A double-layer robust computing method for multi-site distributed energy system of expressway

Abstract

This application provides a two-layer robust computational method for multi-site distributed energy systems along highways. The method includes: acquiring historical meteorological and traffic flow data from multiple stations along the highway; extracting typical days and their probabilities through cluster analysis; constructing a multi-level coupled uncertainty set for each typical day; using infinite norm and 1-norm constraints to limit the deviation amplitude of a single device, the combined wind-solar deviation of a single station, and the combined wind-solar deviation of multiple stations, respectively; constructing an upper-level capacity configuration model with the objective of minimizing the annual cost over the entire lifecycle; and searching for the most unfavorable scenario and optimizing energy management strategies within the uncertainty set for the lower-level subproblems; and using a nested iteration algorithm of column and constraint generation to obtain the optimal capacity configuration scheme. This application uses a multi-level uncertainty set to finely characterize the spatiotemporal correlation of wind and solar power output, avoiding overly conservative configuration, introducing a cross-site mutual assistance mechanism to improve collaborative efficiency, achieving coordinated optimization of planning and operation, and effectively improving the system's economy and reliability.

Description

Technical Field

[0001] This invention relates to the field of power system operation and control technology, specifically to a two-layer robust computation method for distributed energy systems at multiple highway stations. Background Technology

[0002] In recent years, with the rapid development of highway construction, the number and length of highway tunnels have continued to increase, leading to a significant rise in energy consumption due to the increasing number of electromechanical equipment such as lighting, ventilation, monitoring, and fire protection systems within the tunnels. Tunnel loads are characterized by high continuity, high rigidity, and high safety sensitivity, placing higher demands on the reliability and stability of the power supply system. Traditional external power grid supply methods are prone to voltage fluctuations or even power outages in remote areas or under severe weather conditions, making it difficult to meet the needs of safe tunnel operation. Therefore, wind power generation, photovoltaic power generation, and energy storage systems are gradually being introduced into the highway tunnel energy supply system, forming a hybrid wind-solar-storage energy system to achieve clean energy complementarity and improve power supply reliability. Summary of the Invention

[0003] This application provides a two-layer robust computing method for a multi-site distributed energy system on highways, including the following steps: The system acquires hourly historical meteorological and traffic flow data for one year from multiple stations along the highway; each station is equipped with wind power, photovoltaic, energy storage, and tunnels; cross-station mutual support is possible between the multiple stations; historical meteorological data includes wind speed and sunshine values; historical traffic flow data includes vehicle flow values. Cluster analysis was used to extract typical day data for S typical days within a year and the annual occurrence probability of each typical day; the data for each typical day included hourly wind speed, light intensity and traffic flow values ​​for each station within 24 hours. For each typical day, the hourly wind speed, solar irradiance, and traffic flow values ​​at each station are used to determine the typical day center point matrix. The typical day center point matrix includes the hourly wind power output, solar power output, and tunnel load values ​​at each station on that typical day. The wind power output is a function of the wind speed and the wind turbine capacity; the solar power output is a function of the solar irradiance and the solar capacity; and the tunnel load value is a function of the traffic flow value. For each typical day, a typical day uncertainty set is constructed based on the typical center point matrix. The typical day uncertainty set describes the fluctuation of wind and solar power output based on the typical day center point matrix. The typical day uncertainty set includes at least three levels of constraints. The first level uses infinite norm constraints to limit the hourly output deviation of wind power equipment and photovoltaic equipment at each site. The second level uses 1-norm constraints to limit the sum of hourly output deviations of combined wind and solar power output at each site. The third level uses 1-norm constraints to limit the sum of hourly output deviations of combined wind and solar power output at all sites. With the goal of minimizing the annualized cost of the entire life cycle of the energy system, a higher-level master problem model for capacity allocation is constructed. The annualized cost includes equipment investment cost, operation and maintenance cost, and annualized operating risk cost. The annualized operating risk cost is the weighted sum of the minimum operating risk cost under the most unfavorable scenario for each typical day and the probability of occurrence for each typical day. The decision variables of the higher-level master problem model are the wind turbine capacity, photovoltaic capacity, and energy storage capacity of each site. For the capacity configuration scheme given by the upper-level master problem model, a lower-level sub-problem model is constructed for the current typical day. The lower-level sub-problem model is used to evaluate the operational risk cost of the capacity configuration scheme under the current typical day. Its input is the capacity configuration scheme given by the upper level and the typical day uncertainty set of the current typical day, and its output is the minimum operational risk cost under the most unfavorable scenario of the current typical day. The lower-level sub-problem model searches for the most unfavorable scenario in the typical day uncertainty set corresponding to the current typical day, and optimizes the energy management strategy under the scenario. The energy management strategy includes the energy storage charging and discharging power of each site, the curtailed power, the power purchased from the external grid, and the cross-site mutual assistance power. The column and constraint generation algorithm is used to perform nested iterative solutions on the upper-level main problem model and the lower-level sub-problem model to obtain the optimal capacity configuration scheme. Attached Figure Description

[0004] Figure 1 This is a schematic diagram of a two-layer robust computing method for a multi-site distributed energy system on a highway, according to an embodiment of this application. Detailed Implementation

[0005] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0006] This invention provides a two-layer robust computation method 10 for multi-site distributed energy systems on highways.

[0007] Through extensive research, the inventors of this application have discovered that most existing wind and solar power output uncertainty models describe a single site or a single energy type, failing to fully consider the spatial correlation between multiple sites and the temporal complementarity between different energy types. Furthermore, they often employ simple prediction error ranges or single norm constraints, making it impossible to finely characterize the fluctuation characteristics of wind and solar power output across multiple spatial dimensions ("single device-single site-multiple sites") and multiple temporal dimensions ("single time period-all day"), thus making it difficult to accurately reflect the true fluctuation patterns of renewable energy power output.

[0008] This application applies to the planning and operation scenario of a multi-site distributed energy system along a highway. In this scenario, multiple sites are distributed along the highway, each equipped with wind power generation equipment, photovoltaic power generation equipment, and energy storage equipment, supplying power to tunnel loads. Due to the different geographical locations of each site, their micro-meteorological environments vary, with wind speed, sunlight, and other meteorological conditions being independent. There is no strong correlation between the wind power output and photovoltaic output of different sites; simultaneously, the traffic flow data of each site is also independent, resulting in distinct characteristics of tunnel loads. The sites are connected by tie lines, enabling cross-site power exchange and achieving spatiotemporal energy complementarity. This scenario is characterized by multi-site spatial distribution, multi-type energy coupling, independent loads at multiple nodes, and feasible multi-line mutual support. Therefore, it is necessary to collaboratively optimize the capacity configuration and operation strategy of the entire system while fully considering the uncertainty of wind and solar power output and the load adjustability.

[0009] The following combination Figure 1 The embodiments of this application will be described in detail below, including specific steps.

[0010] Method 10 includes: Step 110: Obtain hourly historical meteorological data and historical traffic flow data for one year from multiple stations along the highway; each of the multiple stations is equipped with wind power equipment, photovoltaic equipment, energy storage equipment and tunnels; the multiple stations can perform cross-station mutual assistance; historical meteorological data includes wind speed value and sunshine value; historical traffic flow data includes vehicle flow value.

[0011] The purpose of this step is to collect the basic input data required by the method of this invention, providing data support for subsequent typical day extraction, uncertainty modeling, and optimized configuration. The multiple stations along the highway refer to several tunnel or service area nodes distributed along the highway. Each station is equipped with wind power equipment, photovoltaic equipment, energy storage equipment, and tunnel loads, and the different stations are connected by connecting lines, enabling cross-site power sharing. Due to the different geographical locations of each station, their micro-meteorological environments differ, and meteorological conditions such as wind speed and sunlight are independent of each other. This is the basis for considering the spatial correlation of multiple stations in this invention.

[0012] The historical meteorological data includes hourly wind speed and solar irradiance values ​​for each station over a year. Wind speed is the core input for calculating wind power output, and solar irradiance is the core input for calculating photovoltaic output. The historical traffic flow data includes hourly vehicle flow values ​​for each station over a year. Vehicle flow is a key factor determining tunnel load demand. Higher vehicle flow requires higher operating levels for tunnel ventilation and other equipment.

[0013] For example, suppose there are two stations, A and B, along a highway. At station A, data collected at 00:00 on January 1st of a certain year shows a wind speed of 3.5 meters per second, solar radiation intensity of 0 watts per square meter, and traffic flow of 20 vehicles per hour; at 01:00 on January 1st, the wind speed is 3.2 meters per second, solar radiation intensity is 0 watts per square meter, and traffic flow is 30 vehicles per hour, and so on until 23:00 on December 31st. The same applies to station B. Each station generates a set of 8760 hours of time-series data. This data comprehensively records the meteorological and traffic conditions at each station at every moment throughout the year, forming the basis for all subsequent analyses.

