Regional integrated energy system adjustable capacity optimization method and system

Abstract

The invention provides a regional integrated energy system adjustable capacity optimization method and system, and belongs to the technical field of regional integrated energy system optimization. The method comprises the following steps: determining upper and lower limits of adjustable capacity of equipment to determine total upward and downward adjustment capacity of the system; according to the upward and downward adjustment total capacity of the system, the total transmission loss and the available transmission capacity of the line, solving the upward and downward adjustable capacity of the region; upward and downward capacity prediction errors, wind and light output errors and load fluctuation errors are determined, and probability feature vectors of up-regulation and down-regulation directions are obtained through Bayesian reasoning modeling; calculating the attention weight of the historical window through a time sequence attention mechanism; and determining a dynamic safety margin to dynamically correct the upward and downward adjustable capacities of the current region to obtain the final upward and downward adjustable capacities of the region. According to the method, the precision of the regional integrated energy system capacity and the system operation reliability are improved, and the regional adjustable capacity is more suitable for the actual network state.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of regional integrated energy system optimization, and more particularly, to a regional integrated energy system adjustable capacity optimization method and system. BACKGROUND

[0002] The current evaluation of the adjustable capacity of urban integrated energy systems is a core link to ensure system flexibility and reliability. However, existing evaluation methods are mostly limited to the adjustment capacity of a single device, lacking a three-level coordinated quantification system covering the energy storage and flexible load at the device level, the multi-energy aggregation at the system level, and the cross-network transmission at the regional level. Secondly, they rely on static parameters or linear aggregation methods, making it difficult to respond to real-time changes such as wind and light output fluctuations and load mutations. Finally, the impact of transmission losses and safety margins of the power grid and heat grid is not fully considered in regional evaluation, leading to a disconnect between theoretical evaluation results and actual available capacity.

[0003] Prior art document 1 (CN119863134A) discloses a regional integrated energy system coordination planning method. Its shortcomings are that the method focuses on the early planning stage of the regional integrated energy system, emphasizing component selection and configuration, network layout design, and operation strategy development, and lacks a three-level coordinated quantification system covering real-time adjustment potential at the device level, multi-energy dynamic aggregation at the system level, and cross-network transmission constraints at the regional level. It uses a particle swarm optimization algorithm to solve device configuration capacity, which is essentially a static parameter optimization. It relies on a hybrid prediction algorithm combining grey prediction and multivariate linear regression, and model predictive control (MPC) rolling optimization, and does not model the multi-source error under dynamic scenarios such as wind and light output fluctuations and load mutations, making it difficult to respond to system state changes in real time. At the same time, its transmission loss and pipeline capacity constraints are only used as fixed constraint conditions in the planning stage, without dynamically correcting the loss and safety margin based on real-time power flow and device operating state, leading to a disconnect between theoretical calculation of adjustable capacity and actual network operating state, and failing to meet the requirements of real-time scheduling for adjustable capacity accuracy and reliability. SUMMARY

[0004] To solve the problems in the prior art, the present application provides a regional comprehensive energy system adjustable capacity optimization method and system.Through establishing three-level evaluation indexes of equipment adjustment potential, system multi-energy aggregation ability and regional transmission loss correction, the actual adjustable range of each equipment is accurately quantified, the complementarity and competitiveness between equipment are coordinated through multi-objective optimization, the capacity overestimation caused by simple superposition is avoided, the accuracy and guidance of the evaluation results are improved, the power grid loss and the heat network loss are calculated synchronously in the regional correction layer to avoid overestimation of the transmission link capacity, the overall adjustment capacity of the comprehensive energy system is truly reflected, the reality and precision of the regional adjustable capacity are improved, the traditional index is based on static capacity calculation, and the time and space variation law of the capacity in the dynamic scene such as wind and light fluctuation and load mutation cannot be reflected, the long short-term memory network model is used to capture the dynamic evolution characteristics of the capacity adjustment, the Bayesian robust optimization modeling and the time sequence attention fusion mechanism are combined, the rolling optimization and online learning closed loop are introduced, the model parameters are updated in real time to adapt to the dynamic changes of the wind and light output and the load, the change from static estimation to dynamic tracking of the adjustable capacity prediction is realized, the capacity evaluation precision in the complex fluctuation scene and the system stability in the extreme scene are improved, the network transmission limitation such as but not limited to line capacity and pipeline loss is often ignored in the traditional evaluation result, so that the theoretical adjustable capacity cannot be realized in the actual physical network, the available transmission capacity of the line and the multi-energy flow loss correction are introduced, the buffer space for responding to sudden faults is reserved through the safety margin coefficient, the evaluation result is ensured to strictly meet the engineering realizability, the network congestion caused by the dispatching instruction is avoided, the pre-evaluation and the three-time compensation in the middle are realized through wind and light fluctuation compensation, load error correction and safety margin adjustment, and the reliability of the capacity evaluation is dynamically improved.

[0005] The present application adopts the following technical solutions.

[0006] The first aspect of the present application provides a regional comprehensive energy system adjustable capacity optimization method, comprising: The device real-time data is input into the deep learning model, and the upper and lower limits of the device adjustable capacity are determined according to the output result; A multi-objective optimization model is constructed with the minimum adjustment error cost, the minimum operation cost and the minimum response delay cost as the target, the optimal adjustment power of the equipment is solved, the optimal adjustment power of all the equipment is corrected to the upper and lower limits of the device adjustable capacity, and the corrected upper and lower limits of the device adjustable capacity are aggregated to determine the total upward and downward adjustment capacity of the system; The net available upward and downward capacities are obtained by subtracting the total transmission loss of the line from the total upward and downward adjustment capacities of the system, and the net available upward and downward capacities are corrected by the available transmission capacity of the line to determine the current regional upward and downward adjustable capacities; Based on the current region upward and downward adjustable capacity, the probability feature vector of upward and downward adjustment direction is obtained by using Bayesian inference modeling; the attention weight of the historical window is determined according to the obtained feature vector through the time sequence attention mechanism; the dynamic safety margin is determined by using the attention weight to determine the final region upward and downward adjustment capacity.

[0007] Preferably, determining the upper and lower limits of the adjustable capacity of the device comprises: According to the convolutional neural network, the long short-term memory network and the attention mechanism, a deep learning model is constructed; The real-time data of the device is input into the deep learning model to predict the minimum power boundary and the maximum power boundary of the device under the preset confidence, and the upper and lower limits of the adjustable capacity of the device are solved according to the predicted minimum power boundary and the maximum power boundary, which are expressed as follows:

[0008]

[0009] In the formula, S i,u represents the upper limit of the adjustable capacity of the i-th device at time period t, S i,d represents the lower limit of the adjustable capacity of the i-th device at time period t, c i represents the preset confidence of the i-th device at time period t, represents the predicted minimum power boundary of the i-th device at time period t, represents the predicted maximum power boundary of the i-th device at time period t, represents the expected scheduling instruction reference power of the i-th device at time period t.

