraints generated by the prediction model, and the optimal speed in the current window that satisfies all physiological constraints and has the shortest completion time is selected.

[0080] After determining the optimal speed for the current time window, the system automatically scrolls to the next time window. Based on the latest collected real-time state data, the CNN-LSTM-Attention prediction model is invoked again to generate updated prediction results, constructing a new state transition equation. The improved dung beetle optimization algorithm is then restarted to solve for the optimal speed strategy for the next window. This process continues iteratively until the cumulative distance run reaches the total length of the track. Finally, the optimal speed sequence for all time windows is output to form a complete long-distance running strategy. This rolling optimization mechanism ensures the real-time adaptability and physiological safety of the strategy. Even in the face of fluctuations in the athlete's condition or changes in the environment, the subsequent strategy can be adjusted in a timely manner through the window rolling mechanism to achieve the unified optimization goal of minimizing the global completion time and ensuring physiological safety.

[0081] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A long-distance running strategy method for teenagers that integrates deep learning and improved swarm intelligence optimization algorithms, characterized in that: The method includes the following steps: Collect multidimensional temporal features from athletes’ historical training and competition processes, and construct and train a CNN-LSTM-Attention prediction model based on the multidimensional temporal features. Establish a framework for a points-based long-distance running strategy optimization model with the objective of minimizing the finishing time; The long-distance running process is divided into multiple consecutive time windows. At the beginning of each window, the CNN-LSTM-Attention model is called to generate state prediction results as dynamic constraints to construct state transition equations. Based on the state transition equation, the optimal speed under the current time window is solved by improving the dung beetle optimization algorithm. After execution, it scrolls to the next window and iterates until the optimal speed under all time windows is output.

2. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 1, characterized in that, The multidimensional time-series features include instantaneous speed sequence, heart rate sequence, energy consumption rate, distance run, environmental data, and individual characteristics. The environmental data includes temperature data, humidity data, and altitude data. The individual characteristics include maximum oxygen uptake and fatigue index.

3. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 1, characterized in that, The CNN-LSTM-Attention prediction model consists of a CNN convolutional neural network, an LSTM long short-term memory network, and attention mechanism units connected in series. The CNN convolutional neural network is used for local feature extraction to capture the local dependencies between multiple features within a short time period. The LSTM (Long Short-Term Memory) network is used to receive the high-level feature sequences extracted by the CNN (Convolutional Neural Network) and capture long-term dependencies within a long time period. The attention mechanism unit is used to assign different weights to the hidden states of the LSTM long short-term memory network at all time steps and output a sequence of state prediction results for the next τ time steps.

4. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 3, characterized in that, A framework for optimizing long-distance running strategies using an integral form, with the objective of minimizing the finishing time, is constructed, specifically including: Define a race time objective function with the total length of the track as the integration interval and the goal of minimizing the race time; An energy consumption constraint based on exercise metabolism is introduced, and the remaining energy prediction result in the CNN-LSTM-Attention prediction model is used as a dynamic upper limit condition. Add velocity smoothness constraints and construct heart rate-velocity nonlinear coupling constraints, and set a safety threshold margin by combining the heart rate prediction results in the CNN-LSTM-Attention prediction model.

5. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 4, characterized in that, The defined objective function for completing the race is expressed as follows: Where v(x) represents the instantaneous velocity at a distance x, and L represents the total length of the track.

6. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 1, characterized in that, The long-distance running process is divided into n consecutive time windows of equal length.

7. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 5, characterized in that, The real-time state data of the athlete is input into the CNN-LSTM-Attention model to obtain a sequence of state prediction results consisting of multiple state prediction results, including the remaining energy prediction result and the heart rate prediction result.

8. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 6, characterized in that, Constructing the state transition equations specifically includes: Construct the state transition equation for the remaining energy: in, This represents the athlete's remaining energy at the start of the (i+1)th time window. Let represent the athlete's remaining energy at the start of the i-th time window, and let a, b, and c represent different energy consumption coefficients. This represents the instantaneous velocity used within the i-th time window. Indicates the length of the time window. For integration operations; Construct the transition equation for the distance already traveled: in, This represents the cumulative distance the athlete has run at the start of the (i+1)th time window. This represents the cumulative distance the athlete has run at the start of the i-th time window; Constructing a real-time heart rate equation: in, For the fitting parameters, This represents the athlete's real-time heart rate at the start of the (i+1)th time window. This indicates the resting heart rate.

9. The method for adolescent long-distance running strategy that integrates deep learning and improved swarm intelligence optimization algorithm according to claim 8, characterized in that, Based on the state transition equation, an improved dung beetle optimization algorithm is used to solve for the optimal velocity under the current time window, specifically including: The initial population is generated using a sinusoidal product mapping. Perform a rolling ball action to update the individual's position; Introduce the Levy flight disturbance for local search; Implement differential evolution strategies to induce population mutation and crossover; Perform constraint processing and fitness calculation to determine the optimal speed for the current window.

Images