[0014] Step 120: Use cluster analysis to extract typical day data for S typical days within one year and the annual occurrence probability of each typical day; each typical day data includes hourly wind speed, light intensity and traffic flow values ​​for each station within 24 hours.

[0015] The purpose of this step is to perform dimensionality reduction processing on the 8760 hours of raw data obtained in step 110 throughout the year, and extract several typical daily scenarios that can represent the operational characteristics of the whole year, so as to provide a basic input of moderate scale and complete information for subsequent optimization.

[0016] A typical day refers to a representative 24-hour time-series scene obtained through cluster analysis, which can summarize the common characteristics of a certain type of similar weather-traffic days. Each typical day's data includes hourly wind speed, light intensity, and traffic flow values ​​for each station within 24 hours, fully preserving the intraday time-series characteristics. The annual occurrence probability refers to the percentage of days in a year that the typical day represents of this type of weather-traffic pattern, used to assign appropriate weights to each typical day in subsequent annualized cost calculations.

[0017] Directly using the raw data of 8760 hours throughout the year for optimization would result in excessive computational scale, making the model difficult to solve. However, simply averaging the data or selecting a few fixed dates cannot accurately reflect the diversity of operational characteristics throughout the year. Cluster analysis can significantly reduce data dimensionality while preserving the main characteristics of the data, grouping similar daily scenarios into one category and using a few typical days to represent the operational conditions of the whole year. It is an effective means of balancing computational accuracy and efficiency.

[0018] This step uses time-series vector clustering instead of total output clustering because optimizing energy storage systems requires complete intraday time-series information. Clustering only the total daily power generation or load would lose the intraday fluctuation patterns and peak-valley characteristics of wind and solar resources, rendering energy storage dispatching unreliable. By concatenating 24-hour daily wind speed, solar radiation, and traffic flow data into a high-dimensional time-series vector for clustering, the intraday variation patterns of each site can be fully preserved, providing a foundation for subsequent optimization considering cross-time-period energy storage adjustments.

[0019] For example, step 120 can be implemented through steps 1201-1205.

[0020] Step 1201: Arrange the wind speed, light intensity, and traffic flow values ​​of all stations for each day in 24-hour time sequence and concatenate them into a high-dimensional daily feature vector.

[0021] In this step, the wind speed, sunlight, and traffic flow data for all stations each day are arranged in 24-hour time sequence and concatenated into a high-dimensional daily feature vector. For example, for a case with M stations, each station has 24-hour wind speed, 24-hour sunlight, and 24-hour traffic flow data per day. Therefore, the dimension of each daily feature vector is M times 3 times 24. This vector completely preserves the original information from multiple stations, multiple data types, and hourly data.

[0022] Step 1202: Normalize the different types of data in the high-dimensional daily eigenvector to obtain the normalized eigenvector.

[0023] In this step, since wind speed, light intensity, and traffic flow have different dimensions and numerical ranges, it is necessary to normalize each type of data separately so that they have equal weights when clustering, and avoid variables with large values ​​dominating the clustering results.

[0024] Step 1203: Cluster the 365 normalized feature vectors to obtain several cluster center vectors. Each cluster center vector represents a typical daily weather-traffic pattern.

[0025] In this step, the K-means clustering algorithm can be used to perform cluster analysis on the 365 normalized daily feature vectors for a year. The optimal number of clusters S is determined by the elbow rule or silhouette coefficient, typically between 4 and 12. The algorithm groups similar daily feature vectors into the same class and calculates the cluster center vector for each class.

[0026] Step 1204: Perform inverse normalization on the cluster center vectors to obtain typical daily data.

[0027] This step restores the cluster center vectors to typical daily data with actual physical meaning through inverse normalization. Each typical day includes hourly wind speed, light intensity, and traffic flow values ​​for each station over 24 hours.

[0028] Step 1205: The ratio of the number of days contained in the cluster containing the cluster center to the total number of days in the year is taken as the annual probability of a typical day, and the sum of the probabilities of all typical days is 1.

[0029] Step 130: For each typical day, determine the typical day center point matrix for each station based on the hourly wind speed, solar radiation, and traffic flow values. The typical day center point matrix includes the hourly wind power output, photovoltaic power output, and tunnel load values ​​for each station on that typical day. The wind power output is a function of the wind speed and the wind turbine capacity; the photovoltaic power output is a function of the solar radiation and the photovoltaic capacity; and the tunnel load value is a function of the traffic flow value.

[0030] The purpose of this step is to convert the typical daily meteorological-traffic data extracted in step 120 into the output-load data required for subsequent optimization, and to establish a typical daily center point matrix as the basis for constructing the uncertainty set and optimizing the operation strategy.

[0031] The typical day center point matrix is ​​a three-dimensional matrix composed of hourly wind speed, solar irradiance, and traffic flow values ​​for each typical day, calculated using corresponding transformation functions to obtain wind power output, solar power output, and tunnel load values. The matrix's dimensions are the number of stations multiplied by 3 and then by 24, where 3 represents the three data types (wind power output, solar power output, and tunnel load), and 24 represents the 24 hours of a day.

[0032] The output of wind power equipment is a function of wind speed and turbine capacity, and the specific relationship can be determined by the power characteristic curve of the turbine. An exemplary power characteristic curve of a wind turbine can be as follows: when the wind speed is lower than the cut-in wind speed or higher than the cut-out wind speed, the output is zero; when the wind speed is between the cut-in wind speed and the rated wind speed, the output increases with the increase of wind speed; when the wind speed is between the rated wind speed and the cut-out wind speed, the output remains unchanged at the rated power.

[0033] The output of photovoltaic (PV) equipment is a function of the amount of sunlight and the PV capacity, and the specific relationship is determined by the photoelectric conversion efficiency of the PV modules. The stronger the sunlight, the greater the output.

[0034] The tunnel load is a function of the traffic volume; the higher the traffic volume, the greater the demand on tunnel ventilation, lighting, and other equipment. Optionally, the tunnel load typically includes two parts: rigid load and flexible load. The rigid load is the part that must be met to ensure the safe operation of the tunnel, while the flexible load can be reduced within a certain range.

[0035] It is worth noting that wind power output and solar power output are functions of wind turbine capacity and solar capacity, respectively. This means that under different capacity configurations, the same wind speed and solar irradiance will result in different power output values. Therefore, the typical daily center point matrix is ​​not a fixed quantity, but rather changes dynamically with the wind turbine and solar capacity configuration.

[0036] Wind power output is a function of wind speed and turbine capacity; solar power output is a function of solar irradiance and photovoltaic capacity; and tunnel load is a function of traffic volume. When a typical day is determined, wind speed, solar irradiance, and traffic volume are also determined. The wind power output in the typical day's center point matrix is ​​a function of turbine capacity, and the solar power output is a function of photovoltaic capacity. When the site and time period are determined, the tunnel load value in the typical day's center point matrix is ​​a fixed value.

[0037] Step 140: For each typical day, construct the typical day uncertainty set corresponding to the typical center point matrix. The typical day uncertainty set is used to describe the fluctuation of wind and solar power output based on the typical day center point matrix. The typical day uncertainty set includes at least three levels of constraints. The first level uses infinite norm constraints to limit the hourly output deviation of wind power equipment and photovoltaic equipment at each site. The second level uses 1-norm constraints to limit the total hourly output deviation of combined wind and solar power output at each site. The third level uses 1-norm constraints to limit the total hourly output deviation of combined wind and solar power output at all sites.

[0038] The purpose of this step is to construct a multi-level coupled uncertainty set that can accurately characterize the random fluctuations in wind and solar power output, providing a search space for the most unfavorable scenario in subsequent sub-problems.

[0039] Uncertainty set is a core concept in robust optimization, used to describe the possible range of values ​​for uncertain parameters. Unlike traditional stochastic optimization, robust optimization does not assume a probability distribution for uncertain parameters, but rather restricts the uncertainty to a set, seeking the optimal solution that can handle all possible situations within the set. In this invention, the typical daily uncertainty set is centered on the typical daily center point matrix obtained in step 130, describing the fluctuation range of actual wind and solar power output relative to the center point value.

[0040] Output deviation refers to the difference between the actual output of wind and solar power equipment and the predicted output (i.e., the center point value). Due to the randomness of wind speed and sunlight, the actual output always fluctuates around the predicted value. The magnitude and range of this fluctuation are what the uncertainty set describes.

