Optimized scheduling method for regional integrated energy system

Abstract

The invention discloses an optimal scheduling method for a regional integrated energy system, and relates to the technical field of energy optimization. Comprising the following steps: step 1, multi-source data hierarchical acquisition and preprocessing; step 2, multi-time scale uncertainty scene modeling is carried out; step 3, constructing a multi-objective optimization model; 4, intelligent algorithm solving and dynamic scheduling are carried out; 5, carrying out constraint processing and resource integration; and 6, carrying out closed-loop correction and strategy optimization. According to the regional integrated energy system optimization scheduling method, a dynamic relaxation factor mu (t) is introduced to adjust equipment climbing constraint in real time, and a virtual energy storage equivalent model is combined to integrate a transferable load, so that when the load of the system suddenly increases, the equipment power adjustment fluctuation is controlled within 10% of rated power, frequent out-of-limit of energy storage is avoided, and meanwhile, the optimal scheduling of the regional integrated energy system is realized. The grading pricing mechanism excitation system of the stepped carbon transaction model preferentially consumes new energy, and the carbon emission is greatly reduced compared with that of a traditional fixed carbon price model.

Description

Technical Field

[0001] The present invention relates to the technical field of energy optimization, and in particular to a method for optimizing and scheduling a regional integrated energy system. Background Art

[0002] As the conflict between energy and the environment deepens, building a safe, efficient, energy-saving, and carbon-reducing new energy system is a key direction for the ambitious "3060" goal. In recent years, the rise of the Energy Internet of Things (IoT) has driven the transformation of traditional energy sources. A key vehicle for this IoT is the District Integrated Energy System (DIES). DIES achieves a balance between energy supply and demand through multi-energy coupling and collaborative interaction.

[0003] The invention with publication number CN118735177A discloses a method for optimizing and dispatching a regional integrated energy system, including: constructing a cloud-pipe-edge-based regional integrated energy system operation communication architecture; establishing a two-layer distributed dispatching model for a regional integrated energy system that takes waste heat recovery into consideration based on a stepped carbon trading mechanism. At the upper layer, the regional integrated energy system is considered to participate in demand response and electricity purchase and sales, and an optimization dispatching model with low carbon and economic efficiency as the goals is established. At the lower layer, the characteristics of waste heat recovery of subsystem energy coupling equipment are considered, and an optimization dispatching model that takes waste heat recovery into consideration is established; consistency constraints and augmented Lagrangian penalty functions are introduced, and the target cascade analysis algorithm is used to achieve decoupling of the two-layer model, achieving efficient solution in multiple iterations. By adjusting the output of flexible resources, users are guided to participate in grid interaction, and the economy and low carbon efficiency of system operation are improved while protecting the privacy and interests of multiple parties.

[0004] The existing patents have built a two-layer distributed scheduling model, but the upper layer optimization only considers the demand response and power purchase and sale under fixed scenarios, and does not fully incorporate the variable operating characteristics of the equipment, resulting in a lag in the response of the scheduling strategy to the real-time operating parameters. The existing technology uses the target cascade analysis method (ATC) combined with the augmented Lagrangian penalty function for two-layer decoupling, but the penalty factor and scaling factor s i The update rules rely on preset empirical values ​​(e.g., k = 1.5) and lack adaptive adjustment to system status. When renewable energy output fluctuates suddenly or the power of multiple parks interacts drastically, the algorithm's convergence rate slows significantly and may even fall into a local optimum, leading to increased system operating costs or unexpected carbon trading costs. Summary of the Invention

[0005] In view of the deficiencies of the existing technology, the present invention provides a method for optimizing and scheduling a regional integrated energy system, which solves the existing problems.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for optimizing and dispatching a regional integrated energy system, comprising the following steps:

[0007] Step 1: Multi-source data layered collection and preprocessing: Real-time collection of distributed power supply, energy storage equipment, multiple loads, and market price data through edge computing terminals;

[0008] Step 2: Multi-time scale uncertainty scenario modeling: Generate initial scenarios based on Monte Carlo simulation for ultra-short-term data, screen typical scenarios through Kullback-Leibler divergence and retain probability weights, build a Bayesian network model for short-term data, input weather variables and output load forecast distribution P(L t |W t ,T t );

[0009] Step 3: Construct a multi-objective optimization model: Establish a multi-objective optimization model including economic cost C op , Tiered carbon trading costs C 碳 , Conditional Value at Risk Cost C risk The multi-objective function F = αC op +βC 碳 +γC risk , where the weight α+β+γ=1 is determined by the analytic hierarchy process;

