120 results about "Function optimization" patented technology

The invention discloses a trajectory optimization method based on a multi-source heterogeneous positioning data fusion algorithm, and relates to the technical field of intelligent navigation and high-precision positioning, multi-modal data are acquired through a multi-modal sensor array, positioning redundancy of scenes such as tunnels and indoor scenes is enhanced, a weight distribution strategy is dynamically adjusted through an Actor-Critic network architecture, and the positioning accuracy is improved. The state space input comprises an environment semantic tag, a historical error sequence and a real-time noise variance, the output action space is continuous weight distribution of each data source, a multi-target reward function optimization strategy is combined, scene adaptability is realized, a local SLAM map, inertial navigation error parameters and a weight distribution strategy are shared in real time based on a V2X protocol, and the real-time performance of the system is improved. According to the method, a single device accumulative error is compensated by using adjacent vehicle data, a terminal locally trains an error compensation model, parameters are uploaded to a cloud end through differential privacy encryption, the cloud end adopts a FedAvg algorithm to aggregate a global model and issue the global model, the error difference between devices is inhibited, and dynamic road network updating and scene differentiation model distribution are supported at the same time.

The invention relates to a task scheduling optimization method and device based on reinforcement learning, equipment and a medium. The method comprises the steps that firstly, system resource state data are collected in real time, dynamic environment characteristics are determined through preprocessing and time sequence analysis, task characteristic data are analyzed at the same time, and a task priority sequence and a resource demand vector are generated through a priority ranking algorithm and a resource evaluation model; and then a state space and an action space are constructed by adopting a reinforcement learning algorithm, an optimal task allocation scheme is generated through strategy iteration and reward function optimization, and if the scheme meets a resource balance threshold, scheduling is executed, and performance indexes are collected. And finally, fusing real-time indexes with historical data, and updating parameters of the reinforcement learning model through experience playback and gradient descent to form a closed-loop optimized improved scheduling strategy. By adopting the method, the accurate mapping of the resource state and the task requirement can be realized, and the problem of insufficient adaptability of the traditional static scheduling to a complex scene is solved.

The invention relates to a model predictive control charging optimization method based on a dynamic power state, which initiates a'dynamic power state collaborative optimization 'mechanism, takes a real-time power upper limit as an active optimization target instead of a fixed constraint condition, and breaks through the technical bottleneck of power limitation passive response in a traditional charging strategy. The method specifically comprises the following steps: constructing an electric-thermal-aging multi-physics field coupling model of the lithium ion battery, updating electric-thermal characteristic parameters in real time through an online parameter identification algorithm, and synchronously estimating a core temperature and an aging state in combination with a double-Kalman filtering state observer; innovatively establishing a four-dimensional objective function optimization model containing a dynamic power state, and performing multi-objective collaborative optimization on a power upper limit, a charging speed, a capacity fading rate and a current fluctuation rate; and designing a dynamic rolling optimization algorithm based on a model prediction control framework, and solving the optimal charging current meeting the dynamic power distribution requirement of the power grid in real time under the hard constraint of ensuring the maximum core temperature and terminal voltage.

The invention discloses a bank loan business risk control system and method based on big data analysis, and relates to the technical field of financial risk control, and the method comprises the steps: collecting and preprocessing real-time behavior data, and obtaining a user behavior feature set; based on the user behavior feature set, calling a behavior map modeling engine to carry out structured mapping, matching with a risk anchor point rule base, identifying a potential risk mode and labeling an initial anchor point risk label; correcting the deviation between the initial risk anchor point tag and the actual default record by adopting a value function optimization method, and predicting the risk grade score of the current behavior of each user in combination with the historical behavior sample data and loan feedback data of the user; predicting probability distribution of migrating to a default state in the future through user risk grade scores and historical state evolution data; and in combination with the potential loss under each behavior path, evaluating the current loan business risk, and generating a risk control strategy through a risk level mapping rule and a strategy decision engine.

