Highway tunnel group-oriented wind-solar complementary confidence interval prediction method and system

Abstract

The application discloses a kind of wind and light complementary confidence interval prediction method and system for highway tunnel group, and relates to microgrid technical field.The method includes: obtaining tunnel group geospatial, meteorological, historical error and real-time energy consumption data;Construct node graph and map meteorological data to each portal micro-meteorological sequence;Output wind and light marginal probability distribution through wind and light prediction model;Update node graph to obtain dynamic dependent topology, reconstruct correlation to generate wind and light joint probability distribution;Extract initial confidence interval, calibrate the interval of boundary node based on meteorological scenario and historical error to obtain target confidence interval;According to energy consumption data, construct safety load constraint, convert target confidence interval into safety power supply interval to generate safety power supply scheduling decision.The application effectively improves the prediction reliability under extreme weather and the power supply safety of physical microgrid.

Description

Technical Field

[0001] This invention relates to the field of microgrid technology, and in particular to a method and system for predicting the confidence interval of wind-solar hybrid systems for highway tunnel groups. Background Technology

[0002] With the green and low-carbon transformation of transportation infrastructure, deploying wind and photovoltaic power generation systems along highways and in tunnel clusters to achieve local energy consumption has become an important development trend. Due to the complex geographical environment and strip-shaped distribution of highway tunnel clusters, the power supply reliability requirements are extremely high. Therefore, accurate prediction of wind and solar power output is fundamental to ensuring the stable operation of the tunnel cluster microgrid and energy management and dispatch.

[0003] Currently, forecasting for wind and solar power generation mainly focuses on deterministic point forecasting or conventional probability interval forecasting. However, in practical applications, existing forecasting methods have certain technical limitations. First, most existing methods directly use mesoscale regional meteorological data as model input, ignoring the cutting and disturbance effects of complex micro-topography such as canyons and mountains where tunnel entrances are located on meteorological conditions, resulting in insufficient accuracy of meteorological input data. Second, existing joint forecasting models often treat each wind and solar power station as an independent unit, or only use static spatial topology matrices for modeling, failing to effectively capture the dynamic spatiotemporal dependencies between different nodes of a tunnel group when complex weather systems move, thus limiting the prediction accuracy of joint probability distributions.

[0004] Furthermore, in constructing prediction intervals, traditional methods typically employ a globally uniform error distribution assumption, lacking refined processing for different specific meteorological scenarios. When faced with drastic changes in meteorological conditions, prediction intervals are prone to consecutive omissions or boundary violations, rendering the interval predictions worthless. Finally, existing wind and solar forecasting systems are often independent of the actual energy consumption scenarios of highways, failing to integrate the predicted power uncertainty with rigid safety loads such as lighting and ventilation inside tunnels. This makes the generated prediction results difficult to directly apply to the safe power supply scheduling of microgrids. Therefore, there is an urgent need for a wind-solar complementary forecasting scheme that can adapt to complex micro-meteorology, characterize dynamic spatiotemporal correlations, and link with tunnel operating loads. Summary of the Invention

[0005] This invention provides a method and system for predicting the confidence interval of wind-solar hybrid systems for highway tunnel groups, which solves the problems of prediction being detached from complex micro-topography, lack of dynamic spatiotemporal correlation and load linkage in existing technologies.

[0006] The first aspect of this invention proposes a method for predicting the confidence interval of wind-solar hybrid power systems for highway tunnel groups, comprising the following steps: Obtain geospatial information, regional meteorological data, historical error data, and real-time energy demand data for highway tunnel groups; An initial tunnel group node map is constructed based on geospatial information, and regional meteorological data is mapped into micro-meteorological sequences for each tunnel entrance node; Input the micro-meteorological sequence into the wind and solar prediction model, and output the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power at each tunnel entrance node; The dynamic dependent topology is obtained by updating the initial tunnel group node map based on regional meteorological data. The correlation between the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power is reconstructed by combining the dynamic dependent topology, and the joint probability distribution of wind and solar power at each tunnel entrance node is generated. The initial confidence interval is extracted from the wind-solar joint probability distribution. The current meteorological scenario is determined based on regional meteorological data. Based on historical error data, it is identified whether each tunnel entrance node has continuously crossed the boundary under the current meteorological scenario. The tunnel entrance nodes that have continuously crossed the boundary are identified as boundary crossing nodes. The initial confidence interval is locally calibrated for the boundary crossing nodes to obtain the target confidence interval. Based on real-time energy demand data, a safe load constraint is constructed, and the target confidence interval is combined with the safe load constraint to transform it into a safe power supply interval, so as to generate a safe power supply scheduling decision.

[0007] In one optional implementation, an initial tunnel group node map is constructed based on geospatial information, and regional meteorological data is mapped to micro-meteorological sequences for each tunnel entrance node, specifically including: Each tunnel entrance node is abstracted as a spatial node. The three-dimensional coordinates of each spatial node are extracted from the geospatial information, and the static connection weights are calculated based on the spatial distance characteristics between each spatial node to construct the initial tunnel group node diagram. Regional meteorological data includes observation data from surrounding meteorological stations and regional numerical weather prediction data. The observation data from surrounding meteorological stations and regional numerical weather prediction data are input into a pre-trained meteorological mapping model. The surrounding terrain features of each tunnel entrance node are extracted from geospatial information. The output of the meteorological mapping model is downscaled and corrected based on the surrounding terrain features to generate a micro-meteorological sequence containing actual meteorological elements.

[0008] In one optional implementation, the micro-meteorological sequence is input into the wind-solar prediction model, and the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node are output, specifically including: The historical actual power output sequence and wind and solar equipment operating status parameters of each tunnel portal node are obtained from the data acquisition terminals deployed at each tunnel portal node; Micro-meteorological sequences, historical actual power output sequences, and wind and solar equipment operating status parameters are input into a wind and solar prediction model that includes wind power prediction and solar power prediction branches. In the wind power prediction branch, continuous distribution inference is performed through a pre-set probability density estimation network, and the output probability density function is used as the marginal probability distribution of wind power. In the photovoltaic prediction branch, discrete quantiles are mapped through a preset quantile network, and multiple conditional quantile sequences under preset confidence levels are output as the photovoltaic marginal probability distribution.

[0009] In one alternative implementation, before outputting the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node in the wind-solar prediction model, the method further includes: In the wind power prediction branch, the local wind speed mutation features of the micro-meteorological sequence in the time dimension are extracted by the preset time series feature extraction module, and the local wind speed mutation features are used as conditional variables input into the probability density estimation network. In the photovoltaic prediction branch, the spatial correlation features of the micro-meteorological sequence on the initial tunnel group node map are extracted using a preset spatial feature extraction module. Extract the correlation of photovoltaic power output delay caused by mountain shadow between nodes at each tunnel entrance. Use the preset attention mechanism module to perform cross-branch feature splicing of spatial correlation features, photovoltaic power output delay correlation and local wind speed change features to form a fused spatiotemporal representation vector. The fused spatiotemporal representation vector is input into the hidden layer of the quantile network to capture spatiotemporal complementarity.