[0041] The infinity norm constraint refers to limiting the deviation of each variable individually, meaning that the output deviation of each device in each time period cannot exceed a certain upper limit. This is the most basic constraint, reflecting the physical limitations of a single device.

[0042] A 1-norm constraint limits the sum of the absolute values ​​of the deviations of multiple variables. Unlike the infinite norm constraint, the 1-norm constraint focuses on the total budget, meaning that multiple variables cannot all reach their maximum deviation simultaneously; their total deviation has an upper limit.

[0043] Traditional descriptions of wind and solar power output uncertainty typically employ a simple interval form, assuming that the output of each device fluctuates independently within a fixed range at each time period. This approach has two problems: first, it is overly conservative, as it allows all devices to reach their maximum deviation simultaneously at all times, which is virtually impossible in physics; second, it ignores the temporal and spatial correlations of wind and solar power output.

[0044] This invention constructs a multi-level coupled uncertainty set to constrain wind and solar power output deviations from three different dimensions, thereby more realistically reflecting actual physical laws. The first level (infinity norm constraint) reflects the basic physical limitations of a single device. Due to the inherent technical characteristics of wind turbines and photovoltaic equipment, their output cannot deviate infinitely from the predicted value; this maximum deviation can be obtained from historical data statistics.

[0045] The second level (1-norm constraint) reflects the complementarity of wind and solar power within the same site. At the same site, wind and solar power outputs have a natural temporal complementarity—high wind speeds often coincide with weak sunlight, and strong sunlight often coincides with low wind speeds, making it impossible for them to reach their maximum negative bias simultaneously. The second-level constraint limits the sum of the joint wind and solar biases within each site, avoiding such an unrealistic extreme combination.

[0046] The third level (1-norm constraint) reflects the spatial clustering effect among multiple stations. Due to the different geographical locations of different stations, their meteorological conditions vary, and it is impossible for all stations to reach the maximum negative bias at the same time. The third-level constraint limits the sum of the joint wind and solar biases of all stations, avoiding such extreme spatial combinations.

[0047] Through these three levels of constraints, the uncertainty set mathematically describes a fluctuation space that is more in line with physical reality, which retains the necessary conservatism to ensure system safety while avoiding the waste of resources caused by excessive conservatism.

[0048] Assume that at noon on a typical day, the predicted wind power output at site A is WA kilowatts, and the predicted solar power output is VA kilowatts. Based on historical data, the maximum deviation coefficient for wind power is 0.3, meaning the actual wind power output may differ from the predicted output. to Between; the maximum deviation coefficient for photovoltaic power is 0.2, meaning the actual output of photovoltaic power may vary. to Between. This is the first level of constraint.

[0049] The second level of constraint restricts the sum of the wind and solar combined deviations at site A. Assuming the budget parameter for this site is 1.2, the sum of the normalized absolute values ​​of the wind and solar deviations cannot exceed 1.2. This means that if wind power takes the maximum negative deviation (normalized value of 1), the normalized deviation of solar power can only take a maximum of 0.2, and cannot also take the maximum deviation. This reflects the complementarity of wind and solar power within the same site.

[0050] The third-level constraint limits the sum of the combined wind and solar deviations across all stations. Assuming there are three stations along the entire route, and the regional budget parameter for this time period is 2.5, the sum of the absolute values ​​of the normalized deviations of the three stations cannot exceed 2.5. This means that it is impossible for all three stations to reach their maximum deviation simultaneously, reflecting the spatial clustering effect.

[0051] In a preferred embodiment, fourth and fifth level constraints can also be added. For example, the fourth level uses a 1-norm constraint to limit the total cumulative deviation of the combined wind and solar power output at each site over 24 hours, reflecting the budget effect of the same site in the time dimension, since it is impossible to reach the maximum deviation every hour. The fifth level uses a 1-norm constraint to limit the total cumulative deviation of the combined wind and solar power output at all sites over 24 hours, reflecting the total limit of the entire region in the time dimension.

[0052] Step 150: With the goal of minimizing the annualized cost of the energy system throughout its entire life cycle, construct the upper-level master problem model for capacity configuration; the annualized cost includes equipment investment cost, operation and maintenance cost, and annualized operating risk cost; the annualized operating risk cost is the weighted sum of the minimum operating risk cost under the most unfavorable scenario for each typical day and the probability of occurrence for each typical day; the decision variables of the upper-level master problem model are the wind turbine capacity, photovoltaic capacity, and energy storage capacity of each site.

[0053] The purpose of this step is to establish an optimization model for capacity configuration, determine the optimal installed capacity of wind turbines, photovoltaics, and energy storage at each site during the system planning phase, so as to minimize the overall cost of the system throughout its entire life cycle.

[0054] Lifecycle cost is an equivalent annual cost that discounts all costs over the entire lifespan of a system to an equivalent annual value for each year, used to measure the system's economic viability. The reason for using equivalent annual costs instead of a one-time total investment is that wind power, solar power, and energy storage equipment have different lifespans; only by discounting costs to annual values ​​can economic comparisons be made on a uniform time scale.

[0055] Equipment investment cost refers to the initial purchase cost of wind turbines, photovoltaic modules, energy storage equipment, etc., and is usually calculated based on the product of the unit capacity cost of each piece of equipment and the installed capacity. This part of the cost is a one-time investment and needs to be amortized to each year using an equal annual value conversion formula.

[0056] Operation and maintenance costs refer to the daily maintenance expenses incurred during the operation of a system. They are usually estimated as a certain percentage of the investment cost and include expenses for equipment inspection, repair, and replacement.

[0057] The annualized operational risk cost is a key innovation of this invention. It is defined as the weighted sum of the minimum operational risk cost under the most unfavorable scenario for each typical day and the probability of occurrence for each typical day. This definition incorporates the worst-case scenario that may be encountered during the operation phase into the decision-making considerations during the planning phase.

[0058] The minimum operational risk cost under the worst-case scenario is calculated by the lower-level subproblem model. It represents the minimum operational cost that the system must incur using the optimal operating strategy under a given capacity configuration, regardless of any uncertainties. This value reflects the capacity configuration scheme's ability to cope with extreme situations.

[0059] The decision variables in the upper-level master problem model are the wind turbine capacity, photovoltaic capacity, and energy storage capacity of each site. Once these variables are determined, they constitute the physical foundation of the system, and all subsequent operational optimizations are based on this.

[0060] This invention employs a method combining robust optimization and typical days, considering both uncertainty and reducing computational scale through typical days. The upper-level master problem does not directly address the details of uncertainty; instead, it indirectly evaluates the capacity scheme's ability to cope with uncertainty through the "minimum operational risk cost under the most unfavorable scenario" returned by the lower-level subproblems. The advantages of this two-layer structure are: First, the upper-level master problem focuses only on capacity decisions, while the lower-level subproblems handle operational details and uncertainties, achieving decoupling and synergy between planning and operation. Second, the annualized operational risk cost is a weighted sum of the minimum operational risk costs under the most unfavorable scenario for each typical day. This means that the capacity scheme must be able to cope with the extreme cases that may occur on each typical day, rather than merely satisfying the average case.

[0061] For example Suppose there are 3 stations along a certain highway. Step 120 extracts 6 typical days, with probabilities of occurrence of 0.2, 0.15, 0.2, 0.1, 0.2, and 0.15 respectively. The upper-level main problem requires deciding on the wind turbine capacity, photovoltaic capacity, and energy storage capacity of stations A and B.

[0062] Equipment investment costs are calculated based on unit capacity costs. For example, the unit capacity cost for wind turbines is 3,000 yuan per kilowatt, for photovoltaics it is 2,500 yuan per kilowatt, and for energy storage it is 1,500 yuan per kilowatt-hour. Operation and maintenance costs are estimated at 3% of the investment cost annually.

[0063] The annualized operational risk cost requires data from lower-level subproblems. Assume that in a certain iteration, for the current candidate capacity scheme, the lower-level subproblems calculate the minimum operational risk costs for the worst-case scenarios on six typical days as 1.2 million yuan, 1.5 million yuan, 1.1 million yuan, 1.8 million yuan, 1.3 million yuan, and 1.4 million yuan, respectively. Then, the annualized operational risk cost equals the weighted sum of these costs and their corresponding probabilities: 0.2 multiplied by 120 plus 0.15 multiplied by 150 plus 0.2 multiplied by 110 plus 0.1 multiplied by 180 plus 0.2 multiplied by 130 plus 0.15 multiplied by 140, resulting in approximately 1.325 million yuan.

[0064] The goal of the upper-level main problem is to find a set of capacity configurations that minimizes the sum of equipment investment costs, operation and maintenance costs, and annualized operating risk costs. This optimization problem will be solved through multiple iterations within the framework of the column and constraint generation algorithm.

