A photovoltaic power station energy efficiency intelligent optimization method based on cloud edge cooperation

Abstract

This invention discloses a cloud-edge collaborative intelligent optimization method for photovoltaic power plant energy efficiency, relating to the field of photovoltaic power generation technology. It includes collecting descriptive information of all pending tasks from the edge of the photovoltaic power plant and aggregating it in the cloud through a cloud-edge collaborative architecture. These tasks include equipment operation adjustment tasks, fault handling tasks, and energy efficiency optimization tasks. The method performs multi-dimensional analysis based on task dependency chain length, cross-device or cross-system collaboration frequency, and data sharing requirements, constructing a preliminary evaluation matrix of task association strength values ​​to obtain the distribution of association strength between tasks. This cloud-edge collaborative intelligent optimization method for photovoltaic power plant energy efficiency solves the problems of insufficient task association analysis, unreasonable priority determination, and insufficient cloud-edge collaborative scheduling capabilities in existing technologies. This improves the collaborative efficiency and resource utilization of multi-task processing in photovoltaic power plants, achieving intelligent optimization and stable improvement of the overall energy efficiency level of the power plant.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and specifically to a method for intelligent optimization of energy efficiency in photovoltaic power plants based on cloud-edge collaboration. Background Technology

[0002] In the field of photovoltaic power plant operation and management, with the continuous expansion of installed capacity and the increasing diversification of equipment types, the number of tasks involved in power plant operation, such as regulation and control, equipment maintenance, and anomaly handling, has increased significantly. To ensure the stable operation of the power generation system and improve overall power generation efficiency, it is usually necessary to manage and schedule multiple types of operational tasks in a unified manner. In existing technologies, centralized monitoring systems or traditional operation and maintenance management platforms are mostly relied upon to record, distribute, and control the execution of various tasks, which can achieve basic operation and maintenance and scheduling functions to a certain extent.

[0003] However, in practical applications, the operation of photovoltaic power plants is characterized by significant complexity and dynamism. On the one hand, different tasks often have dependencies or mutual influences. For example, equipment regulation tasks may be coupled with fault handling tasks; if the processing order is unreasonable, it may adversely affect power generation efficiency. On the other hand, during the execution of tasks, their urgency and scope of impact dynamically change with changes in equipment status, environmental conditions, and operating load, increasing the difficulty of unified scheduling. Furthermore, existing technologies, when processing multiple tasks in parallel, typically lack effective analysis mechanisms for the degree of correlation between tasks, making it difficult to accurately identify task conflict risks and coordination needs, resulting in some critical tasks not being executed first. Simultaneously, the process of determining task priorities relies heavily on fixed rules or manual experience, lacking quantitative evaluation methods that incorporate factors affecting power generation efficiency (such as power loss and equipment impact range), easily leading to unreasonable priority determinations and thus affecting overall energy efficiency. Moreover, with the development of cloud computing and edge computing technologies, photovoltaic power plants are gradually introducing cloud-edge collaborative architectures to improve data processing and response capabilities; however, in existing solutions, the task division and collaboration mechanisms between the cloud and the edge are still imperfect. Especially during task scheduling and execution, there is still a lack of effective means to combine the real-time response capabilities of the edge with the global optimization capabilities of the cloud to dynamically adjust the task execution order. In multi-task, high-concurrency scenarios, this can easily lead to uneven resource allocation and decreased scheduling efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a smart optimization method for the energy efficiency of photovoltaic power plants based on cloud-edge collaboration, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a smart energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration, comprising: S1. Through the cloud-edge collaboration architecture, the description information of all tasks to be processed is collected from the edge of the photovoltaic power station and aggregated in the cloud. The tasks include equipment operation adjustment tasks, fault handling tasks and energy efficiency optimization tasks. The task dependency chain length, cross-device or cross-system collaboration frequency and data sharing requirements are analyzed in multiple dimensions to construct a preliminary evaluation matrix of task association strength values ​​and obtain the association strength distribution between tasks. S2. Based on the distribution of task association intensity, analyze the probability of task conflict and the progress synchronization requirements. Combined with information transmission delay and feedback loop count, generate a cross-influence map between tasks and determine the task association intensity level. S3. If the task association intensity level is higher than the preset threshold, the urgency level of the task will be determined by collecting data on the fault response time and power loss duration of the photovoltaic power station, combined with the range of equipment affecting power generation efficiency and the frequency of fault occurrence. S4. Based on the urgency level, extract key business parameters and adjustment resource demand information that affect the energy efficiency of photovoltaic power plants, integrate background data on task source type and expected recovery time, generate a comprehensive task priority score, and obtain a task priority ranking for energy efficiency optimization. S5. By prioritizing tasks, and combining the real-time processing capabilities of the edge side with the global scheduling capabilities of the cloud, analyze the frequency of cross-device collaboration and the number of associated tasks for priority tasks, construct a resource allocation priority queue, and determine the resource scheduling sequence for cloud-edge collaboration.