[0010] Step 4: Intelligent algorithm solution and dynamic scheduling: Using the improved deep reinforcement learning (DRL) algorithm, define the state space S that includes energy storage charge state, load deviation, and electricity price fluctuations. t and the action space A including energy storage charging and discharging power, interconnection line interaction power, and equipment start and stop status t ;

[0011] Step 5: Constraint processing and resource integration: Introduce a dynamic relaxation factor μ(t) to adjust the power upper and lower limits for the equipment ramp constraint, equate the transferable load to virtual energy storage, and establish a state equation. Realize coordinated dispatch of load and energy storage;

[0012] Step 6. Closed-loop correction and strategy optimization: Real-time collection of scheduling execution data to calculate the prediction error ∈ t , if ∈ t >5%, dynamically adjust model parameters through momentum update rules to form a closed loop of "acquisition-optimization-execution-feedback".

[0013] Preferably, the preprocessing in step 1 is to construct a three-dimensional data set D = {d t ,i,p}, the adaptive Kalman filter algorithm is used to denoise the time series data and dynamically update the measurement noise covariance matrix R tAnd extract the high-frequency fluctuation and low-frequency trend components of renewable energy output through variational mode decomposition (VMD);

[0014] The noise covariance update formula of the adaptive Kalman filter is:

[0015]

[0016] Among them, R t represents the measurement noise covariance matrix at time t, γ is the forgetting factor (0.8≤γ≤0.95), M is the amount of sliding window data, which is used to filter out abnormal data with a signal-to-noise ratio of less than 3dB, z k represents the kth measurement value, represents the estimated value of the kth measurement, It represents the average value of the square of the error between the measured value and the estimated value in the sliding window of size M, reflecting the magnitude of the current measurement error;

[0017] The variational mode decomposition (VMD) obtains the modal components by solving the following constrained optimization problem:

[0018]

[0019] Among them, u k is the kth modal component, ω k is the center frequency, K (2≤K≤5) indicates that the optimal decomposition parameters are determined by non-recursive iteration. represents the derivative with respect to time t, j is the imaginary unit, Indicates constraints.

[0020] Preferably, in the ultra-short-term scenario modeling, typical scenarios are screened by Kullback-Leibler divergence, and the probability weight w is retained. s satisfy And the scene error is ≤5%; the Bayesian network model is based on the conditional probability matrix Correlate weather variables with load forecasts.

[0021] Preferably, in step 4, the reward function is designed as follows through the experience replay buffer and the target network optimization strategy:

[0022] r t =-C t +λ1ΔSOC+λ2ΔP grid ;

[0023] Among them, r t represents the reward value at time t, C t represents the real-time operating cost, ΔSOC represents the balance of energy storage charge state, and represents the amount of grid interaction power fluctuation suppression.

[0024] Preferably, the tiered carbon trading cost is calculated based on the difference between carbon emissions and quotas, with a carbon price gradient of ≥ 3 levels, and the formula is:

[0025]

[0026] Among them, C 碳 represents the total carbon trading cost, t represents the time step, C t is the basic carbon price, a is the gradient coefficient, d is the length of the quota interval, E emit.t represents the actual carbon emissions at time t, E 配额,t represents the carbon emission quota allocated to the energy system at time t.

[0027] Preferably, in the reward function of the deep reinforcement learning (DRL), Indicates the balance of energy storage charge state, where SOC t Indicates the state of charge of the energy storage device at time t, SOC nom is the nominal state of charge of the energy storage device, SOC max is the maximum state of charge of the energy storage device, represents the amount of suppression of grid interaction power fluctuations, and the weights λ1 and λ2 are determined by Pareto frontier optimization, where P tie,t represents the interaction power with the grid at time t, P tie,max Indicates the maximum interactive power with the grid.

[0028] Preferably, the dynamic relaxation factor μ(t) is calculated as follows:

[0029]

[0030] Among them, D t is the real-time load deviation rate, k=10 is the adjustment coefficient, and D0=0.1 is the load deviation threshold, ensuring that the equipment power adjustment fluctuation is ≤10% of the rated power.

[0031] Preferably, in the virtual energy storage equivalent model, S DR,max is the maximum capacity of virtual energy storage, which is set to 15% of the load peak, and the response efficiency η DR =0.9, the state update satisfies 0≤S DR,t ≤S DR,max , realizing the coordinated scheduling of demand response load and physical energy storage.