[0010] Preferably, the adjustment error cost is expressed as follows:

[0011] In the formula, represents the adjustment power of the i-th device at the k-th time period in the scheduling period, c penalty represents the power error penalty rate, and K represents the total number of time periods in the scheduling period, represents the adjustment target vector of the system at the k-th time period in the scheduling period, represents the total number of devices.

[0012] Preferably, the adjustment target vector of the system at the k-th time period in the scheduling period is expressed as follows:

[0013]

[0014]

[0015] wherein, denotes the upward regulation demand of the system in the kth period of the dispatch cycle, denotes the downward regulation demand of the system in the kth period of the dispatch cycle, denotes the positive load fluctuation amount of the system in the kth period of the dispatch cycle, denotes the negative load fluctuation amount of the system in the kth period of the dispatch cycle, denotes the positive wind-solar fluctuation amount of the system in the kth period of the dispatch cycle, denotes the negative wind-solar fluctuation amount of the system in the kth period of the dispatch cycle.

[0016] Preferably, determining the current regional upward and downward adjustable capacity comprises:

[0017]

[0018] wherein, denotes the upward adjustable capacity of the region at period t, denotes the downward adjustable capacity of the region at period t, sys,u denotes the total upward regulation capacity of the system at period t, sys,d denotes the total downward regulation capacity of the system at period t, denotes the total transmission loss, and denotes the safety margin coefficient, is the line l in the current available transmission capacity, denotes all lines.

[0019] Preferably, determining the final regional upward and downward regulation capacity comprises: inputting the normalized historical data into a long short-term memory network model to predict the upward and downward adjustable capacity prediction values of the region, determining the upward and downward capacity prediction errors through the upward and downward adjustable capacity prediction values of the region and the current regional upward and downward adjustable capacity, determining the wind-solar output error and the load fluctuation error; based on Bayesian maximum a posteriori estimation, respectively modeling the probability distribution of the upward and downward capacity prediction errors, the wind-solar output error and the load fluctuation error, extracting the distribution parameters thereof and the load mutation rate to construct the probability feature vectors in the upward and downward directions; calculating a similarity score function of the probability feature vectors in the upward and downward directions, and calculating the attention weight of the historical window according to the similarity score function; based on the upward and downward adjustable capacity prediction values of the region, solving the dynamic compensation amount of the upward and downward adjustable capacity through Bayesian robust optimization; The historical dynamic safety margin is determined according to the attention weight of the historical window, and the final upward and downward adjustment capacities are determined according to the dynamic compensation amount of the upward and downward adjustable capacities and the historical dynamic safety margin.

[0020] Preferably, constructing the probability feature vector of the upward and downward adjustment directions comprises: Assuming that the upward and downward capacity prediction error is subject to a t-distribution, the mean and standard deviation of the upward and downward capacity prediction error are updated by Bayesian maximum a posteriori estimation; Assuming that the wind and light output error is subject to a truncated normal distribution, the mean and standard deviation of the wind and light output error are updated by Bayesian maximum a posteriori estimation; Assuming that the load fluctuation error is subject to a Poisson distribution, the mean, standard deviation and load mutation rate per unit time of the load fluctuation error are updated by Bayesian maximum a posteriori estimation; According to the mean and standard deviation of the upward and downward capacity prediction error, the mean and standard deviation of the wind and light output error, the mean and standard deviation of the load fluctuation error, and the load mutation rate per unit time of the load fluctuation error, the probability feature vector of the upward and downward adjustment directions is constructed.

[0021] Preferably, updating the mean and standard deviation of the upward and downward capacity prediction error comprises: The mean of the wind and light output error is updated, and is expressed by the following formula:

[0022] In the formula, represents the updated mean of the wind and light error, represents the prior mean of the wind and light error, represents the confidence coefficient of the wind and light error, represents the sample mean of the wind and light error, represents the sample number; The standard deviation of the wind and light output error is updated, and is expressed by the following formula:

[0023] In the formula, represents the updated standard deviation of the wind and light error, represents the prior standard deviation of the wind and light error, represents the sample variance of the wind and light error.

[0024] Preferably, the attention weight of the historical window is calculated according to a similarity score function, and is expressed by the following formula:

[0025]

[0026] In the formula, and representing history window attention weight of up-regulation and down-regulation direction, , representing history window number, representing sum of similarity scores of each history window , and respectively represent similarity scores of up-regulation and down-regulation direction probability feature vectors and history window features, and are expressed as follows:

[0027]

[0028] In the formula, representing weight of the th feature, , respectively represent the th component of up-regulation and down-regulation direction probability feature vectors of history window , , respectively represent the th component of up-regulation and down-regulation direction probability feature vectors of current time t, representing a parameter for preventing division by zero, representing feature dimension.

[0029] The second aspect of the present application provides a regional integrated energy system adjustable capacity optimization system, which runs the regional integrated energy system adjustable capacity optimization method of the first aspect, and comprises: An equipment upper and lower limit solving module is configured to input equipment real-time data into a deep learning model, and determine equipment adjustable capacity upper and lower limits according to an output result; A system upper and lower limit solving module is configured to construct a multi-objective optimization model with the minimum adjustment error cost, the minimum operation cost and the minimum response delay cost as targets, and solve the optimal adjustment power of the equipment, correct the equipment adjustable capacity upper and lower limits by the optimal adjustment power of all equipment, and aggregate the corrected equipment adjustable capacity upper and lower limits to determine system upward and downward adjustment total capacity; A regional upper and lower limit solving module is configured to subtract line total transmission loss from the system upward and downward adjustment total capacity to obtain net available upward and downward capacity, correct the net available upward and downward capacity by line available transmission capacity to determine current regional upward and downward adjustable capacity; The output module is used to obtain the probability feature vectors of the upward and downward adjustment directions based on the current region's upward and downward adjustable capacity using Bayesian inference modeling; determine the attention weights of the historical window through a temporal attention mechanism based on the obtained feature vectors; and use the attention weights to determine the dynamic safety margin to determine the final upward and downward adjustment capacity of the region.

