Tower type heliostat field multi-tower cooperative central tower calibration scheduling method

Abstract

The application belongs to the technical field of heliostat calibration, and particularly relates to a tower type optical thermal mirror field multi-tower cooperative heliostat calibration scheduling method, the core of which is to model the selection problem of each batch of heliostats to be calibrated as a bipartite graph maximum flow problem with capacity constraints, and iteratively solve it, so that the number of heliostats calibrated simultaneously in each batch is maximized, thereby globally minimizing the total batch number and accelerating the heliostat calibration efficiency.

Description

Technical Field

[0001] This application belongs to the field of heliostat calibration technology, specifically relating to a method for scheduling multi-tower coordinated heliostat calibration in a tower-type photothermal mirror field. Background Technology

[0002] Traditional calibration techniques rely on daytime solar light, with a single project typically involving tens of thousands to hundreds of thousands of heliostats. Even under ideal weather conditions, completing a full-field calibration cycle takes more than two months. Furthermore, the construction and commissioning progress of the receiver tower directly constrains calibration operations, further extending the overall project commissioning cycle and becoming a common efficiency bottleneck in the industry. Traditional calibration techniques involve installing a calibration camera above the receiver tower; the heliostats reflect sunlight into the camera's field of view, and machine vision-level image processing corrects the pointing deviation of the heliostats. However, reflecting sunlight into the camera during the day can easily cause internal overheating, limiting the number of heliostats that can be calibrated simultaneously. Additionally, the number of cameras on the receiver tower is also limited, resulting in insufficient calibration efficiency for large-scale heliostat fields. Summary of the Invention

[0003] This invention, through dynamic prioritization of heliostats and dynamic capacity calculation of cameras, maximizes the utilization of camera and lamp resources within the heliostat field, enabling the highest concurrency calibration execution, improving calibration efficiency, and reducing calibration cycle time. The technical solution is as follows: A method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field includes the following steps: S1. Divide the heliostat into sections to form a heliostat-camera-lamp combination; S2. Establish the camera dynamic capacity model and start the calibration time prediction model; S3. Construct a weighted bipartite graph model; S4. Select the heliostat to be tested; S5. Calculate dynamic priority; S6. Determine if a camera capacity update has been triggered. If yes, update the camera capacity; otherwise, proceed to S7. S7. Solve for the maximum weight matching; S8. Perform calibration.

[0004] Preferably, in step S1, based on the mirror field layout diagram, the location of the calibration tower, the installation parameters of the cameras on each tower, and the position of the lamps, geometric optical tracing is performed to pre-generate an initial usable "camera-lamp" combination for each heliostat; for any heliostat h, its combination with camera c and lamps... The combination must simultaneously satisfy the following three conditions: Field of view conditions: The line connecting the center of the heliostat and the optical center of the camera is within the camera's field of view. Reflection conditions: The direction of the heliostat's normal must be able to reflect light from the lamp. The light is reflected to camera c; Lighting conditions: The direction of the light beam should cover the heliostat, meaning the heliostat is within the illumination range of the light.

[0005] Preferably, in step S2, a camera dynamic capacity model is established: Let Let the set of heliostats using camera c at time t define the effective distance attenuation factor of camera c at time t. for: ; For a single heliostat h using camera c, the distance attenuation factor... Defined as: ; The distance from the heliostat h to the camera c; : Camera reference distance, usually taken as the camera's design reference distance (e.g., 200 meters); Distance attenuation constant; The set of heliostats using camera c at time t is empty.

[0006] Preferably, a dynamic capacity model for the camera is established, based on the capacity update mechanism of the farthest heliostat: Then the real-time concurrent capacity of the camera at time t for: ; Signal-to-noise ratio factor, which is related to ambient lighting, is defined as: ; The current ambient light level; For reference ambient illuminance; Light distance attenuation factor; According to the inverse square law of optics, light intensity is inversely proportional to the square of the distance. Therefore, the light distance attenuation factor is defined as: ; The reference distance for the lights is set at 200 meters. For the heliostat h to the light The distance; The minimum angular resolution required for single-sided heliostat spot identification; The field of view of camera c, centered on the optical axis of the camera lens, is the maximum spatial angle range in which the camera's imaging sensor can effectively receive light, and is an inherent optical parameter of the camera.

