Method and system for coordinated regulation of multi-capacity coupled middle-deep buried pipe groups and power grid

Abstract

The application discloses a multi-time scale coordinated regulation method and system for multiple functionally coupled middle-deep buried pipe groups and power grids, which comprises a hierarchical regulation architecture designed as "bottom single pipe real-time control-middle pipe group heat balance-top system planning", and a multi-time scale regulation is performed by combining model predictive control (MPC) and deep reinforcement learning algorithm; a complementary and coordinated model of geothermal-wind-solar energy storage is constructed, based on system dynamics and energy quality gradient matching theory, an energy conversion and distribution mechanism under multiple time scales is analyzed, an effective coordination among different energy forms is performed through the establishment of a heat-electricity-storage dynamic balance strategy, a carbon trading cost prediction module and a power spot price prediction module are integrated; an action space dimension reduction is performed by adopting a deep reinforcement learning based physical constraint embedding strategy, a solution tool integration is performed by combining a reinforcement learning optimization method and an efficient solver, a multi-disturbance scene verification platform is constructed, complex working conditions of power grid fluctuation and penetration rate mutation are simulated and robustness test is performed, and a target model is output.

Description

Technical Field

[0001] This invention relates to the field of coordinated control, and in particular to a method and system for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids. Background Technology

[0002] Significant progress has been made in recent years in the research of operation strategies, particularly in the regulation and parameter optimization of single-tube thermal storage. However, existing research indicates that relying solely on the number of start-ups and shutdowns can lead to an average annual energy loss of 12%-15%. This demonstrates the critical importance of the operation strategy for medium-deep thermal extraction on system performance.

[0003] Current regulation strategies face three major bottlenecks: First, multi-timescale response mismatch and the lack of a cross-level physical-information bidirectional coupling mechanism lead to poor information flow between multi-level regulation, affecting the synergy of global energy allocation. Second, insufficient understanding of the energy level gradient coordination mechanism between geothermal and wind / solar energy, coupled with carbon trading and electricity spot market price fluctuations, further increases scheduling complexity, and an optimization scheme that balances economic efficiency, low carbon emissions, and stability has not yet been formed. Third, the solvability barrier of the high-dimensional dynamic optimization decision space makes it difficult for conventional algorithms to meet real-time regulation requirements under time-varying geological parameters, complex pipeline topology, and multi-objective constraints. It is necessary to establish optimization and solution methods that combine intelligent control theory with multi-objective coordination to achieve adaptive regulation of pipeline groups and participate in efficient coordination and stable interaction between diverse renewable energy sources and the power grid.

[0004] For the operation of medium-deep geothermal closed-loop buried pipe systems and multi-energy systems, evaluation indicators such as annual sustainability, seasonal non-guarantee rate, and life-cycle cost have been proposed. Furthermore, optimization configuration methods based on source-side temperature and flow have been developed, along with hierarchical control strategies for multi-energy systems considering multiple objectives such as renewable energy consumption, grid interaction, and economics. However, existing strategies mostly focus on the single-pipe scale, and pipe-group level control faces the problem of mismatch between minute-level network fluctuations and multi-timescale variations in cross-seasonal thermal storage cycles. Regarding energy system synergy, preliminary results have been achieved in the intraday complementarity research of medium-deep geothermal with photovoltaic and shallow geothermal systems, and existing coupled energy storage system design methods have raised the clean power consumption rate to a new level. However, the dynamic interaction mechanism between medium-deep pipe groups and the power grid still lacks theoretical research, and it is urgent to overcome the optimization barriers of multi-level gradient synergy and market-physics dual constraints. Summary of the Invention

[0005] The purpose of this invention is to provide a method for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids.

[0006] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0007] This invention includes the following steps:

[0008] A hierarchical control architecture is designed, consisting of "bottom-level single-pipe real-time control - middle-level pipe group thermal balance - top-level system planning," integrating Model Predictive Control (MPC) and deep reinforcement learning algorithms for multi-timescale control. This architecture comprises a bottom layer, a middle layer, and a top layer. The bottom layer employs MPC to optimize the heat load distribution between energy stations and satellite stations based on heat load prediction, achieving real-time and precise control of individual pipes. The middle layer coordinates heat distribution among pipe groups through a thermal balance algorithm. The top layer utilizes system planning methods for overall energy management and optimization, establishing a rolling time-domain optimization feedback mechanism.

[0009] Construct a complementary and synergistic model of geothermal-wind-solar-energy storage, analyze the energy conversion and distribution mechanism under multiple time scales based on system dynamics and energy quality cascade matching theory, and achieve effective synergy among different energy forms by establishing a dynamic balance strategy of heat-electricity-storage, and integrate carbon trading cost prediction and electricity spot price prediction modules.

[0010] We employ a deep reinforcement learning-based approach combined with a physical constraint embedding strategy to reduce the dimensionality of the action space, thereby achieving efficiency and physical feasibility. We integrate the solution tools by combining reinforcement learning optimization methods and efficient solvers, construct a multi-perturbation scenario verification platform, simulate complex operating conditions such as power grid fluctuations and sudden changes in penetration rate, conduct robustness tests, and output the target model.

[0011] Furthermore, the underlying method is designed as follows:

[0012] A model-based prediction-based system scheduling and control method is adopted. First, the heat load demand of each heating station in the area is predicted, and then optimization analysis and decision-making are carried out based on the prediction. The scheduling and control method is classified according to the time and spatial dimensions: the spatial dimension mainly includes the equipment receiving control commands, ground source heat pumps, satellite stations and energy stations; in the time dimension, it is divided into optimization control under the short time scale and scheduling optimization under the medium and long time scale according to the optimization analysis scale.