[0030] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention constructs a deep learning model using convolutional neural networks, long short-term memory networks, and attention mechanisms to predict the upper and lower limits of adjustable capacity at the device level in real time, and introduces a confidence mechanism to suppress uncertainty propagation. With minimizing adjustment error costs, operating costs, and response delay costs as multiple objectives, it employs the augmented Lagrangian method to achieve coordinated aggregation of system-level capacity. Simultaneously, this invention accurately calculates transmission losses in the power grid and heating network, dynamically assesses available transmission capacity of lines, and improves the accuracy of regional adjustable capacity. This invention constructs a three-level evaluation system of "equipment-system-region" and adopts technical means that combine deep learning with multi-objective optimization collaboration, dynamic compensation and adaptive safety margin to improve evaluation accuracy, system flexibility and decision reliability, and reduce prediction errors, operational risks and manual intervention costs. This invention overcomes the limitations of single-dimensional analysis through a three-level progressive evaluation of equipment-level adjustment potential mining, system-level multi-energy collaborative aggregation and regional-level transmission efficiency correction. By compensating for network transmission losses in regional transmission efficiency, it truly reflects the available capacity of cross-regional energy supply and improves the utilization efficiency and operational reliability of regional integrated energy system capacity. This invention achieves basic prediction of regional adjustable capacity through a long short-term memory network model, and combines Bayesian inference to probabilistically model capacity prediction errors, wind and solar power output errors, and load fluctuation errors. A temporal attention mechanism is introduced to mine the correlation between historical error features and the current scenario, generating a temporally enhanced error feature vector. Based on this, Bayesian robust optimization is used to solve for dynamic compensation amounts and robust compensation intervals, coupled with risk-related dynamic safety margins, forming a multi-level correction mechanism. Simultaneously, relying on a "feature-decision-result" triplet experience pool, model parameters are updated in real time based on the final prediction error, improving the dynamic adaptability of capacity adjustment under wind and solar power output fluctuations and load abrupt changes, effectively reducing assessment errors in complex scenarios. Through online learning and real-time parameter iteration, the system's rapid response capability to dynamic changes in energy supply and demand is strengthened, further enhancing the robustness of capacity assessment. Furthermore, robust interval boundary verification is introduced in the final capacity calculation, combined with previously accurately calculated transmission losses and line available capacity constraints, ensuring that the regional adjustable capacity closely matches the actual network operating state, avoiding a disconnect between theoretical assessment values ​​and engineering feasibility. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the adjustable capacity optimization process of a regional integrated energy system provided in accordance with an embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0033] like Figure 1 As shown, Embodiment 1 of the present invention provides a method for optimizing the adjustable capacity of a regional integrated energy system, comprising the following steps: Step 1: Input real-time device data into the constructed deep learning model and output the upper and lower limits of the device's adjustable capacity.

[0034] In a preferred but non-limiting embodiment of the present invention, step 1 includes: Step 1.1: Construct a deep learning model based on convolutional neural networks, long short-term memory networks, and attention mechanisms.

[0035] More preferably, step 1.1 includes: Step 1.1.1: Input the device's historical data into the convolutional neural network of the deep learning model to extract spatial features and generate a feature map.

[0036] More preferably, step 1.1.1 includes: Acquire historical data of the devices, including the data of the i-th device at time step. The energy storage state of charge, the i-th device at time step Ambient temperature, the i-th device at time step The voltage of the i-th device at time step The scheduling instruction references power, device location parameters, and the time step of the i-th device. Meteorological data; Historical device data is input and spatial correlation features of devices are captured using a 3×3 convolution kernel to generate feature map H. spatial Dimension R N×L Where N is the number of devices and L is the spatial feature dimension.

[0037] Step 1.1.2, transfer the feature map H from step 1.1.1. spatial The input Long Short-Term Memory (LSTM) network captures the periodicity of time series data through bidirectional LSM network units, and outputs the temporal features H. temporal Dimension R N×T×MWhere T is the total number of time steps and M is the dimension of the hidden layer.

[0038] Step 1.1.3, convert the output time-series feature H from step 1.1.2 into... temporal Time step weights are calculated using an attention mechanism. Generate weighted feature H final .

[0039] More preferably, step 1.1.3 includes: By combining the attention mechanism with the temporal features H of the output from step 1.2 temporal Calculate time step weights It can be expressed by the following formula: (1) In the formula, For time steps The attention weights, w, are used to reflect the importance of the features in that step. a W is the attention weight vector. h The feature transformation weight matrix, For time steps The temporal characteristics, Indicates from time step Time to step Summation.

[0040] The weighted feature H is solved by using temporal features and time step weights. final It can be expressed by the following formula: (2) Step 1.1.4: Input the weighted features from Step 1.1.3 into the fully connected layer to obtain the predicted values ​​of the minimum and maximum power boundaries and confidence scores of each device. Calculate the multi-task loss function by comparing the predicted values ​​of each device with the true labels. Update the parameters of the multi-task loss function using gradient descent. Repeat Steps 1.1.1-1.1.3 until the multi-task loss function converges to obtain the trained deep learning model.

[0041] More preferably, step 1.1.4 includes: Based on the weighted features obtained in step 1.1.3, the predicted minimum and maximum power boundaries and prediction confidence of the fully connected layer output device are expressed by the following formula: (3) In the formula, Let be the predicted minimum power boundary for the i-th device in time period t. Let c be the predicted maximum power boundary for the i-th device in time period t. i (t) represents the prediction confidence of the i-th device in time period t. W represents the activation function.min,i Calculate the weight matrix W for the minimum power boundary of the i-th device. max,i Calculate the weight matrix W for the maximum power boundary of the i-th device. c,i Calculate the confidence weight matrix for the i-th device, b min,i b max,i b c,i For the bias term of the i-th device, This is the activation function.

[0042] Boundary loss is calculated by comparing the predicted minimum and maximum power boundaries of the device with the actual labels. This is combined with confidence loss and Huber loss (comparing predicted and actual typical operating power values) as auxiliary losses to construct a multi-task loss function, expressed as follows: (4) In the formula, L is the loss function, λ1, λ2, and λ3 are weight coefficients, and λ1 + λ2 + λ3 = 1. The boundary loss value is expressed by the following formula: (5) In the formula, Indicates the total number of devices. This represents the minimum power that the i-th device can actually achieve during time period t, and is the true label of this invention. This represents the maximum power that the i-th device can actually achieve during time period t, and is the true label of this invention.

[0043] The confidence loss value is expressed by the following formula: (6) The auxiliary loss value is expressed by the following formula: (7) In the formula, The typical operating power baseline value of the i-th device during time period t is expressed by the following formula: (8) The weight matrix represents the typical operating power baseline value of the i-th device. The bias term represents the typical operating power baseline value of the i-th device. HuberLoss is a robust loss function used to suppress the influence of outliers, expressed by the following formula: (9) In the formula, This represents the reference power for the expected scheduling command of the i-th device in time period t. deltaIt is a hyperparameter, usually set to 1.0. When the error is small, it is similar to MSE, and when the error is large, it switches to MAE to suppress outlier interference.