[0032] Preferably, in the two-layer iterative solution, the upper layer uses the alternating direction multiplier method (ADMM) to decouple multi-region collaboration, and the objective function is:

[0033]

[0034] Where N is the number of regions, F i represents the objective function value of the i-th region, ρ = 10 is the penalty factor, μ i represents the dual variable of the i-th region, represents the average value of the dual variable, δ = 0.5 is the sparsity regularization parameter, and the power balance between regions is achieved by updating the dual variable.

[0035] Preferably, in the closed-loop correction mechanism, the momentum update rule is where θ new represents the updated model parameters, θ old Represents the model parameters before updating, α=0.01 represents the learning rate, which controls the step size of each parameter update. Denotes the loss function L with respect to θ old The gradient of , β = 0.9 represents the momentum factor, which is used to accelerate the convergence speed of parameter updates and represents the historical optimal parameters.

[0036] Beneficial effects

[0037] The present invention provides a method for optimizing and dispatching a regional integrated energy system. Compared with the prior art, it has the following advantages:

[0038] 1. The regional integrated energy system optimization scheduling method combines VMD with Bayesian networks to significantly improve the accuracy of new energy output and load forecasting, reduce scheduling deviations, and combine multi-objective functions with DRL algorithms to reduce system operating costs and carbon trading costs.

[0039] 2. The optimized scheduling method for the integrated energy system in this region introduces a dynamic relaxation factor μ(t) to adjust the equipment ramping constraints in real time, and integrates the transferable load with a virtual energy storage equivalent model. This allows the system to control the equipment power adjustment fluctuation within 10% of the rated power when the load suddenly increases, thus avoiding frequent energy storage exceeding the limit. At the same time, the tiered pricing mechanism of the ladder-type carbon trading model encourages the system to give priority to the consumption of new energy, and carbon emissions are significantly reduced compared to the traditional fixed carbon price model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0042] See for example Figure 1 , the present invention provides the following two technical solutions:

[0043] The first embodiment: a method for optimizing and dispatching a regional integrated energy system, comprising the following steps:

[0044] Step 1: Multi-source data layered collection and preprocessing:

[0045] The edge computing terminal collects real-time data on distributed power sources (such as photovoltaics and wind power), energy storage devices (such as batteries and thermal storage tanks), multiple loads (electricity, heat, and gas), and market prices. Construct a three-dimensional data set D = {d t ,i,p}, where t is the time point, i is the device number, and p is the power type (active, reactive, thermal power). The adaptive Kalman filter algorithm is used to denoise the collected time series data and dynamically update the measurement noise covariance matrix R t , the formula is:

[0046]

[0047] Among them, R t represents the measurement noise covariance matrix at time t, γ is the forgetting factor (0.8≤γ≤0.95), M is the amount of sliding window data, which is used to filter out abnormal data with a signal-to-noise ratio of less than 3dB, z k represents the kth measurement value, represents the estimated value of the kth measurement, It represents the average of the square of the error between the measured value and the estimated value within a sliding window of size M, reflecting the magnitude of the current measurement error. At the same time, variational mode decomposition (VMD) is performed on the preprocessed renewable energy output data to solve the constrained optimization problem to obtain K modal components. The objective function is:

[0048]

[0049] Among them, u k is the kth modal component, ω k is the center frequency, K (2≤K≤5) indicates that the optimal decomposition parameters are determined by non-recursive iteration. represents the derivative with respect to time t, j is the imaginary unit, Indicates constraints.

[0050] Step 2: Multi-timescale uncertainty scenario modeling:

[0051] Ultra-short term (15 minutes): For the high-frequency components after VMD decomposition, Monte Carlo simulation is used to generate 1000+ random scenarios, and 20 typical scenarios are screened by Kullback-Leibler divergence, retaining the probability weight w s .

[0052] Short term (1h): Build a Bayesian network model, input weather variables (wind speed, radiation, temperature), and output load forecast distribution P(L t |W t ,T t ), dynamic prediction is achieved by updating the conditional probability matrix, and the prediction accuracy is improved to more than 92%.