[0065] Step 160: For the capacity configuration scheme given by the upper-level master problem model, construct a lower-level sub-problem model for the current typical day. The lower-level sub-problem model is used to evaluate the operational risk cost of the capacity configuration scheme under the current typical day. Its input is the capacity configuration scheme given by the upper level and the typical day uncertainty set of the current typical day, and its output is the minimum operational risk cost under the most unfavorable scenario of the current typical day. The lower-level sub-problem model searches for the most unfavorable scenario in the typical day uncertainty set corresponding to the current typical day, and optimizes the energy management strategy under the scenario. The energy management strategy includes the energy storage charging and discharging power of each site, the power curtailment, the power purchased from the external grid, and the power mutual assistance between sites.

[0066] Step 160: For the capacity configuration scheme given by the upper-level master problem model, construct a lower-level sub-problem model for the current typical day. The lower-level sub-problem model is used to evaluate the operational risk cost of the capacity configuration scheme under the current typical day. Its input is the capacity configuration scheme given by the upper level and the typical day uncertainty set of the current typical day, and its output is the minimum operational risk cost under the most unfavorable scenario of the current typical day. The lower-level sub-problem model searches for the most unfavorable scenario in the typical day uncertainty set corresponding to the current typical day, and optimizes the energy management strategy under the scenario. The energy management strategy includes the energy storage charging and discharging power of each site, the power curtailment, the power purchased from the external grid, and the power mutual assistance between sites.

[0067] The purpose of this step is to establish a model that can evaluate the performance of a given capacity scheme on a specific typical day. By solving the model, we can obtain the most unfavorable scenario for that typical day and its corresponding minimum operating risk cost, thus providing a basis for evaluating the economics of the capacity scheme for the upper-level master problem.

[0068] The lower-level subproblem model is the lower-level model in the two-level optimization framework. It takes the capacity configuration scheme given by the upper level as input and outputs the operating risk cost of the capacity scheme under the current typical day.

[0069] Specifically, the lower-level sub-problem model adopts a max-min two-layer optimization structure. In the typical day uncertainty set corresponding to the current typical day, it searches for the most unfavorable scenario that maximizes the operating risk cost of the energy system. Under the given most unfavorable scenario in the outer layer, it optimizes the energy management strategy to minimize the operating risk cost.

[0070] The worst-case scenario refers to the set of wind and solar power output deviations that maximize the system's operational risk and cost within the typical daily uncertainty set constructed in step 140. It represents the most severe natural conditions the system might encounter under the current capacity configuration, such as when both wind and solar power outputs are close to the lower limit of the predicted values, resulting in the minimum available power generation for the system.

[0071] Minimum operating risk cost refers to the lowest operating cost achievable through optimized energy management strategies under the most unfavorable scenario. It represents the best economic performance that the system can achieve through optimal scheduling when facing the worst conditions.

[0072] Energy management strategy is a set of operational decision variables, including the energy storage charging and discharging power, curtailed power, power purchased from the external grid, and cross-site mutual assistance power at each site and time period. These variables together determine the actual operation mode of the system under a given scenario.

[0073] The lower-level subproblem uses a max-min two-level optimization structure because it needs to handle two conflicting problems simultaneously: The outermost maximization represents the perspective of "nature" or "uncertainty," aiming to select a set of wind and solar power output deviations within the uncertainty set to maximize the system's operating cost. This embodies the core idea of ​​robust optimization: instead of assuming that uncertain parameters will take an average value, we prepare for the worst-case scenario.

[0074] The inner-layer minimization represents the perspective of the "operators" or "energy management system," which means minimizing operating costs by optimizing energy storage charging and discharging, power purchase, and mutual assistance strategies under the most unfavorable scenario already selected in the outer layer. This reflects the system's adaptive capability in the face of adverse conditions.

[0075] Combining the two, the output of the lower-level subproblem, "minimum operational risk cost under the worst-case scenario," has a clear physical meaning: given a capacity configuration, regardless of the uncertainty scenario, the minimum operational cost the system must incur through the optimal operating strategy. The larger this value, the worse the capacity scheme's ability to cope with uncertainty; the smaller this value, the more economical and reliable the capacity scheme.

[0076] The energy management strategy for optimizing the inner minimization problem includes the following decision variables: The charging and discharging power of energy storage determines the charging and discharging capacity of the energy storage system at each site during each time period. Energy storage systems can charge and store electrical energy when wind and solar power output is excessive, and discharge to fill the gap when wind and solar power output is insufficient, making it a key means of achieving energy time-shifting. Energy storage charging and discharging are dynamically constrained by energy storage capacity, maximum charging and discharging power, and state of charge.

[0077] Curtailed power refers to clean energy power that is forcibly abandoned when wind and solar power output is excessive, energy storage is full, and power cannot be transmitted to other sites through mutual transfer. Curtailment incurs penalty costs, so the inner layer will try to avoid curtailment as much as possible.

[0078] External grid power purchase refers to the power purchased from the external grid when local wind and solar power output is insufficient, energy storage capacity is inadequate, and sufficient mutual assistance cannot be obtained from other sites. Purchasing electricity incurs costs, but it is sometimes a necessary means to ensure load supply.

[0079] Inter-site power sharing refers to the power transmitted between different sites via tie lines. When a site's wind and solar power output is insufficient, it can obtain power from other sites with surplus power; conversely, when a site has surplus wind and solar power, it can transmit power to other sites. Power sharing is constrained by the transmission capacity of tie lines and is a key mechanism for achieving energy sharing among multiple sites.

[0080] By optimizing these decision variables, the inner-level minimization problem minimizes operational risk costs while satisfying all operational constraints. Operational risk costs include: the cost of purchasing electricity from the external grid, penalties for wind and solar power curtailment, and load gap penalties. Load gap penalties are calculated by multiplying the tunnel load deficit at each site for each time period by a preset gap penalty coefficient.

[0081] The input to this step is the capacity configuration scheme given by the upper-level main problem model, and the typical day uncertainty set of the current typical day constructed in step 140; the output is the minimum operating risk cost under the most unfavorable scenario of the current typical day, which will be used to calculate the annualized operating risk cost in step 150.

[0082] Suppose that the capacity scheme given by the upper layer in the current iteration is as follows: Site A has a wind turbine of 2000 kW, a photovoltaic system of 1000 kW, and an energy storage capacity of 500 kWh; Site B has a wind turbine of 1500 kW, a photovoltaic system of 800 kW, and an energy storage capacity of 400 kWh. The current process is for the first typical day, and its uncertainty set has been constructed.

[0083] The outer layer of the lower-level subproblem begins searching for the worst-case scenario within the uncertainty set. It attempts different combinations of wind power and solar power deviations, with each combination being passed to the inner layer to calculate the minimum operating cost. After the search, it discovers that when site A has the maximum negative deviation for both wind and solar power, and site B also has a similar worst-case combination, the inner layer optimizes the operating cost to the highest level, reaching 1.2 million yuan. This combination represents the worst-case scenario for that typical day, and 1.2 million yuan is the minimum operating risk cost under the worst-case scenario.

[0084] In the inner-layer optimization, given the worst-case scenario in the outer layer, the solver needs to decide: when to charge, when to discharge, and how much to discharge at site A's energy storage; whether to purchase electricity from site B; whether to purchase electricity from the external grid; and whether to discard some of the excess wind and solar power output. By optimizing these decisions, the total operating cost is minimized while meeting load demand.

[0085] This figure of 1.2 million yuan will be recorded and used to calculate the annualized operating risk cost in step 150. Performing the same operation for each typical day yields the minimum operating risk cost under the most unfavorable scenario for all typical days.

[0086] Step 170: The column and constraint generation algorithm is used to perform nested iterative solutions on the upper-level main problem model and the lower-level sub-problem model to obtain the optimal capacity configuration scheme.

[0087] The purpose of this step is to coordinate the coupling between upper-level capacity planning and lower-level energy management through iterative solutions, and ultimately obtain the optimal capacity configuration scheme that can cope with all potential worst-case scenarios.

[0088] The column and constraint generation algorithm is a classic algorithm for solving two-stage robust optimization problems. Its core idea is to progressively add constraints corresponding to the most unfavorable scenarios identified by subproblems to the main problem, gradually shrinking the feasible region of the main problem until it converges to an optimal solution that can handle all potential worst-case scenarios. Compared to the traditional Benders decomposition algorithm, the column and constraint generation algorithm adds complete constraints to the main problem in each iteration rather than cutting planes, thus achieving faster convergence and making it particularly suitable for optimization problems with complex constraints.

