6 results about "Maximum entropy method" patented technology

The invention relates to a low-altitude Internet of Things dynamic spectrum allocation and access method and system based on deep reinforcement learning. The method comprises the following steps: S1, constructing an unmanned aerial vehicle network of the low-altitude Internet of Things; s2, each unmanned aerial vehicle carries an act-critic network, a maximum entropy method is introduced into the act-critic network for optimization, and an MAEAC reinforcement learning model is formed; and S3, training the MAEAC reinforcement learning model to obtain a trained MAEAC reinforcement learning model, and obtaining an optimal strategy of spectrum allocation and access. The invention provides a low-altitude Internet of Things dynamic spectrum allocation and access method and system based on deep reinforcement learning, and the algorithm improves the randomness of a strategy and accelerates the convergence speed through adding the updating of an entropy item execution strategy. The problem of how to realize stable communication of the unmanned aerial vehicle network under limited spectrum resources in the low-altitude Internet of Things is solved, and a good effect is achieved.

The invention provides a reinforcement learning method skilled at multiple targets and multiple tasks, which comprises the steps of constructing a deep reinforcement learning framework comprising a strategy network and an evaluation network, introducing a candidate set expansion and offset coefficient weighting mechanism in an action sampling stage, and performing multi-target and multi-task evaluation on the basis of probability distribution of multiple groups of parameterization in the strategy network; diversified candidate solutions are formed in the action space; a cosine similarity regular term is introduced into the loss function so as to maintain the difference between different action candidates; the evaluation network adopts a double-Q network, and random disturbance is applied to a Q branch, so that the over-estimation problem is relieved, and the value evaluation robustness is improved. The interaction experience of the intelligent agent and the environment supports coexistence of various experiences, including a master experience of a strategy network through normal distribution sampling, a slave experience through Cauchy distribution sampling and a random experience of random sampling. Compared with a traditional maximum entropy method, the method shows better exploration efficiency, strategy generalization performance and task migration ability in a complex high-dimensional task environment.

The invention discloses an intake tower anti-seismic reliability maximum entropy calculation method based on a dynamic integral boundary. The method comprises the following steps: firstly, determining a plurality of anti-seismic performance levels of an intake tower and corresponding tower top displacement thresholds; taking the seismic oscillation peak acceleration as a random variable, and generating sample points by using a layered Latin hypercube sampling method; adopting a C-P power spectrum model to generate a random seismic wave sample; obtaining a structure response value through finite element dynamic analysis, and constructing a performance function, thereby calculating first four-order statistical moments of the performance function; dynamically determining an integral boundary for each performance level; and by taking the statistical moment as a constraint, fitting a probability density function of the performance function by applying a maximum entropy method in a dynamic boundary, and carrying out integral calculation on the probability density function to obtain a failure probability and a reliable index. Thus, the problem that the fixed integral boundary and the performance level of a traditional maximum entropy method are disjointed is solved, the calculation efficiency is guaranteed, and meanwhile an accurate quantification means is provided for evaluating the anti-seismic reliability of the water inlet tower.

This invention discloses a maximum entropy reinforcement learning method based on covariance parameterization, belonging to the fields of intelligent control and machine learning. This method achieves explicit control and automatic adjustment of policy entropy by parameterizing the covariance matrix of the policy distribution, thereby improving the policy's ability to explore the continuous action space. This invention further overcomes the limitation of traditional maximum entropy methods that rely solely on standard deviation methods, realizing a more advanced maximum entropy policy optimization method. This invention includes policy covariance parameterization and a positive definite covariance constraint mechanism. Experiments show that this invention is effective and performs better than existing methods in experimental environments. This invention is applicable to applications such as robot control and intelligent agent navigation.

The present invention relates to the technical field of new energy vehicle collisions, and discloses an optimization method for modeling the uncertainty of a frontal collision of an electric bus. First, a finite element model of the frontal collision of the electric bus is established to obtain design variables. Based on the polynomial chaos expansion method, the relationship between the design variables and the responses is established. Combining the Sobol’ index method for sensitivity analysis, the design variables with high sensitivity are obtained. An uncertainty optimization mathematical model for the frontal collision of the electric bus is established, and it is solved by combining a multi-objective optimization algorithm. This method optimizes the frontal collision cross-section with a 100% overlap rate considering uncertainty for the front end of a certain electric bus. Taking the cross-sectional dimensions of the tube beam components as the optimization variables and the collision intrusion speed as the optimization objective, the uncertainty analysis and optimization design research of the electric bus are carried out by combining the polynomial chaos expansion method and the maximum entropy method, reducing the influence of uncertainty factors on the optimization results, and ensuring the robustness and reliability of the design scheme.

This invention discloses a method and system for simulating stationary non-Gaussian wind pressure based on the maximum entropy method and moment transformation function. The method calculates the marginal probability density function (PDF) based on the k-th primitive moment of the target standard non-Gaussian wind pressure using the maximum entropy method; and uses the marginal probability density function (PDF) f Xj (x) Determine the transfer function; calculate the cross-correlation matrix based on the target cross-power spectral density matrix; divide the transformation relationship into region I and region II based on the error boundary line in the non-Gaussian correlation coefficient estimation; calculate the cross-correlation matrix ρ of the latent standard non-Gaussian vector for region I and region II respectively. G (τ) and the latent standard non-Gaussian vector cross power spectral density (PSD) matrix S G (ω); based on the latent Gaussian cross-power spectral density matrix S G (ω) Obtain Gaussian samples Z(t), and then obtain samples of non-Gaussian wind pressure vector X(t) based on the Gaussian samples Z(t). This method can provide better simulation accuracy, especially for wind pressure with strong non-Gaussian characteristics; since this method derives the Gaussian correlation function as a function of the non-Gaussian correlation function, it has higher simulation efficiency and better simulation accuracy.