The invention discloses an intelligent optimization system based on a self-adaptive control hydraulic cylinder, and relates to the technical field of hydraulic cylinder control, and the system comprises a monitoring data collection module which collects data of a flow field, a temperature field and a stress field in real time by means of various sensors arranged at different parts of the hydraulic cylinder, synchronously carries out the preprocessing and coding of the collected data, and transmits the data to a data processing module; the data is transmitted to the data processing and analyzing module in real time; detailed data of the hydraulic cylinder in different physical fields are acquired in real time through the monitoring data acquisition module, and the coupling coefficient is accurately calculated by using the multi-physical field coupling model, so that the system can deeply understand the specific influence of each physical field on the performance of the hydraulic cylinder; according to the method, the system can construct a more accurate hydraulic cylinder dynamic model in the model prediction control module, so that the future state of the hydraulic cylinder is accurately predicted, further, the system can obtain an optimal control input sequence through an objective function optimization algorithm, and the pertinence and effectiveness of a control strategy are remarkably improved.

The invention discloses a distributed energy system source load coordinated optimization method based on a quasi-potential game method, and the method comprises the steps: constructing a distributed energy system model, inputting system parameters, and predicting renewable energy power generation and initial load demands. In a source side optimization stage, a leader layer potential function of a quasi-potential game is established by taking minimization of source side cost # imgabs0 # as a target, and an initial power generation plan is generated by comprehensively considering economical efficiency and carbon emission constraints; and then, based on a scheduling result, calculating a carbon potential epsilon t of each node and a dynamic carbon emission factor # imgabs1 # of each stage, optimizing a load side response based on an LCDR scheme, constructing a local potential function of a follower layer, reflecting a relationship between a user profit maximization target and carbon emission, and adjusting user behaviors through a distributed decision. And the updated load demand is fed back to the source side, and the source side optimizes the output plan of each unit based on the updated load, so that an iterative process of source side potential function optimization-load side equilibrium response is formed, and an optimal scheduling strategy and scheduling result of the energy supply side are obtained. According to the framework, the global consistency requirement of a traditional potential game is relaxed, independent optimization of source-load two sides under the guidance of respective potential functions is allowed, and a Nash equilibrium state is finally achieved only by ensuring monotonous convergence of total potential energy of a system in an iteration process. According to the method, the convergence advantage of the potential game is reserved, the method is also adapted to the characteristics of a source-load heterogeneous decision subject, efficient consumption of renewable energy and collaborative optimization of carbon emission are realized through bidirectional transmission of the carbon potential signal, and the overall efficiency of the system is remarkably improved.

The invention discloses a large model driven automobile model modeling method based on a Modelica language, and belongs to the technical field of intelligent modeling and automobile simulation. The method comprises the following steps: firstly, accurately analyzing a natural language demand into a structured triple by adopting a BERT-CRF (domain knowledge enhanced) multi-task model; matching an optimal component combination through a multi-objective optimization algorithm driven by a graph neural network, and cooperatively predicting an interdisciplinary parameter feasible region in combination with symbolic mathematical derivation and machine learning; a topological connection matrix is innovatively optimized by using a graph attention network, and intelligent generation and dynamic verification of simulation codes are realized by fusing a template engine and syntax tree analysis; and finally, constructing a multi-target reward function optimization control strategy through reinforcement learning, and establishing a closed-loop knowledge iteration mechanism. Compared with a traditional modeling method, through deep combination of the large model and Modelica, the technical difficulty of automobile system modeling is remarkably reduced while the preciseness of physical modeling is kept, and the method is particularly suitable for complex scenes such as new energy vehicle model development and intelligent driving system integration.

The invention relates to the technical field of intelligent traffic, in particular to an On-Ramp scene trajectory planning method based on deep inverse reinforcement learning, and the method comprises the following steps: S100, candidate trajectory generation: generating a plurality of candidate trajectories based on vehicle state information and traffic environment information; s200, track authenticity filtering, wherein the generated candidate tracks are evaluated and screened through a track authenticity filtering framework, so that abnormal tracks are eliminated, and credible tracks are reserved; and S300, reward function training: utilizing reward function optimization of deep inverse reinforcement learning to realize continuous adjustment and improvement of a trajectory evaluation standard so as to carry out quantitative analysis on rationality and optimality of the trajectory. By constructing a trajectory authenticity filtering framework, optimizing a reward function design and training strategy, and introducing a multi-dimensional scoring mechanism, the trajectory authenticity filtering framework is optimized, so that the accuracy of the trajectory authenticity filtering framework is improved. According to the invention, efficient, safe and dynamically adaptive trajectory planning is realized, and reliable support is provided for real-time decision making of an automatic driving vehicle in a complex scene.

