157 results about "Bilevel optimization" patented technology

The invention relates to the technical field of power data processing, in particular to an energy storage configuration optimization method, which comprises the following steps: acquiring new energy output time sequence data and computing power load characteristic data; generating a space-time correlation coupling evaluation result of the new energy output volatility and the computing power load volatility; inputting a result into a hybrid power supply double-layer optimization model, recursively correcting parameters through a two-stage collaborative solution algorithm, and outputting a Pareto optimal solution set; a computing power task elastic regulation and control mechanism is embedded, and the task priority is dynamically adjusted according to the energy storage charge state and the new energy output level to generate a scheduling strategy; finally, an energy storage configuration scheme and a dynamic scheduling strategy are output, and collaborative optimization of cost effectiveness and power supply reliability is achieved. The method breaks through the coupling conflict of the economic target and the robust constraint in the traditional bilevel planning, remarkably reduces the energy storage configuration cost, and improves the system stability.

The invention relates to a double-layer double-objective optimization scheduling method based on a large language model, and the method comprises the steps: building a double-layer double-objective optimization model in a server, and building an upper-layer model and a lower-layer model for an AISTS problem based on a plurality of different variables, different constraint conditions and different objective functions; and prompting a dual-objective optimization model of adjustment and algorithm evolution to form a two-stage method based on LLM, the method can overcome the defects that a traditional algorithm needs to design the algorithm manually and adjust parameters manually, the calculation cost is high, the efficiency is low, complex problems are difficult to effectively decompose and coordinate double-layer optimization, a dynamic collaborative optimization framework is lacked, and the flexibility of algorithm combination is low.

The invention discloses a power distribution network adaptive topological structure optimization method, system and device and a medium, and the method comprises the steps: obtaining related parameters of a power distribution network system, dividing the power distribution network system into different layers and different regions based on the related parameters of the power distribution network system, and constructing a layering and partitioning model; coordinating and solving the layering and partitioning model through an intelligent algorithm, and obtaining control variables of different regions; constructing an objective function based on the control variable, and solving the objective function by adopting an improved particle swarm optimization method to obtain a topological structure; and optimizing the topological structure by adopting a double-layer optimization strategy to obtain an optimization result, evaluating the optimization result, and feeding back the evaluation result to the target function. According to the method, the randomness and volatility challenges caused by multi-source resource access can be effectively dealt with, and the limitation of the traditional separation design is broken through the collaborative optimization of topology and control, so that the operation efficiency, the stability and the anti-interference capability of the power distribution network are improved, and the collaborative evolution of a topological structure and a control strategy is realized.

The invention discloses a park energy double-layer optimization scheduling method considering electric vehicles and demand response, and the method comprises the following steps: constructing a park integrated energy system which comprises an energy supply side, an energy conversion side and a demand side; a double-layer optimization model is established, the upper layer is a multi-target optimization model with the target of minimizing the total cost of the system, minimizing the carbon emission and maximizing the renewable energy consumption rate, and the lower layer is a single-target optimization model with the target of minimizing the charging cost of the electric vehicle user and the load fluctuation of the power grid; based on the Pareto theory, designing a multi-objective litsea coreana optimization algorithm to solve an upper-layer multi-objective model; introducing chaos initialization, adaptive step length adjustment and a Gaussian disturbance strategy improved litsea coreana optimization algorithm to solve a lower-layer single-target model; and through a vehicle network interaction and demand response cooperation mechanism, real-time scheduling is executed, a day-ahead plan is fed back and corrected, and finally an optimal park energy optimization scheduling strategy is solved.

The invention provides an electric vehicle charging station planning method and system, and the method comprises the steps: obtaining road network data and power distribution network topology information of a planning region, and constructing a time and space distribution model of an electric vehicle distributed load; based on historical travel behaviors and path probability distribution of the electric vehicle, constructing an electric vehicle behavior prediction model in combination with a graph neural network, and determining a candidate charging station service radius and a clustering region; according to the distributed photovoltaic access, the dynamic carbon emission factor and the vehicle charging and discharging behavior, constructing a dynamic optimization model of the coupling energy flow and the carbon flow; constructing an upper-layer charging station site selection model in the candidate charging station nodes, and determining an optimal site selection result by adopting a multi-objective optimization algorithm of an attention-based self-evolution Transform structure; establishing a lower-layer dynamic mixed integer programming configuration model based on the site selection result; and outputting an optimal deployment scheme considering energy efficiency, green low carbon and supply and demand balance by jointly solving the double-layer optimization model.