[0006] Preferably, S1 includes: Acquire real-time operational data from the edge of the photovoltaic power station and aggregate it in the cloud. Use semantic parsing technology to process the real-time operational data to generate standardized task descriptions. Identify the predecessor and successor relationships between task nodes based on standardized task descriptions, calculate the dependency chain length, and count the collaboration frequency and data sharing volume. The dependency chain length, collaboration frequency, and data sharing amount are used as multi-dimensional feature inputs for numerical weighted calculation to construct a preliminary evaluation matrix. Numerical density analysis was performed on the preliminary evaluation matrix to obtain the distribution of inter-task correlation strength.

[0007] Preferably, S2 includes: Obtain the raw execution time sequence and node dependency data from the task management database, calculate the topological path length and coupling depth between nodes in the directed acyclic graph, and obtain the task association strength. The probability of conflict is determined by multi-dimensional weighted calculation based on the task association strength and the real-time resource occupancy rate, and by logical mapping using the communication frequency per unit time. If the probability of conflict is higher than the preset risk threshold, the information transmission delay and feedback loop count in the underlying protocol stack are extracted, and the buffer time in the task chain is dynamically corrected through a time delay compensation algorithm. The progress synchronization model is input with buffer time as a spatiotemporal constraint. The logical topology and interference paths between tasks are mapped in a multidimensional vector space to generate a cross-association graph. Based on the edge weight distribution and node clustering characteristics in the cross-association graph, the task association strength level is divided by clustering analysis algorithm.

[0008] Preferably, S3 includes: Obtain the task association strength value. If the task association strength value is higher than the preset threshold, collect the fault response time limit value and the power loss duration value. The range of damaged equipment is located based on the fault response time limit value and the power loss duration value, and the proportion of affected installed capacity corresponding to the range of damaged equipment is calculated. Based on the frequency of failures within the scope of damaged equipment, and combined with the proportion of affected installed capacity, a quantitative index of urgency is constructed. The urgency level of a task is determined based on the numerical value of its urgency quantification index.

[0009] Preferably, S4 includes: Acquire monitoring data and extract component status; Determine the deviation value based on the component status; If the deviation value exceeds the limit, an initial urgency level is generated; Adjust the gap based on the initial emergency quantification; Determine whether the energy storage margin meets the regulation gap to generate an intervention signal.

[0010] Preferably, S4 further includes: Regarding the intervention signal matching recovery time; Calculate the loss rate using recovery time and adjustment gap; A comprehensive score is generated by weighting the loss rate; The priority ranking of tasks for energy efficiency optimization is obtained based on the comprehensive score.

[0011] Preferably, S5 includes: Extract the collaboration frequency of priority tasks, and quantify the cross-device collaboration frequency based on the collaboration frequency; The edge load is obtained by matching the number of associated devices based on cross-device collaborative frequency, and the cloud bandwidth is obtained based on the edge load.

[0012] Preferably, S5 further includes: The device energy consumption is obtained based on the cloud bandwidth, and the device energy consumption is input into the resource allocation model to build a resource allocation priority queue; If the resource allocation priority queue meets the coordination cycle requirements, then the resource scheduling sequence for cloud-edge coordination is determined.

[0013] Preferably, it also includes S6, dynamically updating the execution order of photovoltaic power plant energy efficiency optimization tasks according to the resource scheduling sequence, monitoring the execution of priority tasks in real time at the edge, and performing global optimization and adjustment in the cloud based on changes in power generation and expected recovery time to obtain the final hierarchical processing sequence, so as to achieve intelligent optimization of the overall energy efficiency of the photovoltaic power plant, specifically including: Obtain the resource scheduling sequence and task set to generate an initial task execution order list; Priority tasks are selected from the initial task execution order list and sent to the edge computing unit, which then monitors the priority tasks in real time and generates monitoring data.

[0014] Preferably, S6 further includes: Analyze the monitoring data to extract characteristic values ​​of power generation changes, and if the characteristic values ​​of power generation changes are abnormal, calculate the expected recovery time. The characteristic values ​​of power generation change and expected recovery time are uploaded to the cloud for global optimization, resulting in a final hierarchical processing sequence to achieve intelligent optimization of the overall energy efficiency of the photovoltaic power station.