[0013] The top layer of the heating system first decomposes tasks based on energy production and orderly energy consumption plans. The operation scheduling module generates control strategies based on the specific conditions of multiple heat sources and the operating status of each device. In the intraday real-time optimization module, control targets are generated through real-time online monitoring of equipment and target optimization combination, and control actions are generated using advanced control methods. After completing the top-down scheduling and control process, the system collects monitoring data from lower-level devices in real time and feeds it back to the upper layer online, realizing a closed loop in the scheduling and control process. The upper-level links make corrections based on the feedback from the lower-level links.

[0014] An optimized control logic is adopted to control the model-based heating system. The optimized control logic is as follows: the future heat load at the heat station level is predicted by load forecasting technology, and the heating load allocation between the energy station and the satellite station is determined by combining the heating capacity and heating cost curves of the energy station and the satellite station. Under the premise of maximizing the use of geothermal heat source heating capacity, the system coordinates with multiple energy station heat sources for efficient heating, so as to achieve low-carbon operation of the heating system. Based on the heat load allocation, the scheduling platform gives the heating flow rate and supply and return water temperature of each energy station.

[0015] Furthermore, the method for designing the middle layer includes:

[0016] To regulate the heat distribution among different pipe groups, the host computer platform calculates the comprehensive physical property coefficient of the room, classifies heat users, determines the regulation cycle, and determines the target return water temperature based on historical heating parameter information.

[0017] When the heat demand of users changes, the return water temperature is adjusted to adapt to the heat demand. The host computer platform issues an adjustment command to change the opening degree of the smart valve, thereby changing the return water temperature of the heat users to achieve the target return water temperature. When the real-time return water temperature meets the requirement... and At that time, the intelligent valve adjustment for the heat user ends; the real-time return water temperature for the heat user is... The target value for real-time return water temperature for heat users is The return water temperature deviation threshold is ξ 2h The actual temperature of the input water is The target temperature of the input water is Temperature deviation threshold is ξ in ;

[0018] Conversely, continue adjusting the valve opening. 2h Different systems may use different threshold values. Changes in the opening degree of the intelligent valve for heat users will cause changes in the differential pressure and impedance of the entire pipe network. At this time, the water pumps at the heating station will perform linked frequency conversion regulation to change the flow rate, restoring the differential pressure and impedance of the pipe network to the set values. When the pipe network impedance and differential pressure meet the requirements... |SS op |<σ s | and the valve opening degree most unfavorable to the heat user is the maximum V z =V max When the frequency conversion stops, it ends; otherwise, it continues. The actual pressure difference in the pipeline network is ΔP, and the set value for the pipeline network pressure difference is... The deviation threshold of the pipeline pressure difference is σ P The current actual impedance of the pipeline network is S, and the set value of the pipeline network impedance is S. op The deviation threshold of the pipeline impedance is σ. s .

[0019] Furthermore, the method for designing the top-level structure includes:

[0020] Multi-energy synergy integrated energy system planning is a process of conversion between multiple energy sources. The system planning method abstracts an integrated energy system into a two-port network of energy input and output, in which multiple energy sources are converted, distributed and stored within the integrated energy system;

[0021] The system planning method connects the input end to the energy network, inputting the corresponding electricity, gas, and oil energy, and outputting energy in the form of electricity, heat, and cooling. Based on the actual situation of the operation phase, an integrated energy system planning and operation optimization model is established.

[0022] The optimization plan involves optimizing the configuration of equipment capacity and quantity, as well as optimizing the cooling, heating, and electrical operation modes of the equipment. Based on the decomposition and coordination principle, the optimization plan is transformed into a two-level planning model. The upper-level plan aims to maximize the net present value within the integrated energy system and optimizes the configuration of equipment capacity and quantity. The lower-level optimization corresponds to the optimization of the integrated energy system's operation strategy. Based on the equipment capacity and quantity optimized in the upper level, the objective function is constructed by objectively weighting the minimum operating cost of the integrated energy system and the maximum proportion of electricity in total energy consumption. According to the equipment operation constraints, the cooling, heating, and electrical operation modes of the equipment are optimized. The optimization objective is passed to the upper-level optimization, the internal rate of return is calculated, and the optimal "planning and operation integrated" integrated energy system optimization plan is obtained through iterative optimization between the upper and lower levels.

[0023] Furthermore, the method for integrating the carbon trading cost forecasting and electricity spot price forecasting modules includes:

[0024] On the TRNSYS platform, geothermal, wind, solar, and energy storage are coupled together. The multi-energy coupled heating and cooling system model includes an energy consumption module, a cost module, and an output module. The energy consumption module is used to calculate the annual energy consumption of each heating and cooling source subsystem. The cost module uses time-of-use electricity and gas prices to calculate the annual operating costs. The output module outputs the capacity ratio, initial investment, operating energy consumption, cooling / heating capacity, total initial investment, annual operating costs, annual carbon emissions, and life cycle cost of each subsystem.

[0025] In the multi-energy coupled heating and cooling system model, variable optimization is performed using the GenOpt optimization tool connected to the TrnOpt module in TRNSYS.