The invention relates to the technical field of alternating-current and direct-current power distribution networks, in particular to an alternating-current and direct-current power distribution network SOP partition interconnection locating and sizing method and a related device, and the method comprises the steps: obtaining GIS multi-source data and power distribution network data of a target region; dividing a power supply area according to the GIS multi-source data and the power distribution network data to obtain a to-be-interconnected partition pair set; generating a high-quality interconnected partition pair set by using the to-be-interconnected partition pair set; constructing a target optimization model by taking the minimum total operation cost as an optimization target and taking a power balance constraint, a node voltage constraint, a branch capacity constraint, an SOP operation constraint, an SOP reactive power constraint and an SOP capacity constraint as constraint conditions; and inputting the high-quality interconnection partition pair set into the target optimization model, cooperatively solving the target optimization model by adopting a double-layer optimization algorithm to obtain the optimal position and the optimal capacity of the SOP, and solving the technical problem that the existing SOP locating and sizing method is difficult to meet the targets of engineering feasibility and optimal operation benefit at the same time.

The invention belongs to the technical field of power grid optimization, and particularly relates to a GAN driven learning-based power grid planning equalization optimization method and device, and the method comprises the steps: constructing a physical constraint-based space-time generative adversarial network PI-ST-GAN, building a composite risk index, and screening out a high-risk scene from a generated source-load probabilistic time sequence scene; constructing a risk-cost balanced double-layer optimization model; reconstructing the double-layer optimization model into a single-layer mixed integer programming (MIP) model based on a strong dual theory, and setting constraint conditions of the single-layer mixed integer programming (MIP) model; based on a single-layer mixed integer programming MIP model, a solver is utilized to carry out joint solution on a line investment scheme and operation scheduling variables, PI-ST-GAN is dynamically guided to enhance high-risk scene generation according to a solution result, an optimization-generation-feedback closed-loop collaborative iteration solution process is realized, and finally an optimal power grid planning scheme is output. The method can improve the toughness and economical efficiency of a power grid under the high permeation of renewable energy sources.

The invention discloses an efficient graph comparison learning method for social recommendation. As an emerging self-supervised learning normal form, graph contrast learning is excellent in response to data sparseness and cold start due to the fact that the graph contrast learning can effectively capture similarity and heterogeneity characteristics in a graph structure, although the learning normal form achieves a good effect in a recommendation system, the graph contrast learning can be used for solving the problems of data sparseness and cold start. However, the method still faces three defects: (1) average neighbor aggregation and a non-adaptive representation reading mechanism are adopted in a message propagation process, and high-quality node representation is difficult to learn; (2) a visual angle is enhanced by depending on a random disturbance generation graph during intervention of comparative learning, which may destroy the inherent structure of graph data and further weaken the accuracy of the model; and (3) equally treating all observation samples during parameter optimization, and neglecting the difference influence of positive samples in different training stages. Specifically, aiming at the problems, the invention provides an efficient graph contrast learning method (EGCL for short). The method comprises the following steps: firstly, designing a graph adaptive propagation module, improving an information propagation rule of a graph neural network by referring to a thermonuclear thought and an attention mechanism, and realizing differentiated aggregation of neighbor nodes by adopting a learnable weight distribution strategy; secondly, designing a double contrast learning normal form which does not need graph enhancement, and realizing mutual promotion of node characterization through intra-domain contrast learning (inter-CL) and inter-domain contrast learning (inter-CL); and finally, introducing a sample weight adaptive efficient optimization algorithm, converting the training process into a double-layer optimization problem, and adaptively adjusting the contribution degree of each sample to model optimization in different stages.