[0007] Preferably, step S3 involves constructing a weighted bipartite graph model: treating each camera slot as an independent node, and constructing a weighted bipartite graph G=(U,V,E): Left node set U: Take the first M heliostats from the waiting queue. Δ is the scheduling redundancy coefficient, which is a preset positive integer constant used to expand the size of the candidate heliostat set, ensuring that there are enough effective edges in the weighted bipartite graph, ensuring that the maximum weight matching can make full use of all available slots of the camera, and avoiding the idle camera resources due to the candidate set being too small.

[0008] The camera slot is defined as follows: Each camera c generates dynamic capacity based on its current moment. An independent, virtual "slot node"; if the current capacity of camera c... So it has 3 slots.

[0009] Each slot can only be allocated to one heliostat at any given time; If different slots come from the same physical camera, they are considered completely independent right nodes in a bipartite graph model.

[0010] The right node set V: all available camera slots, generated for each camera c. Each slot node ; Edge set E: If the heliostat h can be combined Then h is related to all the slot nodes of that camera. Connect the edges between them; Edge weight is defined as: ; Heliostat current priority; Heliostat h uses combination Predicted calibration time; Waiting time; Overall signal-to-noise ratio factor; , , , Weighting coefficients, satisfying .

[0011] Preferred dynamic priority of each heliostat Defined as: ; Basic priority, all heliostats have the same initial value; Failure penalty coefficient; Total number of failures; Waiting time coefficient; Time since the last calibration attempt; Emergency mission reward coefficient; Emergency task indicator function: 1 if the heliostat is marked as emergency, 0 otherwise; The capacity is recalculated at the following times: Before assigning a new task: When attempting to assign a new task to camera c, it is necessary to first determine the current task assignment method. ,calculate To determine the available slots; When the task is completed: When the calibration of a heliostat on camera C is completed, from After removing from the middle, recalculate. This is to reflect potential increases in capacity; When the environment changes: When the change exceeds a certain threshold, the capacity of all cameras is recalculated.

[0012] Preferably, the maximum weight matching is solved: For each camera c, the number of slots currently occupied. The number of available slots is: ; Each available slot for a camera is added to a weighted bipartite graph model. The Hungarian algorithm is then used to find the maximum weighted matching in the weighted bipartite graph, with a time complexity of [missing information]. The matching results provide the heliostats that should be calibrated, their assigned camera slots, and the corresponding lighting combinations.

[0013] Preferably, calibration is performed for each successfully matched heliostat h: S81. Remove from the waiting queue; S82. Add it to the service set of the corresponding camera c. ; S83. Record the camera c used, lighting. Calibration execution time and other calibration execution information; S84. Immediately issue a command to the mirror field controller to begin calibration; S85. After calibration, release resources, save sample information, time information, and camera and light information for calibration time prediction.

[0014] Preferably, when heliostat calibration fails: Step 1: Immediately from Remove from the middle, record failure information, and release resources; Step 2: Update the number of failures f h ←f h+1 Recalculate priority ; Step 3: Re-add to the waiting queue and trigger a new event.

[0015] Compared with the prior art, the beneficial effects of this application are as follows: Parallel scheduling under camera capacity constraints: Each camera can only process a limited number of heliostats at any given time. How can we maximize the number of heliostats calibrated in each batch while meeting the capacity constraints?

[0016] Global batch minimization: Arrange the calibration sequence and camera allocation of all heliostats to minimize the total number of batches required to complete the full-site calibration.

[0017] Dynamic environmental adaptability: The camera's concurrent capacity is dynamically adjusted based on factors such as ambient light and heliostat distance, and retrying of heliostats that fail calibration is handled in real time. Attached Figure Description

[0018] Figure 1 A diagram showing the distribution of calibration towers; Figure 2 Diagram of the calibration tower structure; Figure 3 Here is a flowchart of the calibration method; Figure 4 To calibrate the camera, the sampling field of view is calibrated simultaneously. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Step S0: Offline Data Preparation – Generating a Set of Cameras and Lights Available for Heliostat Use: Based on the mirror field layout diagram, calibration tower locations, camera installation parameters on each tower, and lamp positions, geometric optical tracing is performed to generate an initial set of usable cameras and lamps for each heliostat. For each heliostat h, there is one camera c and one lamp. To be usable, the following conditions must be met: The line connecting the center of the heliostat and the optical center of the camera lies within the camera's field of view; The direction of the heliostat's normal can reflect light (light from the tower where the camera is located) back to the camera; The direction of the light beam can cover the heliostat (i.e., the heliostat is located within the illumination range of the light).