[0010] In one optional implementation, the initial tunnel group node graph is updated based on regional meteorological data to obtain a dynamic dependent topology. The correlation between the marginal probability distributions of wind power and photovoltaic power is then reconstructed using this dynamic dependent topology to generate the joint wind-solar probability distribution for each tunnel entrance node. Specifically, this includes: Extract cloud movement trajectories, prevailing wind direction vectors, and real-time weather type identifiers from regional meteorological data; Based on the real-time weather type identifier, the corresponding preset map update strategy is activated. Combined with the cloud movement trajectory and the prevailing wind direction vector, the edge weights of the initial tunnel group node graph are dynamically adjusted to obtain dynamic edge weights. Based on the dynamic edge weights, a dynamic adjacency matrix as a dynamic dependent topology is generated. Based on the real-time weather type identifier, the corresponding conditional joint distribution function is selected from the preset joint distribution function candidate library, and the wind-solar complementarity coefficient between adjacent tunnel portal nodes and the consistency of tunnel portal wind direction are calculated based on the micro-meteorological sequence. By utilizing the dynamic adjacency matrix, the wind-solar complementarity coefficient, and the wind direction consistency at the tunnel entrance, the correlation parameters of the selected conditional joint distribution function are dynamically corrected. By using the modified conditional joint distribution function, multidimensional joint sampling of the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node is performed to obtain the joint probability distribution of wind and solar power.

[0011] In one optional implementation, an initial confidence interval is extracted from the wind-solar joint probability distribution, specifically including: Extract the prediction time-domain features, power output fluctuation rate, and error variance features of the wind-solar joint probability distribution; Obtain a pre-built library of candidate interval models; The severity index of meteorological change is extracted from regional meteorological data. Based on the severity index of meteorological change, prediction time domain characteristics, power output fluctuation rate, and error variance characteristics, the fitness score of each candidate interval model in the candidate interval model library is calculated. Select the target interval model with the highest fitness score for each tunnel entrance node; The probability density integral of the wind-solar joint probability distribution is solved by using the target interval model, and the upper and lower limits of the specified confidence level are extracted to form the initial confidence interval.

[0012] In one optional implementation, the current meteorological scenario is determined based on regional meteorological data. Historical error data is used to identify whether each tunnel entrance node has continuously crossed boundaries under the current meteorological scenario. Tunnel entrance nodes that have continuously crossed boundaries are identified as boundary-crossing nodes. The initial confidence interval is then locally calibrated for the boundary-crossing nodes to obtain the target confidence interval, specifically including: Obtain a pre-established database of typical weather scenarios, perform pattern matching on regional meteorological data based on the database of typical weather scenarios, and determine the current meteorological scenario to which the current meteorological conditions belong. Historical actual output data is extracted from historical error data, and the actual interval coverage and average interval width of each tunnel portal node under the current weather scenario are calculated. The number of missed packets in the historical actual output data of each tunnel entrance node within the preset time window that exceeds the initial confidence interval is counted. When the number of missed packets of a specific tunnel entrance node under the current weather scenario exceeds the preset tolerance threshold, the specific tunnel entrance node is determined to be an out-of-bounds node. The interval scale magnification factor of the out-of-bounds node under the current meteorological scenario is increased online, and the upper and lower limits of the initial confidence interval are shifted and corrected using the updated interval scale magnification factor to obtain the target confidence interval.

[0013] In one optional implementation, a safe load constraint is constructed based on real-time energy demand data. The target confidence interval is then combined with the safe load constraint to transform it into a safe power supply interval, thereby generating a safe power supply scheduling decision. Specifically, this includes: Obtain the operating status and backup power capacity of energy storage devices configured in a microgrid system located in a highway tunnel complex; Real-time energy demand data is divided into loads of different priorities according to safety importance, and the rated power corresponding to the rigid load with the highest priority is extracted to construct safety load constraints and determine the rigid load bottom line corresponding to the safety load constraints. Subtract the rigid load floor from the lower and upper limits of the target confidence interval to obtain the safe power supply interval; The lower limit of the safe power supply range is compared with the preset safe power consumption threshold. Based on the shortage difference and the operating status of the energy storage device and the backup power capacity, the discharge command and the start-up capacity allocation command are calculated. The upper limit of the safe power supply range is compared with the preset grid absorption limit, and the charging command and energy curtailment control command are calculated based on the surplus difference of the comparison. It summarizes and coordinates discharge commands, start-up capacity allocation commands, charging commands, and energy curtailment control commands to generate safe power supply scheduling decisions to guide energy flow in microgrid systems.

[0014] In one alternative implementation, after generating a secure power supply scheduling decision, the method further includes: The safe power supply dispatch decision is sent to the microgrid control terminal in the microgrid system for physical execution; During the physical execution of the microgrid control terminal, the actual charging and discharging power of the energy storage device and the frequency and voltage fluctuation data of the microgrid are collected in real time. The deviation between actual charging and discharging power and safe power supply dispatch decisions is compared and analyzed, and the microgrid frequency and voltage fluctuation data are transformed into over-limit penalty feedback items; The deviation and the over-limit penalty feedback term are combined to construct a closed-loop loss function; Backpropagation updates are performed using a closed-loop loss function to dynamically reduce the redundancy width of the target confidence interval for each tunnel entrance node in the next scheduling period, thus forming a physical control closed loop.

[0015] A second aspect of this invention proposes a wind-solar hybrid confidence interval prediction system for highway tunnel groups, comprising: The data acquisition module is used to acquire geospatial information of the highway tunnel group, regional meteorological data, historical error data, and real-time energy demand data. The micro-meteorological mapping module is used to construct an initial tunnel group node map based on geospatial information and map regional meteorological data into micro-meteorological sequences for each tunnel entrance node; The marginal distribution prediction module is used to input micro-meteorological sequences into the wind and solar prediction model and output the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node. The joint probability generation module is used to update the initial tunnel group node map based on regional meteorological data to obtain dynamic dependent topology, and to reconstruct the correlation between the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power based on the dynamic dependent topology, thereby generating the joint probability distribution of wind and solar power at each tunnel entrance node. The interval local calibration module is used to extract the initial confidence interval from the wind-solar joint probability distribution, determine the current meteorological scenario based on regional meteorological data, identify whether each tunnel portal node has continuously crossed the boundary under the current meteorological scenario based on historical error data, determine the tunnel portal nodes that have continuously crossed the boundary as boundary crossing nodes, and perform local calibration on the initial confidence interval for the boundary crossing nodes to obtain the target confidence interval. The safe power supply decision module is used to construct safe load constraints based on real-time energy demand data, and to convert the target confidence interval into a safe power supply interval by combining the safe load constraints, so as to generate safe power supply scheduling decisions.

[0016] Beneficial effects: This invention provides a method and system for predicting the confidence interval of wind-solar hybrid power generation in highway tunnel clusters. By introducing a micro-meteorological mapping mechanism from regional numerical weather prediction data and constructing a node graph using geospatial information, it improves the distortion of meteorological input under complex terrain and enhances the accuracy of basic data. By updating the dynamic dependent topology using meteorological data, it effectively captures the dynamic spatiotemporal correlation characteristics of wind and solar power output between different nodes in the tunnel cluster, improving the prediction accuracy of the joint probability distribution. A two-layer local calibration mechanism based on meteorological scenarios and historical errors is established, enabling directional correction for nodes experiencing continuous boundary violations. This alleviates the problem of missing packets in a single static interval under complex weather conditions and enhances the reliability and adaptability of the confidence interval. The calibrated target confidence interval is integrated with the real-time energy load constraints of the highway tunnels, transforming uncertain power prediction into a safe and achievable power supply interval. This provides a reliable basis for making reasonable power supply scheduling decisions and effectively ensures the stable operation of the energy system of the highway tunnel cluster. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 A flowchart illustrating the steps of a method for predicting the confidence interval of wind-solar hybrid tunnel groups for highways, provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the functional modules of a wind-solar hybrid confidence interval prediction system for highway tunnel groups provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] The present invention will be further described below with reference to the accompanying drawings.

[0020] Reference Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0021] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (Wi-Fi). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0022] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0023] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and computer programs.