[0053] Step 3: Build a multi-objective optimization model:

[0054] Establishment includes economic cost C op , Tiered carbon trading costs C 碳 , Conditional Value at Risk Cost C risk The multi-objective function F = αC op +βC 碳 +γC risk , where the weight α+β+γ=1 is determined by the analytic hierarchy process. The economic cost Cop includes the cost of purchasing energy and equipment operation and maintenance; the tiered carbon trading cost carbon is calculated based on the difference between carbon emissions and quotas, with a carbon price gradient of ≥3 levels, and the formula is:

[0055]

[0056] Among them, C 碳 represents the total carbon trading cost, t represents the time step, C t is the basic carbon price, a is the gradient coefficient, d is the length of the quota interval, E emit.t represents the actual carbon emissions at time t, E 配额,t represents the carbon emission quota allocated to the energy system at time t.

[0057] Step 4: Intelligent algorithm solution and dynamic scheduling: Use the improved deep reinforcement learning (DRL) algorithm to solve the multi-objective optimization model. Define the state space St to include the energy storage charge state, load deviation, and electricity price fluctuation; action space A t This includes energy storage charging and discharging power, interconnection line interaction power, and equipment start and stop status. By optimizing the strategy using the experience replay buffer (capacity ≥ 10^5) and the target network (update cycle 500 steps), the reward function is designed as follows:

[0058] rt =-C t +λ1ΔSOC+λ2ΔP grid ;

[0059] Among them, r t represents the reward value at time t, C t represents the real-time operating cost, ΔSOC represents the balance of energy storage charge state, and represents the suppression of grid interaction power fluctuation. Indicates the balance of energy storage charge state, where SOC t Indicates the state of charge of the energy storage device at time t, SOC nom is the nominal state of charge of the energy storage device, SOC max is the maximum state of charge of the energy storage device, represents the amount of suppression of grid interaction power fluctuations, λ1 and λ2 are weight coefficients, which are determined by Pareto frontier optimization, where P tie,t represents the interaction power with the grid at time t, P tie,max Indicates the maximum interactive power with the grid.

[0060] Step 5: Constraint processing and resource integration:

[0061] A dynamic relaxation factor μ(t) is introduced to adjust the upper and lower limits of power according to the equipment climbing constraint. The formula is:

[0062]

[0063] Among them, D t is the real-time load deviation rate, k = 10 is the adjustment coefficient, and D0 = 0.1 is the load deviation threshold, ensuring that the equipment power adjustment fluctuation is ≤ 10% of the rated power. At the same time, the transferable load is equivalent to virtual energy storage, and the state equation is established:

[0064]

[0065] in, is the load charge / discharge power, η DR =0.9 is the response efficiency, S DR,max The maximum capacity of virtual energy storage is set at 15% of the load peak, which enables coordinated scheduling of demand response load and physical energy storage.

[0066] Step 6: Closed-loop correction and strategy optimization:

[0067] Collect scheduling execution data in real time and calculate prediction errors:

[0068]

[0069] If ∈ t >5%, dynamically adjust model parameters through momentum update rule:

[0070]

[0071] where θ new represents the updated model parameters, θ old Represents the model parameters before updating, α=0.01 represents the learning rate, which controls the step size of each parameter update. Denotes the loss function L with respect to θ old The gradient of , β = 0.9 represents the momentum factor, which is used to accelerate the convergence speed of parameter updates and represents the historical optimal parameters.

[0072] Second implementation method: Example, taking a 50MW photovoltaic + 20MWh energy storage + regional load system as an example, the implementation process of the present invention is described in detail:

[0073] 1. Multi-source data layered collection and preprocessing

[0074] PV output, energy storage SOC, load power, and electricity price data were collected every 30 seconds using edge computing terminals to construct a dataset D containing over 1,000 nodes. An adaptive Kalman filter algorithm was used to denoise the collected data, setting the forgetting factor γ to 0.9 and the sliding window size M to 10 to filter out abnormal data. VMD decomposition was performed on the PV output data, and the optimal decomposition parameter K = 3 was determined, decomposing the PV output into three modal components: high-frequency fluctuations, medium-frequency fluctuations, and low-frequency trends.

[0075] 2. Multi-timescale uncertainty scenario modeling

[0076] Ultra-short term (15 minutes): For the high-frequency components after VMD decomposition, Monte Carlo simulation is used to generate 1500 random scenarios, and 20 typical scenarios are screened out through Kullback-Leibler divergence, retaining the probability weight w s , ensuring that the scene error is ≤5%.

[0077] Short term (1h): Build a Bayesian network model, input weather variables such as wind speed, radiation, and temperature, and output load forecast distribution P(L t |W t ,T t By training the model with historical data and updating the conditional probability matrix, dynamic forecasting is achieved, with a load forecast accuracy of 93%.