[0089] The specific algorithm flow is as follows: In the initialization phase, the lower bound is set to negative infinity, the upper bound is set to positive infinity, the number of iterations is initialized to zero, and the worst-case scenario set is initially set to an empty set.

[0090] In the main problem-solving phase, the upper-level main problem model is solved under the current set of worst-case scenarios to obtain candidate capacity configuration schemes and their objective function values. Since the current main problem only considers some worst-case scenarios, its objective function value is a lower bound of the original problem. Therefore, the lower bound is updated to the objective function value of the current main problem.

[0091] In the sub-problem solving phase, for each typical day, the lower-level sub-problem model is solved under the candidate capacity configuration scheme. The lower-level sub-problem first searches for the most unfavorable scenario that maximizes the operational risk cost within the uncertainty set of the typical day. Then, under this most unfavorable scenario, the energy management strategy is optimized to obtain the minimum operational risk cost. The minimum operational risk cost under the most unfavorable scenario for each typical day is weighted and summed with the corresponding probability of occurrence. This sum is then added to the investment cost and operation and maintenance cost of the current candidate capacity configuration scheme to obtain the upper bound of the total cost corresponding to the current capacity configuration scheme.

[0092] During the convergence judgment phase, the relative error between the current upper bound and the lower bound is calculated. If the error is less than the preset convergence threshold, it means that the current solution is close enough to the optimal solution. The iteration is stopped and the current candidate capacity configuration solution is output as the optimal capacity configuration solution.

[0093] During the constraint addition phase, if convergence is not achieved, the worst-case scenarios for all newly identified typical days in this iteration are added as new constraints to the set of worst-case scenarios of the upper-level main problem model. These new constraints require the upper-level model to ensure that a feasible operating strategy exists under this scenario in subsequent optimizations, thereby gradually shrinking the feasible region of the main problem. Then, the iteration count is incremented by one, and the model returns to the main problem solving phase to continue iterating.

[0094] In the upper-level main problem model in step 150 and the lower-level sub-problem model in step 160, in addition to the objective function, a series of constraints need to be considered. These constraints together constitute the feasible region of the optimization model, ensuring that the solved capacity configuration scheme and operation strategy are feasible and reasonable in engineering.

[0095] The upper-level master problem model primarily involves constraints on capacity configuration, ensuring that decision variables remain within a reasonable range. The installed capacity of wind turbines, photovoltaic (PV) modules, and energy storage at each site is typically limited by factors such as site conditions and investment budgets, requiring values ​​to be taken within certain ranges. For example, wind turbine capacity cannot exceed the product of the maximum number of turbines that can be installed at the site and the rated power of a single turbine; PV capacity cannot exceed the maximum installed capacity corresponding to the available area. Furthermore, in actual projects, wind turbines, PV modules, and energy storage units usually exist in standard specifications, and capacity configuration may need to conform to discrete equipment specifications. In this case, integer variables or discrete value constraints need to be introduced. If the project has a total investment cap, a total investment budget constraint can also be added to ensure that the total equipment investment across all sites does not exceed available funds.

[0096] The lower-level sub-problem model involves various operational constraints of energy management, ensuring that, given capacity configuration and uncertainty scenarios, the optimized energy management strategy is physically realizable and meets operational requirements. This may include at least one of the following constraints.

[0097] Power balance constraints are the most fundamental constraints in energy management, requiring that the input power and output power of each site be equal in every time period. Input power includes actual wind power output, actual photovoltaic power output, energy storage discharge power, power purchased from the external grid, and mutual assistance power input from other sites; output power includes tunnel load, energy storage charging power, mutual assistance power output to other sites, and curtailed power. The sum of input power must equal the sum of output power.

[0098] Energy storage system constraints involve multiple aspects: dynamic state of charge (SOC) update constraints describe the relationship between the energy storage system's charge and time, coupling operational decisions at different times; SOC upper and lower limit constraints require the energy storage system's charge to always be within a safe range, preventing overcharging that could damage the equipment or over-discharging that could affect its lifespan; and charging and discharging power constraints limit the charging and discharging power of the energy storage system at each time period to ensure that it does not exceed its maximum allowable value.

[0099] The tie-line transmission capacity constraint requires that the mutual power transmitted between two sites in each time period cannot exceed the maximum transmission capacity of the tie-line. The external grid power purchase constraint requires that the power purchased by each site from the external grid in each time period cannot exceed the maximum transmission capacity of the line connecting that site to the grid. The curtailment power constraint requires that the curtailed power be non-negative and cannot exceed the sum of the actual output of wind power and solar power in that time period. Furthermore, all decision variables have reasonable value ranges; for example, power variables are typically non-negative, and the state of charge of energy storage is between 0% and 100%.

[0100] The upper-level master problem and lower-level subproblems achieve collaborative optimization through a specific coupling mechanism. After determining the capacity configuration scheme, the upper-level master problem passes it to the lower-level subproblems as fixed parameters. The lower-level subproblems perform operational optimization under this capacity scheme and return the minimum operational risk cost under the worst-case scenario to the upper level for calculating the annualized operational risk cost. Simultaneously, the worst-case scenario identified by the lower-level subproblems is transformed into new constraints added to the upper-level master problem. These new constraints require the upper level to ensure a feasible operational strategy exists under this scenario in subsequent iterations. This coupling mechanism allows the capacity configuration scheme to directly consider the worst-case scenarios that may be encountered during the operational phase, avoiding the disconnect between planning and execution in traditional methods. Through nested iterative solutions using column and constraint generation algorithms, the upper-level master problem and lower-level subproblems continuously exchange information and correct each other, ultimately converging to an optimal capacity configuration scheme capable of handling all potential worst-case scenarios, achieving collaborative optimization of system economy and robustness.

[0101] This embodiment constructs a multi-level coupled uncertainty set to achieve a refined description of the uncertainty in wind and solar power output, thereby improving the economic efficiency of capacity configuration. The uncertainty set constructed in this embodiment includes at least three levels of constraints: the first level uses an infinite norm constraint to limit the output deviation of wind and solar power equipment within a single site during a single time period; the second level uses a 1-norm constraint to limit the sum of deviations in the combined wind and solar power output within each site during the same time period; and the third level uses a 1-norm constraint to limit the sum of deviations in the combined wind and solar power output across all sites during the same time period. These three levels respectively reflect the physical limitations of individual equipment, the complementarity of wind and solar power within the same site, and the spatial clustering effect between multiple sites, avoiding the overly conservative configuration caused by ignoring these physical laws in traditional methods. This effectively reduces equipment investment costs and total lifecycle costs while ensuring safe system operation. Simultaneously, this embodiment introduces a cross-site mutual assistance mechanism, improving the collaborative efficiency between multiple sites and the level of renewable energy consumption. The energy management strategy for the lower-level sub-problem in claim 1 includes cross-site mutual assistance power, allowing power transmission between different sites via tie lines. When a power station's wind and solar power output is insufficient, it can draw power from other stations with surplus power; conversely, when a power station has surplus wind and solar power, it can transmit power to other stations. This multi-site collaborative mechanism reduces dependence on the external power grid and improves the overall utilization efficiency of renewable energy.

[0102] Steps 110 to 170 belong to the planning phase. Based on historical meteorological and traffic flow data, this phase extracts typical daily scenarios through cluster analysis, constructs a multi-level coupled uncertainty set, and employs a two-layer robust optimization framework for capacity configuration optimization. The purpose of the planning phase is to determine the optimal installed capacity of wind turbines, photovoltaics, and energy storage at each site before system construction, enabling the system to cope with various potential uncertainties at the lowest cost throughout its lifecycle. The optimal capacity configuration scheme output in this phase forms the basis for subsequent operation phases and remains unchanged throughout the system's lifespan once determined.

[0103] After planning is completed and capacity configuration is known, the operation phase begins. This phase, based on short-term weather and traffic flow forecasts, and taking the optimal capacity configuration obtained in the planning phase as a premise, constructs an operational uncertainty set and optimizes the actual operation strategy for the next 24 hours through an energy management model. The purpose of the operation phase is to dynamically adjust energy storage charging and discharging, power purchase, and mutual assistance operations based on the latest forecast information, given the known capacity configuration, to optimally address real-time fluctuations in wind and solar power output and ensure a reliable supply of tunnel load. The energy management strategy output in this phase can be directly distributed to the energy management systems of each site for execution. For example, the actual operation strategy can still be determined on a 24-hour basis during the operation phase.

[0104] In some embodiments, method 10 further includes: Step 180: During the operation phase, acquire hourly weather forecast data and traffic flow forecast data for multiple stations along the highway for the next 24 hours; the weather forecast data includes wind speed and sunshine value; the traffic flow forecast data includes vehicle flow value.

