Dynamic power management and multi-energy collaborative optimization method combined with AI technology

Abstract

The invention discloses a dynamic power management and multi-energy collaborative optimization method combined with an AI technology, and the method comprises the steps: training an AI prediction model to predict a future network load through collecting historical network flow, equipment power consumption and energy supply data; generating a power consumption optimization strategy according to the predicted network load; through a double-layer optimization algorithm, multi-energy dynamic distribution is realized; and resource scheduling is performed by using a genetic algorithm, so that resource redundancy is reduced. The method has the advantages that the traditional deep learning algorithm and the symbol AI are combined, the interpretation of the power consumption adjustment strategy is realized, the transparency of AI scheduling is improved, and the controllability and reliability are enhanced. Meanwhile, a power consumption optimization strategy is dynamically generated through dynamic programming and a heuristic algorithm, the utilization rate of renewable energy sources is increased, and traditional energy dependence is reduced. And network resources are abstracted through a software defined network technology, and a genetic algorithm is introduced for resource scheduling, so that resource redundancy is remarkably reduced, and the network utilization rate is improved.

Description

Technical Field

[0001] The present invention relates to the field of energy optimization management technology, and is particularly suitable for dynamic power management and multi-energy collaborative optimization methods combined with AI technology. Background Art

[0002] With the rapid development of 5G communication technology, the energy consumption of network base stations and data centers has increased significantly, and the requirements for sustainable development have continued to increase. In order to improve the energy utilization efficiency of network base stations and data centers, reduce energy waste, and implement a green and low-carbon sustainable development strategy, most network base stations and data centers are equipped with corresponding power optimization management strategies. The current common solution is to configure a static power management strategy or apply an AI optimization model locally in network base stations and data centers.

[0003] Static power management strategies usually do not consider the dynamics of multiple energy flows and multiple time scales. It is difficult to balance accuracy and calculation speed in long-time optimization, resulting in the optimization results may not be globally optimal, unable to effectively cope with real-time changes in network loads, and insufficient to adapt to complex network environments. Moreover, static power configuration strategies cannot fully utilize multiple energy sources, limiting the potential of network base stations and data centers for the comprehensive use of multiple and environmentally friendly energy sources.

[0004] The local configuration and application of AI optimization models in network base stations and data centers focuses on local explanations and lacks an effective global coordination mechanism, resulting in a disconnect between resource allocation and energy management, and failing to accurately represent overall behavior. At the same time, for network base stations and data centers, uneven optimization strategies can cause some parts of the network to perform better than others, causing data traffic congestion and reduced service quality. Summary of the invention

[0005] The purpose of the present invention is to provide a dynamic power management and multi-energy collaborative optimization method combined with AI technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions: The carbon emission data prediction method for park building scenarios described in the present invention comprises the following steps: S1, collects historical network traffic, device power consumption and energy supply data to train AI prediction models; S2, based on real-time network traffic, device power consumption and energy supply data, uses AI prediction models to predict future network loads; S3, generating a power consumption optimization strategy based on the predicted network load; S4, achieves dynamic allocation of multiple energy sources through a two-layer optimization algorithm; S5, use genetic algorithm to schedule resources and reduce resource redundancy.

[0007] Furthermore, the AI ​​prediction model in step S1 includes a decision tree and a linear regression model.

[0008] Furthermore, the power consumption optimization strategy includes identifying peak and valley load periods and determining energy usage principles during peak and valley load periods.

[0009] Furthermore, the upper layer of the two-layer optimization algorithm adopts the MILP algorithm to optimize the long-term allocation of energy resources with the goal of reducing costs and improving the utilization rate of renewable energy. The lower layer dynamically adjusts energy use and intelligently switches power sources based on load forecasts and current energy supply conditions.

[0010] The advantage of the present invention is that it combines traditional deep learning algorithms with symbolic AI to achieve the interpretability of power consumption adjustment strategies, improve the transparency of AI scheduling, and enhance controllability and reliability. At the same time, the present invention dynamically generates power consumption optimization strategies through dynamic programming and heuristic algorithms to improve the utilization rate of renewable energy and reduce dependence on traditional energy. It also abstracts network resources through software-defined networking (SDN) technology and introduces genetic algorithms for resource scheduling, significantly reducing resource redundancy and improving network utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a flow chart of the dynamic power management and multi-energy collaborative optimization method combined with AI technology in the present invention. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] The dynamic power management and multi-energy collaborative optimization method combined with AI technology described in the present invention is as follows: Figure 1 As shown, the following steps are included: S1, collects historical network traffic, device power consumption and energy supply data to train AI prediction models.

[0014] Data cleaning, missing value filling and feature normalization were performed on the collected historical network traffic, device power consumption and energy supply data. The processed sample data set was divided into training set, validation set and test set. Model training was performed on the training set, and grid search was used to optimize hyperparameters. Model performance was evaluated using the validation set, and model parameters were adjusted to improve generalization ability. A final evaluation was performed on the test set to ensure the interpretability and prediction accuracy of the model.