[0044] The multi-task loss function L is minimized using the gradient descent algorithm, and the model parameters (including the weights and biases of CNN, Long Short-Term Memory network, attention mechanism and fully connected layer) are iteratively updated until the multi-task loss function converges, resulting in a trained deep learning model.

[0045] With the actual power output and power absorption of the equipment as the target, historical operation and adjustment data of the equipment unit are collected. State parameters, such as but not limited to equipment operating status and environmental parameters, are used as input variables. With the actual power output as the target parameter, a deep learning model is constructed. By training the training network, the adjustment capability of a single device is analyzed and evaluated, so as to achieve a precise characterization of the device's adjustment capability and clarify the adjustment potential of a single device under different operating conditions.

[0046] Step 1.2: Input real-time device data into the trained deep learning model to obtain the device's predicted minimum power boundary, maximum power boundary, and confidence level. Based on the device's predicted minimum power boundary, maximum power boundary, and confidence level, calculate the upper and lower limits of the device's adjustable capacity, expressed by the following formula: (10) (11) In the formula, S i,u (t) represents the upward adjustment capacity of the i-th device in time period t, S i,d (t) represents the downward adjustment capacity of the i-th device during time period t.

[0047] It is worth noting that, in response to the technical problems of amplified equipment-level prediction errors during system aggregation, the upward propagation and accumulation of uncertainty along the scheduling hierarchy, and the final system-level decision based on unreliable underlying data, this invention constructs a deep learning model using convolutional neural networks, long short-term neural networks, and attention mechanisms. It introduces a confidence assessment mechanism, using confidence as a weighting factor in the calculation of equipment regulation capacity. Low-confidence predictions automatically have their weight reduced in aggregation, effectively suppressing the uncertainty of equipment-level capacity propagation to the system level, improving the accuracy of regional regulation capacity, reducing scheduling over-limit risks and equipment overload risks, and enhancing the safety and stability of the regional integrated energy system.

[0048] Step 2: Construct a multi-objective optimization model with the objectives of minimizing adjustment error cost, minimizing operating cost, and minimizing response delay cost. Solve the multi-objective optimization model to obtain the optimal adjustment power of the equipment. Based on the optimal adjustment power of the equipment and the upper and lower limits of the adjustable capacity of the equipment, solve for the total upward and downward adjustment capacity of the system.

[0049] In a preferred but non-limiting embodiment of the present invention, step 2 includes: Step 2.1: Construct a multi-objective optimization model with the goals of minimizing regulation error, minimizing operating cost, and minimizing response delay, so as to achieve the synergistic aggregation of system-level regulation capabilities.

[0050] More preferably, step 2.1 includes: Step 2.1.1: Construct a multi-objective optimization model with the objectives of minimizing adjustment error, minimizing operating cost, and minimizing response delay. The objective function of the multi-objective optimization model is... It can be expressed by the following formula: (12) In the formula, ω1, ω2, and ω3 are weighting coefficients. Indicates minimization. F 1 represents the adjustment error cost, expressed by the following formula: (13) In the formula, Let be the adjusted power of the i-th device in the k-th time period within the scheduling cycle. c penalty The power error penalty rate is K, which represents the total number of time periods in the scheduling cycle. Let be the adjustment target vector of the system in the k-th time period within the scheduling cycle, expressed by the following formula:

[0051]

[0052] (14) In the formula, This represents the upward adjustment demand of the system in the k-th time period within the scheduling cycle. P target,d (k) This represents the downward adjustment demand of the system in the k-th time period within the scheduling cycle. This represents the positive load fluctuation of the system during the k-th time period within the scheduling cycle. This represents the negative load fluctuation of the system during the k-th time period within the scheduling cycle. This represents the positive fluctuation of wind and solar power in the k-th time period of the system's scheduling cycle (actual output > predicted output). The negative fluctuation of wind and solar power in the k-th time period of the system within the scheduling cycle (actual output < predicted output) is expressed by the following formula: (15) (16) (17) (18) In the formula, For the first m The actual power of a load in the k-th time period within the scheduling cycle P Lm,pre ( k ) is the first m The historical average power of a load during the k-th time period within the scheduling cycle M Total load This represents the positive fluctuation of total wind and solar power output during the k-th time period within the scheduling cycle. For the first j The predicted output of a photovoltaic unit in the k-th time period within the scheduling cycle. For the first The predicted output of a wind turbine in the k-th time period within the scheduling cycle. No. j The actual power output of a photovoltaic unit during the k-th time period within the dispatch cycle. For the first The actual output of a wind turbine in the k-th time period within the scheduling cycle, where W represents the wind turbine, V represents the photovoltaic power, R is the total number of wind turbines, and J is the total number of photovoltaic power.

[0053] F 2 represents operating costs, including equipment start-up, shutdown, and depreciation costs, expressed by the following formula: (19) In the formula, c i,op Let i be the unit power operating cost of the i-th device. c i,start Let i be the start-up and shutdown cost of the i-th device. u i (t) This indicates the device's start / stop status. {0, 1}, =1 indicates that a start / stop action occurs in the k-th time period. =0 indicates that no start / stop action occurred in the k-th time period.

[0054] F 3 represents the response latency cost, expressed by the following formula: (20) In the formula, c i,delay Let t be the penalty cost per unit power delay per unit time for the i-th device. i Let t be the actual response time of the i-th device. i,ref Δt represents the device's reference response time, and Δt represents the length of the scheduling period.

[0055] The constraints include equipment power regulation constraints, expressed by the following formula: (twenty one) Step 2.2: The augmented Lagrange method combined with the alternating direction multiplier method is used to solve the multi-objective optimization model of Step 2.1, obtaining the optimal adjustment power of each device in each time period within the scheduling cycle. Based on the upper and lower limits of the adjustable capacity of the devices, the total upward and downward adjustment capacity of the system is calculated, expressed by the following formula:

[0056] (twenty two) In the formula, C sys,u (t) represents the total upward adjustment capacity of the system during time period t, used to describe how much additional power output the entire system, including all equipment, can provide to cope with load increases, decreases in wind and solar power output, etc., under the current state. C sys,d (t) represents the total downward adjustment capacity of the system in time period t, used to describe how much power output the entire system can still reduce in the current state to cope with situations such as sudden load drops and excess wind and solar power output. P i,* (t k The ) represents the time period of the i-th device within the scheduling period corresponding to time period t, calculated in step 2.1. The optimal adjustment power.

[0057] It is worth noting that, in response to the technical problem that traditional methods often focus on a single objective and optimize each objective independently, resulting in mutual conflicts, this invention reduces capacity regulation error through multi-objective optimization.