The invention discloses a method for constructing an energy storage optimal scheduling uncertainty set fusing estimation and optimization, and the method comprises the following steps: constructing a risk-avoiding energy storage optimal scheduling model, constructing a robust optimization model based on an electricity price uncertainty error vector, and enabling an uncertainty set to cover the error distribution under a preset probability threshold; constructing a double-layer optimization model taking a decision as a center, and jointly optimizing related parameters and variables to form a nested structure; executing a statistical feasible size calibration method, determining a minimum feasible set radius based on mahalanobis distance sorting of historical samples, and ensuring that coverage probability constraints are met; and calculating an objective function gradient, solving the gradient by applying an implicit function theorem and a KKT condition, iteratively optimizing parameters by utilizing a gradient descent algorithm after synthesizing a total gradient, and obtaining an optimal uncertainty set in combination with an early stop mechanism. According to the method, the problem that a model and a strategy are disjointed in a traditional two-stage method of estimation first and optimization second is solved, and the robustness and the economical efficiency of energy storage scheduling can be effectively improved.

The invention discloses an industrial park virtual power plant low-carbon optimization scheduling method based on augmented Lagrange safety reinforcement learning, and belongs to the technical field of virtual power plant low-carbon optimization scheduling related to carbon emission transaction. The problem that in the prior art, a traditional virtual power plant low-carbon optimization scheduling method is poor in scheduling performance due to the fact that the optimization result deviates from reality is solved. According to the method, a constrained Markov decision process solving framework based on an augmented Lagrange soft actor-commentator algorithm is proposed, a double-layer optimization problem is converted into a constrained Markov decision process by performing multi-dimensional modeling and establishing a double-layer optimization framework, and an augmented Lagrange function is introduced, so that the constraint Markov decision process is obtained. And completing the design of an augmented Lagrange reinforcement learning algorithm, and obtaining a trained low-carbon optimization scheduling neural network model. According to the method, the low-carbon optimal scheduling efficiency of the virtual power plant is effectively improved, and the method can be applied to optimization of the EV aggregator charging and discharging strategy in the virtual power plant in the industrial park.

An SMES joint planning method considering carbon emission comprises the following steps: step 1, building a ship integrated energy system model based on a multi-energy coupling principle, the model comprising an AC / DC hybrid ship power system architecture model, a ship energy equipment model and a ship navigation model; 2, establishing a double-layer optimization planning model considering carbon emission by adopting the ship integrated energy system model obtained in the step 1, optimizing equipment configuration at the upper layer of the model by taking the minimum total investment cost as a target, and optimizing an equipment scheduling strategy at the lower layer by taking the operating cost and the carbon emission as double targets; and step 3, solving the double-layer optimization planning model considering the carbon emission obtained in the step 2 based on a mixed integer linear programming method, and realizing collaborative optimization of optimal capacity configuration and an operation strategy of the system through a CPLEX solver.

The invention discloses a game theory-based capacity configuration optimization method for a water, wind and light storage micro-grid, and relates to the technical field of energy system operation optimization configuration. According to the method, four types of energy operators including water, wind, light and storage serve as game subjects, and the problems of multi-subject benefit conflict and capacity configuration are solved through a cooperative game model and a double-layer optimization framework. The method comprises the following steps: constructing an alliance total revenue function and verifying the super-additivity; fairly distributing earnings by adopting a Shapley value in combination with a capacity correction factor; an economical, stable and environment-friendly multi-objective optimization model is established, an improved NSGA-III algorithm is utilized to dynamically generate a Pareto optimal solution set, and an optimal capacity configuration scheme is selected through three-dimensional tradeoff. According to the invention, the economical efficiency of the system is improved by 51.6%, the output fluctuation is reduced by 47.4%, the carbon emission is reduced by 43.2%, and an efficient and balanced capacity optimization method is provided for the multi-energy microgrid.