[0015] As can be seen from the above technical solution, the present invention has the following beneficial effects: This cloud-edge collaborative intelligent optimization method for photovoltaic power plant energy efficiency unifies the modeling and multi-dimensional analysis of various tasks during the operation of the photovoltaic power plant under a cloud-edge collaborative architecture. It constructs a task correlation strength assessment mechanism to accurately identify the dependencies and cross-influences between tasks. Based on this, it quantitatively assesses the urgency of tasks by combining energy efficiency-related factors such as power loss, equipment impact range, and fault frequency, thereby generating a more scientific and reasonable task priority ranking. Simultaneously, by integrating real-time processing capabilities on the edge side with global optimization capabilities on the cloud, it constructs a dynamic resource scheduling mechanism to prioritize and monitor high-priority tasks in real time, dynamically adjusting based on changes in power generation and expected recovery time. This effectively solves the problems of insufficient task correlation analysis, unreasonable priority determination, and insufficient cloud-edge collaborative scheduling capabilities in existing technologies, thereby improving the collaborative efficiency and resource utilization of multi-task processing in photovoltaic power plants and achieving intelligent optimization and stable improvement of the overall energy efficiency level of the power plant. Attached Figure Description

[0016] Figure 1 This is a flowchart of the intelligent energy efficiency optimization method for photovoltaic power plants according to the present invention. Detailed Implementation

[0017] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, this invention provides a technical solution: a smart optimization method for the energy efficiency of photovoltaic power plants based on cloud-edge collaboration, comprising: S1. Through the cloud-edge collaboration architecture, the description information of all tasks to be processed is collected from the edge of the photovoltaic power station and aggregated in the cloud. The tasks include equipment operation adjustment tasks, fault handling tasks and energy efficiency optimization tasks. The task dependency chain length, cross-device or cross-system collaboration frequency and data sharing requirements are analyzed in multiple dimensions to construct a preliminary evaluation matrix of task association strength values ​​and obtain the association strength distribution between tasks. S2. Based on the distribution of task association intensity, analyze the probability of task conflict and the progress synchronization requirements. Combined with information transmission delay and feedback loop count, generate a cross-influence map between tasks and determine the task association intensity level. S3. If the task association intensity level is higher than the preset threshold, the urgency level of the task will be determined by collecting data on the fault response time and power loss duration of the photovoltaic power station, combined with the range of equipment affecting power generation efficiency and the frequency of fault occurrence. S4. Based on the urgency level, extract key business parameters and adjustment resource demand information that affect the energy efficiency of photovoltaic power plants, integrate background data on task source type and expected recovery time, generate a comprehensive task priority score, and obtain a task priority ranking for energy efficiency optimization. S5. By prioritizing tasks, and combining the real-time processing capabilities of the edge side with the global scheduling capabilities of the cloud, analyze the frequency of cross-device collaboration and the number of associated tasks for priority tasks, construct a resource allocation priority queue, and determine the resource scheduling sequence for cloud-edge collaboration. S6. Based on the resource scheduling sequence, dynamically update the execution order of photovoltaic power plant energy efficiency optimization tasks, monitor the execution of priority tasks in real time on the edge side, and perform global optimization and adjustment in the cloud in combination with changes in power generation and expected recovery time to obtain the final hierarchical processing sequence, so as to realize intelligent optimization of the overall energy efficiency of photovoltaic power plants.

[0019] This implementation is based on a cloud-edge collaborative computing framework. By deploying a data acquisition and preliminary processing module at the edge of the photovoltaic power station and a global scheduling and analysis module in the cloud, it achieves closed-loop control of task awareness, correlation analysis, and scheduling optimization. At the edge, data from inverters, combiner boxes, tracking brackets, and environmental monitoring equipment is collected via industrial communication protocols (such as Modbus and IEC 61850). Tasks are initially classified and tagged, and task description information (including task type, timestamp, involved equipment, and data requirements) is uploaded to the cloud.

[0020] In the cloud, a task association analysis model is first constructed. By defining parameters such as task dependency chain length (e.g., representing task dependencies based on a directed acyclic graph), cross-device collaboration frequency (statistically counting the number of times a task calls different devices), and data sharing requirements (based on data access overlap), a weighted matrix method is used to establish a task association strength evaluation matrix, and the distribution of task association strength is obtained through normalization.

[0021] Furthermore, by establishing a task cross-influence graph, tasks are regarded as nodes and the correlation strength is used as edge weight. Combining queuing theory model and time delay analysis model, the probability of task conflict and synchronization constraint relationship are calculated. At the same time, information transmission delay (such as network delay) and feedback loop number (control loop number) are introduced as correction factors to classify the task correlation strength level (such as low, medium and high).

[0022] For highly relevant tasks, the system introduces an urgency assessment model, which integrates fault response time (e.g., second-level or minute-level requirements), power loss duration (unit: kWh loss time), affected equipment range (single device or multiple arrays), and fault occurrence frequency (number of times within a statistical period) to construct a multi-parameter weighted scoring function, thereby quantifying the urgency level.

[0023] Subsequently, by integrating key business parameters (such as power generation deviation, equipment load rate, MPPT efficiency) and resource requirements (computing resources, communication bandwidth, control permissions), and combining the task source (automatic detection or manual scheduling) and expected recovery time, a task priority model is constructed (e.g., using the analytic hierarchy process or machine learning ranking model), and the task priority ranking is output.