[0105] The purpose of this step is to obtain forecast data for the next 24 hours during the actual operation of the system, providing input for intraday energy management. Unlike the planning phase, which uses historical data for cluster analysis, the operation phase uses short-term forecast data provided by meteorological and traffic management departments. This forecast data has higher timeliness and accuracy, reflecting the changing trends of wind speed, sunshine, and traffic flow within the next day. The accuracy of the forecast data directly affects the effectiveness of energy management; therefore, it is usually obtained using numerical weather prediction models or time series forecasting models. The input of this step is the short-term forecast data source for each station, and the output is hourly wind speed, sunshine, and traffic flow values ​​for the next 24 hours, used for the calculation of the operation center point matrix in the next step.

[0106] Step 190: Determine the operation center point matrix based on meteorological forecast data and traffic flow forecast data; the operation center point matrix includes hourly wind power output, photovoltaic power output, and tunnel load values ​​for the next 24 hours.

[0107] The purpose of this step is to convert the predicted meteorological-traffic data into power output-load data and construct a center point matrix for the operational phase. Similar to step 130, wind power output is a function of wind speed and turbine capacity, photovoltaic power output is a function of solar irradiance and photovoltaic capacity, and tunnel load is a function of traffic volume. The turbine capacity, photovoltaic capacity, and energy storage capacity here have already been obtained in step 170 and are deterministic values ​​in the optimal capacity configuration scheme. Therefore, the operational center point matrix is ​​a deterministic power output-load sequence calculated based on the predicted data, given the known capacity configuration, and serves as the benchmark for subsequently constructing the operational uncertainty set. The inputs to this step are the predicted data obtained in step 180 and the optimal capacity configuration scheme obtained in step 170, and the outputs are the hourly wind power output, photovoltaic power output, and tunnel load values ​​for the next 24 hours.

[0108] Step 200: Construct an operational uncertainty set for the operational center point matrix; the operational uncertainty set is used to describe the degree of fluctuation of wind and solar power output based on the operational center point matrix; the operational uncertainty set has the same constraints as the typical daily uncertainty set.

[0109] The purpose of this step is to construct the uncertainty set for the operational phase, which describes the possible fluctuation range of wind and solar power output over the next 24 hours. Similar to step 140, the operational uncertainty set also adopts a multi-level coupled structure: the first level uses infinite norm constraints to limit the hourly output deviation of wind and solar power equipment at each site; the second level uses 1-norm constraints to limit the sum of hourly deviations of combined wind and solar power output at each site; the third level uses 1-norm constraints to limit the sum of hourly deviations of combined wind and solar power output across all sites. Fourth and fifth level constraints may also be included. Unlike the planning phase, the parameters of the operational uncertainty set can be dynamically adjusted based on the latest meteorological forecast error statistics to better match the current forecast accuracy. The input to this step can be the operational center point matrix obtained in step 190, or the operational center point matrix obtained in step 190 and uncertainty parameters based on short-term forecast error statistics; the output is the operational uncertainty set corresponding to the next 24 hours.

[0110] Step 210: Construct an energy management model. The input to the energy management model is the optimal capacity configuration scheme and the set of operational uncertainties. The output is the energy management strategy with the minimum operational risk cost under the most unfavorable scenario in the next 24 hours. The energy management model searches for the most unfavorable scenario in the set of operational uncertainties and optimizes the energy management strategy under this scenario. The energy management strategy includes the energy storage charging and discharging power, curtailed power, power purchased from the external grid, and cross-site mutual assistance power at each site.

[0111] The purpose of this step is to establish an energy management optimization model for the operational phase, used to formulate the actual operational plan for the next 24 hours. Similar to the lower-level sub-problem model in step 160, the energy management model also adopts a max-min two-layer optimization structure: the outer layer searches for the most unfavorable scenario that maximizes operational risk cost within the set of operational uncertainties, while the inner layer optimizes the energy management strategy to minimize operational risk cost under this most unfavorable scenario. The difference between the two is that the lower-level sub-problem in the planning phase runs on multiple typical days to evaluate the economics of the capacity plan; while the energy management model in the operational phase runs on current forecast data to formulate actual dispatch instructions. The energy management strategy also includes the energy storage charging and discharging power, curtailed power, power purchased from the external grid, and cross-site mutual assistance power at each site. The inputs to this step are the optimal capacity configuration scheme obtained in step 170 and the set of operational uncertainties obtained in step 200, and the output is the energy management optimization model to be solved.

[0112] Step 220: Solve the energy management model to obtain the energy management strategy with the minimum operating risk cost under the most unfavorable scenario in the next 24 hours.

[0113] The purpose of this step is to solve the energy management model constructed in step 210 to obtain an energy management strategy that can be used to guide actual operation. This step differs from the solution in step 170 of the planning phase in that the energy management model in the operation phase does not involve capacity decisions, as the optimal capacity configuration scheme has already been determined in step 170 and used as a fixed input. During the solution process, a column and constraint generation algorithm can be used to iterate between the operational uncertainty set and the operational strategy, or the max-min problem can be transformed into an equivalent single-level optimization problem for direct solution. The energy management strategy corresponding to the minimum operational risk cost under the most unfavorable scenario obtained from the solution is the operational plan that the system should execute in the next 24 hours, including energy storage charging and discharging instructions, power curtailment instructions, power purchase plans, and inter-site power transfer instructions for each site and time period. These instructions can be issued to the energy management systems of each site for execution, realizing a dynamic response to fluctuations in wind and solar power output.

[0114] This embodiment has the following beneficial effects: First, it achieves an organic connection between planning and operation, ensuring the effective implementation of capacity allocation schemes in actual operation. The operation phase is based on the optimal capacity allocation scheme obtained in the planning phase, and all operational decisions are optimized within the physical capacity defined by this scheme. This closed-loop mechanism of "planning guides operation, and operation verifies planning" avoids the problem of a disconnect between planning schemes and operational strategies in traditional methods, enabling early optimization investments to truly translate into actual operational benefits.

[0115] Second, by dynamically optimizing operational strategies based on short-term forecast data, the system's adaptability to wind and solar power fluctuations is improved. During operation, short-term weather and traffic flow forecasts for the next 24 hours are used, offering higher timeliness and accuracy compared to historical statistical data used during the planning phase. The energy management model dynamically adjusts energy storage charging and discharging, power purchase, and mutual assistance operations based on the latest forecast information, enabling more precise responses to real-time fluctuations in wind and solar power output, reducing wind and solar curtailment, and lowering operating costs.

[0116] Third, it continues the detailed description of multi-level uncertainty sets, enhancing the robustness of operational strategies. The operational uncertainty set constructed during the operation phase maintains the same multi-level coupling structure as the planning phase, also considering the physical limitations of individual devices, the complementarity of wind and solar power within the same site, and the spatial clustering effect between multiple sites. The energy management model searches for the most unfavorable scenario and optimizes the operational strategy within this uncertainty set, ensuring that the formulated scheduling instructions can still guarantee load supply even if forecasts deviate, thus improving the reliability of system operation.

[0117] Fourth, the rolling optimization mechanism supports continuous operation and adapts to long-term operational needs. In practical applications, steps 180 to 220 can be executed on a rolling basis according to a fixed time cycle (e.g., daily). Each time new forecast data is obtained, the energy management model is re-solved, and the control commands for the next 24 hours are updated. This rolling optimization mechanism enables the system to continuously adapt to changes in weather conditions and traffic flow, achieving long-term stable operation.

[0118] In some embodiments, the tunnel load value includes a rigid load and a flexible load, wherein the flexible load is a reduceable load and the rigid load is a non-reduceable load; method 10 further includes: Step 230: Determine the safety margin for each typical day. The safety margin is the difference between the normal power of the flexible load at each site and the minimum safe power required for safe tunnel operation. Step 240: Based on the safety margin of each typical day, determine the feasible scheduling interval for that typical day. The feasible scheduling interval is the range of tunnel load values ​​for each station. The upper limit of the feasible scheduling interval for that typical day is the tunnel load value of the typical day center point matrix, and the lower limit of the feasible scheduling interval for that typical day is the upper limit of the feasible scheduling interval for that typical day minus the safety margin constraint of the typical day. Among them, the range of values ​​for the concentrated tunnel load on a typical day is the feasible scheduling range for a typical day.

[0119] In this embodiment, in order to make full use of the adjustable capability of the load side and further improve the system economy and operational flexibility, the present invention performs refined modeling of tunnel load and introduces the concept of safe operating margin.