[0058] Step 3: Calculate the total transmission loss based on the power grid transmission loss and heat energy transmission loss. Solve for the regional upward and downward adjustable capacity based on the total transmission loss, the available transmission capacity of the line, and the total upward and downward adjustment capacity of the system.

[0059] In a preferred but non-limiting embodiment of the present invention, step 3 includes: Step 3.1: Calculate the total transmission loss based on the power grid transmission loss and the heat energy transmission loss.

[0060] More preferably, step 3.1 includes: Calculate the power grid transmission loss L net It can be expressed by the following formula: (twenty three) In the formula, To connect node p and node The equivalent resistance of the transmission line, The current actual flow through the connection node p and node The effective value of the current in the line, For the node The transformer equivalent impedance, For the current actual node The effective value of the voltage, N p This represents the total number of nodes in the distribution network.

[0061] Heat transfer loss L heat (t), expressed by the following formula: (twenty four) In the formula, N P β is the collection of all transport pipelines in the regional heating network. p0 For pipelines p0 The heat loss coefficient, Q p0 For current actual pipelines p0 heat flow, This indicates the length of the time period corresponding to the calculated loss.

[0062] Total transmission loss L loss =L net +L heat .

[0063] It is worth noting that, in response to the technical problems of existing technologies that simply superimpose equipment adjustment capabilities, ignore energy transmission losses, and use fixed percentage estimates for loss estimation which cannot reflect the actual network status, resulting in scheduling instructions exceeding the actual transmission capacity and causing execution errors, this invention is based on the accurate calculation of power grid transmission losses based on line resistance and actual power flow, and the dynamic calculation of heat energy transmission losses taking into account pipeline heat loss coefficients. The total loss is dynamically updated as the network topology and load change, thereby improving the accuracy of capacity assessment.

[0064] Step 3.2, calculate the available transmission capacity of the line, expressed by the following formula: (25) In the formula, For the line l With the current available transmission capacity, Q l For the line l Maximum allowed transmission capacity For the line l For the first b Power transfer distribution factor of the power supply unit. For the line l For the first r The power transfer distribution factor of each wind turbine. For the line l For the firstm Power transfer distribution factor of each load node For the first b The current output power of the power generation unit in Taiwan For the first r The current actual output of each wind turbine For the first m The current electricity demand of each load node B , K , M These represent the total number of power supply units, wind turbines, and load nodes, respectively.

[0065] It is worth noting that existing technologies only consider static line capacity constraints, and the weakest link determines the overall capacity adjustment capability. This invention accurately calculates the impact of equipment output on line power flow, dynamically tracks the available transmission capacity of the line, reduces the risk of line overload and voltage overrun events, reduces N-1 safety violations, and improves the accuracy of capacity assessment.

[0066] Step 3.3: Based on the available transmission capacity of the line obtained in Step 3.2 and the total transmission loss obtained in Step 3.1, calculate the area-upward adjustable capacity and the area-downward adjustable capacity, expressed by the following formula: (26) (27) In the formula, This indicates the upward adjustable capacity of the region during time period t. This represents the downward adjustable capacity of the region over time period t, where η represents the safety margin coefficient. This indicates all lines.

[0067] It is worth noting that existing technologies have fixed adjustable capacity, which cannot adapt to changes in network structure or intelligently adjust regional capacity according to network status. This invention dynamically senses changes in network topology by using available transmission capacity and total transmission loss, and adaptively corrects adjustable capacity, reducing the assessment error of regional adjustable capacity. It increases adjustable capacity during light load periods and adaptively optimizes adjustable capacity during normal periods. Through transmission management, it prevents network congestion and avoids the waste of capacity in the regional integrated energy system.

[0068] Step 4: Based on the current adjustable capacity upward and downward in the region, determine the upward and downward capacity prediction errors, wind and solar power output errors, and load fluctuation errors; based on the obtained errors, use Bayesian inference modeling to obtain the probability feature vectors of the upward and downward adjustment directions; based on the obtained feature vectors, calculate the attention weights of the historical windows through a temporal attention mechanism; combine the feature vectors and attention weights to calculate the dynamic safety margin to dynamically correct the current adjustable capacity upward and downward in the region, and obtain the final upward and downward adjustment capacity of the region.

[0069] In a preferred but non-limiting embodiment of the present invention, step 4.1 includes: Step 4.1: Input the normalized historical data into the Long Short-Term Memory Network model to predict the upward and downward adjustable capacity forecast values ​​of the region. Determine the upward and downward capacity forecast errors based on the upward and downward adjustable capacity forecast values ​​of the region, and determine the wind and solar power output error and load fluctuation error.

[0070] More preferably, step 4.1 includes: The predicted value of the regional adjustable capacity for the current time period is generated using a long short-term memory network model, expressed by the following formula: (28) (29) (30) In the formula, Normalized historical data includes: historical regional adjustable capacity, wind and solar power output, load, temporal characteristics, and weather data. This represents the hidden state of the Long Short-Term Memory network model during time period t. This represents all the internal parameters of the Long Short-Term Memory network model. This represents the upwardly adjustable capacity prediction value for region t predicted by the Long Short-Term Memory network model. This represents the downwardly adjustable capacity prediction value for the region t predicted by the Long Short-Term Memory network model. and This represents the output layer weight matrix. and This represents the output layer bias term.

[0071] The upward and downward adjustable capacity forecasts for the region are calculated based on the upward and downward capacity forecasts, the wind farm output error, and the load fluctuation error, expressed by the following formula: (31) (32) (33) (34) In the formula, This indicates the upward capacity prediction error. This indicates the downward capacity prediction error. This represents the power output error of the wind farm, which is the actual total power output of the wind farm. With the predicted total output of the wind farm The absolute value of the difference This represents the load fluctuation error, which is the actual total load of the system. Compared with the reference load The load fluctuation error.

[0072] Step 4.2: Based on MAP (Maximum A Posteriori Probability), the probability distribution models of upward and downward capacity prediction errors, wind and solar power output errors, and load fluctuation errors are performed respectively, and their distribution parameters are extracted and combined with the load mutation rate to construct the probability feature vectors of the upward and downward adjustment directions.

[0073] Step 4.2.1: Assuming that the upward and downward capacity prediction errors follow a t-distribution, update the mean and standard deviation of the upward and downward capacity prediction errors using MAP.

[0074] The mean of the upward and downward capacity prediction errors is updated, expressed by the following formula: (35) (36) In the formula, This represents the mean of the updated upward capacity prediction error. This represents the mean of the updated downward capacity prediction error. and These represent the confidence coefficients for upward and downward capacity prediction errors, respectively. =0.7, This represents the prior mean of capacity error. This indicates the sample size, using error data from the past 60 time points. This represents the sample mean of the upward capacity prediction error. The sample mean representing the downward capacity prediction error is expressed by the following formula: (37) (38) In the formula, Indicates time Upward capacity error, Indicates time Downward capacity error.