The invention discloses an underground carry-scraper path planning method and system, and the method comprises the steps: obtaining environment data information of a target underground mine roadway, constructing an environment map, and carrying out the preprocessing of the environment map, and obtaining a binary grid map; constructing a generalized Voronoi diagram, extracting a roadway center line and carrying out region division; performing path guidance and segmentation by adopting an algorithm to obtain a reference state set of the target underground mine roadway; and based on a hybrid algorithm, a regional dynamic node expansion strategy and a multi-target heuristic cost function, realizing path planning of the underground carry-scraper, and realizing path planning of the underground carry-scraper. According to the invention, through data acquisition and data processing of a target underground mine roadway, a roadway center line and divided areas are extracted, then path guidance is carried out through an algorithm, and based on a divided area dynamic node expansion strategy and a multi-target heuristic cost function optimization hybrid algorithm, path guidance is carried out. By combining a hybrid algorithm, a Dubins curve, a collision detection scheme, path smoothness punishment and path distance punishment, the path planning of the underground carry-scraper is realized, the reliability is higher, the accuracy is better, and the effect is better.

The invention discloses an intelligent vibration isolation control method and system based on vibration monitoring, and the method comprises the steps: collecting vibration data, carrying out the preprocessing, carrying out the fuzzy conversion according to the preprocessed data, and calculating a preliminary control quantity; constructing an optimization model and defining a target function, using a locust optimization algorithm to iteratively optimize the population position until the target function value converges, and using the initial control quantity as the final optimization model input to obtain the final control quantity; driving force signals are calculated based on the final control quantity, a difference equation is further constructed based on accumulation of the driving force signals, then a predicted value is output, conversion is carried out according to the obtained predicted value, and instantaneous data are obtained. According to the method, efficient adaptive control over a complex nonlinear vibration system is achieved, the optimization process is rapidly converged, the target function optimization result is directly used for dynamic adjustment of the control quantity, the time for calculation of the control quantity is remarkably shortened, the future vibration trend is predicted through a difference equation, and the calculation efficiency is improved. And the dynamic response capability is further improved.

The invention discloses an automatic driving carrying equipment scheduling system and method for an industrial robot. The system comprises a plurality of automatic driving carrying devices, a central dispatching device and corresponding communication modules. The system adopts a multi-agent reinforcement learning framework and combines a graph neural network processing environment topological structure to realize dynamic path planning and multi-device collaborative scheduling; integrating an energy consumption prediction mechanism based on a Kalman filter, and bringing energy consumption factors into a task allocation decision; meanwhile, a fault-tolerant management mechanism based on a distributed account book and federated learning is established, and fault detection and rapid recovery are achieved. The technical modules are deeply coupled, and a unified collaborative optimization framework is formed through reward function design, utility function optimization and fault probability calculation of multi-agent reinforcement learning. According to the method, the problems of poor dynamic environment adaptability, isolated decision making of each module, extensive energy consumption management and the like in the prior art are effectively solved, and the overall efficiency, energy efficiency and reliability of a scheduling system are remarkably improved.

The invention provides a social robot detection method and system, computer equipment and a storage medium, and belongs to the technical field of social robot detection.The method comprises the steps that user multi-source data on a social platform and the social relation between users are collected; a hybrid encoder is adopted to capture local dependence and global time sequence dynamic states of behaviors, and user behavior characteristics are generated; aggregating structure attention features between the initial node features of the target account and the initial node features of the social relation account to obtain multi-modal relation aggregated structure features; introducing a multi-modal adversarial training strategy, adaptively adjusting the disturbance intensity of each modal based on gradient sensitivity, and aligning the characterization of the clean sample and the adversarial sample in combination with a content discriminator and a behavior discriminator; and outputting a target user classification result through the multi-task target function optimization model. According to the method, the problems of insufficient modal fusion and insufficient adversarial robustness of an existing method can be effectively solved, and the accuracy and stability of social robot detection in a complex and adversarial scene are improved.