[0021] This step involves offline static planning, and the results are stored in tabular form. Since each tower has only one light, and there is a one-to-one correspondence between cameras and towers, the available camera set implicitly contains information about the available towers and lights.

[0022] Step S1: Start the calibration time prediction model The XGBoost regression model is used for fitting and prediction. Standardized feature values ​​are the basis of the model input, and the predicted values ​​are... It is the model's final output of the input features. The two are strongly correlated through the hierarchical splitting, leaf node mapping, and ensemble accumulation of the CART regression tree. Standardization only applies to the feature input stage and does not change the dimensions or physical meaning of the target variable T. The final output is the prediction calibration time. It will be restored to the actual calibration time unit of seconds, rather than the standardized value.

[0023] Actual calibration time for each heliostat The heliostat, camera, and light combination are affected by various factors and are subject to uncertainty. This invention uses a gradient boosting tree (XGBoost) regression model for prediction.

[0024] For each training sample (one historical calibration record), extract the following features: : Distance (meters) from the heliostat to the camera, continuous value.

[0025] : Ambient illuminance (lux) during calibration, continuous value.

[0026] : Mirror cleanliness coefficient, obtained from maintenance records or online testing, with a value range of [0,1][0,1], a continuous value.

[0027] : The cumulative number of failures of this heliostat in history, in integer form.

[0028] Camera identifier, category characteristics.

[0029] Light identification, category characteristics.

[0030] time_of_daytime: time (hour), which can be segmented (e.g., day / night).

[0031] weather: weather type (sunny, cloudy, overcast), category characteristics.

[0032] Model training and prediction: The XGBoost model is trained using historical calibration data (including actual calibration time), with the actual calibration time T as the target variable.

[0033] The historical calibration dataset was divided into a training set (model fitting) and a validation set (preliminary parameter validation) in a 9:1 ratio. During the division, the feature distribution and target variable distribution of each subset were kept consistent with the original dataset.

[0034] 1. Feature normalization stage: For all continuous features , , , The time_of_daytime is standardized using the following formula: To eliminate the impact of dimensional differences on the model, camera labels, light labels, and weather types are not normalized but are classified and coded through features to facilitate software learning and fitting.

[0035] Original eigenvalues; The mean of this feature on the training set; The standard deviation of this feature on the training set;

[0036] : The normalized eigenvalues.

[0037] 2. Fitting Phase: The core of fitting is to enable the XGBoost model to learn the inherent relationship between features and the actual calibration time T through standardized feature values, and to solidify this relationship in the splitting rules and leaf node outputs of each CART regression tree. Standardized feature values ​​are the data foundation for model learning, and the correlation with the actual calibration time T is achieved through tree splitting and leaf node fitting. The specific steps are as follows: 1) Traverse all standardized features (such as distance d, illuminance) ) and encoded category features (such as camera, light, weather, time).

[0038] 2) For each standardized feature, iterate through all possible split thresholds. The threshold is defined based on the maximum and minimum interval of the standardized data as 0.1. For categorical features, iterate through all possible split categories (such as sunny / not sunny, camera and light). Here, it is not necessary to follow the established heliostat-camera-light combination. The grouping here is only based on the feature type.

[0039] 3) With the goal of minimizing the squared error, the optimal splitting feature and splitting threshold are selected to divide the training set samples into different child nodes. Core connection: The numerical difference of standardized feature values ​​is the basis for judging the splitting of tree nodes. The model divides the samples into subsets with similar features based on the size / category of the feature values, and each subset corresponds to a similar actual calibration time T (e.g., T is generally smaller for close samples and larger for distant samples).

[0040] 4) The sample subsets divided by standardized feature values ​​are strongly bound to the mean / residual of the actual calibration time T of the samples (fitting subset → actual calibration time T pattern).

[0041] After each node split, each resulting child node is a set of samples with similar feature values. The model calculates the mean (first tree) or residual (subsequent trees) of the actual calibration time T of the samples within each child node, which is used as the fitted value for that node.

[0042] First tree (fitting global residuals):

[0043] Input: The standardized feature vector x' of the sample.

[0044] Objective: To find the true calibration time value T of the fitted sample and the global sample mean. The residuals between them.

[0045] The second tree and subsequent trees (fitting the residuals of the previous predictions):

[0046] Input: The same standardized feature vector x'.

[0047] Objective: To find the residual between the true calibration time of the fitted sample and the sum of the predicted values ​​of all preceding trees for that sample.