[0120] In this embodiment, tunnel loads are divided into two categories: rigid loads and flexible loads. Rigid loads refer to loads that must be unconditionally met to ensure the safe operation of the tunnel, such as emergency lighting, monitoring systems, and fire-fighting equipment. These loads cannot be reduced under any circumstances. Flexible loads refer to loads that can be adjusted within a certain range according to the system's operating status, such as dimmable lighting systems and ventilation equipment that can adjust airflow according to traffic volume. These loads have a certain degree of adjustment range while ensuring safety.

[0121] By classifying and modeling the load, we can more accurately depict the actual power consumption characteristics of the tunnel, creating conditions for utilizing load-side regulation capabilities to cope with the uncertainties of wind and solar power.

[0122] This embodiment introduces the concept of a safety margin to quantify the adjustability of the flexible load at each site on a typical day. The safety margin is defined as the difference between the normal power of the flexible load at each site and the minimum safe power required for the safe operation of the tunnel. Physically, the safety margin represents the maximum power that the flexible load can be reduced to while ensuring the safe operation of the tunnel. A larger safety margin indicates a stronger load adjustment capability at that site on a typical day; a smaller safety margin indicates a more limited load adjustment space.

[0123] The normal power of the flexible load refers to the conventional operating power calculated based on typical daily traffic flow data, which is the flexible load portion corresponding to the tunnel load value in the typical daily center point matrix. The minimum safe power required for safe tunnel operation refers to the minimum power that the flexible load must maintain under the premise of ensuring the basic safety functions of the tunnel, such as the minimum illuminance requirement of the lighting system and the minimum air change rate requirement of the ventilation system.

[0124] Based on the safety margin, the feasible scheduling range of tunnel load for each station on each typical day can be determined. The upper limit of the feasible scheduling range is the tunnel load value in the center point matrix of that typical day, that is, the tunnel load value when the flexible load is running at normal operating power; the lower limit of the feasible scheduling range is the upper limit minus the safety margin of that typical day, that is, the lowest power level that the flexible load can be reduced to under the premise of ensuring safety.

[0125] The mathematical expression for the feasible scheduling interval is as follows: the actual value of the tunnel load can continuously take values ​​between the lower and upper limits, and the specific value is determined by the energy management strategy based on the system's operational needs. When the system's wind and solar power output is sufficient, the load can be allowed to operate near the upper limit to provide better service; when the system's wind and solar power output is insufficient, the flexible load can be appropriately reduced to near the lower limit to reduce dependence on energy storage and electricity purchase.

[0126] In this embodiment, the typical daily uncertainty set not only includes the uncertainty dimension of wind and solar power output, but also incorporates tunnel load into the uncertainty search space. Specifically, the range of tunnel load values ​​in the uncertainty set is the aforementioned feasible scheduling range.

[0127] This means that when searching for the worst-case scenario in the outer layer of the lower-level subproblem, not only can the output deviations of wind power and photovoltaics be selected, but also the actual value of the tunnel load. The worst-case scenario can be a combination of "worst wind power output, worst photovoltaic output, and maximum load value," or other combinations that maximize system operating costs. This design is closer to reality; for system operation, the most difficult situation is often when supply is at its lowest and demand is at its highest.

[0128] Meanwhile, since reducing flexible loads can be costly, a load reduction penalty cost needs to be introduced into the inner-layer minimization problem. When the inner layer chooses to reduce flexible loads, a corresponding penalty cost can be incurred, thus balancing the economics of load reduction with other operational methods (such as electricity purchase and energy storage discharge) in the optimization process. The penalty cost for reducing flexible loads can be calculated based on the actual amount of flexible load reduction and the reduction penalty coefficient.

[0129] This embodiment fully utilizes the load-side adjustment potential, providing a new means for the system to cope with the uncertainties of wind and solar power. When wind and solar power output is insufficient, the demand for electricity purchases and the pressure on energy storage discharge can be reduced by appropriately cutting flexible loads, thereby improving the system's self-balancing capability. The safety boundary of load regulation is quantified through a safety operation margin, ensuring that load reduction always occurs within the range that guarantees the safe operation of the tunnel, avoiding the problem of sacrificing safety for the sake of economy. Including the load in the search space of the uncertainty set makes the characterization of the worst-case scenario more complete and realistic. The system needs to consider the extreme combination of "worst supply and greatest demand" during the planning stage, resulting in a more robust capacity configuration. When operating costs include load reduction penalty costs, the relationship between economy and service quality is balanced through the load reduction penalty cost, enabling the optimization model to make the optimal trade-off between load reduction and other operating methods based on actual conditions.

[0130] Correspondingly, a safety margin can also be introduced during the runtime phase. In some embodiments, method 10 further includes: Step 250: Determine the safety margin for the next 24 hours.

[0131] Step 260: Based on the safety margin for operation in the next 24 hours, determine the feasible scheduling interval for the next 24 hours. The upper limit of the feasible scheduling interval for the next 24 hours is the tunnel load value of the typical day center point matrix of the typical day, and the lower limit of the feasible scheduling interval for the next 24 hours is the upper limit of the feasible scheduling interval for the typical day minus the safety margin constraint for operation in the next 24 hours.

[0132] Among them, the range of values ​​for the concentrated tunnel load value with operational uncertainty is the feasible scheduling range within the next 24 hours.

[0133] In step 250, similar to the planning phase, the safe operating margin in the operation phase is defined as the difference between the normal power of the flexible load at each site and the minimum safe power required for the safe operation of the tunnel.

[0134] Step 260: Determine the feasible scheduling interval for the next 24 hours based on the safety margin for operation within the next 24 hours. The upper limit of the feasible scheduling interval is the tunnel load value in the operation center point matrix, that is, the tunnel load value when the flexible load is running at normal operating power; the lower limit of the feasible scheduling interval is the upper limit minus the safety margin for that period, that is, the lowest power level that the flexible load can be reduced to under the premise of ensuring safety.

[0135] Unlike the planning phase, the feasible scheduling range in the operation phase is dynamically calculated based on short-term forecast data, reflecting changes in load adjustment capacity at different times within the next 24 hours. For example, during peak traffic periods, the safety margin may be small, and the load adjustment space may be limited; during off-peak traffic periods, the safety margin may be large, and the load adjustment space may be ample. This dynamic characteristic enables energy management models to utilize load-side resources more accurately.

[0136] In this embodiment, the operational uncertainty set not only includes the uncertainty dimension of wind and solar power output, but also incorporates the actual value of tunnel load into the uncertainty search space. Specifically, the range of tunnel load values ​​in the operational uncertainty set is the feasible scheduling range for the next 24 hours determined in step 260.

[0137] This means that when searching for the worst-case scenario, the outer layer of the energy management model can select not only the output deviation of wind and solar power, but also the actual value of the tunnel load. The worst-case scenario can be a combination of "worst wind power output, worst solar power output, and maximum load value," or other combinations that maximize system operating costs. This design allows the energy management model to fully consider the impact of load-side uncertainties on system operation and formulate more robust energy management strategies.

[0138] Meanwhile, since the reduction of flexible loads can generate corresponding penalty costs (such as affecting service quality or user comfort), the inner layer of the energy management model needs to make an economic trade-off between load reduction and other operating methods (such as electricity purchase, energy storage discharge, and cross-site mutual assistance) when optimizing operating strategies, so as to minimize the total operating cost.

[0139] This embodiment incorporates safe operating margins and feasible dispatch intervals into the planning and operation phases, integrating the adjustability of flexible loads into the optimization framework. This has the following beneficial effects: In the planning phase, by considering load-side adjustment capabilities, the system's capacity configuration redundancy is further reduced, improving investment economics; in the operation phase, by dynamically adjusting flexible loads, a new control method is provided to address real-time fluctuations in wind and solar power, reducing electricity purchase demand and energy storage call-up; simultaneously, by including loads in the search space of the uncertainty set, the most unfavorable scenario is more comprehensively characterized, significantly enhancing the system's safety assurance capability under extreme operating conditions.