[0024] exist Figure 1In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the wind-solar complementary confidence interval prediction system for highway tunnel groups stored in the memory 1005 through the processor 1001, and executes the wind-solar complementary confidence interval prediction method for highway tunnel groups provided in the embodiment of the present invention.

[0025] like Figure 2 As shown, this embodiment provides a wind-solar hybrid confidence interval prediction method for highway tunnel groups, focusing on solving technical problems in existing technologies such as wind-solar prediction being detached from complex micro-topographical environments, lacking dynamic spatiotemporal dependence characterization, and failing to effectively link with the physical and electrical constraints of microgrids. It includes the following steps S101 to S106: S101 acquires geospatial information, regional meteorological data, historical error data, and real-time energy demand data for the highway tunnel group.

[0026] In the actual implementation process, the acquisition of various heterogeneous data from multiple sources must ensure timeliness and accuracy. All meteorological, geographical, and electricity load data involved have been encrypted, anonymized, and desensitized using the Secure Hash Algorithm (SHA) at the edge acquisition side to remove any attribute identifiers that may be associated with specific individuals or sensitive vehicle identities. The data is strictly limited to maintaining the power supply stability of the public transportation microgrid and improving the utilization rate of renewable energy.

[0027] Step S101 specifically includes the following sub-steps S1011 to S1014: S1011. Obtain geospatial information of the highway tunnel group. Data fusion and extraction are performed using a high-precision Digital Elevation Model (DEM) database and airborne LiDAR (Light Detection and Ranging) scanned point cloud data. The extracted geospatial information covers the three-dimensional coordinates (including longitude, latitude, and elevation) of each tunnel entrance along the highway, the azimuth angle of the entrance (the angle of deviation clockwise from true north as 0 degrees, ranging from [0, 360) degrees), slope aspect (i.e., the projection direction of the slope normal onto the horizontal plane), the slope of the surrounding terrain, and physical characteristic parameters such as the mountain shading angle. To balance the precision of terrain depiction with computational resource consumption, the spatial resolution of the digital elevation model is preferably set to 30m×30m to 90m×90m. During the extraction process, a bilinear interpolation algorithm is used to smooth the discrete elevation points, eliminating noise interference from surface vegetation and man-made structures. The extracted high-precision physical feature parameters will provide physical boundary conditions for subsequent microclimate wind speed downscaling and light and shadow occlusion calculations.

[0028] S1012. Obtain regional meteorological data. Regional meteorological data originates from the public meteorological data interface (Application Programming Interface, API) provided by the meteorological bureau, meteorological satellite remote sensing data streams, and observation data from surrounding meteorological stations deployed within a predetermined physical distance (e.g., a radius of 50 kilometers) along the highway. The regional meteorological data includes mesoscale regional numerical weather prediction (NWP) data, with a spatial grid resolution of 30km×30km or 9km×9km, and a time forecast span covering the next 24 to 72 hours. In addition, it includes macro-meteorological elements observed in real-time by surrounding meteorological stations, including but not limited to: wind speed (m / s) at a height of 10 meters above the ground, wind direction (range: [0, 360) degrees), ambient temperature (°C) at a height of 2 meters above the ground, atmospheric pressure (hPa), relative humidity (%), and total cloud cover. The acquired data sequences were initially spatially gridded and aligned using either Inverse Distance Weighting (IDW) or Kriging Interpolation, and the time sampling step of all meteorological time series data was uniformly resampled to 15 minutes. For missing data caused by communication delays, forward imputation based on time series or linear interpolation methods were used to complete the data.

[0029] S1013. Obtain historical error data. Retrieve data archives from the historical database of the highway microgrid management system for past preset operating cycles (e.g., the past 12 to 36 months). Historical error data includes not only physical measurement records (unit: kW) of the actual active power output of wind power and photovoltaic power generation equipment corresponding to each tunnel entrance node at past times, but also the expected predicted power output of the model within the previous prediction scheduling cycle, and the upper and lower limits of the prediction confidence interval under a set confidence level. A prediction deviation sequence is formed by calculating the difference between the actual output and the expected predicted value at each time step; simultaneously, a sequence of actual interval coverage states composed of Boolean values ​​is generated by determining whether the actual output value falls within the corresponding prediction confidence interval. This historical error data serves as an important data foundation for driving the dynamic local calibration mechanism of the prediction interval in subsequent steps.

[0030] S1014. Acquire Real-Time Energy Demand Data. Real-time energy demand data is acquired using Supervisory Control and Data Acquisition (SCADA), high-precision smart meters, and various sensors deployed within the substations of each highway tunnel and the physical environment of the microgrid. The sampling rate is between 1Hz and 10Hz. This data includes the total active and reactive power consumed by the highway tunnel complex at the current moment. Due to the significant physical rigidity and safety constraints of the tunnel energy load, the load power of each subsystem is analyzed, specifically including the basic lighting system load, the enhanced lighting system load, the jet fan ventilation system load, the fire monitoring system load, and the emergency communication base station load (all in kW). Furthermore, the rate of change of energy consumption trend based on the bidirectional traffic density feedback within the tunnel during the current period is extracted. This high-precision subsystem energy consumption data will provide physical-level numerical support for the final steps of calculating safe load constraints and microgrid power supply dispatch.

[0031] S102, construct an initial tunnel group node map based on geospatial information, and map regional meteorological data into micro-meteorological sequences for each tunnel entrance node.

[0032] Since large-scale macro-climate forecasts cannot directly reflect the microclimate variations caused by topographic relief, this step combines graph theory topology with physical downscaling models to scale regional weather forecasts down to the local space of tunnel entrances, which have complex micro-topographic features. Specifically, this includes sub-steps S1021 to S1023: S1021. Abstract each tunnel entrance node into a spatial node, extract the three-dimensional coordinates of each spatial node from the geospatial information, and calculate the static connection weights based on the spatial distance characteristics between each spatial node to construct an initial tunnel group node diagram.

[0033] When performing graph topology modeling, each tunnel entrance equipped with a wind turbine or distributed photovoltaic array is defined as an independent vertex in the graph data structure. Let the set of vertices be... Extract any two distinct spatial nodes. and The three-dimensional coordinate vectors corresponding to the two nodes are extracted from the geospatial information obtained in step S101. Based on the three-dimensional coordinates, the Euclidean distance between the two nodes is calculated. As a spatial distance feature (unit: km), the static connection weights are calculated using a Gaussian kernel function with distance decay characteristics to describe the correlation decay law of meteorological conditions in spatial distribution. Nodes are defined. With nodes Static connection weights between The calculation formula is as follows:

[0034] In the formula, This represents the static spatial correlation strength between two tunnel portal nodes, with a value range of (0,1]. For nodes With nodes The Euclidean distance between them; This represents the distance attenuation control constant. Based on the empirical law in meteorology that the correlation between microclimates and geographical distance decreases, this constant is used to determine the attenuation of microclimate correlation. The value range is set to 5 to 15. If the physical distance between nodes... If the set meteorological correlation cutoff threshold is exceeded (e.g., a cutoff distance of 20 kilometers), the static connection weight of that edge will be forced. By traversing all node pairs, a static adjacency matrix reflecting the physical spatial proximity of each tunnel entrance node and its correlation with meteorological diffusion is generated, thus completing the construction of the initial tunnel group node graph. This graph structure effectively reduces the sparsity and complexity of subsequent spatial feature extraction calculations.

[0035] S1022. Regional meteorological data includes observation data from surrounding meteorological stations and regional numerical weather prediction data. The observation data from surrounding meteorological stations and regional numerical weather prediction data are input into a pre-trained meteorological mapping model.