[0024] During the resource scheduling phase, the system combines the real-time computing capabilities (CPU utilization, memory usage) of edge nodes with the global resource pool capabilities of the cloud. By analyzing the cross-device collaboration frequency of tasks and the number of associated tasks, it constructs a priority-based resource allocation queue and uses scheduling algorithms (such as priority queue scheduling or reinforcement learning scheduling) to generate a cloud-edge collaborative resource scheduling sequence.

[0025] Finally, the task execution order is dynamically adjusted according to the scheduling sequence. Real-time monitoring and rapid response control are performed on the edge side, while the cloud performs rolling optimization of the task execution strategy based on the global power generation change trend and task recovery time prediction, thereby forming a hierarchical processing sequence to achieve optimal energy efficiency.

[0026] S1 includes acquiring real-time operational data from the edge of the photovoltaic power station and aggregating it in the cloud; processing the real-time operational data using semantic parsing technology to generate standardized task descriptions; identifying predecessor and successor relationships between task nodes based on the standardized task descriptions; calculating dependency chain lengths and statistically analyzing collaboration frequency and data sharing volume; using dependency chain lengths, collaboration frequency, and the data sharing volume as multi-dimensional feature inputs for numerical weighted calculations to construct a preliminary evaluation matrix; and performing numerical density analysis on the preliminary evaluation matrix to obtain the distribution of inter-task correlation strength.

[0027] In one possible implementation, the process of acquiring and modeling the real-time operation data of the photovoltaic power station edge side involves firstly continuously collecting the operating status of various devices at the edge side through a data acquisition unit. These devices include inverters, combiner boxes, meteorological monitoring devices, and protection units. The collected data includes DC-side voltage, current, AC output power, device operating status codes, and alarm information. The acquisition cycle is set to 1 to 5 seconds, determined based on the dynamic change rate of the devices. For inverters with rapid power fluctuations, a 1-second cycle is used, while for environmental parameters, a 5-second cycle is used to balance data accuracy and communication load. After acquisition, the data is encapsulated in chronological order through the edge-side communication module and sent to the cloud server via a message transmission protocol. During transmission, timestamp alignment is used to ensure data synchronization consistency. The timestamp accuracy is set to milliseconds, determined based on the required fine-grained scheduling.

[0028] After receiving data in the cloud, the raw data is first processed through semantic parsing. Specifically, the data from different devices is mapped to fields according to a preset data dictionary. Voltage, current, power, and status codes are uniformly converted into standard field names. At the same time, task trigger identifiers are generated based on status codes and operating threshold rules. For example, when the power deviates from the rated value by more than 10%, it is marked as an adjustment task. This 10% threshold is determined based on the upper limit of the normal range of power fluctuation in historical operating statistics. Subsequently, a task type label, the number of the device involved, and the trigger time are added to each piece of data to form a task description unit with a unified structure.

[0029] During the task relationship identification process, all task description units are arranged in chronological order and a directed relationship structure is constructed. The predecessor and successor tasks are determined by judging the trigger sequence and data dependency relationship between tasks. For example, when the output data of task A is used as the input condition of task B, task A is defined as the predecessor task and task B is defined as the successor task. On this basis, each task path is traversed and the number of nodes traversed from the starting task to the current task is counted to obtain the dependency chain length. This length is used to reflect the complexity of the task execution path, and its maximum statistical range is limited to 10 levels. This upper limit is determined based on the controllable range of system scheduling complexity.

[0030] During the collaborative frequency statistics process, the number of devices involved in the execution of each task is counted, and the number of times the task calls different devices per unit time is counted. The statistical time window is set to 10 minutes, which is determined based on the cycle of changes in the operating status of the photovoltaic power station to ensure the stability of the statistical results. During the data sharing calculation process, the number of data fields shared between tasks is counted, and the proportion of shared fields in the total number of fields is calculated. This proportion ranges from 0 to 1 and is used to reflect the degree of data coupling between tasks.

[0031] In the multi-dimensional feature fusion stage, the dependency chain length, collaboration frequency, and data sharing amount are processed with unified dimensions. Specifically, the dependency chain length is normalized to the maximum value of 10, the collaboration frequency is normalized to the maximum observation value within the statistical window, and the data sharing amount is directly used as a proportion in the calculation. Then, weights are assigned to the three features respectively: the dependency chain length weight is set to 0.4, the collaboration frequency weight is set to 0.3, and the data sharing amount weight is set to 0.3. The weight values ​​are determined based on the degree of influence of each factor on task coupling in the historical scheduling effect evaluation. By weighted summing of the three normalized features, the correlation strength value between any two tasks is calculated, and the pairwise calculation results of all tasks are constructed into a matrix structure to form a preliminary evaluation matrix.