The invention discloses an energy storage type intelligent soft switch double-layer optimization site selection method based on improved whale algorithm and second-order cone programming cooperation, and belongs to the technical field of power distribution networking and optimization operation, and the method comprises the steps: collecting and preprocessing the basic data of a power distribution network; a double-layer optimization model with the minimum daily operation cost of the system as the target is established, the upper layer takes the installation position of the energy storage type intelligent soft switch as a decision variable, and the lower layer takes the three-port power transmission value as a decision variable and constructs constraints based on the power flow equation after second-order cone relaxation and the energy storage operation limitation; an upper layer site selection problem is solved by adopting an improved whale optimization algorithm introducing a self-adaptive inertia weight, a lower layer operation problem is solved by adopting a second-order cone programming method, and collaborative optimization is realized through double-layer interaction. According to the method, global collaborative optimization of planning and operation is realized, the solving precision and efficiency of the site selection problem are improved, and the system operation cost and the network loss are effectively reduced.

The invention discloses a polyhedral equivalent modeling method and device for a distributed energy cluster, and the method comprises the steps: building corresponding mathematical models for description according to the operation characteristics of different distributed energy sources; a generalized model of the distributed energy is established based on the established mathematical model for unified description, and an individual polyhedron feasible region model of the distributed energy is obtained according to variable controllable characteristics; performing internal approximation by using a homogeneous polyhedron in individual polyhedron feasible region models of all distributed energy sources, and calculating Minkowski sum of approximation results to obtain an aggregation model; an aggregation model construction process is expressed as a multi-objective optimization process by adopting a double-layer optimization method, an optimal homogeneous polyhedron shape is searched on an upper layer, internal approximation results of all individual feasible regions are searched on a lower layer, optimization is executed by using a moth fire suppression algorithm, and the accuracy of an aggregation model is maximized. According to the method, the problems that distributed energy sources with different characteristics cannot be described by using a unified model and the model precision is poor can be solved.

The invention relates to the field of edge computing and task scheduling, and particularly discloses a reinforcement learning and heuristic driven edge computing dependent task scheduling method, which comprises the following steps of: firstly, merging tasks of multiple users, and adding a starting node for all tasks without precursor tasks; in upper-layer optimization, an artificial intelligence algorithm based on reinforcement learning is used for modifying the dependency relationship of tasks, namely, a new edge is added, so that a new constraint condition is added; in the lower-layer optimization, the tasks are unloaded and solved by adopting a traditional heuristic algorithm to obtain a scheduling sequence and a scheduling place of the tasks; and iteratively optimizing the upper layer and the lower layer to minimize the target function. According to the edge computing dependency task scheduling method based on reinforcement learning and heuristic driving, high-quality user experience is achieved by dynamically increasing the dependency relationship between the tasks, and the method can be generalized to a large-scale scene.

The invention discloses a double-layer optimization method and system for real-time flood control scheduling of a series reservoir group. The method comprises the steps that the maximum feasible impounding boundary of each time period is calculated based on the discharge capacity and storage capacity constraint of each reservoir; constructing a space-time risk response kernel matrix, encoding lag response and attenuation intensity of flood routing, constructing a dynamic risk potential energy index in combination with real-time reservoir capacity, and solving space distribution water quantity of each reservoir by taking minimization of the index as a target under physical boundary constraint; the space distribution water quantity is used as a guide target, an evolutionary algorithm embedded with a physical consistent projection operator is adopted, and iterative optimization is carried out on hourly output flow of each reservoir. According to the method, through matrix coding of the space-time risk and introduction of manifold projection solution, the problem that upstream and downstream hydraulic space-time coupling and strong physical constraint are difficult to solve is effectively solved, and the physical consistency and peak clipping efficiency of a scheduling scheme are improved.

The invention relates to the technical field of three-dimensional point cloud measurement and intelligent path planning, solves the technical problems that a traditional method is strong in shielding, large in normal change, limited in reachability and difficult in path execution in point cloud measurement, and particularly relates to a double-layer optimization measurement planning method for a complex model measurement task. Calculating and caching a visible point set of each viewpoint under the conditions of angle constraint and shielding elimination based on the adjacency graph of the candidate viewpoints and a visibility evaluation mechanism; in the inner layer optimization, an initial measurement route and a viewpoint set are generated from an initial viewpoint, and a continuous executable measurement route is reconstructed after full coverage constraint is met; in the outer layer optimization, a genetic algorithm is adopted to carry out global search on a starting viewpoint, and the coverage rate and viewpoint scale cost are used as fitness to output an optimal measurement viewpoint set and a measurement path. According to the method, high-coverage-rate measurement planning can be realized under complex shielding and strict view angle constraint conditions, the number of viewpoints and path redundancy are reduced, and the planning efficiency and robustness are improved.