[0036] Since coarse-resolution NWP data cannot directly characterize complex mountainous environments, this step utilizes deep learning algorithms to construct a nonlinear meteorological feature transformation and temporal series extrapolation network. The meteorological mapping model employs a deep architecture combining a multilayer feedforward neural network (FNN) and a long short-term memory (LSTM) network. The model's input layer receives standardized real-time observation data feature vectors from surrounding meteorological stations, as well as the meteorological element forecast sequence output by NWP at the corresponding grid points. The hidden layer contains at least two LSTM layers (each with 64 to 128 hidden units) to extract the evolution patterns and long-range dependencies of multivariate meteorological features over time. The LSTM layers are followed by fully connected layers, and nonlinearity is introduced through a rectified linear unit (ReLU) activation function. This meteorological mapping model utilizes historically accumulated meteorological observation and forecast datasets. It underwent offline supervised training using the Adaptive Moment Estimation (Adam) optimizer, employing mean squared error as the loss function and an initial learning rate of 0.001. During online execution, the pre-trained meteorological mapping model performs forward propagation calculations, outputting a preliminary downscaled sequence of meteorological baseline predictions (including baseline wind speed, baseline ambient temperature, and baseline irradiance) to the latitude and longitude positions of each spatial node.

[0037] S1023. Extract the surrounding terrain features of each tunnel entrance node from geospatial information, and downscale the output of the meteorological mapping model based on the surrounding terrain features to generate a micro-meteorological sequence containing actual meteorological elements.

[0038] This step involves physical correction of the output of the purely data-driven model by combining actual fluid dynamics and optical geometry mechanisms.

[0039] First, local dynamic corrections are applied to the wind speed. The azimuth angle of the canyon where the target tunnel entrance is located is extracted from the geospatial information and denoted as... The prevailing wind direction of the regional weather system is denoted as Calculate the absolute angle deviation between the two, denoted as . Because narrow terrain can compress and guide airflow, the formula for calculating wind speed downscaling correction is as follows:

[0040] In the formula, The predicted local micro-wind speed at the tunnel entrance is after physical correction. This is the baseline wind speed prediction value output by the meteorological mapping model; The canyon narrowing effect coefficient is extracted based on the topographic contour contraction and the canyon width-to-height ratio. The value ranges from 0.1 to 0.4, representing the proportion of the squeezing and amplification effect of the narrow canyon topography on wind speed. The wind speed wake attenuation coefficient is determined based on the relationship between the slope aspect of the mountain where the node is located and the prevailing wind direction: if the node is on the windward slope and the airflow is unobstructed, then... Take 1.0; if the node is on the leeward slope, the airflow will experience wake separation and kinetic energy dissipation due to the obstruction of the mountain, then The value ranges from 0.6 to 0.8.

[0041] Secondly, geometric shading correction is applied to the photovoltaic irradiance. Based on the local latitude, longitude, and timestamp, astronomical algorithms are used to calculate the solar altitude angle and solar azimuth angle at the target time. Combining the three-dimensional mountain shading matrix of each tunnel entrance, a ray tracing algorithm is used to determine whether the direct sunlight is blocked by surrounding mountains at the current time. If it is determined to be completely blocked, the direct irradiance component in the total irradiance output by the meteorological mapping model is forcibly reduced to zero, retaining only the scattered irradiance component due to atmospheric scattering; if it is determined to be partially blocked, the direct irradiance component is linearly reduced according to the shading ratio.

[0042] After the above-mentioned downscaling correction, a high-frequency micro-meteorological sequence that closely matches the microenvironment is generated for each tunnel entrance node, including local corrected wind speed, local wind direction, local temperature, and corrected irradiance.

[0043] S103 inputs the micro-meteorological sequence into the wind and solar prediction model and outputs the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power at each tunnel entrance node.

[0044] Because wind power generation exhibits a highly nonlinear fluctuation mechanism due to the cubic relationship of wind speed, while photovoltaic power generation shows a clear periodic cutoff pattern due to the influence of the Earth's rotation and the day-night cycle, a dual-branch structure is adopted to infer the marginal distribution of wind power and photovoltaic power respectively, specifically including sub-steps S1031 to S1038: S1031. Obtain the historical actual power output sequence and wind and solar equipment operation status parameters of each tunnel entrance node from the data acquisition terminals deployed at each tunnel entrance node.

[0045] The system acquires real-time hub speed (rpm), yaw system status and pitch angle (degrees), generator nacelle ambient temperature (°C), and inverter maximum power point tracking DC voltage (V), operating current (A), and photovoltaic module surface photoelectric effect operating temperature (°C) of the wind turbines at each tunnel entrance. These operating status parameters objectively reflect the current electromechanical conversion efficiency and health of the power generation equipment.

[0046] S1032. Input the micro-meteorological sequence, historical actual power output sequence and wind and solar equipment operating status parameters into the wind and solar prediction model that includes wind power prediction branch and photovoltaic prediction branch.

[0047] To avoid gradient explosion during model training, all input data are uniformly normalized using maximum and minimum values ​​before being fed into the dual-branch structure of the model, and are linearly mapped to the standard data range of [-1, 1].

[0048] S1033. In the wind power prediction branch, continuous distribution inference is performed through a preset probability density estimation network, and the output probability density function is used as the marginal probability distribution of wind power.

[0049] The preset probability density estimation network employs the Conditional Normalizing Flow (CNF) algorithm. The core principle of this model is to transform a simple prior basic distribution (such as the standard normal distribution) into a simple basic distribution through a series of invertible and easily computed nonlinear affine transformations using the Jacobian determinant. The gradual mapping transformation results in a complex probability density distribution variable of wind power. The transformation formula for a single-layer affine coupling layer of conditionally normalized flow is as follows:

[0050] In the formula, These are the implicit variables output by the previous layer's transformation; These are the implicit variables output by the current layer transformation; This represents the Hadamard product (i.e., the element-wise multiplication operation of a matrix). This is the normalized feature vector of micrometeorological and equipment condition conditions; and Let represent the scaling factor and translation factor networks obtained by fitting a multilayer perceptron, respectively. After multilayer flow mapping operations, the network calculates the sum of the logarithms of the Jacobian determinants for each data point, and outputs a continuous probability density function equation for wind power output falling within the entire rated power domain, thus forming the marginal probability distribution of wind power.

[0051] S1034. In the photovoltaic prediction branch, discrete quantile points are mapped through a preset quantile network, and multiple conditional quantile sequences under preset confidence levels are output as the photovoltaic marginal probability distribution.

[0052] Considering the physical boundary cutoff characteristic of photovoltaic power output at night, the photovoltaic prediction branch employs a parameter-free quantile regression network. A series of discrete confidence level quantiles are defined (e.g., 19 quantiles). The final output of the quantile network is for each specified quantile. Power prediction values ​​below The network is trained using a quantile pinball loss function that incorporates an asymmetric penalty mechanism. The loss calculation formula is as follows:

[0053] In the formula, For quantiles The loss value below; These are historical actual power output observations for photovoltaic systems. The quantile output values ​​predicted by the model; The quantile values ​​are defined. By minimizing the total bouncing loss function at each quantile, the network parametrically plots the photovoltaic marginal probability distribution curve with zero-point truncation.

