CN-121980489-A - Teenager long-distance running strategy method integrating deep learning and improved swarm intelligent optimization algorithm

CN121980489ACN 121980489 ACN121980489 ACN 121980489ACN-121980489-A

Abstract

The invention discloses a teenager long-distance running strategy method integrating deep learning and an improved swarm intelligent optimization algorithm, which comprises the following steps of collecting multidimensional time sequence characteristics in the process of athlete historical training and competition, constructing and training a CNN-LSTM-Attention prediction model based on the multidimensional time sequence characteristics, establishing an integral form long-distance running strategy optimization model frame aiming at minimizing the time of finishing the competition, dividing the long-distance running process into a plurality of continuous time sequence windows, calling the CNN-LSTM-Attention model at the starting moment of each window to generate a state prediction result as dynamic constraint, constructing a state transition equation, solving the optimal speed under the current time sequence window through an improved dung beetle optimization algorithm based on the state transition equation, rolling to the next window after execution, and iteratively executing until the optimal speed under all the time sequence windows is output.

Inventors

Hao Kaicheng
WANG LEI
WANG YANLI

Assignees

湖北邮电规划设计有限公司

Dates

Publication Date: 20260505
Application Date: 20251211

Claims (9)

1. The teenager long-distance running strategy method integrating deep learning and improved intelligent group optimization algorithm is characterized by comprising the following steps: collecting multidimensional time sequence characteristics of athlete history training and competition, and constructing and training a CNN-LSTM-Attention prediction model based on the multidimensional time sequence characteristics; establishing an integral form long-distance running strategy optimization model framework with the aim of minimizing the race completion time; Dividing the long-distance running process into a plurality of continuous time sequence windows, calling a CNN-LSTM-attribute model at the starting moment of each window to generate a state prediction result as dynamic constraint, and constructing a state transition equation; based on a state transition equation, solving the optimal speed under the current time sequence window by improving a dung beetle optimization algorithm, scrolling to the next window after executing, and iteratively executing until the optimal speed under all time sequence windows is output.
2. The teenager sprinting strategy method of claim 1, wherein the multi-dimensional time series features include an instantaneous speed sequence, a heart rate sequence, an energy consumption rate, a distance travelled, environmental data including temperature data, humidity data, altitude data, and individual features including a maximum oxygen uptake and a fatigue index.
3. The teenager long-distance running strategy method integrating deep learning and improved intelligent group optimization algorithm according to claim 1, wherein the CNN-LSTM-Attention prediction model is formed by connecting a CNN convolutional neural network, an LSTM long-term memory network and an Attention mechanism unit in series, The CNN convolutional neural network is used for extracting local features and capturing local dependency relations among a plurality of features in a short-time sequence period; The LSTM long-term memory network is used for receiving the advanced characteristic sequences extracted by the CNN convolutional neural network and capturing long-term dependency relationship in a long time sequence period; the attention mechanism unit is used for distributing different weights to the hidden states of all time steps of the LSTM long-term memory network and outputting a state prediction result sequence of tau time steps in the future.
4. The teenager long-distance race strategy method of claim 3, wherein constructing the integrated long-distance race strategy optimization model framework with the goal of minimizing the race time, specifically comprises: Defining a race completion time objective function taking the total track length as an integral interval and taking the minimum race completion time as an objective; introducing energy consumption constraint based on motion metabolism, and taking a residual energy prediction result in a CNN-LSTM-Attention prediction model as a dynamic upper limit condition; And adding a speed smoothness constraint, constructing a heart rate-speed nonlinear coupling constraint, and setting a safety threshold margin by combining a heart rate prediction result in a CNN-LSTM-Attention prediction model.
5. The teenager long-distance race strategy method of claim 4, wherein the expression of the defined race completion objective function is: Where v (x) represents the instantaneous speed at a distance x and L represents the overall track length.
6. The teenager long-distance running strategy method of claim 1, wherein the long-distance running process is divided into n consecutive time sequence windows of equal length.
7. The teenager long-distance running strategy method integrating deep learning and improved swarm intelligent optimization algorithm according to claim 5, wherein real-time state data of athletes are input into a CNN-LSTM-Attention model to obtain a state prediction result sequence composed of a plurality of state prediction results, wherein the state prediction results comprise residual energy prediction results and heart rate prediction results.
8. The teenager long-distance running strategy method integrating deep learning and improved swarm intelligent optimization algorithm according to claim 6, wherein the construction of the state transition equation specifically comprises the following steps: Constructing a residual energy state transfer equation: Wherein, the Representing the remaining energy mobilized at the beginning of the (i + 1) th timing window, Representing the remaining energy mobilized at the beginning of the ith timing window, a, b, c representing different energy expenditure coefficients, Indicating the instantaneous speed taken in the ith time window, The length of the time window is indicated, Is an integrating operation; Constructing a run distance transfer equation: Wherein, the Indicating the cumulative distance that the mobilizer has run at the beginning of the (i + 1) th timing window, Indicating the accumulated distance that the mobilizer has run at the beginning of the ith timing window; constructing a real-time heart rate equation: Wherein, the In order to fit the parameters of the model, Representing the real-time heart rate of the mobilization at the beginning of the (i + 1) th timing window, Representing the resting heart rate.
9. The teenager long-distance running strategy method integrating deep learning and improved swarm intelligent optimization algorithm according to claim 8, wherein the method is characterized in that based on a state transfer equation, the optimal speed under the current time sequence window is solved by improving a dung beetle optimization algorithm, and specifically comprises the following steps: Generating an initial population by adopting sine product mapping; Performing a rolling ball action to update the individual position; Introducing Lewy flight disturbance to perform local search; implementing a differential evolution strategy to perform population variation and crossover; and (5) carrying out constraint processing and fitness calculation, and determining the optimal speed of the current window.