[0026] By incorporating carbon emissions into the geothermal-wind-solar-energy storage system, a carbon emission cost model is established, expressed as follows:

[0027]

[0028] Among them, the basic carbon emission quota is Q0, the equivalent actual carbon emission is Q1, and the carbon emission cost is F co2 , the carbon emission cost coefficients are C1, C2, C3, C4 respectively, the first-level boundary quota of carbon emission is, and the second-level boundary quota of carbon emission is Q 02 ;

[0029] When the emission Q1 is lower than Q0, the geothermal-wind-energy storage system sells the excess emission quota; when the emission Q1 is higher than Q0, the geothermal-wind-energy storage system needs to purchase carbon emission quotas, and as the purchased quota increases, the unit purchase cost increases; when Q1 < Q0, it means that when the emission does not exceed the basic quota Q0, the surplus quota is sold; when the emission exceeds different quotas, the excess part purchases carbon emission quotas according to the corresponding emission costs;

[0030] The electricity spot market conducts real-time trading of electricity in a short time cycle. The electricity spot price reflects the dynamic balance of electricity supply and demand and is affected by multiple factors, including weather conditions, load demand, fuel prices, generation capacity, network constraints, and market participants;

[0031] According to the advantages of different methods, a hybrid model of time series models, machine learning models, multi-model integration, and decomposition-based methods is proposed. The hybrid model effectively captures the complex characteristics of electricity spot prices; among them, the hybrid of time series models and machine learning models: first, use the time series model to extract the linear characteristics of price data, and then use the machine learning model to learn the non-linear characteristics of the residuals; multi-model integration: perform weighted averaging on the prediction results of multiple different models; decomposition-based method: first decompose the original price data into components of different frequencies, then establish prediction models for each component, and finally combine the prediction results of each component;

[0032] By establishing a thermal-electricity-energy storage dynamic balance strategy, effective coordination between different energy forms is achieved, and a carbon trading cost prediction and electricity spot price prediction module is integrated to ensure the optimal scheduling of the energy system under different market environments and operating conditions.

[0033] Furthermore, the method for reducing the dimensionality of the action space includes:

[0034] Based on the general framework of multi-objective evolutionary algorithms for online target dimensionality reduction, the ε-MOSFLA algorithm is combined with SORA1 and SORA2 target dimensionality reduction algorithms to construct online target dimensionality reduction algorithms to meet different needs. The online target dimensionality reduction algorithms include SO1-MOSFLA and SO2-MOSFLA algorithms. The SO1-MOSFLA algorithm uses SORA1 to reduce a fixed number of targets each time until the desired number of targets is reached, at which point no more targets are deleted. The SO2-MOSFLA algorithm uses SORA2 to adaptively find unimportant targets for deletion based on a given error threshold until no more targets can be deleted or the number of targets is reduced to two.

[0035] Based on the online objective dimensionality reduction algorithm, a strategy based on objective integration is proposed. The objectives to be deleted are integrated into a single objective by weighted summation based on their importance index and added to the objective subset. This is achieved by solving two types of high-dimensional multi-objective optimization problems: redundant and non-redundant.

[0036] We propose a sparse feature selection-based method that utilizes the geometric structure properties of the approximate solution set and the Pareto dominance relationship to measure the importance of the target and construct a target dimensionality reduction algorithm for solving two different needs.

[0037] Based on the idea of ​​sparse feature selection, a target preference ranking and evaluation algorithm is proposed. When the error of changing the original problem does not exceed the error threshold, the target preference ranking and evaluation algorithm starts from the original target set and reduces one target at a time. The algorithm calculates the comprehensive ranking of the targets in terms of importance in each dimensionality reduction process to determine the ranking of each target.

[0038] Furthermore, the robustness testing method includes:

[0039] Obtain the labels obtained from the classification of the U-shaped medium-deep underground heat exchange model, construct a local alternative model that is highly similar to the classification ability of the U-shaped medium-deep underground heat exchange model, construct adversarial examples based on the output of the alternative model, and use the transferability of the adversarial examples to test the robustness of the target model.

[0040] Secondly, the multi-energy coupled medium-deep underground pipeline network and power grid coordinated control system includes:

[0041] The hierarchical multi-scale control module is used to design a hierarchical control architecture of "bottom-level single-pipe real-time control - middle-level pipe group thermal balance - top-level system planning". It integrates model predictive control (MPC) and deep reinforcement learning algorithms for multi-time-scale control. The hierarchical control architecture includes a bottom layer, a middle layer, and a top layer. The bottom layer uses model predictive control to optimize the heat distribution between the energy station and the satellite station based on heat load prediction, achieving real-time and precise control of a single pipe. The middle layer coordinates the heat distribution among the pipe groups through a thermal balance algorithm. The top layer uses a system planning method to perform overall energy management and optimization, and establishes a rolling time-domain optimization feedback mechanism.

[0042] Collaborative Prediction Module: Used to construct a complementary and collaborative model of geothermal-wind-solar-energy storage. Based on system dynamics and energy quality cascade matching theory, it analyzes the energy conversion and distribution mechanism under multiple time scales. It establishes a dynamic balance strategy of heat-electricity-storage to achieve effective collaboration among different energy forms and integrates carbon trading cost prediction and electricity spot price prediction modules.

[0043] Dimensionality Reduction Testing Module: This module is used to reduce the dimensionality of the action space using a deep reinforcement learning-based strategy combined with physical constraint embedding to obtain efficiency and physical feasibility. It integrates reinforcement learning optimization methods and efficient solvers to build a multi-perturbation scenario verification platform, simulates complex operating conditions such as power grid fluctuations and sudden changes in penetration rate, conducts robustness tests, and outputs the target model.

[0044] The beneficial effects of this invention are:

[0045] This invention relates to a method and system for coordinated control of multi-energy coupled medium-deep underground pipeline groups and power grids. Compared with existing technologies, this invention has the following technical advantages:

[0046] This invention improves the accuracy of constructing a U-shaped medium-deep underground heat exchange model for buried pipes through hierarchical regulation, multi-timescale regulation, construction of a complementary and synergistic model, energy conversion and distribution mechanisms, effective coordination, integrated prediction modules, action space dimensionality reduction, and robustness testing. This optimization of the U-shaped medium-deep underground heat exchange model construction significantly saves resources, improves work efficiency, and enables the scientific construction of such models, allowing for real-time monitoring of the U-shaped medium-deep underground heat exchange model. The underground heat exchange model of buried pipes is constructed with hierarchical control and integrated prediction modules. By building a three-level intelligent control architecture of "single pipe-pipe group-system", integrating multivariate control theory and efficient optimization algorithms, the problem of real-time solution of high-dimensional decision-making is overcome, and minute-level thermal equilibrium response is achieved. At the same time, in view of the complementary characteristics of geothermal-wind-solar-energy storage, a multi-timescale collaborative model covering day-ahead, intraday and real-time is established, which greatly improves the overall energy efficiency of the system and effectively optimizes the peak-valley regulation capability of the power grid, providing cutting-edge technical support for the safe and efficient utilization of high proportion of renewable energy. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the steps of the multi-energy coupled medium-deep underground pipe network and power grid coordinated control method of the present invention.