[0015] The AI ​​models used in the present invention include decision trees, linear regression or random forests.

[0016] The present invention also uses SHAP value (Shapley Additive Explanations) and LIME (Local Interpretable Modelagnostic Explanations) technology to provide detailed feature contribution analysis and local explanations to improve the interpretability of AI prediction models.

[0017] S2,uses AI prediction model to generate prediction of future network load based on real-time network traffic, device power consumption and energy supply data.

[0018] Real-time network traffic, equipment power consumption and energy supply data can be automatically collected through IoT devices, and integrated and analyzed through the cloud platform to form a real-time dynamic load view.

[0019] S3, based on the predicted network load results, uses dynamic programming and heuristic algorithms to dynamically generate power consumption optimization strategies to achieve the optimal balance between energy efficiency and performance. Strategies include: identifying peak and valley load periods, and determining energy usage principles during peak and valley load periods. During peak load periods: give priority to the use of renewable energy to avoid excessive use of traditional energy. During low load periods: reduce the power output of the equipment and enter energy-saving mode. The present invention can manage network base station equipment, such as 5G base station equipment, etc.; data center equipment, such as servers, storage devices, network equipment, etc.; energy supply equipment, such as traditional power supply equipment, renewable energy equipment such as photovoltaic power generation equipment, energy storage systems, etc.; and Internet of Things data acquisition equipment.

[0020] S4, through a two-layer optimization algorithm, dynamic allocation of multiple energy sources is achieved.

[0021] The upper layer of the two-layer optimization algorithm adopts the MILP algorithm, which takes a 24-hour cycle, aims to reduce costs and improve the utilization rate of renewable energy, and optimizes the ratio of traditional energy and renewable energy according to the predicted load data, current energy prices and other data to achieve long-term allocation of energy resources. Specifically, the optimization objective function of the MILP algorithm is established as: Min Z = Σ(C_t * X_t + C_r * Y_r + C_s * S_s + C_switch * Z_switch +C_carbon * E_carbon) Among them: C_t, X_t are the unit cost and usage of traditional energy; C_r, Y_r are the unit cost and usage of renewable energy; C_s, S_s are the unit cost and usage of energy storage system; C_switch is the switching loss cost; Z_switch is the switching loss; C_carbon is the unit cost of carbon emissions; E_carbon is the carbon emissions.

[0022] Then, under the constraints of the core constraint equations of the optimization objective function, the optimal solution and the optimal energy configuration plan are solved.

[0023] The lower layer uses a dynamic programming algorithm to solve the optimal execution path of the optimal energy configuration solution based on the optimal energy configuration solution output by the upper layer, combined with real-time network traffic, equipment power consumption, and energy supply data, in preset minutes or hours, such as 5 minutes or 1 hour, and generates scheduling instructions to achieve dynamic and real-time energy scheduling. When there is a surplus of energy, the two-layer optimization algorithm stores excess energy in the energy storage system. When energy is scarce, energy in the energy storage system is used first to cooperate with the traditional power grid.

[0024] S5, use genetic algorithm to schedule resources and reduce resource redundancy.

[0025] The specific steps include: S5.1, initialize the population to represent different resource scheduling schemes.

[0026] S5.2, evaluate the power consumption, load balancing and response time of each solution.

[0027] S5.3,generate a new generation of solutions through selection, crossover and mutation operations.

[0028] S5.4, repeat the iteration until the optimal resource scheduling solution that meets the constraints is found.

[0029] When executing the optimal resource scheduling plan, different data centers are connected through the SDN controller to achieve global abstraction and management of network resources.

Claims

1. A dynamic power management and multi-energy collaborative optimization method combining AI technology, characterized in that: The following steps are involved: S1, collects historical network traffic, device power consumption and energy supply data to train AI prediction models; S2, uses AI prediction models to predict future network loads based on real-time network traffic, device power consumption, and energy supply data; S3, generating a power consumption optimization strategy based on the predicted network load; S4, achieves dynamic allocation of multiple energy sources through a two-layer optimization algorithm; S5, use genetic algorithm to schedule resources and reduce resource redundancy.

2. The method for dynamic power management and multi-energy collaborative optimization combined with AI technology according to claim 1, characterized in that: The AI ​​prediction models in step S1 include decision trees and linear regression models.

3. The method for dynamic power management and multi-energy collaborative optimization combined with AI technology according to claim 1, characterized in that: The power consumption optimization strategy includes identifying peak and valley load periods and determining energy usage principles during peak and valley load periods.

4. The method for dynamic power management and multi-energy collaborative optimization combined with AI technology according to claim 1, characterized in that: The upper layer of the two-layer optimization algorithm adopts the MILP algorithm to optimize the long-term allocation of energy resources with the goal of reducing costs and improving the utilization rate of renewable energy; the lower layer dynamically adjusts energy use and intelligently switches power sources based on load forecasts and current energy supply conditions.

Images