The invention discloses a traffic flow prediction method and device based on a space-time diagram convolutional network, and the method comprises the steps: collecting road traffic flow data, and constructing a traffic congestion index matrix model; calculating a road traffic congestion index based on the collected road traffic flow data, and constructing a traffic flow adjacency matrix by using the road traffic congestion index; performing fast time convolution processing on the traffic flow adjacency matrix to generate a traffic flow data sequence containing spatio-temporal information; inputting the traffic flow data sequence into a space-time synchronization graph convolutional network model STSGCN, further learning a spatial topological structure, and deeply extracting space-time features by stacking multiple layers of space-time synchronization graph convolutional layers; and inputting the spatial-temporal characteristics into a wave loss function-based random vector function link network model Wave-RVFL optimized by an index and trigonometric function optimization algorithm, and predicting the traffic flow. The method can improve the precision and efficiency of traffic flow prediction.

The invention discloses a three-dimensional model wall thickness measuring method, device and equipment combining a ray method and a spherical surface method, and a storage medium, and is applied to the field of three-dimensional models. Processing the input three-dimensional model by adopting a self-adaptive subdivision algorithm of a quadtree to generate an optimized triangular mesh; the method comprises the following steps: constructing an axial alignment bounding box (AABB) hierarchical tree structure optimized based on a superficial area heuristic (SAH) cost function, and in a wall thickness calculation stage, combining the advantages of a ray method and a spherical method: performing dynamic judgment according to an included angle between a normal vector at an intersection point of a ray and the inner surface of a model and an initial ray direction; the hybrid strategy ensures that reliable and efficient measurement results can be obtained in different geometric feature regions; a visual distribution diagram of color coding can be generated; according to the method, through adaptive grid optimization, efficient SAH-AABB acceleration and ray / spherical surface method mixed measurement, high-precision and efficient global measurement of the wall thickness of any three-dimensional model, especially a complex structure model, is realized.

The invention provides a cascade flow field prediction method and device based on sparse promotion modal feature prediction, and belongs to the technical field of gas turbine flow field prediction. According to the method, a flow field snapshot is decomposed into a DMD mode, a mode amplitude matrix and a DMD eigenvalue matrix, and a flow field reconstruction expression is constructed; constructing an objective function, including promoting a sparse term to produce a more sparse solution; the optimal position of the non-zero modal amplitude is solved based on target function optimization, and a modal order corresponding to the non-zero modal amplitude forms a dominant modal representing the dynamic characteristics of the flow field; the modal amplitude and the DMD characteristic value of the unknown working condition are predicted based on the modal amplitude and the DMD characteristic value of the dominant modal of the flow field under the known working condition, and then the flow field of the unknown working condition is predicted by using the inverse process of the DMD. According to the method, control over the number of sparse modes is added into the target function, the mode sparsity is enhanced, and then the mode subset which has remarkable influence on the flow field is better identified.

A method includes ingesting by a set of computing nodes of a parallelized database, a dataset for storage therein. The method further includes formatting, by a lead computing node of the set, the dataset in a primary data format to produce a primary data formatted dataset, and storing the primary data formatted dataset in system state data. The method further includes receiving, by a first computing node of the set, a query request involving the dataset and a first data function, determining that the first data function triggers a first data conversion optimization, accessing the primary data formatted dataset from the system state data, converting the primary data formatted dataset from the primary data format to a secondary data format to produce a secondary data formatted dataset, and processing the secondary data formatted dataset in accordance with the first data function to produce a first function output.