The invention discloses a dynamic time-of-use electricity price pricing method based on two-stage robust optimization. According to the method, a two-stage time-of-use electricity price robust optimization model is constructed; according to a duality theory, converting the problem into a double-layer optimization problem which can be iteratively solved, further splitting the double-layer optimization problem which can be iteratively solved into a main problem structure and a sub-problem structure, and iteratively approaching an optimal solution by using a parameterized CCG algorithm; the optimal solution of the two-stage time-of-use electricity price robust optimization model is used as the basic electricity price, a deep reinforcement learning framework is constructed to dynamically adjust the time-of-use electricity price, the dynamic supply and demand environment is matched, and an efficient, flexible and sustainable pricing scheme is provided for a novel power system.

The invention provides a park light storage and charging shared energy storage pricing method based on a Stackelberg game, and relates to the technical field of novel power systems, and the method comprises the steps: constructing a park light storage and charging system, constructing an upper layer model taking a shared energy storage power station as a leader, and constructing a lower layer model taking a user group as a follower, the upper layer model and the lower layer model form a double-layer optimization model by using a Stackelberg game mathematical model, the double-layer optimization model is converted into a single-layer optimization model by using a KKT optimality condition, the single-layer optimization model is solved by using a solver, and a shared energy storage price is obtained. According to the invention, behaviors of leaders and followers in the market can be simulated, a more reasonable and efficient pricing strategy is realized, a globally optimal solution is found under a KKT optimality condition, and the energy consumption cost of a user group is reduced. And meanwhile, a differential privacy disturbance mechanism is introduced, individual data leakage is prevented, and the balance of privacy protection and optimization effectiveness is realized.

The invention provides a man-machine collaborative dynamic scheduling method fused with a double-layer optimization mechanism, and relates to the field of man-machine collaborative dynamic scheduling. According to a double-layer optimization mechanism provided by the invention, an offline optimization layer and an online real-time scheduling layer are organically combined: firstly, the offline optimization layer is based on a workpiece set, stations and robot resources, and is combined with a simulated annealing algorithm for adaptive optimization to obtain a global optimal solution; and then, taking a global optimal solution output by the offline optimization layer as an initial scheduling scheme of the online real-time scheduling layer, and performing real-time adjustment by adopting a multi-agent adaptive near-end strategy optimization algorithm. According to the method, the low efficiency of online scheduling from zero exploration is avoided, and the high efficiency of decision making is ensured. Besides, the online real-time scheduling layer constructs a collaborative decision-making system composed of a task allocation agent, a resource scheduling agent and a disturbance response agent, decision-making dimension pressure faced by a single agent is remarkably reduced through specialized labor division, and a complex interaction relation in man-machine collaborative scheduling is delicately processed.

The invention relates to the technical field of electric energy metering, and discloses an electric energy meter metering error dynamic compensation method. According to the method, original metering data in the running process of the electric energy meter are collected and subjected to principal component analysis processing to obtain a metering principal component sequence of the electric energy meter. Performing wavelet transform processing on the principal component sequence to obtain an approximation coefficient and a detail coefficient; a resistor and an electric reactor are utilized to form a dynamic compensation circuit, an approximation coefficient is used as a metering reference value, detail coefficients are distributed in the dynamic compensation circuit, and a distribution coefficient is obtained. And constructing an electric energy meter measurement error compensation double-layer optimization model, wherein the model comprises an upper layer optimization model and a lower layer optimization model. And after constraint conditions of the two models are determined, inputting the distribution coefficient into the models, solving through gradient descent according to the constraint conditions to obtain metering error compensation parameters of the electric energy meter, and performing dynamic error compensation on the electric energy meter according to the parameters. The method effectively improves the metering precision of the electric energy meter, and has an important practical value.