[0048] Figure 2 This is a logic diagram of the heating system optimization and control based on model prediction in this embodiment;

[0049] Figure 3 This is a flowchart illustrating the heat distribution control process among different pipe groups in this embodiment;

[0050] Figure 4 This is a flowchart of the integrated energy system optimization planning scheme in this embodiment. Detailed Implementation

[0051] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0052] The present invention provides a method and system for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids, comprising the following steps:

[0053] like Figure 1 As shown, this embodiment includes the following steps:

[0054] A hierarchical control architecture is designed, consisting of "bottom-level single-pipe real-time control - middle-level pipe group thermal balance - top-level system planning," integrating Model Predictive Control (MPC) and deep reinforcement learning algorithms for multi-timescale control. This architecture comprises a bottom layer, a middle layer, and a top layer. The bottom layer employs MPC to optimize the heat load distribution between energy stations and satellite stations based on heat load prediction, achieving real-time and precise control of individual pipes. The middle layer coordinates heat distribution among pipe groups through a thermal balance algorithm. The top layer utilizes system planning methods for overall energy management and optimization, establishing a rolling time-domain optimization feedback mechanism.

[0055] Construct a complementary and synergistic model of geothermal-wind-solar-energy storage, analyze the energy conversion and distribution mechanism under multiple time scales based on system dynamics and energy quality cascade matching theory, and achieve effective synergy among different energy forms by establishing a dynamic balance strategy of heat-electricity-storage, and integrate carbon trading cost prediction and electricity spot price prediction modules.

[0056] We employ a deep reinforcement learning-based approach combined with a physical constraint embedding strategy to reduce the dimensionality of the action space, thereby achieving efficiency and physical feasibility. We integrate the solution tools by combining reinforcement learning optimization methods and efficient solvers, construct a multi-perturbation scenario verification platform, simulate complex operating conditions such as power grid fluctuations and sudden changes in penetration rate, conduct robustness tests, and output the target model.

[0057] In this embodiment, the method for designing the underlying layer includes:

[0058] A model-based prediction-based system scheduling and control method is adopted. First, the heat load demand of each heating station in the area is predicted, and then optimization analysis and decision-making are carried out based on the prediction. The scheduling and control method is classified according to the time and spatial dimensions: the spatial dimension mainly includes the equipment receiving control commands, ground source heat pumps, satellite stations and energy stations; in the time dimension, it is divided into optimization control under the short time scale and scheduling optimization under the medium and long time scale according to the optimization analysis scale.

[0059] The top layer of the heating system first decomposes tasks based on energy production and orderly energy consumption plans. The operation scheduling module generates control strategies based on the specific conditions of multiple heat sources and the operating status of each device. In the intraday real-time optimization module, control targets are generated through real-time online monitoring of equipment and target optimization combination, and control actions are generated using advanced control methods. After completing the top-down scheduling and control process, the system collects monitoring data from lower-level devices in real time and feeds it back to the upper layer online, realizing a closed loop in the scheduling and control process. The upper-level links make corrections based on the feedback from the lower-level links.

[0060] An optimized control logic is employed to manage the model-based heating system. This optimized control logic involves: predicting future heat loads at the heat station level using load forecasting technology; determining the heat load allocation between the energy station and satellite stations based on their heating capacity and cost curves; and coordinating with multiple energy stations for efficient heating while maximizing the utilization of geothermal heat source capacity to achieve low-carbon operation of the heating system. Based on the heat load allocation, the scheduling platform assigns heating flow rates and supply / return water temperatures to each energy station.

[0061] In actual assessments, the multi-unit load allocation optimization method improves the overall heating efficiency and economic benefits of the units while reducing carbon emissions.

[0062] In this embodiment, the method for designing the middle layer includes:

[0063] To regulate the heat distribution among different pipe groups, the host computer platform calculates the comprehensive physical property coefficient of the room, classifies heat users, determines the regulation cycle, and determines the target return water temperature based on historical heating parameter information.

[0064] When the heat demand of users changes, the return water temperature is adjusted to adapt to the heat demand. The host computer platform issues an adjustment command to change the opening degree of the smart valve, thereby changing the return water temperature of the heat users to achieve the target return water temperature. When the real-time return water temperature meets the requirement... and At that time, the intelligent valve adjustment for the heat user ends; the real-time return water temperature for the heat user is... The target value for real-time return water temperature for heat users is The return water temperature deviation threshold is ξ 2h The actual temperature of the input water is The target temperature of the input water is Temperature deviation threshold is ξ in ;

[0065] Conversely, continue adjusting the valve opening. 2h Different systems may use different threshold values. Changes in the opening degree of the intelligent valve for heat users will cause changes in the differential pressure and impedance of the entire pipe network. At this time, the water pumps at the heating station will perform linked frequency conversion regulation to change the flow rate, restoring the differential pressure and impedance of the pipe network to the set values. When the pipe network impedance and differential pressure meet the requirements... |SS op |<σ s | and the valve opening degree most unfavorable to the heat user is the maximum V z =V max When the frequency conversion stops, it ends; otherwise, it continues. The actual pressure difference in the pipeline network is ΔP, and the set value for the pipeline network pressure difference is... The deviation threshold of the pipeline pressure difference is σ P The current actual impedance of the pipeline network is S, and the set value of the pipeline network impedance is S. op The deviation threshold of the pipeline impedance is σ. s .