[0048] When constructing the second tree, the sample will be assigned to a leaf node j. This leaf node may contain multiple training samples (not just the current one). The formula for calculating the residual value of the node output is: ; λ is the regularization coefficient (e.g., λ penalizes the weight of leaf nodes). To optimize the regularization coefficient and prevent overfitting, λ=3 was ultimately selected. Let i be the residual of sample i. This represents the number of samples at this node.

[0049] Subsequent tree (Kth tree): Repeat the above process, with each new tree working to fit the residuals of the predictions from the previous K-1 trees. The residual output of each tree (i.e., the correction amount of the prediction) is multiplied by a learning rate η (e.g., 0.1) and then summed to prevent overfitting. Ultimately, the model's prediction is the sum of the "global mean" and the "weighted corrections for each tree".

[0050] Core Relationship: Standardized feature values ​​divide samples by "feature similarity", so that each sample subset has a one-to-one correspondence with the numerical pattern of T. The model remembers "which combinations of standardized feature values ​​correspond to what calibration time T".

[0051] 5) The "feature-T mapping rules" of all trees are integrated to form a global model (single tree → ensemble model). Each CART tree learns a set of local rules for "feature value combination → T fitting value" based on standardized feature values. XGBoost weighted sums of the local rules of all trees form a global mapping model of standardized feature values ​​and encoded features → actual T, where: ; The standardized feature vector; The output residual value of the k-th tree for the standardized eigenvalue x′; The global mean of the training set T; η: learning rate; η controls the update step size of each tree. The optimal value is determined based on the principle of minimizing the root mean square error of the validation set. In this application, η=0.1 is preferred.

[0052] : The predicted actual calibration time of the standardized feature value after integrating m trees (in seconds, unstandardized).

[0053] Core Relationship: The standardized feature value x′ is the input variable of the ensemble model, and the model... The summation directly outputs the predicted value of the actual calibration time T corresponding to x′. Standardization is only to make the feature values ​​more suitable for model learning and does not affect the dimensions and physical meaning of the output results.

[0054] After fitting, for any heliostat h and camera c, the lamp By combining the current feature values ​​at the current time t and under weather conditions, the model is obtained by inputting the current feature values. That is, the heliostat h and the camera c. The model combines the predicted calibration time values ​​under the current feature values, and uses the predicted calibration time as a reference for subsequent scheduling. The model is updated incrementally periodically (e.g., daily).

[0055] Step S2: Establish the camera dynamic capacity model: In actual scheduling, the same camera may simultaneously serve multiple heliostats at different distances. In this case, the effective capacity of the camera should be determined by the most unfavorable heliostat among all those currently being served or about to be served, because the farthest heliostat determines the overall image quality and the number of concurrent heliostats that the camera can serve.

[0056] set up Let be the set of heliostats using camera c at time t (i.e., the task that has started calibration but has not yet been completed). Define the effective distance attenuation factor of camera c at time t as: ; For a single heliostat h using camera c, the distance attenuation factor is defined as: ; : Distance (in meters) from heliostat h to camera c.

[0057] Reference distance (meters), usually taken as the camera's design reference distance (e.g., 200 meters).

[0058] Distance attenuation constant (meters) characterizes the rate at which the signal-to-noise ratio decreases with increasing distance, and is usually taken as 500 meters.

[0059] This distance attenuation factor reflects the decrease in signal-to-noise ratio caused by light spot diffusion in long-distance heliostats, thus affecting the number of observations the camera can simultaneously observe.

[0060] Calculation of camera dynamic capacity – Based on the capacity update mechanism of the farthest heliostat, the real-time concurrent capacity of the camera at time t is: ; The camera's actual concurrent capacity is limited by the furthest heliostat currently being served. If a more distant heliostat is added, the capacity may decrease; if the furthest heliostat is completed, the capacity may increase.

[0061] Signal-to-noise ratio factor, which is related to ambient lighting, is defined as: ; : Current ambient illuminance (lux); : For reference ambient illuminance; This is the light distance attenuation factor.

[0062] The distance between the light source and the heliostat directly affects the intensity of light hitting the heliostat's surface, thus affecting the brightness of the reflected light spot. According to the inverse square law of optics, light intensity is inversely proportional to the square of the distance. The light source distance attenuation factor is defined as: ; For reference distance (meters), take 200 meters.

[0063] For the heliostat h to the light The distance (meters).

[0064] The minimum angular resolution (in radians) required for single-sided heliostat spot identification is typically 2.5 mrad.