The invention discloses a trajectory planning method based on double sampling optimization, and belongs to the technical field of automatic driving. The method aims at solving the technical problems that in a multi-branch unstructured scene, a traditional planning method is prone to falling into local optimum, planning fails or calculation time is too long, and system reliability is affected. According to the scheme, sampling optimization is carried out on a guide path and a smooth path at the same time, an improved WRRT *-S algorithm is adopted for searching, and an optimal guide path is selected in combination with a risk potential field, path distance cost and steering cost; generating a smooth path candidate cluster by using a quintic Bezier curve, and selecting an optimal smooth path through multi-objective function optimization of path smoothness, economy, collision penalty and the like; and discretizing a speed planning problem into a nonlinear planning problem for solving by taking the maximum speed as a constraint and utilizing a Gaussian pseudo-spectral method, so as to realize collaborative optimization of the path and the speed. According to the invention, the vehicle can plan a safe, stable and efficient track in a complex scene.

The invention discloses a photovoltaic power prediction method based on an adaptive modified quantile regression neural network. Feature extraction is realized by constructing a double-flow hybrid neural network so as to improve prediction accuracy, a branch, combined with a multi-head attention mechanism, of the convolutional neural network is responsible for extracting long-term features, a branch of a bidirectional gating circulation unit focuses on identifying short-term fluctuation, and the double-flow hybrid neural network is combined with quantile regression. In order to solve the problems of quantile crossing and non-differentiable zero point of a loss function, a self-adaptive correction marble loss function is provided, and smooth function optimization is introduced to ensure monotone increasing of predicted quantiles and whole-domain differentiable of the loss function. According to the method, point prediction, interval prediction and probability density prediction can be realized, the prediction effect is verified through a multi-dimensional evaluation index, potential information of photovoltaic power is fully mined, and the method has practical engineering application value.

The invention provides an unmanned excavator track planning and control method, and relates to the field of artificial intelligence vehicle control. According to the system, aiming at engineering requirements of linear excavation and the like, an excavator working device is simplified into a four-degree-of-freedom mechanical arm, and a target excavation track is planned through an RRT-Connect algorithm and an S-type speed interpolation function; constructing the nonlinear system into a second-order linear system, and identifying model parameters on line by adopting a recursive least square method; and then an augmentation system containing error integration is constructed, model prediction control is combined, and high-precision trajectory tracking is realized through discretization processing, objective function optimization and a QP solver. The problem that a traditional control strategy is insufficient in precision is solved, the operation precision and adaptability of the unmanned excavator under complex working conditions are improved, the method is suitable for scenes such as groove excavation and slope leveling, and intelligent development of the excavator is promoted.

The invention discloses a numerical control machine tool spindle bearing feature extraction method based on improved FMD, and belongs to the technical field of rotating machine fault diagnosis. The method aims at solving the problems that traditional feature mode decomposition is high in parameter dependency and fault feature extraction is difficult under the noise background. The core of the method is that firstly, noise is added into a collected bearing vibration signal, and an ETO-FMD model is input; secondly, using an exponential trigonometric function optimization algorithm to take weighted envelope spectrum kurtosis as a fitness function, performing adaptive global optimization on the mode number M of the FMD and the length L of a filter, and automatically obtaining an optimal parameter combination; and finally, calculating a weighted envelope spectrum kurtosis value of each IMF component after FMD decomposition, and screening out the most critical component to perform signal reconstruction so as to realize accurate extraction of fault features. According to the method, the limitation of manually setting parameters is overcome, the accuracy, the adaptability and the robustness of feature extraction are remarkably improved, and the method is suitable for diagnosis of various faults of the spindle bearing of the high-end numerical control machine tool.

The invention belongs to the field of artificial intelligence, and discloses a distributed data acquisition system based on multi-agent collaborative decision and a construction method, and the method comprises the steps: 1, collecting heterogeneous data of production equipment, an environment sensor, a man-machine interaction terminal and an enterprise information system in real time through the deployment of a distributed data acquisition network, cleaning, de-noising and standardizing the edge computing nodes to generate a structured data stream; 2, automatically extracting features and laws of structured data through a multi-dimensional engine analysis model, and completing semantic alignment and fusion of heterogeneous data; 3, multiple agents are adopted to form empirical data according to a deterministic strategy gradient model in combination with a multi-dimensional reward function optimization model; and step 4, performing incremental learning on the empirical data to optimize the anomaly detection model, realizing multi-dimensional information visualization display of first-line decision and center command, and deploying corresponding functional agents according to the multi-dimensional information. The problems that a multi-agent system is low in control precision and low in efficiency are solved.