[0054] To fully explore the spatiotemporal dependence of micro-topography under the zonal geographical structure of tunnel clusters, this method also includes sub-steps S1035 to S1038 for deep spatiotemporal feature extraction and cross-branch fusion: S1035. In the wind power prediction branch, a pre-defined time-series feature extraction module is used to extract local wind speed abrupt change features in the time dimension of the micrometeorological sequence, and these local wind speed abrupt change features are input as conditional variables into the probability density estimation network. The pre-defined time-series feature extraction module adopts a one-dimensional dilated causal convolutional network (1D Dilated Causal Convolutional Network). The convolutional layers follow causal constraints to prevent information leakage, and the expansion rate increases exponentially by 1, 2, 4, and 8. This design expands the receptive field on the time axis while maintaining low computational burden, capturing abrupt signals of wind speed increases or decreases, and generating local wind speed abrupt change feature vectors.

[0055] S1036. In the photovoltaic prediction branch, a pre-defined spatial feature extraction module is used to extract the spatial correlation features of the micro-meteorological sequence on the initial tunnel group node graph. The spatial feature extraction module adopts a graph convolutional network (GCN). Using the constructed initial tunnel group node graph as the graph topology base, the micro-meteorological parameter matrix of each node at the current time is used as the node feature. Through Laplacian smoothing and neighborhood aggregation operations of the graph convolution operator, a spatial correlation feature vector reflecting the global light diffusion pattern is output.

[0056] S1037. Extract the correlation of photovoltaic power output delay caused by mountain shadow between nodes at each tunnel entrance. Use the preset attention mechanism module to perform cross-branch feature splicing of spatial correlation features, photovoltaic power output delay correlation and local wind speed change features to form a fused spatiotemporal representation vector.

[0057] The delay peak is found by calculating the cross-correlation function of the historical photovoltaic output time series of different nodes, and the photovoltaic output delay correlation vector matrix is ​​extracted. A multi-head attention mechanism module is introduced to calculate the feature interaction influence weights, and the calculation formula is as follows:

[0058] In the formula, (Query matrix) (Key matrix) The (value matrix) is obtained by linear mapping of spatial correlation features, temporal delay correlation features, and wind speed change features, respectively; Calculate the feature relevance score; To scale the feature dimensions; The function transforms the scores into normalized probability distribution weights. The attention weights are multiplied by the value matrix and cross-branch feature concatenation is performed to output a fused spatiotemporal representation vector.

[0059] S1038. The fused spatiotemporal representation vector is input into the hidden layer of the quantile network to capture the spatiotemporal complementarity relationship. This fused spatiotemporal representation vector, as additional high-level contextual prior information, is injected into the feature mapping of the fully connected layer of the continuous flow network and the discrete quantile network, ensuring that the marginal probability distribution of the final output implicitly includes the spatiotemporal complementarity constraint relationship between wind energy and solar energy.

[0060] S104. Based on regional meteorological data, the initial tunnel group node map is updated to obtain a dynamic dependent topology. The correlation between the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power is reconstructed by combining the dynamic dependent topology, and the joint probability distribution of wind and solar power at each tunnel entrance node is generated.

[0061] Weather systems in real physical environments (such as cold fronts) are dynamically moving. Step S104 employs a technique combining dynamic graph weight updates and condition-dependent structural modeling. Specifically, it includes sub-steps S1041 to S1045: S1041. Extract cloud movement trajectories, prevailing wind direction vectors, and real-time weather type identifiers from regional meteorological data. Obtain continuous time-series meteorological satellite cloud image sequences, apply the Dense Optical Flow Method to calculate the spatial displacement difference of cloud pixels over time, and synthesize large-scale cloud movement velocity and translation direction vectors within the region. The prevailing wind direction vector for the region is obtained by weighted average of wind directions. Real-time weather type labels are divided into discrete types such as "clear sky", "cloudy", "showers", "continuous precipitation", and "strong winter front" based on meteorological preset judgment rules.

[0062] S1042. Based on the real-time weather type identifier, activate the corresponding preset map update strategy, combine the cloud movement trajectory and the dominant wind direction vector, dynamically adjust the edge weights of the initial tunnel group node graph to obtain dynamic edge weights, and generate a dynamic adjacency matrix as a dynamic dependent topology based on the dynamic edge weights.

[0063] Taking the cloudy and windy weather identifier as an example, the static connection weights are... The dynamic edge weights have been corrected and adjusted. The calculation is based on the following formula:

[0064] In the formula, From geometric space nodes Pointing to node The geographic direction vector; This indicates solving for the angle between two vectors; The weather influence weighting adjustment coefficient is set between 0.2 and 0.8. As the angle between the weather system's moving vector and the line connecting the nodes approaches 0 degrees, the correlation strengthens, and the dynamic edge weights increase. This generates a dynamically dependent topology with physical time-series orientation.

[0065] S1043. Based on the real-time weather type identifier, select the corresponding conditional joint distribution function from the preset joint distribution function candidate library, and calculate the wind-solar complementarity coefficient between adjacent tunnel entrance nodes and the consistency of tunnel entrance wind direction based on the micro-meteorological sequence.

[0066] The candidate library of joint distribution functions stores various Copula functions. For example, Gaussian Copula is used to match stationary scenarios; Gumbel or t-joint distribution functions are used to match extreme weather events. The Pearson correlation coefficient between the wind speed series and the photovoltaic irradiance series of adjacent nodes within a time sliding window is calculated using micrometeorological sequences as the wind-solar complementarity coefficient (value range [-1,1]). The wind direction consistency at the cave entrance is obtained by comparing the deviation of the local canyon wind direction angle from the regional prevailing wind direction.

[0067] S1044. Using the dynamic adjacency matrix, wind-solar complementarity coefficient, and wind direction consistency at the tunnel entrance, the correlation parameters of the selected conditional joint distribution function are dynamically corrected. The dynamic adjacency matrix determines the non-zero sparse network distribution of the correlation matrix of the function model; the wind-solar complementarity coefficient dynamically adjusts the size of the corresponding covariance elements between wind power and photovoltaics in the joint matrix; and the wind direction consistency at the tunnel entrance weights the correlation coefficients of homogeneous energy in the matrix.

[0068] S1045. Using the modified conditional joint distribution function, multi-dimensional joint sampling is performed on the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node to obtain the joint wind-solar probability distribution. Gibbs sampling based on the Markov Chain Monte Carlo (MCMC) algorithm is employed. The output marginal probability distribution function is used as the input to the inverse probability integral transform to extract tens of thousands of scene sample points (e.g., N=10000 sample sequences) in the multi-dimensional space, forming a probability density surface describing the probability of coordinated power output, thus generating the joint wind-solar probability distribution.

[0069] S105, extract the initial confidence interval from the wind-solar joint probability distribution, determine the current meteorological scenario based on regional meteorological data, identify whether each tunnel entrance node has continuously crossed the boundary under the current meteorological scenario based on historical error data, determine the tunnel entrance node that has continuously crossed the boundary as the boundary node, and perform local calibration on the initial confidence interval for the boundary node to obtain the target confidence interval.

[0070] It includes initial interval extraction sub-steps S1051 to S1055, and local range expansion calibration sub-steps S1056 to S1059: S1051. Extract the prediction time-domain characteristics, power output fluctuation rate, and error variance characteristics of the wind-solar joint probability distribution. Calculate the expected mean curve and variance curve from the massive sample points of the generated joint probability distribution, and obtain the power output fluctuation rate by taking the first difference of the expected value curve; analyze the dispersion of the prediction variance over time to obtain the error variance characteristics.

[0071] S1052. Obtain the pre-built candidate interval model library. The library integrates different interval boundary generation models based on variance scaling and quantile direct truncation methods.

[0072] S1053. Extract indicators of the severity of meteorological changes (such as wind speed range) from regional meteorological data. Based on these indicators, prediction time-domain characteristics, power output fluctuation rate, and error variance characteristics, calculate the fitness score of each candidate interval model in the candidate interval model library. The fitness score is calculated using a pre-trained Support Vector Regression (SVR) model.