The invention provides a double-layer optimization scheduling method for source network load storage collaborative loss reduction, and belongs to the technical field of source network load storage collaborative scheduling. A double-layer optimization framework comprising an upper layer planning model and a lower layer operation model is constructed, and a key node set is screened by using a node importance comprehensive scoring method based on a graph theory; the improved wolf pack algorithm is adopted to solve a double-layer coupling problem, a lower-layer operation effect is fed back to an upper layer to serve as a fitness evaluation basis, and in cooperation with population aggregation degree monitoring, a wandering wolf random jumping mechanism and a reverse learning strategy, deep collaboration and global optimal solution of a planning layer and an operation layer are achieved. The technical problem of poor network loss optimization effect caused by lack of an effective double-layer coupling solution mechanism for collaborative optimization planning and operation scheduling of the distributed photovoltaic and energy storage system in the power distribution network is solved.

The invention provides a scheduling method and device for source-network load-storage coordination of an active power distribution network, and computer equipment, and belongs to the technical field of power grid control. The method comprises the following steps: acquiring real-time data of a regulation and control main body, wherein the regulation and control main body comprises a power distribution main network and new energy, an energy storage unit and an adjustable load which are accessed to the power distribution main network; based on real-time data, optimal solutions of an outer-layer optimization problem and an inner-layer optimization problem of the double-layer optimization model are solved, and an optimal scheduling strategy in a scheduling period is obtained; generating a control instruction according to the optimal scheduling strategy, and issuing the control instruction to the regulation main body; wherein the double-layer optimization model is used for scheduling optimization of source-network-load-storage cooperative operation, peak clipping and valley filling maximization and new energy consumption maximization are taken as objective functions of the outer-layer optimization problem, and new energy output fluctuation and load curve fluctuation minimization are taken as objective functions of the inner-layer optimization problem. Through the method, source network load storage collaborative optimization scheduling is realized.

The invention is suitable for the technical field of market transaction optimization, and provides a double-layer optimization method and system for a virtual power plant to participate in market balanced transaction, and the method comprises the steps: constructing an upper-layer optimization model and a lower-layer optimization model, and enabling the upper-layer optimization model to achieve the maximization of the total profit of the virtual power plant as a target; the lower-layer optimization model takes market transaction subject benefit balance and market resource allocation optimization as targets; setting constraint conditions of the upper-layer model, wherein the constraint conditions comprise an energy market segmented quotation constraint, a standby market quotation constraint, a standby market capacity constraint and a distributed energy standby availability probability constraint; setting constraint conditions of the lower-layer model, wherein the constraint conditions comprise a node power balance constraint, a standby power consumption balance constraint, a transaction volume constraint, a capacity coupling constraint and a network topology constraint; and solving the solutions of the upper and lower optimization models by adopting a distributed iterative algorithm, and outputting the optimal transaction strategy and the market clearing price of the virtual power plant, so that the model adaptability is improved, and meanwhile, the balance of the benefits of the source-load double-side market subjects can be realized.

The invention relates to a micro-grid optimization scheduling method with electric vehicle and multi-source uncertainty. The micro-grid optimization scheduling method comprises the following steps: S1, constructing a multi-target optimization scheduling model of a micro-grid system layer; s2, performing robust modeling on multi-source uncertainty in the multi-target optimization scheduling model of the micro-grid system layer based on a multi-target confidence gap decision theory, and establishing a robust optimization model of the micro-grid system layer; s3, constructing an optimal scheduling model of an electric vehicle user layer based on a foreground theory, and taking electric vehicle user comprehensive foreground value maximization as a target function; and S4, constructing a double-layer segment model of day-ahead scheduling, carrying out cooperative solution on an upper layer and a lower layer by adopting a double-layer optimization algorithm, and outputting a day-ahead scheduling plan of each unit in the micro-grid and the electric vehicle. According to the method, the comprehensive operation cost, the load fluctuation level, the electric vehicle owner utility and the source load uncertainty of the micro-grid are integrated, and the internal load stability and the energy utilization rate of the micro-grid are improved by excavating the scheduling potential of the source load side.