[0078] 3. Build a multi-objective optimization model

[0079] Establish multi-objective function F = αC op +βC 碳 +γC risk, the weights α=0.6, β=0.2, γ=0.2 are determined by the analytic hierarchy process. Calculate the economic cost C op Including photovoltaic power purchase costs, energy storage charging and discharging costs and equipment operation and maintenance costs; calculate the ladder carbon trading cost C 碳 , set the basic carbon price ct = 20 yuan / ton, the gradient coefficient a = 0.5, and the quota interval length d = 10 tons; calculate the conditional risk value cost Crisk, with a confidence level of 95%.

[0080] 4. Intelligent algorithm solution and dynamic scheduling

[0081] The improved deep reinforcement learning (DRL) algorithm is used to solve the multi-objective optimization model. Define the state space S t Including energy storage charge state, load deviation, electricity price fluctuation; action space A t Including energy storage charging and discharging power, interconnection line interaction power, equipment start and stop status. Set the experience playback buffer capacity to 10 6 The target network update cycle is 500 steps. The optimal scheduling strategy is obtained by training the DRL algorithm. The weight coefficients in the reward function are λ1 = 0.3 and λ2 = 0.2.

[0082] 5. Constraint processing and resource integration

[0083] In view of the equipment climbing constraint, the dynamic relaxation factor μ(t) is introduced to adjust the power upper and lower limits. Real-time calculation of the load deviation rate D t , according to the formula Adjust the upper and lower limits of power to ensure that the power adjustment fluctuation of the equipment is ≤10% of the rated power. Equivalently treat the transferable load as virtual energy storage and set the maximum capacity of the virtual energy storage S DR,max The response efficiency is 15% of the peak load. DR =0.9, realizing the coordinated dispatch of load and energy storage.

[0084] 6. Closed-loop correction and strategy optimization

[0085] Collect scheduling execution data in real time and calculate prediction error ∈ t If ∈ t >5%, via momentum update rule Dynamically adjust model parameters. Generate the next cycle's scheduling strategy based on the revised model, forming a closed-loop feedback loop. In actual operation, the system's response time to sudden operating conditions such as PV dips and load surges has been shortened to 2 minutes, and power fluctuations have been controlled within 10% of rated power.

[0086] In summary, through the above six core steps, the present invention achieves efficient scheduling of regional integrated energy systems, significantly improves the economy, reliability and robustness of the system, and is suitable for complex scenarios of multi-energy coupling.

[0087] At the same time, the contents not described in detail in this specification belong to the existing technology known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited, and conventional equipment can be used.

[0088] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0089] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing and dispatching a regional integrated energy system, characterized in that: The following steps are involved: Step 1: Multi-source data layered collection and preprocessing: Real-time collection of distributed power supply, energy storage equipment, multiple loads, and market price data through edge computing terminals; Step 2: Multi-time scale uncertainty scenario modeling: Generate initial scenarios based on Monte Carlo simulation for ultra-short-term data, screen typical scenarios through Kullback-Leibler divergence and retain probability weights, build a Bayesian network model for short-term data, input weather variables and output load forecast distribution P(L t |W t ,T t ); Step 3: Construct a multi-objective optimization model: Establish a multi-objective optimization model including economic cost C op , Tiered carbon trading costs C 碳 , Conditional Value at Risk Cost C risk The multi-objective function F = αC op +βC 碳 +γC risk , where the weight α+β+γ=1 is determined by the analytic hierarchy process; Step 4: Intelligent algorithm solution and dynamic scheduling: Using an improved deep reinforcement learning algorithm, define the state space S that includes energy storage charge state, load deviation, and electricity price fluctuations t and the action space A including energy storage charging and discharging power, interconnection line interaction power, and equipment start and stop status t ; Step 5: Constraint processing and resource integration: Introduce a dynamic relaxation factor μ(t) to adjust the power upper and lower limits for the equipment ramp constraint, equate the transferable load to virtual energy storage, and establish a state equation. Realize coordinated dispatch of load and energy storage; Step 6. Closed-loop correction and strategy optimization: Real-time collection of scheduling execution data to calculate the prediction error ∈ t , if ∈ t >5%, dynamically adjust model parameters through momentum update rules to form a "collection-optimization-execution-feedback" closed loop.