[0066] In this embodiment, the method for designing the top layer includes:

[0067] Multi-energy synergy integrated energy system planning is a process of conversion between multiple energy sources. The system planning method abstracts an integrated energy system into a two-port network of energy input and output, in which multiple energy sources are converted, distributed and stored within the integrated energy system;

[0068] The system planning method connects the input end to the energy network, inputting the corresponding electricity, gas, and oil energy, and outputting energy in the form of electricity, heat, and cooling. Based on the actual situation of the operation phase, an integrated energy system planning and operation optimization model is established.

[0069] The optimization planning involves an optimization configuration scheme for equipment capacity and quantity, as well as optimizing the cold, heat, and power operation modes of equipment. According to the idea of decomposition and coordination, the optimization planning is transformed into a two-layer planning model. The upper-layer planning aims to maximize the internal net present value within the integrated energy system and carry out the optimization configuration of equipment capacity and quantity. The lower-layer optimization corresponds to the optimization of the operation strategy of the integrated energy system. Based on the equipment capacity and quantity optimized by the upper layer, an objective function is objectively weighted by the minimum operation cost of the integrated energy system and the maximum proportion of electric energy in the total energy consumption. According to the equipment operation constraints, the cold, heat, and power operation modes of the equipment are optimized, and the optimization objective is transmitted to the upper-layer optimization. The internal rate of return is calculated, and the optimal "planning-operation integration" integrated energy system optimization planning scheme is obtained through iterative optimization of the upper and lower layers.

[0070] In this embodiment, the method of the integrated carbon trading cost prediction and electricity spot price prediction module includes:

[0071] On the TRNSYS platform, geothermal - wind - energy storage is coupled. The multi - energy coupled heating and cooling system model includes an energy consumption module, a cost module, and an output module. Among them, the energy consumption module is used to count the annual energy consumption of each cold and heat source subsystem. The cost module calculates the annual operation cost using time - of - use electricity prices and gas prices. The output module outputs the capacity ratio, initial investment, operation energy consumption, cooling / heating capacity, total initial investment, annual operation cost, annual carbon emissions, and life - cycle cost of each subsystem.

[0072] In the multi - energy coupled heating and cooling system model, with the help of the GenOpt optimization tool externally connected to the TrnOpt module in TRNSYS, variable optimization is carried out.

[0073] Carbon emissions are incorporated into the geothermal - wind - energy storage system, and a carbon emission cost model is established. The expression is:

[0074]

[0075] Where the basic carbon emission quota is Q0, the equivalent actual carbon emission is Q1, and the carbon emission cost is F co2 , the carbon emission cost coefficients are C1, C2, C3, C4 respectively, the first - level boundary quota of carbon emission is, and the second - level boundary quota of carbon emission is Q 02 ;

[0076] When the emission Q1 is lower than Q0, the geothermal - wind - energy storage system sells the excess emission quota. When the emission Q1 is higher than Q0, the geothermal - wind - energy storage system needs to purchase carbon emission quotas, and as the purchased quota increases, the unit purchase cost increases. When Q < Q0, it means that when the emission does not exceed the basic quota Q0, the excess quota is sold. When the emission exceeds different quotas, the excess part purchases carbon emission quotas according to the corresponding emission cost.

[0077] The electricity spot market facilitates real-time electricity trading over short time periods. Electricity spot prices reflect the dynamic balance of electricity supply and demand and are influenced by a variety of factors, including weather conditions, load demand, fuel prices, generation capacity, grid constraints, and market participants.

[0078] Based on the advantages of different methods, a hybrid model is proposed, which combines time series models, machine learning models, multi-model ensemble, and decomposition-based methods. The hybrid model effectively captures the complex characteristics of electricity spot prices. The hybrid model combines time series models and machine learning models by first extracting the linear features of price data using a time series model and then using a machine learning model to learn the nonlinear features of the residuals. The multi-model ensemble involves weighted averaging of the prediction results from multiple different models. The decomposition-based method first decomposes the original price data into components of different frequencies, then builds a prediction model for each component, and finally combines the prediction results of each component.

[0079] By establishing a dynamic balance strategy for heat, electricity, and storage, effective synergy among different energy forms is achieved. By integrating carbon trading cost prediction and electricity spot price prediction modules, the energy system can be optimally dispatched under different market environments and operating conditions.

[0080] In practical assessments, hybrid models effectively capture the complex characteristics of electricity spot prices, exhibiting higher prediction accuracy and better generalization capabilities. By establishing a dynamic balance strategy across heat, electricity, and storage, effective synergy among different energy forms is achieved, improving the overall system operating efficiency. Integrating carbon trading cost prediction and electricity spot price prediction modules, the optimized model targets encompass economic efficiency, low carbon emissions, and stability, ensuring optimal dispatch of the energy system under different market environments and operating conditions.

[0081] In this embodiment, the method for action space dimensionality reduction includes:

[0082] Based on the general framework of multi-objective evolutionary algorithms for online target dimensionality reduction, the ε-MOSFLA algorithm is combined with SORA1 and SORA2 target dimensionality reduction algorithms to construct online target dimensionality reduction algorithms to meet different needs. The online target dimensionality reduction algorithms include SO1-MOSFLA and SO2-MOSFLA algorithms. The SO1-MOSFLA algorithm uses SORA1 to reduce a fixed number of targets each time until the desired number of targets is reached, at which point no more targets are deleted. The SO2-MOSFLA algorithm uses SORA2 to adaptively find unimportant targets for deletion based on a given error threshold until no more targets can be deleted or the number of targets is reduced to two.