[0032] In the numerical density analysis process, the distribution statistics of all correlation strength values ​​in the preliminary evaluation matrix are performed. First, the values ​​are sorted according to their size, and then divided into several continuous intervals. The width of each interval is set to 0.05, which is determined according to the required level of detail in the numerical distribution. The frequency of the values ​​in each interval is counted to form a density distribution curve. The main distribution range of task correlation is identified based on the density concentration area. Further, the correlation strength level intervals are divided according to the distribution results. For example, the numerical range corresponding to the interval with the highest density is defined as the medium correlation interval, the interval above it is defined as the high correlation interval, and the interval below it is defined as the low correlation interval. This division is determined based on the density peak position in the statistical results, thus obtaining the correlation strength distribution between tasks.

[0033] S2 includes acquiring the execution time sequence and node dependency raw data from the task management database; calculating the topological path length and coupling depth between nodes in a directed acyclic graph to obtain the task association strength; performing multi-dimensional weighted calculation based on the task association strength and real-time collected resource occupancy rate; using the communication frequency per unit time for logical mapping to determine the conflict probability; if the conflict probability is higher than a preset risk threshold, extracting the information transmission delay and feedback loop count from the underlying protocol stack; dynamically correcting the buffer time in the task chain using a latency compensation algorithm; inputting the buffer time as a spatiotemporal constraint into the progress synchronization model; mapping the logical topological structure and interference paths between tasks in a multi-dimensional vector space to generate a cross-association graph; and classifying the task association strength level using a clustering analysis algorithm based on the edge weight distribution and node clustering characteristics in the cross-association graph.

[0034] In one possible implementation, for the task association strength analysis and conflict identification process, execution time sequence data and node dependency raw data are first extracted from the task management database. Specifically, the start time, end time, and execution sequence number of each task are read in chronological order, and the call relationship records and data transfer records between tasks are extracted. Based on this, each task is defined as a node, and the records of dependencies between tasks are converted into directed connections, thereby constructing a directed acyclic structure. During the construction process, by comparing the task execution times one by one, when the start time of one task is later than the end time of another task and there is a data call record, the former task is marked as the predecessor node and the latter task is marked as the successor node to ensure the accuracy of the dependency relationship.

[0035] In the topology path length calculation process, starting from any target task node, all predecessor nodes are traced forward level by level along the dependency relationship until no predecessor node exists. The number of nodes traversed in each path is counted, and the value with the largest number of nodes in the path is taken as the topology path length of the task. The maximum upper limit of the path length is set to 15. This value is determined based on the statistical upper limit of the task serialization level in the actual operation of the photovoltaic power station. When the actual calculation result exceeds 15, it is treated as 15 to avoid abnormal data affecting the result. In the coupling depth calculation process, the number of predecessor nodes and successor nodes directly associated with each node are counted. The two are added together to obtain the number of connections of the node. Then, the number of connections of all nodes in the same task path is accumulated one by one and divided by the total number of nodes in the path to obtain the average connection level. This average value is the coupling depth. This parameter is determined based on the density of connections between nodes.

[0036] Subsequently, the topology path length and coupling depth are numerically unified. Specifically, the topology path length is divided by the maximum upper limit of 15 to convert it into a proportional value between 0 and 1. This processing method is determined based on the comparability requirements between different tasks. The coupling depth is divided by the maximum number of node connections observed during system operation. This maximum number of connections is obtained through statistics from historical operation data and is usually between 8 and 12, depending on the specific power plant scale. After the unified processing is completed, the two values ​​are weighted and summed. The weight of the topology path length is set to 0.5, which is determined based on the degree of influence of task chain depth on scheduling complexity. The weight of the coupling depth is set to 0.5, which is determined based on the degree of influence of the interaction density between tasks on system load. The task association strength value is obtained by multiplying each value and summing them.

[0037] During the conflict probability calculation process, real-time resource utilization data is first acquired, including processor utilization, memory utilization, and communication bandwidth utilization. Processor utilization is obtained by reading the current processor usage ratio through the system monitoring interface, memory utilization is obtained by reading the current used memory as a percentage of total memory through the operating system's memory management module, and communication bandwidth utilization is obtained by calculating the proportion of data transmission volume per unit time to the maximum bandwidth through the network traffic monitoring module. Subsequently, the three resource utilization rates are averaged to obtain the comprehensive resource utilization level. This average value is used to reflect the current overall load status of the system.

[0038] Simultaneously, the communication frequency between tasks is statistically analyzed. Specifically, the total number of data interactions between tasks is calculated within a continuous 60-second time window. This time window is determined based on the system's communication change cycle to ensure the stability of the statistical results. The communication frequency is then divided by the maximum number of communications obtained from historical statistics to convert it into a proportional value between 0 and 1. This maximum number of communications is obtained through long-term operational data statistics.