The invention provides a smart building collaborative optimization operation method considering toughness constraint, and relates to the technical field of building comprehensive energy scheduling, comprising the steps of constructing a source network load storage integrated module of a smart building comprehensive energy system, integrating distributed energy, an energy storage unit and load data through an energy router, and establishing a source network load storage integrated module of the smart building comprehensive energy system; gradient utilization of electric energy, heat energy and cold energy is achieved based on the multi-energy complementation principle; a toughness index is constructed to quantify the ability of the intelligent building comprehensive energy system to deal with emergencies; constructing a double-layer optimization scheduling model based on the intelligent building comprehensive energy system and the toughness index; and solving the double-layer optimization scheduling model by using an improved grey wolf optimization algorithm to obtain an optimal energy scheduling strategy. According to the method, system economy and toughness indexes can be considered as a whole, the total power generation cost of the whole building comprehensive energy system is reduced as much as possible while the power utilization safety of the building is ensured and the new energy consumption capability is improved, and the optimization problem is solved through the improved grey wolf algorithm.

The invention relates to the technical field of computers, in particular to a double-layer optimization model training method and denoising method based on regularization outer gradient. The method comprises the steps that a training sample set is acquired, a to-be-trained model is subjected to a first preset number of times of iterative training based on the training sample set, each time of iterative training comprises multiple times of iterative updating processes, and a global target and a local target are considered through optimization structures of two levels of iterative training and iterative updating. And by introducing a variance reduction technology, a non-negligible calculation error introduced by a random sampling calculation gradient is avoided, the training speed of the model is improved, and the training cost of model training is reduced.

The invention discloses an optimal preset output solving method based on a minimum-maximum criterion and global robustness, and relates to the technical field of power system optimization scheduling. An existing scheduling method mainly focuses on robustness improvement of a local scene, cost deviation controllability of a global extreme scene is not achieved, the solving efficiency of a multi-layer optimization model is low, and the real-time scheduling requirement of an actual power system is difficult to meet. Establishing a power system economic dispatching minimum cost objective function, and establishing a linearized KKT condition constraint objective function; presetting a segmented cost function and a segmented constraint condition, and constraining the linearized preset cost function by using linear constraint; establishing an optimization target according to the minimum cost target function and the linearized preset cost function, processing the optimization target into a multi-layer optimization model, converting the multi-layer optimization model into a double-layer optimization model, and solving the double-layer optimization model by using a Benders decomposition method to obtain optimal solutions including preset wind power output, preset photovoltaic output and thermal power generating unit output. The method is used for obtaining power system output.

The present invention relates to the field of edge computing and task scheduling, and specifically discloses a reinforcement learning and heuristic driven edge computing dependent task scheduling method, comprising the following steps: first merging the tasks of multiple users, and adding a start node for all tasks without predecessor tasks; converting the edge computing and dependent task scheduling problem into a two-layer optimization problem: in the upper layer optimization, using an artificial intelligence algorithm based on reinforcement learning to modify the dependency relationship of the tasks, that is, adding new edges, thereby adding new constraints; in the lower layer optimization, using a traditional heuristic algorithm to offload and solve the tasks, and obtain the scheduling order and scheduling location of the tasks; iteratively optimizing the upper and lower layers to minimize the objective function. The present invention adopts the above-mentioned reinforcement learning and heuristic driven edge computing dependent task scheduling method, and achieves high-quality user experience by dynamically increasing the dependency relationship between tasks, and can be generalized to large-scale scenarios.

Decentralized bilevel optimization techniques for personalized learning over a heterogenous network are provided. In one aspect, a decentralized learning system includes: a distributed machine learning network with multiple nodes, and datasets associated with the nodes; and a bilevel learning structure at each of the nodes for optimizing one or more features from each of the datasets using a decentralized bilevel optimization solver, while maintaining distinct features from each of the datasets. A method for decentralized learning is also provided.