[0083] Based on the online objective dimensionality reduction algorithm, a strategy based on objective integration is proposed. The objectives to be deleted are integrated into a single objective by weighted summation based on their importance index and added to the objective subset. This is achieved by solving two types of high-dimensional multi-objective optimization problems: redundant and non-redundant.

[0084] We propose a sparse feature selection-based method that utilizes the geometric structure properties of the approximate solution set and the Pareto dominance relationship to measure the importance of the target and construct a target dimensionality reduction algorithm for solving two different needs.

[0085] Based on the idea of ​​sparse feature selection, a target preference ranking and evaluation algorithm is proposed. When the error of changing the original problem does not exceed the error threshold, the target preference ranking and evaluation algorithm starts from the original target set and reduces one target at a time. The comprehensive ranking of the targets with respect to importance is calculated in each dimensionality reduction process to determine the ranking of each target.

[0086] In actual evaluation, simulation results show that the target dimensionality reduction algorithm proposed in this patent can accurately remove redundant targets for high-dimensional multi-objective optimization problems with different redundancy, and its dimensionality reduction accuracy is almost unaffected by the quality of the approximate solution set, demonstrating strong robustness.

[0087] In this embodiment, the robustness test method includes:

[0088] Obtain the labels obtained from the classification of the U-shaped medium-deep underground heat exchange model, construct a local alternative model that is highly similar to the classification ability of the U-shaped medium-deep underground heat exchange model, construct adversarial examples based on the output of the alternative model, and use the transferability of the adversarial examples to test the robustness of the target model.

[0089] Secondly, the multi-energy coupled medium-deep underground pipeline network and power grid coordinated control system includes:

[0090] The hierarchical multi-scale control module is used to design a hierarchical control architecture of "bottom-level single-pipe real-time control - middle-level pipe group thermal balance - top-level system planning". It integrates model predictive control (MPC) and deep reinforcement learning algorithms for multi-time-scale control. The hierarchical control architecture includes a bottom layer, a middle layer, and a top layer. The bottom layer uses model predictive control to optimize the heat distribution between the energy station and the satellite station based on heat load prediction, achieving real-time and precise control of a single pipe. The middle layer coordinates the heat distribution among the pipe groups through a thermal balance algorithm. The top layer uses a system planning method to perform overall energy management and optimization, and establishes a rolling time-domain optimization feedback mechanism.

[0091] Collaborative Prediction Module: Used to construct a complementary and collaborative model of geothermal-wind-solar-energy storage. Based on system dynamics and energy quality cascade matching theory, it analyzes the energy conversion and distribution mechanism under multiple time scales. It establishes a dynamic balance strategy of heat-electricity-storage to achieve effective collaboration among different energy forms and integrates carbon trading cost prediction and electricity spot price prediction modules.

[0092] Dimensionality Reduction Testing Module: This module is used to reduce the dimensionality of the action space using a deep reinforcement learning-based strategy combined with physical constraint embedding to obtain efficiency and physical feasibility. It integrates reinforcement learning optimization methods and efficient solvers to build a multi-perturbation scenario verification platform, simulates complex operating conditions such as power grid fluctuations and sudden changes in penetration rate, conducts robustness tests, and outputs the target model.

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for coordinated control of multi-energy coupled medium-deep underground pipe networks and power grids, characterized in that, Includes the following steps: A hierarchical control architecture of "bottom-level single-pipe real-time control - middle-level pipe group thermal balance - top-level system planning" is designed, integrating Model Predictive Control (MPC) and deep reinforcement learning algorithms for multi-timescale control. The hierarchical control architecture includes a bottom layer, a middle layer, and a top layer. The bottom layer adopts a model-predictive system scheduling and control method, first predicting the heat load demand of each heating station within the area, and then performing optimization analysis and decision-making based on the prediction in both time and spatial dimensions. Based on the heat load prediction, the load allocation between the energy station and the satellite station is optimized to achieve real-time and precise control of a single pipe. The middle layer, through a host computer... The system determines the target return water temperature based on historical heating parameters, adjusts the opening of smart valves to change the return water temperature, and combines the frequency conversion of water pumps in the heating station to regulate the pressure difference and impedance of the pipeline network. It also coordinates the heat distribution among the pipe groups through a heat balance algorithm. The top layer abstracts the integrated energy system into a dual-port network of energy input and output, and establishes a two-layer optimization planning model that integrates planning and operation. The upper layer optimizes the equipment capacity and quantity by maximizing the internal net present value, while the lower layer optimizes the equipment operation mode by minimizing operating costs and maximizing the proportion of electricity. The system planning method is used for overall energy management and optimization, and a rolling time-domain optimization feedback mechanism is established. Construct a complementary and synergistic model of geothermal-wind-solar-energy storage, analyze the energy conversion and distribution mechanism under multiple time scales based on system dynamics and energy quality cascade matching theory, and achieve effective synergy among different energy forms by establishing a dynamic balance strategy of heat-electricity-storage, and integrate carbon trading cost prediction and electricity spot price prediction modules. We employ a deep reinforcement learning-based approach combined with a physical constraint embedding strategy to reduce the dimensionality of the action space, thereby obtaining efficiency and physical feasibility. We integrate the solution tools by combining reinforcement learning optimization methods and efficient solvers, construct a multi-perturbation scenario verification platform, simulate complex operating conditions such as power grid fluctuations and sudden changes in penetration rate, conduct robustness tests, and output the control results.