[0039] In determining the conflict probability, the task association strength and the overall resource occupancy level are weighted. The task association strength weight is set to 0.6, which is determined based on the dominant role of the relationship between tasks in the formation of conflict. The overall resource occupancy level weight is set to 0.4, which is determined based on the auxiliary role of system load in the formation of conflict. The two are multiplied by their respective weights and then summed. The communication frequency ratio is used as an adjustment factor for correction. When the communication frequency is high, the conflict probability value is increased, thus obtaining the final conflict probability.

[0040] When the probability of a conflict exceeds a risk threshold of 0.7, an adjustment mechanism is triggered. This threshold is statistically derived from the critical point of probability of significant task conflicts in historical operational data. After triggering, information transmission delay data is extracted from the underlying communication protocol processing module. Specifically, the time difference between the data packet sending time and the receiving time is recorded, and the average value of multiple consecutive communication results is taken to eliminate random fluctuations. At the same time, the number of feedback loops is counted, which is the number of complete round trips from sending to receiving confirmation during task execution. This number is obtained by counting each item in the system log.

[0041] During the buffer time correction process, the initial buffer time between tasks is first determined, which is set to 2 seconds. This value is determined based on the normal level of device response time and communication latency. Then, the buffer time is increased based on the measured information transmission latency. When the average latency is higher than the original set value, the latency difference is directly added to the buffer time. At the same time, the buffer time is further adjusted based on the number of feedback loops. The number of feedback loops is multiplied by the average latency per loop and then added to the buffer time to obtain the corrected buffer time, so that the waiting time between tasks is consistent with the actual communication situation.

[0042] In the process of constructing the progress synchronization model, each task is converted into a point in space. First, the task start time is used as the first dimension coordinate, the corrected buffer time is used as the second dimension coordinate, and the comprehensive resource occupancy level is used as the third dimension coordinate. The position of the task in space is determined by the above three data. Then, the connection relationship between points is established according to the dependency relationship between tasks. At the same time, the task pairs that may cause interference are identified according to the resource competition relationship, and corresponding connections are established in space, thus forming a structure that includes logical dependency paths and interference paths.

[0043] During the generation of the cross-association graph, the above spatial structure is organized, and the task nodes and their connections are mapped into a graph structure, where nodes represent tasks, connections represent dependencies or interference relationships between tasks, and each connection is assigned a weight, which is determined by the task association strength and conflict probability. The importance of the connection is reflected by the combination of these two values.

[0044] In the process of classifying the association strength level, all connection weights in the graph structure are sorted and divided into three intervals according to their numerical distribution. The number of these intervals is determined based on the task complexity layering requirements. Then, the nodes are clustered, with nodes having high and concentrated connection weights classified as high association category, nodes with weights in the middle range classified as medium association category, and nodes with low weights classified as low association category, thus completing the task association strength level classification.

[0045] S3 includes obtaining the task association strength value; if the task association strength value is higher than a preset threshold, collecting the fault response time limit value and the power loss duration value; locating the range of damaged equipment based on the fault response time limit value and the power loss duration value, calculating the proportion of affected installed capacity corresponding to the range of damaged equipment; statistically analyzing the fault occurrence frequency value based on the range of damaged equipment, and constructing an urgency level quantification index value in combination with the proportion of affected installed capacity value; and determining the task urgency level based on the urgency level quantification index value.

[0046] In one possible implementation, regarding the task urgency determination process, firstly, the task association strength values ​​obtained in the previous steps are read one by one and organized according to time sequence and task identifiers. After organization, all task association strength values ​​are uniformly sorted to obtain a sequence from largest to smallest. In this sequence, the association strength judgment threshold is set to 0.65. This threshold is determined based on the statistical results of association strength when significant linkage effects occur between tasks in the historical operation data of the photovoltaic power station. That is, by statistically analyzing the distribution of association strength values ​​corresponding to historical conflict events, a critical value that can distinguish obviously coupled tasks from general tasks is selected as the judgment criterion. Subsequently, the association strength value of each task is compared one by one. When the value of a certain task is greater than 0.65, the task is marked as a highly associated task, and the marking result is passed to the subsequent processing stage.

[0047] After completing the screening of highly correlated tasks, the device number involved in the corresponding task is located from the operation monitoring system, and the operation record of the device at the time of the fault is retrieved according to the device number. By reading the time record in the fault log, the fault occurrence time and the system response completion time are extracted. The time difference between the two is the fault response time limit value, which is in seconds. Its specific value is determined according to the device protection setting parameters. For example, the protection response time of the inverter is usually set to 30 seconds. This value comes from the device's factory protection strategy and operation safety specifications. At the same time, power change data is extracted from the power monitoring curve, and the power value in continuous time period is compared point by point. Timing starts when the power is lower than 90% of the rated power and ends when the power recovers to above 90% of the rated power. The 90% criterion is determined based on the statistical results of the lower limit of the photovoltaic system under normal fluctuation conditions. By statistically analyzing the duration of this time period, the power loss duration value is obtained.