[0073] S1054. Select the target interval model with the highest fitness score for each tunnel entrance node and perform matching and sorting.

[0074] S1055. Using the target interval model, the probability density integral of the wind-solar joint probability distribution is solved inversely to extract the upper and lower limits of the specified confidence level to form the initial confidence interval. Setting the confidence level to 90% or 95%, the boundary of the cumulative area integral along the coordinate axis is used to truncate extremely low-probability tail data points, extracting the upper time-series vector of the envelope's main probability. and lower bound time series vector This forms the initial confidence interval.

[0075] Perform a local penalty calibration procedure: S1056. Obtain a pre-established typical weather scenario database. Based on this database, perform pattern matching on regional meteorological data to determine the current meteorological scenario to which the current weather conditions belong. The typical weather scenario database is obtained by partitioning the meteorological feature vector dataset using the K-Means clustering analysis algorithm (number of cluster centers). (Set between 4 and 8). Calculate the Euclidean distance from the real-time regional multidimensional meteorological feature vector to the cluster centers in the scene library, find the cluster center with the smallest metric value, and determine whether it is classified as the current meteorological scene.

[0076] S1057. Extract historical actual output data from historical error data, and calculate the actual interval coverage and average interval width of each tunnel portal node under the current weather scenario. Retrieve historical cross-sections of the same "current weather scenario", count the number of samples whose actual output values ​​fall within the historical initial interval envelope, and calculate the actual interval coverage.

[0077] S1058. Count the number of missed packets in the historical actual output data of each tunnel entrance node that exceeds the initial confidence interval within a preset time window. When the number of missed packets for a specific tunnel entrance node under the current meteorological scenario exceeds a preset tolerance threshold, the specific tunnel entrance node is identified as an out-of-bounds node. The preset time window is the past 24-hour moving observation period. The tolerance threshold is set to "three consecutive 15-minute sampling points exceeding the boundary". The specific tunnel entrance node that exceeds the threshold is confirmed as an out-of-bounds node.

[0078] S1059. Increase the interval scale magnification factor of the boundary-crossing node in the current meteorological scenario online, and use the updated interval scale magnification factor to shift and correct the upper and lower limits of the initial confidence interval to obtain the target confidence interval.

[0079] System preset interval scale magnification factor The initial constant value is assigned to 1.0. When an out-of-bounds error occurs and correction is needed, The elevation is adaptively adjusted online according to a preset proportional step size (fixed step value 0.05). The technical algorithm for translation correction of out-of-bounds nodes is based on the following formula:

[0080]

[0081] In the formula, This represents the mean expected power calculated by the model for the joint probability distribution of the current time period; and The upper and lower limits of the initial confidence interval are to be extracted; This is the updated interval scale magnification factor; and These are the upper and lower limits of the target confidence interval after the correction and expansion, respectively.

[0082] S106: Construct safe load constraints based on real-time energy demand data, and transform the target confidence interval into a safe power supply interval by combining the safe load constraints, so as to generate safe power supply scheduling decisions.

[0083] Step S106 performs closed-loop coupling mapping between the probability interval and the physical energy control layer of the highway microgrid, including sub-steps S1061 to S1066: S1061. Obtain the operating status and backup power capacity of energy storage devices configured in the microgrid system of the highway tunnel group. Interact with the Battery Management System (BMS) and diesel generator controller via industrial Ethernet to obtain the real-time State of Charge (SOC) of the battery energy storage system. The scale range is [0, 100%]), the maximum rechargeable and dischargeable power extreme threshold, and the upper limit of the backup power cold start backup capacity and time constant.

[0084] S1062. Real-time energy demand data is divided into loads of different priorities according to safety importance. The rated power corresponding to the highest priority rigid loads is extracted to construct safety load constraints, and the rigid load floor corresponding to the safety load constraints is determined. The highest priority rigid loads include emergency basic lighting systems, video and environmental monitoring system loads, and critical communication base stations. The total rated power of the highest priority rigid loads is extracted and fixed as a safety load constraint that prevents the microgrid from being disconnected, and this is determined as the rigid load floor. .

[0085] S1063. Subtract the rigid load floor from the lower and upper limits of the target confidence interval to obtain the safe power supply interval. Physically, this equates the renewable energy generation output interval to the net usable energy confidence interval after deducting the safe rigid floor maintenance power. The mathematical derivation formula is as follows:

[0086]

[0087] In the formula, and Define the lower and upper limits of the target confidence interval; This serves as a rigid load floor. and These are the lower and upper limits of the generated safe power supply range, respectively.

[0088] S1064. Compare the lower limit of the safe power supply range with the preset safe power consumption threshold. Based on the shortage difference and the operating status of the energy storage device and the backup power capacity, calculate the discharge command and the start-up capacity allocation command. The preset safe power consumption threshold is set to 0kW.

[0089] like This means that a power outage is imminent. Calculate the absolute value of the power shortage. Read Determine if the battery's state of charge is higher than the over-discharge protection limit (e.g., ...). If the conditions are met, the difference amount is allocated, and the constant power discharge command is calculated; if the remaining energy of the energy storage device is insufficient, the cold start logic sequence of the backup diesel generator is triggered, and the diesel engine start capacity allocation command is issued.

[0090] S1065. Compare the upper limit of the safe power supply range with the preset grid absorption limit, and calculate the charging command and energy curtailment control command based on the surplus difference of the comparison.

[0091] For the active power surplus exceeding the grid absorption limit of the distribution transformer, a charging command is generated to charge the energy storage battery; if it is detected that the battery has reached full charge ( In order to protect electrical components, wind curtailment commands that adjust the pitch angle of wind turbines or solar curtailment commands that control the derating power of photovoltaic inverters are calculated and generated, collectively referred to as curtailment control commands.

[0092] S1066: Summarize and coordinate discharge commands, capacity allocation start commands, charging commands, and energy curtailment control commands to generate safe power supply scheduling decisions to guide energy flow in the microgrid system. Transform the full-scenario response command matrix into a unified time-step control action lookup table (Schedule Table), and merge them to form a set of safe power supply scheduling decision messages directly output to the energy management system for execution.

[0093] To form a complete physical closed-loop feedback system, after generating the safe power supply scheduling decision output, the method also includes the following physical feedback update steps S1067 to S106B: S1067. The safe power supply scheduling decision is sent to the microgrid control terminal in the microgrid system for physical execution. The computing device encapsulates the generated safe power supply scheduling decision and power setpoint sequence in a formatted manner, and transmits it to the microgrid control terminal (such as a programmable logic controller PLC) via industrial Ethernet using the Modbus protocol or the IEC 61850 standard communication protocol. This enables underlying physical responses such as relay action and IGBT pulse width modulation duty cycle adjustment.

[0094] S1068. During the physical execution of the microgrid control terminal, the actual charging and discharging power of the energy storage device and the microgrid frequency and voltage fluctuation data are collected in real time. A phasor measurement unit (PMU) deployed at the grid connection point collects and calculates the actual charging and discharging active power in real time using an industrial high-frequency sampling rate of no less than 100Hz. (Unit: kW), and microgrid operating frequency and root mean square voltage fluctuation data. .

[0095] S1069. Compare and analyze the deviation between the actual charging and discharging power and the safe power supply dispatch decision, and transform the microgrid frequency and voltage fluctuation data into over-limit penalty feedback items. The actual charging and discharging power feedback data... , and the target set value of the scheduling decision-making instruction Perform subtraction to extract the absolute value and calculate the control response residual. If the microgrid frequency deviates from the reference rated value, or the bus voltage deviation exceeds the safety limit, the algorithm amplifies the root mean square error and converts it into an over-limit penalty feedback term index through a preset penalty function.