[0075] The standard deviations of the upward and downward capacity forecast errors are updated using the following formula: (39) (40) In the formula, This represents the standard deviation of the updated upward capacity prediction error. This represents the standard deviation of the updated downward capacity prediction error. The prior standard deviation of capacity error is expressed by the following formula: (41) (42) In the formula, The sample variance representing the upward capacity error. This represents the sample variance of the downward capacity error.

[0076] Step 4.2.2: Assuming that the wind and solar power output error follows a truncated normal distribution, update the mean and standard deviation of the wind and solar power output error using MAP.

[0077] The mean value of the updated wind and solar power output error is expressed by the following formula: (43) In the formula, This represents the updated mean error of the landscape. This represents the prior mean of the wind and light error. The confidence coefficient represents the error in the wind and light perception. The sample mean of the wind and light error is expressed by the following formula: (44) The standard deviation of the updated wind power output error is expressed by the following formula: (45) In the formula, This represents the updated standard deviation of the landscape error. This represents the prior standard deviation of the wind and light error. The sample variance representing the error in wind and light conditions is expressed by the following formula: (46) Step 4.2.3: Assuming the load fluctuation error follows a Poisson distribution, update the mean, standard deviation, and load mutation rate per unit time of the load fluctuation error using MAP. The mean and standard deviation of the load fluctuation error are expressed by the following formulas: (47) (48) In the formula, The sample mean of load fluctuation error. This represents the sample standard deviation of load fluctuation error.

[0078] The load fluctuation error per unit time is expressed by the following formula: (49) In the formula, The load mutation rate is used to describe the average number of load mutations per hour. and Represents the prior parameter of Gamma. Indicates the duration of the window. This represents the number of load abrupt changes, used to describe the number of times the load power change exceeds the abrupt change threshold in the past n time points, expressed by the following formula: (50) In the formula, This is an indicator function; its value is 1 if the condition is true, and 0 otherwise. Indicates time The actual load value, Indicates time The baseline value, This represents the mutation threshold.

[0079] Step 4.2.4: Based on the mean and standard deviation of the upward and downward capacity prediction errors, the mean and standard deviation of the wind and solar power output errors, the mean and standard deviation of the load fluctuation errors, and the unit-time load mutation rate of the load fluctuation errors from steps 4.2.1-4.2.3, construct the probability feature vectors for the upward and downward adjustment directions, expressed by the following formula: (51) (52) In the formula, , The probabilistic feature vectors represent the directions of adjustment (up and down).

[0080] Step 4.3: Calculate the similarity scoring function of the probability feature vectors of the up and down directions, and calculate the attention weight of the historical window based on the similarity scoring function.

[0081] The similarity scoring function for calculating the probability feature vectors of upward and downward adjustment directions is expressed by the following formula: (53) (54) In the formula, and These represent the probabilistic feature vectors for upward and downward adjustments and the historical window, respectively. Feature similarity score, Indicates the first The weights of each feature, , These represent the history window. The probability eigenvectors of the upward and downward adjustment directions One portion, Indicates a historical window, The first and second eigenvectors represent the probabilistic eigenvectors of the upward and downward directions at the current time t, respectively. One portion, This indicates a parameter to prevent division by zero. Indicates the feature dimension.

[0082] The attention weights for the historical windows are calculated based on the similarity scoring function, expressed by the following formula: (55) (56) In the formula, and Represents the history window Attention weights for upward and downward adjustments , Indicates the number of historical windows.

[0083] Step 4.4: Based on the predicted upward and downward adjustable capacity values ​​for the region, solve for the dynamic compensation amount of the upward and downward adjustable capacity through Bayesian robust optimization.

[0084] Step 4.4.1: Construct the upward capacity optimization model. The objective function of the upward capacity optimization model is expressed by the following formula: (57) In the formula, This represents the dynamic compensation amount for the upward adjustable capacity. express The expected value.

[0085] The constraints of the upward capacity optimization model include equipment regulation capability constraints, power balance constraints, and robustness constraints.

[0086] The equipment adjustability constraint is expressed by the following formula: (58) In the formula, This represents the output power of the i-th device at the current time t. and Let represent the minimum and maximum output power of the i-th device, respectively. This represents the power adjustment amount when the i-th device is adjusted upwards, and N represents the number of devices.

[0087] The power balance constraint is expressed by the following formula: (59) Robustness constraints are expressed by the following formula: (60) In the formula, This represents a probability operator used to describe the probability of an event occurring. This represents the robust probability.

[0088] Step 4.4.2: Construct the downward capacity optimization model. The objective function of the downward capacity optimization model is expressed by the following formula: (61) In the formula, This represents the dynamic compensation amount for downward adjustable capacity.

[0089] The constraints of the downward capacity optimization model include equipment regulation capability constraints, power balance constraints, and robustness constraints.

[0090] The equipment adjustability constraint is expressed by the following formula: (62) In the formula, This represents the amount of power adjustment when the i-th device is adjusted downwards.

[0091] The power balance constraint is expressed by the following formula: (63) Robustness constraints are expressed by the following formula: (64) Step 4.4.3: Both the upward capacity optimization model and the downward model are solved using Bayesian optimization to obtain the dynamic compensation amount of the optimal upward adjustable capacity and the dynamic compensation amount of the optimal downward adjustable capacity.

[0092] Step 4.4.4, calculate the robust interval in the upward direction, expressed by the following formula: (65) In the formula, This indicates the lower limit of the upward compensation amount. This indicates the upper limit of the upward compensation amount. Indicates the lower bound of the confidence level. This indicates the upper bound of the confidence level.

[0093] The robust interval in the downward direction is calculated using the following formula: (66) In the formula, This indicates the lower limit of the downward compensation amount. This indicates the upper limit of the downward compensation amount. Indicates the lower bound of the confidence level. This indicates the upper bound of the confidence level.

[0094] Step 4.5: Determine the historical dynamic safety margin based on the attention weight of the historical window.

[0095] Historical weighted risk is determined based on the attention weights of historical windows, expressed by the following formula: (67) (68) In the formula, , This indicates the historical weighted risk of upward or downward adjustments. , Represents the history window Attention weights for upward and downward adjustments , Represents the history window The upward and downward adjustment risk coefficients are expressed by the following formula: (69) (70) In the formula, Represents the history window The rate of mutation in the load, Represents the history window The average error of the scenery, Represents the history window The average load error, Represents the history window The area's capacity is adjustable upwards. Represents the history window The capacity is adjustable downwards for the region. This represents the daily average load mutation rate benchmark.