[0048] In determining the scope of damaged equipment, an equipment list is first established based on the set of equipment associated with the task, and a joint judgment is made by combining the fault response time limit value and the power loss duration value. When the fault response time limit is less than 30 seconds but the power loss duration is greater than 300 seconds, it is determined that there is a multi-equipment diffusion effect. The 300-second threshold is determined based on the significant change point in the power recovery time distribution in historical data. When the fault response time limit is greater than 30 seconds and the power loss duration is less than 300 seconds, it is determined to be a single-equipment centralized fault. Subsequently, based on the judgment results, the equipment with electrical connection or control association with the faulty equipment is searched layer by layer according to the power plant equipment topology, and all related equipment is included in the scope of damaged equipment. After the scope is determined, the rated installed capacity value of each equipment is read one by one. This value comes from the power plant design parameter table. The installed capacity of all damaged equipment is summed to obtain the total damaged installed capacity. At the same time, the overall installed capacity value of the power plant is obtained. This value also comes from the design parameters. The total damaged installed capacity is divided by the overall installed capacity value to obtain the proportion of affected installed capacity, which is used to quantify the scope of the fault impact.

[0049] In the process of calculating the frequency of failures, historical failure records are read in chronological order and classified according to failure type, with failures of the same type grouped together. The statistical time window is set to 24 hours, which is determined based on the daily cycle operation characteristics of the photovoltaic power station. The number of times the same type of failure occurs within each 24-hour cycle is counted to obtain the daily failure frequency value. At the same time, the data for the previous 7 consecutive days is traced back, and the 7 statistical results are accumulated item by item and divided by 7 to obtain the average failure frequency value. This 7-day cycle is determined based on the stability requirements of the operating data and is used to reduce the impact of occasional fluctuations on the results.

[0050] In the process of quantifying the urgency level, the proportion of affected installed capacity and the average failure frequency are processed in a unified manner. Specifically, the failure frequency is divided by the maximum number of failures obtained from historical statistics, converting it into a proportion value between 0 and 1. The maximum number of failures is obtained through long-term operation records. Then, the two values ​​are weighted, with the weight of the proportion of affected installed capacity set at 0.6, which is determined based on the degree of impact of the failure on power generation capacity, and the weight of the failure frequency set at 0.4, which is determined based on the degree of impact of failure repetition on system stability. The urgency level quantification index is obtained by multiplying the two values ​​by their respective weights and then summing them up.

[0051] In the process of classifying the urgency level, the quantitative indicators of urgency for all tasks are sorted and the level ranges are divided according to the distribution of the values. When the quantitative indicator value is greater than 0.7, it is classified as high urgency level. This threshold is obtained by statistically analyzing the range of task indicators that have a significant impact on power generation efficiency in historical operation. When the quantitative indicator value is between 0.4 and 0.7, it is classified as medium urgency level. This range is determined based on the fluctuation range of normal system operation. When the quantitative indicator value is less than 0.4, it is classified as low urgency level. By comparing the quantitative indicator values ​​of each task one by one and completing the level marking, the urgency level of the tasks can be accurately classified.

[0052] S4 includes acquiring monitoring data and extracting component status; determining deviation values ​​based on component status; generating an initial urgency level if the deviation value exceeds the limit; quantifying the adjustment gap based on the initial urgency level; determining whether the energy storage margin meets the adjustment gap to generate an intervention signal; matching recovery time to the intervention signal; calculating the loss rate using the recovery time and the adjustment gap; generating a comprehensive score by adding weights to the loss rate; and obtaining a task priority ranking for energy efficiency optimization based on the comprehensive score.

[0053] In one possible implementation, regarding the task priority generation process, real-time operating data is first obtained from the photovoltaic power station monitoring system line by line. Specifically, the output power, voltage, current, and temperature data of each photovoltaic module are read sequentially according to the equipment number, and the data at the same time point are aligned to ensure the comparability of the data of each module through a unified timestamp. After the data is processed, the current output power of each module is compared with its rated power. The rated power value is determined according to the equipment's factory parameters and is imported into the equipment parameter library during the system initialization phase. The operating status ratio is obtained by dividing the current output power by the rated power, which is used to reflect the current performance level of the module.

[0054] During the component status determination process, the voltage and current data are checked for range. The current voltage is compared with the rated voltage, and the current current is compared with the rated current. When any parameter deviates from the rated value by more than 10, it is marked as abnormal. The deviation range of 10 is determined based on the upper limit of normal fluctuation in the long-term operation statistics of the equipment. Then, a comprehensive judgment is made in combination with the power operation ratio. When the operation ratio is greater than 0.9 and the electrical parameters are within the allowable range, it is judged as a normal state. When the operation ratio is between 0.7 and 0.9 or a single electrical parameter deviates, it is judged as a slightly abnormal state. When the operation ratio is less than 0.7 or multiple electrical parameters deviate, it is judged as a seriously abnormal state.