[0096] S106A. The bias and over-limit penalty feedback terms are fused to construct the closed-loop loss function. A closed-loop loss function equation is then constructed to guide the adaptive adjustment and optimization of the network model. The calculation formula is as follows:

[0097] In the formula, Calculate the prediction bias loss for the probability density of the model; The non-negative penalty weight setting determined by the microgrid hardware security requirements (e.g.) When the actual physical hardware output deviates from the instruction or the frequency fluctuation increases, It will increase due to physical feedback penalties.

[0098] S106B: Backpropagation updates are performed using a closed-loop loss function to dynamically reduce the redundancy width of the target confidence interval for each tunnel entrance node in the next scheduling period, forming a physical control closed loop. When the closed-loop loss function value... When the value remains consistently small, it is inferred that the interval scale magnification factor is used to prevent the amplification of missing packets. This resulted in excessive redundancy. The control algorithm dynamically suppresses and attenuates the previously amplified physical gradient online in real time, according to a pre-set attenuation physical gradient setting (adjusted in steps of 0.01). The amplification coefficient value is used to dynamically reduce the redundancy safety width envelope range margin of the target confidence interval in the next scheduling period, thus forming an electrical-physical feedback control closed-loop link.

[0099] This invention also provides a wind-solar hybrid confidence interval prediction system for highway tunnel groups, referring to... Figure 3 The diagram illustrates a functional block diagram of a wind-solar hybrid confidence interval prediction system 300 for highway tunnel groups according to the present invention. The system may include the following modules: The data acquisition module 301 is used to acquire geospatial information, regional meteorological data, historical error data and real-time energy demand data of the highway tunnel group; The micro-meteorological mapping module 302 is used to construct an initial tunnel group node map based on geospatial information and map regional meteorological data into micro-meteorological sequences for each tunnel entrance node; The marginal distribution prediction module 303 is used to input the micro-meteorological sequence into the wind and solar prediction model and output the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power at each tunnel entrance node. The joint probability generation module 304 is used to update the initial tunnel group node map based on regional meteorological data to obtain dynamic dependent topology, and to reconstruct the correlation between the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power based on the dynamic dependent topology, thereby generating the joint probability distribution of wind and solar power at each tunnel entrance node. The interval local calibration module 305 is used to extract the initial confidence interval from the wind-solar joint probability distribution, determine the current meteorological scenario based on regional meteorological data, identify whether each tunnel portal node has continuously crossed the boundary under the current meteorological scenario based on historical error data, determine the tunnel portal nodes that have continuously crossed the boundary as boundary crossing nodes, and perform local calibration on the initial confidence interval for the boundary crossing nodes to obtain the target confidence interval. The safe power supply decision module 306 is used to construct safe load constraints based on real-time energy demand data, and to convert the target confidence interval into a safe power supply interval by combining the safe load constraints, so as to generate a safe power supply scheduling decision.

[0100] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. Memory, used to store computer programs; The processor, when executing the program stored in the memory, implements the wind-solar hybrid confidence interval prediction method for highway tunnel groups of the present invention.

[0101] The communication bus mentioned in the aforementioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include RAM, or it can include NVM, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0102] The processors mentioned above can be general-purpose processors, including CPUs, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0103] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the wind-solar complementary confidence interval prediction method for highway tunnel groups according to embodiments of the present invention.

[0104] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the confidence interval of wind-solar hybrid power transmission in highway tunnel groups, characterized in that, Includes the following steps: Obtain geospatial information, regional meteorological data, historical error data, and real-time energy demand data for highway tunnel groups; An initial tunnel group node map is constructed based on the geospatial information, and the regional meteorological data is mapped into micro-meteorological sequences for each tunnel entrance node; The micrometeorological sequence is input into the wind and solar prediction model, and the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node are output. The initial tunnel group node map is updated based on the regional meteorological data to obtain a dynamic dependent topology. The correlation between the wind power marginal probability distribution and the photovoltaic marginal probability distribution is reconstructed by combining the dynamic dependent topology, and the wind and solar joint probability distribution of each tunnel entrance node is generated. An initial confidence interval is extracted from the wind-solar joint probability distribution. The current meteorological scenario is determined based on the regional meteorological data. The historical error data is used to identify whether each tunnel entrance node has continuously crossed the boundary under the current meteorological scenario. The tunnel entrance nodes that have continuously crossed the boundary are identified as boundary crossing nodes. The initial confidence interval is locally calibrated for the boundary crossing nodes to obtain the target confidence interval. Based on the real-time energy demand data, a safe load constraint is constructed, and the target confidence interval is combined with the safe load constraint to convert it into a safe power supply interval, so as to generate a safe power supply scheduling decision.

2. The method for predicting the confidence interval of wind-solar hybrid power transmission for highway tunnel groups according to claim 1, characterized in that, The step of constructing an initial tunnel group node map based on the geospatial information and mapping the regional meteorological data into micro-meteorological sequences for each tunnel entrance node specifically includes: Each tunnel entrance node is abstracted as a spatial node. The three-dimensional coordinates of each spatial node are extracted from the geospatial information. Static connection weights are calculated based on the spatial distance characteristics between each spatial node to construct the initial tunnel group node diagram. The regional meteorological data includes observation data from surrounding meteorological stations and regional numerical weather forecast data. The observation data from surrounding meteorological stations and the regional numerical weather forecast data are input into a pre-trained meteorological mapping model. The surrounding terrain features of each tunnel entrance node are extracted from the geospatial information, and the output of the meteorological mapping model is downscaled and corrected based on the surrounding terrain features to generate the micro-meteorological sequence containing actual meteorological elements.

3. The method for predicting the confidence interval of wind-solar hybrid power transmission for highway tunnel groups according to claim 1, characterized in that, The step of inputting the micro-meteorological sequence into the wind-solar prediction model and outputting the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node specifically includes: The historical actual power output sequence and wind and solar equipment operating status parameters of each tunnel portal node are obtained from the data acquisition terminals deployed at each of the tunnel portal nodes. The micro-meteorological sequence, the historical actual power output sequence, and the wind and solar equipment operating status parameters are input into the wind and solar prediction model, which includes wind power prediction and photovoltaic prediction branches. In the wind power prediction branch, continuous distribution inference is performed through a preset probability density estimation network, and the output probability density function is used as the marginal probability distribution of wind power. In the photovoltaic prediction branch, discrete quantiles are mapped through a preset quantile network, and multiple conditional quantile sequences under preset confidence levels are output as the photovoltaic marginal probability distribution.

4. The method for predicting the confidence interval of wind-solar hybrid power transmission for highway tunnel groups according to claim 3, characterized in that, In the wind and solar prediction model, before outputting the marginal probability distributions of wind power and photovoltaic power at each tunnel entrance node, the method further includes: In the wind power prediction branch, a preset time series feature extraction module is used to extract the local wind speed mutation features of the micro-meteorological sequence in the time dimension, and the local wind speed mutation features are input as conditional variables into the probability density estimation network. In the photovoltaic prediction branch, the spatial correlation features of the micro-meteorological sequence on the initial tunnel group node map are extracted using a preset spatial feature extraction module; Extract the photovoltaic output delay correlation caused by mountain shadows between each of the tunnel entrance nodes, and use a preset attention mechanism module to perform cross-branch feature splicing of the spatial correlation features, the photovoltaic output delay correlation and the local wind speed change features to form a fused spatiotemporal representation vector; The fused spatiotemporal representation vector is input into the hidden layer of the quantile network to capture spatiotemporal complementarity.