[0096] Historical dynamic safety margin is expressed by the following formula: (71) (72) In the formula, , This indicates the safety margin for upward and downward adjustments. Indicates the maximum safety margin. Indicates the minimum safety margin. Indicates the basic margin, Indicates the current risk weight. This represents the historical risk weight.

[0097] Step 4.6: Determine the final upward and downward adjustment capacity based on the robust intervals in the upward and downward directions and the historical dynamic safety margin.

[0098] The final upward and downward capacity adjustments are expressed by the following formula: (73) (74) In the formula, , These represent the final upward and downward adjustment capacities, respectively. and The adjusted upward and downward compensation amounts after verification are expressed by the following formula: (75) (76) In the formula, , This represents the dynamic compensation amount for the optimal upward adjustable capacity and the dynamic compensation amount for the optimal downward adjustable capacity.

[0099] Embodiment 2 of the present invention provides a regional integrated energy system adjustable capacity optimization system, which runs the regional integrated energy system adjustable capacity optimization method described in Embodiment 1, and includes: The upper and lower limit calculation module is used to input real-time data of the device into the deep learning model and determine the upper and lower limits of the adjustable capacity of the device based on the output results. The system upper and lower limit solution module is used to construct a multi-objective optimization model with the objectives of minimizing adjustment error cost, minimizing operating cost, and minimizing response delay cost, and solve for the optimal adjustment power of the equipment. By correcting the upper and lower limits of the adjustable capacity of the equipment by the optimal adjustment power of all equipment, the corrected upper and lower limits of the adjustable capacity of the equipment are aggregated to determine the total upward and downward adjustment capacity of the system. The region upper and lower limit solution module is used to obtain the net available upward and downward capacity by subtracting the total transmission loss of the line from the total upward and downward adjustment capacity of the system, and then corrects the net available upward and downward capacity with the available transmission capacity of the line to determine the current region's upward and downward adjustable capacity. The output module is used to obtain the probability feature vectors of the upward and downward adjustment directions based on the current region's upward and downward adjustable capacity using Bayesian inference modeling; determine the attention weights of the historical window through a temporal attention mechanism based on the obtained feature vectors; and use the attention weights to determine the dynamic safety margin to determine the final upward and downward adjustment capacity of the region.

[0100] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention constructs a deep learning model using convolutional neural networks, long short-term memory networks, and attention mechanisms to predict the upper and lower limits of adjustable capacity at the device level in real time, and introduces a confidence mechanism to suppress uncertainty propagation. With minimizing adjustment error costs, operating costs, and response delay costs as multiple objectives, it employs the augmented Lagrangian method to achieve coordinated aggregation of system-level capacity. Simultaneously, this invention accurately calculates transmission losses in the power grid and heating network, dynamically assesses available transmission capacity of lines, and improves the accuracy of regional adjustable capacity. This invention constructs a three-level evaluation system of "equipment-system-region" and adopts technical means that combine deep learning with multi-objective optimization collaboration, dynamic compensation and adaptive safety margin to improve evaluation accuracy, system flexibility and decision reliability, and reduce prediction errors, operational risks and manual intervention costs. This invention overcomes the limitations of single-dimensional analysis through a three-level progressive evaluation of equipment-level adjustment potential mining, system-level multi-energy collaborative aggregation and regional-level transmission efficiency correction. By compensating for network transmission losses in regional transmission efficiency, it truly reflects the available capacity of cross-regional energy supply and improves the utilization efficiency and operational reliability of regional integrated energy system capacity. This invention achieves basic prediction of regional adjustable capacity through a long short-term memory network model, and combines Bayesian inference to probabilistically model capacity prediction errors, wind and solar power output errors, and load fluctuation errors. A temporal attention mechanism is introduced to mine the correlation between historical error features and the current scenario, generating a temporally enhanced error feature vector. Based on this, Bayesian robust optimization is used to solve for dynamic compensation amounts and robust compensation intervals, coupled with risk-related dynamic safety margins, forming a multi-level correction mechanism. Simultaneously, relying on a "feature-decision-result" triplet experience pool, model parameters are updated in real time based on the final prediction error, improving the dynamic adaptability of capacity adjustment under wind and solar power output fluctuations and load abrupt changes, effectively reducing assessment errors in complex scenarios. Through online learning and real-time parameter iteration, the system's rapid response capability to dynamic changes in energy supply and demand is strengthened, further enhancing the robustness of capacity assessment. Furthermore, robust interval boundary verification is introduced in the final capacity calculation, combined with previously accurately calculated transmission losses and line available capacity constraints, ensuring that the regional adjustable capacity closely matches the actual network operating state, avoiding a disconnect between theoretical assessment values ​​and engineering feasibility.

[0101] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for optimizing the adjustable capacity of a regional integrated energy system, characterized in that, include: Input real-time data from the device into a deep learning model, and determine the upper and lower limits of the device's adjustable capacity based on the output results; A multi-objective optimization model is constructed with the objectives of minimizing adjustment error cost, operating cost, and response delay cost. The optimal adjustment power of the equipment is obtained by solving the optimal adjustment power of all equipment. The upper and lower limits of the adjustable capacity of the equipment are corrected by adjusting the optimal adjustment power of all equipment. The corrected upper and lower limits of the adjustable capacity of the equipment are aggregated to determine the total upward and downward adjustment capacity of the system. The net available upward and downward capacities are obtained by subtracting the total line transmission loss from the total upward and downward adjustment capacity of the system, respectively. The net available upward and downward capacities are then corrected using the available line transmission capacity to determine the current area's adjustable upward and downward capacities. Based on the current adjustable capacity of the region upward and downward, Bayesian inference modeling is used to obtain the probability feature vectors of the upward and downward adjustment directions. The attention weights of the historical window are determined based on the obtained feature vectors using a temporal attention mechanism. Attention weights are used to determine dynamic safety margins to determine the final up and down adjustment capacity of the region.

2. The method for adjusting the capacity of a regional integrated energy system according to claim 1, characterized in that: Determining the upper and lower limits of the adjustable capacity of the equipment includes: Deep learning models are constructed based on convolutional neural networks, long short-term memory networks, and attention mechanisms. Real-time data from the device is input into a deep learning model to predict the minimum and maximum power boundaries of the device under a pre-set confidence level. Based on the predicted minimum and maximum power boundaries, the upper and lower limits of the device's adjustable capacity are calculated, as expressed by the following formula: In the formula, S i,u (t) represents the adjustable capacity limit of the i-th device during time period t, S i,d (t) represents the lower limit of the adjustable capacity of the i-th device in time period t, c i (t) represents the preset confidence level of the i-th device in time period t. This is the minimum power boundary for the predicted i-th device in time period t. This is the predicted maximum power boundary for the i-th device in time period t. This represents the reference power for the expected scheduling command of the i-th device during time period t.