2. The method for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids according to claim 1, characterized in that, The underlying method is designed as follows: A model-based prediction-based system scheduling and control method is adopted. First, the heat load demand of each heating station in the area is predicted, and then optimization analysis and decision-making are carried out based on the prediction. The scheduling and control method is classified according to the time and spatial dimensions: the spatial dimension mainly includes the equipment receiving control commands, ground source heat pumps, satellite stations and energy stations; in the time dimension, it is divided into optimization control under the short time scale and scheduling optimization under the medium and long time scale according to the optimization analysis scale. The top layer of the heating system first decomposes tasks based on energy production and orderly energy consumption plans. The operation scheduling module generates control strategies based on the specific conditions of multiple heat sources and the operating status of each device. In the intraday real-time optimization module, control targets are generated through real-time online monitoring of equipment and target optimization combination, and control actions are generated using advanced control methods. After completing the top-down scheduling and control process, the system collects monitoring data from lower-level devices in real time and feeds it back to the upper layer online, realizing a closed loop in the scheduling and control process. The upper-level links make corrections based on the feedback from the lower-level links. An optimized control logic is adopted to control the model-based heating system. The optimized control logic is as follows: the future heat load at the heat station level is predicted by load forecasting technology, and the heating load allocation between the energy station and the satellite station is determined by combining the heating capacity and heating cost curves of the energy station and the satellite station. Under the premise of maximizing the use of geothermal heating capacity, the system coordinates with multiple energy station heat sources for efficient heating, so as to achieve low-carbon operation of the heating system. Based on the heat load allocation, the scheduling platform gives the heating flow rate and supply and return water temperature of each energy station.

3. The method for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids according to claim 1, characterized in that, The method for designing the middle layer includes: To regulate the heat distribution among different pipe groups, the host computer platform calculates the comprehensive physical property coefficient of the room, classifies heat users, determines the regulation cycle, and determines the target return water temperature based on historical heating parameter information. When the heat demand of users changes, the return water temperature is adjusted to adapt to the heat demand. The host computer platform issues an adjustment command to change the opening degree of the smart valve, thereby changing the return water temperature of the heat users to achieve the target return water temperature. When the real-time return water temperature meets the requirement... ,and At that time, the intelligent valve adjustment for the heat user ends; the real-time return water temperature for the heat user is... The target value for the real-time return water temperature of heat users is The return water temperature deviation threshold is The actual temperature of the input water is The target temperature of the input water is Temperature deviation threshold is ; Conversely, continue adjusting the valve opening. Different systems can use different threshold values; changes in the opening degree of the intelligent valve for heat users will cause changes in the differential pressure and impedance of the entire pipe network. At this time, the water pumps of the heating station will perform linked frequency conversion regulation to change the flow rate, so that the differential pressure and impedance of the pipe network can be restored to the set value; when the pipe network impedance and differential pressure meet the requirements... , | and the valve opening for the least favorable heat user is at its maximum. When the frequency conversion stops, it ends; otherwise, it continues. The actual pressure difference in the pipeline network is... The set value for pipeline differential pressure is The deviation threshold for pipeline pressure difference is The current actual impedance of the pipeline network is S, and the set value of the pipeline network impedance is... The deviation threshold of the pipeline impedance is .

4. The method for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids according to claim 1, characterized in that, The method for designing the top-level structure includes: Multi-energy synergy integrated energy system planning is a process of conversion between multiple energy sources. The system planning method abstracts an integrated energy system into a two-port network of energy input and output, in which multiple energy sources are converted, distributed and stored within the integrated energy system; The system planning method connects the input end to the energy network, inputting the corresponding electricity, gas, and oil energy, and outputting energy in the form of electricity, heat, and cooling. Based on the actual situation of the operation phase, an integrated energy system planning and operation optimization model is established. The optimization plan involves optimizing the configuration of equipment capacity and quantity, as well as optimizing the cooling, heating, and electrical operation modes of the equipment. Based on the decomposition and coordination principle, the optimization plan is transformed into a two-level planning model. The upper-level plan aims to maximize the net present value within the integrated energy system and optimizes the configuration of equipment capacity and quantity. The lower-level optimization corresponds to the optimization of the integrated energy system's operation strategy. Based on the equipment capacity and quantity optimized in the upper level, the objective function is constructed by objectively weighting the minimum operating cost of the integrated energy system and the maximum proportion of electricity in total energy consumption. According to the equipment operation constraints, the cooling, heating, and electrical operation modes of the equipment are optimized. The optimization objective is passed to the upper-level optimization, the internal rate of return is calculated, and the optimal "planning and operation integrated" integrated energy system optimization plan is obtained through iterative optimization between the upper and lower levels.

5. The method for coordinated control of multi-energy coupled medium-deep underground pipeline groups and power grids according to claim 1, characterized in that, The method for integrating carbon trading cost forecasting and electricity spot price forecasting modules includes: On the TRNSYS platform, geothermal, wind, solar, and energy storage are coupled together. The multi-energy coupled heating and cooling system model includes an energy consumption module, a cost module, and an output module. The energy consumption module is used to calculate the annual energy consumption of each heating and cooling source subsystem. The cost module uses time-of-use electricity and gas prices to calculate the annual operating costs. The output module outputs the capacity ratio, initial investment, operating energy consumption, cooling / heating capacity, total initial investment, annual operating costs, annual carbon emissions, and life cycle cost of each subsystem. In the multi-energy coupled heating and cooling system model, variable optimization is performed using the GenOpt optimization tool connected to the TrnOpt module in TRNSYS. By incorporating carbon emissions into the geothermal-wind-solar-energy storage system, a carbon emission cost model is established, expressed as follows: The basic carbon emission allowance is The equivalent actual carbon emissions are The carbon emission cost is The carbon emission cost coefficients are respectively , , , The primary threshold for carbon emissions is The secondary boundary allowance for carbon emissions is ; When emissions Below In this case, the geothermal-wind-solar-energy storage system will sell excess emission allowances; when emissions... Higher than At that time, geothermal-wind-solar-energy storage systems require the purchase of carbon emission credits, and the unit cost increases with the increase in the amount of credits purchased; when When, it means that the emissions do not exceed the basic limit. The excess carbon emission credits will be sold; when emissions exceed different credit limits, the excess portion will be used to purchase carbon emission credits at the corresponding emission cost. The electricity spot market facilitates real-time electricity trading over short time periods. Electricity spot prices reflect the dynamic balance of electricity supply and demand and are influenced by a variety of factors, including weather conditions, load demand, fuel prices, generation capacity, grid constraints, and market participants. Based on the advantages of different methods, a hybrid model is proposed, which combines time series models, machine learning models, multi-model ensemble, and decomposition-based methods. The hybrid model effectively captures the complex characteristics of electricity spot prices. The hybrid model combines time series models and machine learning models by first extracting the linear features of price data using a time series model and then using a machine learning model to learn the nonlinear features of the residuals. The multi-model ensemble involves weighted averaging of the prediction results from multiple different models. The decomposition-based method first decomposes the original price data into components of different frequencies, then builds a prediction model for each component, and finally combines the prediction results of each component. By establishing a dynamic balance strategy for heat, electricity, and storage, effective synergy among different energy forms is achieved. By integrating carbon trading cost prediction and electricity spot price prediction modules, the energy system can be optimally dispatched under different market environments and operating conditions.