Description

Teenager long-distance running strategy method integrating deep learning and improved swarm intelligent optimization algorithm Technical Field The invention relates to the technical field of intelligent planning of sports, in particular to a teenager long-distance running strategy method integrating deep learning and improved swarm intelligent optimization algorithm. Background With the rapid development of competitive sports, scientific optimization of young long-distance running strategies becomes a core topic for improving athlete performance, preventing sports injury and optimizing training plans. The conventional method relies on training experience or a simplified segmentation model, and has the limitation that experience dependence is strong, and subjective experience is difficult to accurately quantify dynamic coupling relation between energy consumption, heart rate variation and speed distribution. The global optimization capability is insufficient, namely, a method based on rule or gradient descent is easy to fall into local optimization, and a global optimal solution cannot be found under complex constraints (such as energy limit and heart rate safety threshold). The real-time performance is poor, the calculation complexity of algorithms such as dynamic planning is high, and the real-time strategy adjustment requirements in the competition are difficult to adapt. In recent years, a swarm intelligent optimization algorithm (such as particle swarm optimization and genetic algorithm) is gradually applied to the field of motion science due to the parallel searching and global optimization capabilities of the swarm intelligent optimization algorithm. For example, particle Swarm Optimization (PSO) achieves multi-objective optimization by simulating the foraging behavior of a bird swarm, but the convergence speed is affected by parameter sensitivity, and Genetic Algorithms (GA) are highly robust but prone to premature convergence in high-dimensional problems. In addition, prior studies have mostly ignored dynamic coupling effects of physiological constraints (e.g., non-linear relationship of heart rate to velocity), resulting in disjointing the optimization results from the actual physiological limits. At the same time, deep learning techniques have made significant progress in the field of timing prediction. In particular to a CNN-LSTM-Attention mixed model, local characteristics are extracted through a convolutional neural network, a long-term and short-term memory network captures long-term dependence, an Attention mechanism focuses on key information, and the change trend of physiological parameters such as heart rate, energy consumption and the like of athletes can be accurately predicted. However, existing studies have not yet fused these accurate predictions with the strategy optimization process depth. Aiming at the problems, some students propose a dung beetle optimization algorithm according to the habit of the dung beetles. However, in the long-distance running strategy optimization, how to balance the global exploration and the local development of the algorithm and efficiently process multi-constraint coupling still remains a technical difficulty. Particularly in terms of real-time policy adjustment in dynamic environments, there is a lack of effective integration with accurate predictive models. The invention provides an improved dung beetle optimization algorithm (HDBO), which solves the defects of the traditional method by fusing SPM chaotic initialization, lewy flight disturbance and differential evolution strategies and combining the output of a CNN-LSTM-Attention prediction model as a dynamic constraint condition. The hybrid method fully utilizes the precise prediction capability of deep learning and the global optimization advantage of the swarm intelligence algorithm, and provides more precise theoretical support and technical support for real-time dynamic optimization of the long-distance running strategy. Disclosure of Invention Aiming at the prior art, the invention aims to provide a teenager long-distance running strategy method and system integrating deep learning and improved swarm intelligent optimization algorithm, which mainly solve the technical problems in the background art. In order to achieve the above purpose, the technical scheme of the embodiment of the invention is realized by a teenager long-distance running strategy method integrating deep learning and an improved group intelligent optimization algorithm, which comprises the following steps: collecting multidimensional time sequence characteristics of athlete history training and competition, and constructing and training a CNN-LSTM-Attention prediction model based on the multidimensional time sequence characteristics; establishing an integral form long-distance running strategy optimization model framework with the aim of minimizing the race completion time; Dividing the long-distance running process into a plur