3. The method for adjusting the capacity of a regional integrated energy system according to claim 1, characterized in that: Adjustment error cost It can be expressed by the following formula: In the formula, Let be the adjusted power of the i-th device in the k-th time period within the scheduling cycle. c penalty The power error penalty rate is K, which represents the total number of time periods in the scheduling cycle. Let K be the adjustment target vector of the system in the k-th time period within the scheduling cycle. This indicates the total number of devices.

4. The adjustable capacity optimization method for a regional integrated energy system according to claim 3, characterized in that: The system's adjustment target vector for the k-th time period within the scheduling cycle is expressed by the following formula: In the formula, This represents the upward adjustment demand of the system in the k-th time period within the scheduling cycle. This represents the downward adjustment demand of the system in the k-th time period within the scheduling cycle. This represents the positive load fluctuation of the system during the k-th time period within the scheduling cycle. This represents the negative load fluctuation of the system during the k-th time period within the scheduling cycle. This represents the positive fluctuation of wind and solar power in the k-th time period within the scheduling cycle. This represents the negative fluctuation of wind and solar power in the k-th time period within the scheduling cycle.

5. The method for adjusting the capacity of a regional integrated energy system according to claim 1, characterized in that: Determining the current region's adjustable capacity upwards and downwards includes: In the formula, This indicates the upward adjustable capacity of the region during time period t. C represents the downward adjustable capacity of a region during time period t. sys,u (t) represents the total upward adjustment capacity of the system during time period t, C sys,d (t) represents the total downward adjustment capacity of the system in time period t. Represents the total transmission loss, and η represents the safety margin factor. For the line l With the current available transmission capacity, This indicates all lines.

6. The adjustable capacity optimization method for a regional integrated energy system according to claim 1, characterized in that: Determining the final upward and downward adjustment capacity for the region includes: Normalized historical data is input into a long short-term memory network model to predict the upward and downward adjustable capacity of the region. The upward and downward capacity prediction errors are determined by comparing the region's upward and downward adjustable capacity with the current region's upward and downward adjustable capacity, thus determining the wind and solar power output error and load fluctuation error. Based on Bayesian maximum a posteriori estimation, probability distribution models are performed for upward and downward capacity prediction errors, wind and solar power output errors, and load fluctuation errors, respectively. Their distribution parameters are extracted and combined with the load mutation rate to construct probability feature vectors for upward and downward adjustment directions. Calculate the similarity scoring function of the probability feature vectors of the up and down directions, and calculate the attention weight of the historical window based on the similarity scoring function; Based on the region's upward and downward adjustable capacity predictions, the dynamic compensation amount of the upward and downward adjustable capacity is solved by Bayesian robust optimization. The historical dynamic safety margin is determined based on the attention weight of the historical window, and the final upward and downward adjustment capacity is determined based on the dynamic compensation amount of the upward and downward adjustable capacity and the historical dynamic safety margin.

7. The adjustable capacity optimization method for a regional integrated energy system according to claim 6, characterized in that: Constructing the probabilistic feature vectors for the up- and down-adjustment directions includes: Assuming that the upward and downward capacity prediction errors follow a t-distribution, the mean and standard deviation of the upward and downward capacity prediction errors are updated by Bayesian maximum a posteriori estimation. Assuming that the wind and solar power output error follows a truncated normal distribution, the mean and standard deviation of the wind and solar power output error are updated by Bayesian maximum a posteriori estimation. Assuming that the load fluctuation error follows a Poisson distribution, the mean, standard deviation, and load mutation rate per unit time of the updated load fluctuation error are estimated using Bayesian maximum a posteriori estimation. Based on the mean and standard deviation of upward and downward capacity prediction errors, the mean and standard deviation of wind and solar power output errors, the mean and standard deviation of load fluctuation errors, and the unit time load mutation rate of load fluctuation errors, probabilistic feature vectors for upward and downward adjustment directions are constructed.

8. The adjustable capacity optimization method for a regional integrated energy system according to claim 7, characterized in that: The update of the mean and standard deviation of the upward and downward capacity prediction errors includes: The mean value of the updated wind and solar power output error is expressed by the following formula: In the formula, This represents the updated mean error of the landscape. This represents the prior mean of the wind and light error. The confidence coefficient represents the error in the wind and light perception. The sample mean of the wind and light error is represented. Indicates the number of samples; The standard deviation of the updated wind power output error is expressed by the following formula: In the formula, This represents the updated standard deviation of the landscape error. This represents the prior standard deviation of the wind and light error. This represents the sample variance of the wind and light error.

9. The adjustable capacity optimization method for a regional integrated energy system according to claim 6, characterized in that: The attention weights for the historical windows are calculated based on the similarity scoring function, expressed by the following formula: In the formula, and Represents the history window Attention weights for upward and downward adjustments , Indicates the number of historical windows. Represents each history window The sum of similarity scores, and These represent the probabilistic feature vectors for upward and downward adjustments and the historical window, respectively. The feature similarity score is expressed by the following formula: In the formula, Indicates the first The weights of each feature, , These represent the history window. The probability eigenvectors of the upward and downward adjustment directions One portion, , The first and second eigenvectors represent the probabilistic eigenvectors of the upward and downward directions at the current time t, respectively. One portion, This indicates a parameter to prevent division by zero. Indicates the feature dimension.

10. A regional integrated energy system adjustable capacity optimization system, operating the regional integrated energy system adjustable capacity optimization method according to any one of claims 1-9, characterized in that: The upper and lower limit calculation module is used to input real-time data of the device into the deep learning model and determine the upper and lower limits of the adjustable capacity of the device based on the output results. The system upper and lower limit solution module is used to construct a multi-objective optimization model with the objectives of minimizing adjustment error cost, minimizing operating cost, and minimizing response delay cost, and solve for the optimal adjustment power of the equipment. By correcting the upper and lower limits of the adjustable capacity of the equipment by the optimal adjustment power of all equipment, the corrected upper and lower limits of the adjustable capacity of the equipment are aggregated to determine the total upward and downward adjustment capacity of the system. The region upper and lower limit solution module is used to obtain the net available upward and downward capacity by subtracting the total transmission loss of the line from the total upward and downward adjustment capacity of the system, and then corrects the net available upward and downward capacity with the available transmission capacity of the line to determine the current region's upward and downward adjustable capacity. The output module is used to obtain the probability feature vectors of the upward and downward adjustment directions based on the current region's adjustable capacity upward and downward using Bayesian inference modeling. The attention weights of the historical window are determined based on the obtained feature vectors using a temporal attention mechanism. Attention weights are used to determine dynamic safety margins to determine the final up and down adjustment capacity of the region.

Images