The invention discloses a railway construction management multi-objective equilibrium optimization method based on an improved MOPSO algorithm, and the method comprises the steps: 1) constructing four objective function optimization models on the premise of meeting the basic constraint conditions of a railway construction project, and enabling each objective function optimization model to correspond to an optimization objective; 2) according to the four objective function optimization models constructed in the step 1), constructing a multi-objective equilibrium optimization model; and 3) performing iterative solution on the multi-objective equilibrium optimization model constructed in the step 2) by using an improved multi-objective particle swarm optimization MOPSO algorithm to obtain an optimal solution set of the project under equilibrium in four aspects of construction period, resource, cost and safety. According to the method, the convergence speed, the diversity maintenance capability and the constraint processing flexibility of the MOPSO algorithm can be remarkably improved while the four optimization objectives of the construction period, the resources, the cost and the safety of the railway construction project are effectively considered, and an efficient and reliable multi-objective comprehensive decision support tool is provided for railway construction management.

The invention discloses a multi-scene agent training method and system fused with a large language model, and relates to the field of agent training, and the method comprises the steps: constructing a three-dimensional state space of an unmanned plane agent, and carrying out the standardization of the three-dimensional state space; initializing the pre-training large language model to obtain task scene adaptation parameters; configuring an intelligent agent based on the task scene adaptation parameters; training the intelligent agent in stages by adopting a hierarchical reinforcement learning algorithm, and recording a training log in real time; optimizing a reward function based on the training log and generating a new reward value; the new reward value is input into a large language model for defect recognition, and a quantitative adjustment suggestion is generated; feeding the quantitative adjustment suggestions back to a training process and a reward function optimization process, and performing iterative training until convergence; and executing a cooperative task in multiple scenes based on the finally trained agent. According to the method, intelligent agent training is realized by fusing a large language model, and the problems of high cross-scene reconstruction cost, low cooperation efficiency and lack of dynamic optimization of training of a traditional intelligent agent are effectively solved.

The invention discloses an AGV global and local fusion navigation method and system based on LT-TD3 reinforcement learning, and the method comprises the steps: extracting environment features based on laser radar data through obstacle edge detection, channel recognition and open space detection, dynamically generating a navigation target point set, and selecting an optimal navigation point through a distance weighted evaluation strategy; a multi-modal input processing mechanism is adopted to fuse laser radar data, target position information and a historical action sequence, and a state vector is formed after processing of a convolutional neural network and a gating circulation unit; a state value network, a double-Q network and a strategy network are adopted for cooperative work, and in combination with Expectile regression, a mixed target updating mechanism and composite strategy target function optimization, continuous linear speed and angular speed control instructions are output; and collision detection and intelligent control right switching are combined to realize AGV intelligent safety navigation, so that the problem that traditional reinforcement learning is easy to fall into local optimum is solved, and the phenomenon of navigation instability caused by distribution offset is effectively relieved.

The invention provides a rock mass fracture sub-pixel level identification and parameter extraction method and system, and belongs to the technical field of computer vision and digital image processing. According to the method, a multi-task convolutional neural network is constructed, and a segmentation probability matrix and a boundary signed distance field matrix are synchronously generated; in combination with energy function optimization, sub-pixel-level fracture boundary extraction is realized; a skeleton line is extracted by adopting constrained Delaunay triangulation and direction weight optimization, and topological distortion is avoided; and finally, geometric parameters are calculated based on the high-precision boundary and skeleton data. The system comprises an image processing module, a boundary optimization module, a skeleton extraction module and a parameter calculation module, and realizes integration of identification and parameter extraction. According to the rock mass fracture sub-pixel-level identification and parameter extraction method and system, the problems of low precision, skeleton distortion and flow splitting of a traditional method are solved, and the precision and reliability of rock mass fracture identification are improved.