[0065] : The field of view (in radians) of camera c.

[0066] Step S3: Construct a weighted bipartite graph: Treat each camera slot as an independent node, and construct a weighted bipartite graph G=(U,V,E):

[0067] Left node set U: Take the first M heliostats from the waiting queue. Δ is taken as 50.

[0068] Right node set V: All available camera slots for each camera c, generated Each slot node .

[0069] Edge set E: If the heliostat h can be combined Then h is related to all the slot nodes of that camera. Connect the edges between them.

[0070] Edge weight is defined as: ; Heliostat current priority; Heliostat h uses combination Predicted calibration time; Waiting time; Overall signal-to-noise ratio factor (the higher the better); , , , Weighting coefficients, satisfying .

[0071] Step S4: Filter all heliostats that need calibration -- Add threshold: Based on information such as heliostat labels indicating reinstallation, re-zeroing, motor malfunction, lack of recalibration for an extended period, or recalibration settings, all heliostats requiring calibration are filtered.

[0072] Step S5: Dynamic Priority Calculation: Dynamic priority of each heliostat Defined as: ;

[0073] : Basic priority, all heliostats have the same initial value (e.g., 50).

[0074] : Failure penalty coefficient, increases by 10 for each failure.

[0075] : Total number of failures.

[0076] Waiting time coefficient, increasing by 0.1 per minute.

[0077] : Time (minutes) since the last calibration attempt.

[0078] Emergency mission reward coefficient, set to 100.

[0079] : Emergency task indicator function, which takes the value 1 if the heliostat is marked as emergency, otherwise 0.

[0080] The higher the priority, the more advantageous it is in resource allocation.

[0081] Step S6: Triggering the capacity update: The capacity is recalculated at the following times: Before assigning a new task: When attempting to assign a new task to camera c, it is necessary to first determine the current task assignment method. ,calculate To determine the available slots.

[0082] When the task is completed: When the calibration of a heliostat on camera C is completed, from After removing from the middle, recalculate. This is to reflect the potential increase in capacity.

[0083] When the environment changes: When the change exceeds a certain threshold (e.g., 5%), the capacity of all cameras is recalculated.

[0084] Step S7: Solve for the maximum weight matching: For each camera c, the number of slots currently occupied. The number of available slots is: ;

[0085] Each available slot for a camera is added to the weighted bipartite graph model. The Kuhn-Munkres algorithm is used to solve for the maximum weight matching in the weighted bipartite graph. The time complexity is... The matching results indicate the heliostats that should be calibrated, their assigned camera slots, and the corresponding lighting combinations.

[0086] Time complexity : Describes the trend of the running time of the Hungarian algorithm as the size of the left node set U changes, usually representing the number of executions in the worst case.

[0087] Step S8: Calibration is performed for each successfully matched heliostat h: 1. Remove from the waiting queue.

[0088] 2. Add it to the service set of the corresponding camera c. .

[0089] 3. Record the camera (c) and lighting used. Calibration execution information, such as calibration execution time.

[0090] 4. Immediately send a command to the mirror field controller to begin calibration.

[0091] 5. After calibration, release resources, save sample information, time information, and camera and light information to predict calibration time.

[0092] When heliostat calibration fails: 1. Immediately from Remove from the list, record failure information, and release resources.

[0093] 2. Number of update failures f h ←f h+1 Recalculate priority .

[0094] 3. Re-add to the waiting queue and trigger a new event.

[0095] This invention utilizes the resources of cameras and lights in the heliostat field to the maximum extent by dynamically prioritizing heliostats and calculating the dynamic capacity of cameras, thereby completing calibration execution with the highest concurrency, improving the efficiency of heliostat field calibration, and reducing the calibration cycle.

[0096] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not limited to the embodiments shown herein, but should be accorded the widest scope of protection consistent with the principles and novel features disclosed herein.

Claims

1. A method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field, characterized in that, Includes the following steps: S1. Divide the heliostat into sections to form a heliostat-camera-lamp combination; S2. Establish the camera dynamic capacity model and start the calibration time prediction model; S3. Construct a weighted bipartite graph model; S4. Select the heliostat to be tested; S5. Calculate dynamic priority; S6. Determine if a camera capacity update has been triggered. If yes, update the camera capacity; otherwise, proceed to S7. S7. Solve for the maximum weight matching; S8. Perform calibration.

2. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, Step S1 involves performing geometric optical tracing based on the mirror field layout diagram, calibration tower locations, camera installation parameters on each tower, and lamp positions to pre-generate an initial usable "camera-lamp" combination for each heliostat; for any heliostat h, its combination with camera c and lamp... The combination must simultaneously satisfy the following three conditions: Field of view conditions: The line connecting the center of the heliostat and the optical center of the camera is within the camera's field of view. Reflection conditions: The direction of the heliostat's normal must be able to reflect light from the lamp. The light is reflected to camera c; Lighting conditions: The direction of the light beam should cover the heliostat, meaning the heliostat is within the illumination range of the light.

3. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, In step S2, a camera dynamic capacity model is established: Let... Let the set of heliostats using camera c at time t define the effective distance attenuation factor of camera c at time t. for: ； For a single heliostat h using camera c, the distance attenuation factor... Defined as: ； The distance from the heliostat h to the camera c; The camera reference distance is the standard calibration distance from the heliostat mirror to the camera's imaging surface, taken as the camera's design reference distance. Distance attenuation constant; The set of heliostats using camera c at time t is empty.

4. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, Establish a dynamic capacity model for the camera, based on the capacity update mechanism of the farthest heliostat: Then the real-time concurrent capacity of the camera at time t for: ； To round down; This represents the maximum concurrent capacity of the camera. Signal-to-noise ratio factor, which is related to ambient lighting, is defined as: ； The current ambient light level; For reference ambient illuminance; Light distance attenuation factor; According to the inverse square law of optics, light intensity is inversely proportional to the square of the distance. Therefore, the light distance attenuation factor is defined as: ； This is the reference distance for the lamp, serving as the benchmark reference value for calibrating the light intensity of the light source; For the heliostat h to the light The distance; The minimum angular resolution required for single-sided heliostat spot identification; : The field of view of camera c.

5. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, Step S3: Construct a weighted bipartite graph model: Treat each camera slot as an independent node and construct a weighted bipartite graph G=(U,V,E): Left node set U: Take the first M heliostats from the waiting queue. Δ is the scheduling redundancy coefficient; The camera slot is defined as follows: Each camera c generates based on its current dynamic capacity. Each slot node; each slot can only be assigned to one heliostat at a time. If different slots come from the same physical camera, but are treated as completely independent right nodes in a bipartite graph model; The right node set V: all available camera slots, generated for each camera c. Each slot node ; Edge set E: If the heliostat h can be combined Then h is related to all the slot nodes of that camera. Connect the edges between them; Edge weight is defined as: ； Heliostat current priority; Heliostat h uses combination Predicted calibration time; Waiting time; Overall signal-to-noise ratio factor; , , , Weighting coefficients, satisfying .

6. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, Dynamic priority of each heliostat Defined as: ； Basic priority, all heliostats have the same initial value; Failure penalty coefficient; Total number of failures; Waiting time coefficient; Time since the last calibration attempt; Emergency mission reward coefficient; Emergency task indicator function: 1 if the heliostat is marked as emergency, 0 otherwise; The capacity is recalculated at the following times: Before assigning a new task: When attempting to assign a new task to camera c, it is necessary to first determine the current task assignment method. ,calculate To determine the available slots; When the task is completed: When the calibration of a heliostat on camera C is completed, from After removing from the middle, recalculate. This is to reflect potential increases in capacity; When the environment changes: When the change exceeds a certain threshold, the capacity of all cameras is recalculated.

7. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, Solve for the maximum weight matching: For each camera c, the number of slots currently occupied. The number of available slots is: ； Each available camera slot is added to the weighted bipartite graph model. The Hungarian algorithm is used to solve the maximum weight matching problem in the weighted bipartite graph. The matching results give the heliostats that should be calibrated, their assigned camera slots, and the corresponding lighting combinations.

8. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, Calibration is performed for each heliostat h that is successfully matched: S81. Remove from the waiting queue; S82. Add it to the service set of the corresponding camera c. ; S83. Record the camera c used, light. Calibration execution time and other calibration execution information; S84. Immediately issue a command to the mirror field controller to begin calibration; S85. After calibration, release resources, save sample information, time information, and camera and light information for calibration time prediction.

9. The method for calibrating and scheduling multi-tower coordinated heliostats in a tower-type photothermal mirror field according to claim 1, characterized in that, When heliostat calibration fails: Step 1: Immediately from Remove from the middle, record failure information, and release resources; Step 2: Update the number of failures f h ←f h+1 Recalculate priority ; Step 3: Re-add to the waiting queue and trigger a new event.

Images