[0140] Principles and steps not explicitly described in this invention are all obtainable by those skilled in the art through conventional technical means, and therefore will not be elaborated upon. Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A two-layer robust computation method for multi-site distributed energy systems on highways, characterized in that, Includes the following steps: The system acquires hourly historical meteorological data and historical traffic flow data for one year from multiple stations along the highway; each of the multiple stations is equipped with wind power equipment, photovoltaic equipment, energy storage equipment, and a tunnel; the multiple stations can perform cross-station mutual assistance; the historical meteorological data includes wind speed values ​​and sunshine values; the historical traffic flow data includes vehicle flow values. Cluster analysis was used to extract typical day data for S typical days within a year and the annual occurrence probability of each typical day; the data for each typical day included hourly wind speed, light intensity and traffic flow values ​​for each station within 24 hours. For each typical day, the hourly wind speed, solar irradiance, and traffic flow values ​​at each station are used to determine the typical day center point matrix. This matrix includes the hourly wind power output, solar PV output, and tunnel load values ​​at each station for that typical day. The wind power output is a function of wind speed and turbine capacity; the solar PV output is a function of solar irradiance and solar PV capacity; and the tunnel load value is a function of traffic flow. For each typical day, a typical day uncertainty set is constructed based on the typical center point matrix. The typical day uncertainty set is used to describe the fluctuation of wind and solar power output based on the typical day center point matrix. The typical day uncertainty set includes at least three levels of constraints. The first level uses infinite norm constraints to limit the hourly output deviation of wind power equipment and photovoltaic equipment at each site respectively; the second level uses 1-norm constraints to limit the total hourly output deviation of combined wind and solar power at each site; the third level uses 1-norm constraints to limit the total hourly output deviation of combined wind and solar power at all sites. With the goal of minimizing the annualized cost of the entire life cycle of the energy system, a higher-level master problem model for capacity allocation is constructed. The annualized cost includes equipment investment cost, operation and maintenance cost, and annualized operating risk cost. The annualized operating risk cost is the weighted sum of the minimum operating risk cost under the most unfavorable scenario for each typical day and the probability of occurrence for each typical day. The decision variables of the higher-level master problem model are the wind turbine capacity, photovoltaic capacity, and energy storage capacity of each site. For the capacity configuration scheme given by the upper-level main problem model, a lower-level sub-problem model is constructed for the current typical day. The lower-level sub-problem model is used to evaluate the operational risk cost of the capacity configuration scheme under the current typical day. Its input is the capacity configuration scheme given by the upper level and the typical day uncertainty set of the current typical day, and its output is the minimum operational risk cost under the most unfavorable scenario of the current typical day. The lower-level sub-problem model searches for the most unfavorable scenario in the typical day uncertainty set corresponding to the current typical day, and optimizes the energy management strategy under the scenario. The energy management strategy includes the energy storage charging and discharging power, abandoned power, external grid power purchase power, and cross-site mutual assistance power of each site. The column and constraint generation algorithm is used to perform nested iterative solutions on the upper-level main problem model and the lower-level sub-problem model to obtain the optimal capacity configuration scheme.

2. The method according to claim 1, characterized in that, The typical daily uncertainty set also includes fourth-level and fifth-level constraints; The fourth level adopts a 1-norm constraint to limit the total output deviation of the combined wind and solar power output of each station within 24 hours; the fifth level adopts a 1-norm constraint to limit the total output deviation of the combined wind and solar power output of all stations within 24 hours.

3. The method according to claim 1, characterized in that, The method further includes: During the operation phase, hourly weather forecast data and traffic flow forecast data for the next 24 hours are acquired from multiple stations along the highway. The weather forecast data includes wind speed and sunshine values, and the traffic flow forecast data includes vehicle flow values. Based on the meteorological forecast data and traffic flow forecast data, an operation center point matrix is ​​determined; the operation center point matrix includes hourly wind power output, photovoltaic power output, and tunnel load values ​​for the next 24 hours. For the aforementioned operational center point matrix, an operational uncertainty set is constructed; the operational uncertainty set is used to describe the degree of fluctuation of wind and solar power output based on the operational center point matrix; the operational uncertainty set has the same constraints as the typical daily uncertainty set; Develop an energy management model; The energy management model takes the optimal capacity configuration scheme and the set of operational uncertainties as input, and outputs the energy management strategy with the minimum operational risk cost under the most unfavorable scenario in the next 24 hours. The energy management model searches for the most unfavorable scenario in the set of operational uncertainties and optimizes the energy management strategy under the scenario. The energy management strategy includes the energy storage charging and discharging power, curtailed power, power purchased from the external grid, and cross-site mutual assistance power at each site.

4. The method according to claim 3, characterized in that, The tunnel load value includes rigid load and flexible load, wherein the flexible load is a reduceable load and the rigid load is a non-reduceable load; the method further includes: Determine the safety margin for each typical day, which is the difference between the normal power of the flexible load at each site and the minimum safe power required for safe tunnel operation; Based on the safety margin of each typical day, the feasible scheduling interval for that typical day is determined. The feasible scheduling interval is the range of tunnel load values ​​for each station. The upper limit of the feasible scheduling interval for that typical day is the tunnel load value of the typical day center point matrix, and the lower limit of the feasible scheduling interval for that typical day is the upper limit of the feasible scheduling interval for that typical day minus the safety margin constraint of that typical day. Among them, the range of values ​​for the concentrated tunnel load on a typical day is the feasible scheduling range for a typical day.

5. The method according to claim 4, characterized in that, The method further includes: Determine the safety margin for the next 24 hours; Based on the safety margin for the next 24 hours, a feasible scheduling interval for the next 24 hours is determined. The upper limit of the feasible scheduling interval for the next 24 hours is the tunnel load value of the typical day center point matrix of the typical day, and the lower limit of the feasible scheduling interval for the next 24 hours is the upper limit of the feasible scheduling interval for the typical day minus the safety margin constraint for the next 24 hours. The range of values ​​for the concentrated tunnel load value with operational uncertainty is the feasible scheduling range for the next 24 hours.

6. The method according to claim 5, characterized in that, The operational risk costs include: the cost of purchasing electricity from the external power grid, the penalty costs for wind and solar power curtailment, the load gap penalty costs, and the penalty costs for reducing flexible loads; the load gap penalty costs are calculated by multiplying the tunnel load deficit of each site at each time period by a preset gap penalty coefficient; the penalty costs for reducing flexible loads are calculated based on the actual reduction amount of flexible loads and the reduction penalty coefficient.

7. The method according to claim 1, characterized in that, The operational risk costs include: the cost of purchasing electricity from the external power grid, the penalty costs for wind and solar power curtailment, and the load gap penalty costs; the load gap penalty costs are calculated by multiplying the tunnel load deficit of each site at each time period by a preset gap penalty coefficient.

8. The method according to claim 1, characterized in that, The process involves acquiring historical meteorological and traffic flow data from multiple stations along the highway, and using cluster analysis to extract S typical days within a year and their corresponding annual occurrence probabilities, including: The wind speed, light intensity, and traffic flow values ​​of all stations for each day are arranged in 24-hour time sequence and concatenated into a high-dimensional daily feature vector. Normalize the different types of data in the high-dimensional daily eigenvector to obtain the normalized eigenvector; Clustering of 365 normalized feature vectors yields several cluster center vectors, each representing a typical daily weather-traffic pattern. The cluster center vectors are denormalized to obtain the typical daily data; The ratio of the number of days contained in the cluster containing the cluster center to the total number of days in the year is used as the annual probability of a typical day.

9. The method according to claim 1, characterized in that, The lower-level sub-problem model searches for the most unfavorable scenario in the typical day uncertainty set corresponding to the current typical day, and optimizes the energy management strategy under this scenario, including: The lower-level sub-problem model adopts a max-min two-level optimization structure, which searches for the most unfavorable scenario that maximizes the risk cost of energy system operation in the typical day uncertainty set corresponding to the current typical day. Under the most unfavorable scenario given in the outer layer, optimize the energy management strategy to minimize operational risk and cost.

10. The method according to claim 1, characterized in that, A column and constraint generation algorithm is used to perform nested iterative solutions on the upper-level main problem model and the lower-level sub-problem model to obtain the optimal capacity configuration scheme, including: During the initialization phase, the lower bound is set to negative infinity, the upper bound to positive infinity, the number of iterations is initialized to zero, and the set of worst-case scenarios is initially empty. In the main problem-solving phase, the upper-level main problem model is solved under the current set of worst-case scenarios to obtain candidate capacity configuration schemes and their objective function values, and the lower bound is updated. In the sub-problem solving phase, for each typical day, the lower-level sub-problem model is solved under the candidate capacity configuration scheme to identify the most unfavorable scenario of the typical day and its corresponding minimum operating risk cost. Based on the weighted sum of the minimum operating risk cost and the probability of occurrence of all typical days, combined with the investment cost and operation and maintenance cost of the current candidate capacity configuration scheme, the upper bound of the total cost corresponding to the current capacity configuration scheme is calculated. During the convergence judgment phase, the relative error between the current upper bound and the lower bound is calculated. If the error is less than the preset convergence threshold, the iteration is stopped and the current candidate capacity configuration scheme is output as the optimal capacity configuration scheme; otherwise, the worst-case scenarios of all typical days newly identified in this iteration are added as new constraints to the worst-case scenario set of the upper-level main problem model for the next round of iteration.

Images