6. The method for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids according to claim 1, characterized in that, The method for reducing the dimensionality of the action space includes: Based on the general framework of multi-objective evolutionary algorithms for online objective dimensionality reduction, The -MOSFLA algorithm is combined with the SORA1 and SORA2 target dimensionality reduction algorithms to construct online target dimensionality reduction algorithms to meet different needs. The online target dimensionality reduction algorithms include the SO1-MOSFLA algorithm and the SO2-MOSFLA algorithm. The SO1-MOSFLA algorithm uses SORA1 to reduce a fixed number of targets each time until the desired number of targets is reached, at which point no more targets are deleted. The SO2-MOSFLA algorithm uses SORA2 to adaptively find unimportant targets to delete based on a given error threshold until no more targets can be deleted or the number of targets is reduced to two. Based on the online objective dimensionality reduction algorithm, a strategy based on objective integration is proposed. The objectives to be deleted are integrated into a single objective by weighted summation based on their importance index and added to the objective subset. This is achieved by solving two types of high-dimensional multi-objective optimization problems: redundant and non-redundant. We propose a sparse feature selection-based method that utilizes the geometric structure properties of the approximate solution set and the Pareto dominance relationship to measure the importance of the target and construct a target dimensionality reduction algorithm for solving two different needs. Based on the idea of ​​sparse feature selection, a target preference ranking and evaluation algorithm is proposed. When the error of changing the original problem does not exceed the error threshold, the target preference ranking and evaluation algorithm starts from the original target set and reduces one target at a time. The algorithm calculates the comprehensive ranking of the targets in terms of importance in each dimensionality reduction process to determine the ranking of each target.

7. The method for coordinated control of multi-energy coupled medium-deep underground pipe groups and power grids according to claim 1, characterized in that, The robustness testing method includes: Obtain the labels obtained from the classification of the U-shaped medium-deep underground heat exchange model, construct a local alternative model that is highly similar to the classification ability of the U-shaped medium-deep underground heat exchange model, construct adversarial examples based on the output of the alternative model, and use the transferability of the adversarial examples to test the robustness of the target model.

8. A multi-energy coupled medium-deep underground pipe network and power grid coordinated control system, used to perform the method according to any one of claims 1-7, characterized in that, include: A hierarchical multi-scale control module is used to design a hierarchical control architecture of "bottom-level single-pipe real-time control - middle-level pipe group heat balance - top-level system planning," integrating model predictive control (MPC) and deep reinforcement learning algorithms for multi-time-scale control. The hierarchical control architecture includes a bottom layer, a middle layer, and a top layer. The bottom layer employs a model-predictive system scheduling control method, first predicting the heat load demand of each heating station within the area, and then performing optimization analysis and decision-making based on the prediction in both time and spatial dimensions. Based on the heat load prediction, it optimizes the load allocation between energy stations and satellite stations, achieving real-time and precise control of a single pipe. The middle layer... The target return water temperature is determined by the host computer platform based on historical heating parameters. The opening of the smart valve is adjusted to change the return water temperature. Combined with the linkage of the water pump of the heating station to adjust the pressure difference and impedance of the pipeline network, the heat distribution among the pipe groups is coordinated by the heat balance algorithm. The top layer abstracts the integrated energy system into a dual-port network of energy input and output, and establishes a two-layer optimization planning model integrating planning and operation. The upper layer optimizes the equipment capacity and quantity by maximizing the internal net present value, and the lower layer optimizes the equipment operation mode by minimizing the operating cost and maximizing the proportion of electricity. The system planning method is used for overall energy management and optimization, and a rolling time-domain optimization feedback mechanism is established. Collaborative Prediction Module: Used to construct a complementary and collaborative model of geothermal-wind-solar-energy storage. Based on system dynamics and energy quality cascade matching theory, it analyzes the energy conversion and distribution mechanism under multiple time scales. It establishes a dynamic balance strategy of heat-electricity-storage to achieve effective collaboration among different energy forms and integrates carbon trading cost prediction and electricity spot price prediction modules. Dimensionality Reduction Testing Module: This module is used to reduce the dimensionality of the action space using a deep reinforcement learning-based strategy combined with physical constraint embedding to obtain efficiency and physical feasibility. It integrates reinforcement learning optimization methods and efficient solvers to build a multi-perturbation scenario verification platform, simulates complex operating conditions such as power grid fluctuations and sudden changes in penetration rate, conducts robustness tests, and outputs the target model.

Images