2. A regional integrated energy system optimization scheduling method according to claim 1, characterized in that: The preprocessing in step 1 is to construct a three-dimensional data set D={d t ,i,p}, the adaptive Kalman filter algorithm is used to denoise the time series data and dynamically update the measurement noise covariance matrix R t And extract the high-frequency fluctuation and low-frequency trend components of renewable energy output through variational mode decomposition; The noise covariance update formula of the adaptive Kalman filter is: Among them, R t represents the measurement noise covariance matrix at time t, γ is the forgetting factor (0.8≤γ≤0.95), M is the amount of sliding window data, which is used to filter out abnormal data with a signal-to-noise ratio of less than 3dB, z k represents the kth measurement value, represents the estimated value of the kth measurement, It represents the average value of the square of the error between the measured value and the estimated value in the sliding window of size M, reflecting the magnitude of the current measurement error; The variational mode decomposition obtains the modal components by solving the following constrained optimization problem: Among them, u k is the kth modal component, ω k is the center frequency, K (2≤K≤5) indicates that the optimal decomposition parameters are determined by non-recursive iteration. represents the derivative with respect to time t, j is the imaginary unit, Indicates constraints.

3. A regional integrated energy system optimization scheduling method according to claim 1, characterized in that: In the ultra-short-term scenario modeling, typical scenarios are screened by Kullback-Leibler divergence, and the probability weight w is retained. s satisfy And the scene error is ≤5%; the Bayesian network model is based on the conditional probability matrix Correlate weather variables with load forecasts.

4. A method for optimizing and dispatching a regional integrated energy system according to claim 1, characterized in that: In step 4, the reward function is designed as follows through the experience replay buffer and the target network optimization strategy: r t =-C t +λ1ΔSOC+λ2ΔP grid ; Among them, r t represents the reward value at time t, C t represents the real-time operating cost, ΔSOC represents the balance of energy storage charge state, and represents the amount of grid interaction power fluctuation suppression.

5. The method for optimizing and dispatching a regional integrated energy system according to claim 1, characterized in that: The tiered carbon trading cost is calculated based on the difference between carbon emissions and quotas, with a carbon price gradient of ≥ 3 levels. The formula is: Among them, C 碳 represents the total carbon trading cost, t represents the time step, C t is the basic carbon price, a is the gradient coefficient, d is the length of the quota interval, E emit.t represents the actual carbon emissions at time t, E 配额,t represents the carbon emission quota allocated to the energy system at time t.

6. A method for optimizing and dispatching a regional integrated energy system according to claim 1, characterized in that: In the reward function of the deep reinforcement learning, Indicates the balance of energy storage charge state, where SOC t Indicates the state of charge of the energy storage device at time t, SOC nom is the nominal state of charge of the energy storage device, SOC max is the maximum state of charge of the energy storage device, represents the amount of suppression of grid interaction power fluctuations, and the weights λ1 and λ2 are determined by Pareto frontier optimization, where P tie,t represents the interaction power with the grid at time t, P tie,max Indicates the maximum interactive power with the grid.

7. A method for optimizing and dispatching a regional integrated energy system according to claim 1, characterized in that: The dynamic relaxation factor μ(t) is calculated as follows: Among them, D t is the real-time load deviation rate, k=10 is the adjustment coefficient, and D0=0.1 is the load deviation threshold, ensuring that the equipment power adjustment fluctuation is ≤10% of the rated power.

8. A method for optimizing and dispatching a regional integrated energy system according to claim 1, characterized in that: In the virtual energy storage equivalent model, S DR,max is the maximum capacity of virtual energy storage, which is set to 15% of the load peak, and the response efficiency η DR =0.9, the state update satisfies 0≤S DR,t ≤S DR,max , realizing the coordinated scheduling of demand response load and physical energy storage.

9. The method for optimizing and dispatching a regional integrated energy system according to claim 1, characterized in that: In the two-layer iterative solution, the upper layer uses the alternating direction multiplier method to decouple multi-region coordination, and the objective function is: Where N is the number of regions, F i represents the objective function value of the i-th region, ρ = 10 is the penalty factor, μ i represents the dual variable of the i-th region, represents the average value of the dual variable, δ = 0.5 is the sparsity regularization parameter, and the power balance between regions is achieved by updating the dual variable.

10. A method for optimizing and dispatching a regional integrated energy system according to claim 1, characterized in that: In the closed-loop correction mechanism, the momentum update rule is: where θ new represents the updated model parameters, θ old Represents the model parameters before updating, α=0.01 represents the learning rate, which controls the step size of each parameter update. Denotes the loss function L with respect to θ old The gradient of , β = 0.9 represents the momentum factor, which is used to accelerate the convergence speed of parameter updates and represents the historical optimal parameters.

Images