This invention discloses a water treatment dosing control method and system based on quadratic programming, belonging to the field of water treatment process control and optimization technology. It collects historical water treatment operation data and constructs a mechanistic feature set, using the mechanistic feature set as the independent variable and turbidity reduction as the target variable. The turbidity reduction is used to represent the change in flocculation or sedimentation of suspended impurities in the water, and a full-variable regression model is constructed. Variables in the mechanistic feature set are screened, and the full-variable regression model is optimized using a stepwise regression method to obtain a simplified prediction model. Real-time influent water quality parameters are acquired and input into the simplified prediction model. The process of maximizing turbidity reduction in the simplified prediction model is transformed into minimizing a convex loss function, and the convex loss function is iteratively optimized using a hierarchical constrained projection gradient descent method to obtain the optimal dosing scheme. Through mechanism-driven modeling, convex function optimization, and hierarchical constrained projection, intelligent, efficient, and reliable control of the dosing process is achieved.

The invention discloses a task-oriented dialogue system reward function optimization method and system, and belongs to the technical field of natural language processing. The method comprises the following steps: acquiring an expert dialogue track from a dialogue system data set, extracting a state-action-reward triple, training an initial reward function by using a maximum entropy inverse reinforcement learning framework, and initializing a strategy network; an Actor-Critic reinforcement learning algorithm is adopted to train a strategy network, and a suboptimal trajectory is collected to dynamically update a reward function; and taking the dynamic reward function as a unified evaluation signal, and optimizing the strategy network to form a dialogue strategy model. Through dynamic reward function optimization, manual rule dependence is reduced, the generalization ability, the task completion rate and the stability of a dialogue system are improved, and the method is suitable for complex dialogue scenes in multiple fields.

The application provides a light steel hot deformation constitutive optimization method based on a neural network and precipitation strengthening coupling, and relates to the technical field of metal material hot working and constitutive model establishment; the neural network model automatically learns the nonlinear relationship among the rheological stress, temperature, strain rate and strain, and introduces a precipitation strengthening correction term in a stress prediction term, which is used for representing the competition effect of cutting mechanism and bypass mechanism; through engineering design of precipitation characteristic parameters, base training and strengthening parameter calibration are performed on the neural network model; a damage function optimization algorithm based on a dislocation slip mechanism is further introduced, so that the double constraints of prediction accuracy and physical consistency are realized; the application can realize high-precision flow stress prediction of light steel under different deformation temperatures and strain rates, significantly improves the generalization and physical interpretability of the model, and provides a new modeling approach and theoretical support for constitutive modeling and microstructure and performance control of light steel and other precipitation strengthening type high-strength alloys.

The invention discloses a construction settlement dynamic monitoring method based on data analysis, and relates to the technical field of construction engineering monitoring and data-driven modeling, and the method comprises the steps: combining a causal diagram modeling method with a multi-source data analysis mechanism, achieving the structural modeling and dynamic correlation recognition of multi-level observation variables in a construction settlement region, and achieving the dynamic monitoring of the construction settlement. According to the method, a causal structure is updated in real time under multi-source heterogeneous data flow through a fast conditional independence test algorithm in combination with a modular skeleton diagram updating mechanism, logic consistency of a monitoring model is kept, and a multi-objective function optimization algorithm is combined with DAG constraint and a direction confidence coefficient matrix, so that a multi-source heterogeneous data flow is optimized. According to the method, automatic judgment and global optimal causal direction reasoning of conflict causal relationships are achieved, a priori knowledge attenuation function is combined with a data-driven confidence fusion model, time sequence dynamic evaluation of causal edge reliability is achieved, and self-evolution of a knowledge system is achieved.

The invention provides an integrated topological optimization photonic device reverse design method. Maxwell equation solution and objective function optimization are put in the same position. And the optimization of a target function is realized while the Maxwell equation is solved. Compared with a traditional topological optimization method which needs to solve a Maxwell equation set for multiple times, the integrated topological optimization algorithm put forward by the invention has the advantages that the Maxwell equation and the objective function are placed at the same position, the limitation that one of the Maxwell equation and the objective function must be established constantly is relaxed, Maxwell equation solving and objective function optimization are realized at the same time, the Maxwell equation does not need to be solved again after parameters are updated each time, and the optimization efficiency of the Maxwell equation and the objective function is improved. And a large amount of computing resources required by topological optimization are greatly saved. The photonic device designed by the invention has the advantages of small physical size, suitability for large-scale integration and the like.