5. The method for predicting the confidence interval of wind-solar hybrid power transmission for highway tunnel groups according to claim 1, characterized in that, The process involves updating the initial tunnel group node map based on the regional meteorological data to obtain a dynamic dependent topology, and then reconstructing the correlation between the wind power marginal probability distribution and the photovoltaic marginal probability distribution using the dynamic dependent topology to generate a joint wind-solar probability distribution for each tunnel entrance node. Specifically, this includes: Extract cloud movement trajectories, prevailing wind direction vectors, and real-time weather type identifiers from the meteorological data of the region. Based on the real-time weather type identifier, the corresponding preset map update strategy is activated. Combining the cloud movement trajectory and the dominant wind direction vector, the edge weights of the initial tunnel group node graph are dynamically adjusted to obtain dynamic edge weights. Based on the dynamic edge weights, a dynamic adjacency matrix 1042 is generated as the dynamic dependent topology. Based on the real-time weather type identifier, a corresponding conditional joint distribution function is selected from a preset joint distribution function candidate library, and the wind-solar complementarity coefficient between adjacent tunnel entrance nodes and the consistency of tunnel entrance wind direction of each tunnel entrance node are calculated based on the micro-meteorological sequence. The correlation parameters of the selected conditional joint distribution function are dynamically corrected using the dynamic adjacency matrix, the wind-solar complementarity coefficient, and the wind direction consistency at the tunnel entrance. The wind power marginal probability distribution and the photovoltaic marginal probability distribution at each tunnel entrance node are sampled in multiple dimensions using the modified conditional joint distribution function to obtain the wind-solar joint probability distribution.

6. The method for predicting the confidence interval of wind-solar hybrid power transmission for highway tunnel groups according to claim 1, characterized in that, The extraction of the initial confidence interval from the wind-solar joint probability distribution specifically includes: Extract the prediction time-domain features, power output fluctuation rate, and error variance features of the wind-solar combined probability distribution; Obtain a pre-built library of candidate interval models; The meteorological change intensity index is extracted from the regional meteorological data. Based on the meteorological change intensity index, the prediction time domain characteristics, the power output fluctuation rate, and the error difference variance characteristics, the fitness score of each candidate interval model in the candidate interval model library is calculated. Select the target interval model with the highest fitness score for each of the tunnel entrance nodes; The probability density integral of the wind-solar joint probability distribution is solved by using the target interval model, and the upper and lower limits of the specified confidence level are extracted to form the initial confidence interval.

7. The method for predicting the confidence interval of wind-solar hybrid power transmission for highway tunnel groups according to claim 1, characterized in that, The process involves determining the current meteorological scenario based on the regional meteorological data, identifying whether each tunnel entrance node has continuously crossed boundaries under the current meteorological scenario based on the historical error data, defining the tunnel entrance nodes that have continuously crossed boundaries as boundary-crossing nodes, and performing local calibration on the initial confidence interval for the boundary-crossing nodes to obtain the target confidence interval. Specifically, this includes: Obtain a pre-established typical weather scenario database, perform pattern matching on the regional meteorological data based on the typical weather scenario database, and determine the current meteorological scenario to which the current meteorological conditions belong. Historical actual output data are extracted from the historical error data, and the actual interval coverage and average interval width of each tunnel portal node under the current meteorological scenario are calculated. The number of missed packets in the historical actual output data of each tunnel entrance node within a preset time window that exceeds the initial confidence interval is counted. When the number of missed packets of a specific tunnel entrance node under the current weather scenario exceeds a preset tolerance threshold, the specific tunnel entrance node is determined to be the boundary-crossing node. The interval scale magnification factor of the out-of-bounds node under the current meteorological scenario is increased online, and the upper and lower limits of the initial confidence interval are shifted and corrected using the updated interval scale magnification factor to obtain the target confidence interval.

8. The method for predicting the confidence interval of wind-solar hybrid power transmission for highway tunnel groups according to claim 1, characterized in that, The step of constructing safe load constraints based on the real-time energy demand data, and converting the target confidence interval into a safe power supply interval by combining the safe load constraints, in order to generate a safe power supply scheduling decision, specifically includes: Obtain the operating status and backup power capacity of the energy storage devices configured in the microgrid system of the highway tunnel group; The real-time energy demand data is divided into loads of different priorities according to safety importance, and the rated power corresponding to the rigid load with the highest priority is extracted to construct the safety load constraint, and the rigid load bottom line corresponding to the safety load constraint is determined. The safe power supply range is obtained by subtracting the rigid load floor from the lower and upper limits of the target confidence interval, respectively. The lower limit of the safe power supply range is compared with the preset safe power consumption threshold. Based on the shortage difference and the operating status of the energy storage device and the backup power capacity, the discharge command and the start-up capacity allocation command are calculated. The upper limit of the safe power supply range is compared with the preset grid absorption limit, and the charging command and energy curtailment control command are calculated based on the surplus difference of the comparison. The discharge command, the start-up capacity allocation command, the charging command, and the energy curtailment control command are summarized and coordinated to generate the safe power supply scheduling decision for guiding the energy flow of the microgrid system.

9. The method for predicting the confidence interval of wind-solar hybrid power systems for highway tunnel groups according to claim 8, characterized in that, After generating the secure power supply scheduling decision, the method further includes: The safe power supply scheduling decision is sent to the microgrid control terminal in the microgrid system for physical execution; During the physical execution of the microgrid control terminal, the actual charging and discharging power of the energy storage device and the frequency and voltage fluctuation data of the microgrid are collected in real time. The deviation between the actual charging and discharging power and the safe power supply scheduling decision is compared and analyzed, and the microgrid frequency and voltage fluctuation data is converted into over-limit penalty feedback items; The deviation is fused with the over-limit penalty feedback term to construct a closed-loop loss function; The closed-loop loss function is used for backpropagation updates to dynamically reduce the redundancy width of the target confidence interval of each tunnel portal node in the next scheduling period, thus forming a physical control closed loop.

10. A wind-solar hybrid confidence interval prediction system for highway tunnel groups, characterized in that, include: The data acquisition module is used to acquire geospatial information of the highway tunnel group, regional meteorological data, historical error data, and real-time energy demand data. The micro-meteorological mapping module is used to construct an initial tunnel group node map based on the geospatial information and map the regional meteorological data into micro-meteorological sequences for each tunnel entrance node; The marginal distribution prediction module is used to input the micro-meteorological sequence into the wind and solar prediction model and output the marginal probability distribution of wind power and the marginal probability distribution of photovoltaic power at each of the tunnel entrance nodes. The joint probability generation module is used to update the initial tunnel group node map based on the regional meteorological data to obtain a dynamic dependent topology, and to reconstruct the correlation between the wind power marginal probability distribution and the photovoltaic marginal probability distribution by combining the dynamic dependent topology, thereby generating the wind and solar joint probability distribution of each tunnel entrance node. The interval local calibration module is used to extract an initial confidence interval from the wind-solar joint probability distribution, determine the current meteorological scenario based on the regional meteorological data, identify whether each of the tunnel entrance nodes has continuously crossed the boundary under the current meteorological scenario according to the historical error data, determine the tunnel entrance nodes that have continuously crossed the boundary as boundary crossing nodes, and perform local calibration on the initial confidence interval for the boundary crossing nodes to obtain the target confidence interval. The safe power supply decision module is used to construct safe load constraints based on the real-time energy demand data, and to convert the target confidence interval into a safe power supply interval by combining the safe load constraints, so as to generate a safe power supply scheduling decision.

Images