[0055] During the deviation calculation process, each abnormal component is processed individually. The rated power of the component is subtracted from the current output power to obtain the power loss value, which is then divided by the rated power to obtain the deviation value of a single component. Subsequently, the deviation values ​​of all abnormal components are accumulated and divided by the number of abnormal components to obtain the average deviation value. This average value is used to reflect the overall deviation of the current task. In the deviation determination stage, the average deviation value is compared with the deviation threshold of 0.15. This threshold is determined based on the statistical results of historical operation where a significant decrease in power generation efficiency occurs when the power deviation exceeds 15. When the average deviation value is greater than 0.15, it is determined to be an over-limit deviation, and the urgency calculation process begins.

[0056] During the initial urgency determination process, the urgency is classified according to the range of the average deviation value. When the average deviation value is greater than 0.15 and less than or equal to 0.3, the initial urgency is marked as 1, and this range is determined based on the range of mild power generation loss. When the average deviation value is greater than 0.3 and less than or equal to 0.5, the initial urgency is marked as 2, and this range is determined based on the range of moderate power generation loss. When the average deviation value is greater than 0.5, the initial urgency is marked as 3, and this range is determined based on the range of severe power generation loss.

[0057] During the adjustment gap calculation process, the power loss values ​​of all abnormal components are accumulated one by one to obtain the total power gap value, which is used to represent the power generation capacity that the current system needs to compensate for. At the same time, the total installed capacity value of the power station is obtained, which comes from the power station design parameters. The total power gap is compared with the total installed capacity to determine the proportion of the gap in the overall system.

[0058] During the energy storage capacity assessment process, the current available energy storage capacity is read from the energy storage management system. This value is calculated using the battery state of charge. The available energy storage capacity is compared with the total power deficit item by item. When the available energy storage capacity is greater than or equal to the total power deficit, it is determined to be in a fully satisfied state, and an intervention signal is generated. When the available energy storage capacity is less than the total power deficit, it is determined to be in a partially satisfied state, and a limited intervention signal is generated. This assessment is based on the matching relationship between the actual discharge capacity of the energy storage system and the current power deficit.

[0059] During the recovery time determination process, the historical operation record database is accessed based on the intervention signal type, and historical events consistent with the current task conditions are filtered out. These conditions include the initial urgency level, power gap range, and energy storage intervention status. When the intervention signal is in a fully satisfied state, the corresponding recovery time data is extracted from the historical records, and all matching records are summed and divided by the number of records to obtain the average recovery time. When the intervention signal is in a restricted state, the recovery time under partial compensation conditions is extracted in the same way. The historical data filtering period is set to 30 days, which is determined based on the stability requirements of the operation data.

[0060] In the process of calculating the loss rate, the recovery time is first standardized and compared with the historical maximum recovery time to obtain the degree of time impact. The maximum recovery time is obtained through historical statistics. Then, the total power gap is compared with the total installed capacity of the power plant to obtain the degree of power loss. The degree of time impact and the degree of power loss are accumulated to obtain the loss rate value, which is used to reflect the intensity of the task's impact on overall energy efficiency.

[0061] In the comprehensive score calculation process, the loss rate is used as the core evaluation indicator, and it is combined with the task source type and expected recovery time for comprehensive processing. The task source type is distinguished according to the way the task is generated: manually triggered tasks are assigned a value of 1, and automatically detected tasks are assigned a value of 0.8. This value is determined based on the principle that tasks with human intervention are of higher importance. The expected recovery time is determined by the ratio of the current recovery time to the historical average recovery time. Subsequently, the three indicators are weighted. The weight of the loss rate is set at 0.5, which is determined based on its impact on energy efficiency. The weight of the task source type is set at 0.2, which is determined based on the task scheduling priority strategy. The weight of the expected recovery time is set at 0.3, which is determined based on the impact of the recovery speed on the system. The comprehensive score is obtained by multiplying and summing the results of each indicator.

[0062] In the task priority ranking process, the comprehensive scores of all tasks are sorted and arranged in descending order of value. The task with the highest comprehensive score is determined as the highest priority task, and the priority of the remaining tasks is determined in turn according to the ranking result, thus forming a task priority sequence for optimizing the energy efficiency of photovoltaic power plants.

[0063] S5 includes extracting the collaboration frequency of priority tasks, quantifying the cross-device collaboration frequency through the collaboration frequency; matching the number of associations based on the cross-device collaboration frequency to obtain the edge load, obtaining the cloud bandwidth based on the edge load; obtaining the device energy consumption based on the cloud bandwidth, inputting the device energy consumption into the resource allocation model to construct a resource allocation priority queue; if the resource allocation priority queue meets the collaboration cycle requirements, then determining the resource scheduling sequence for cloud-edge collaboration.

[0064] In one possible implementation, for the resource scheduling sequence generation process, priority tasks are first read sequentially from the task priority ranking results, and the device interaction records of each task during its historical operation are extracted based on the task execution log. Specifically, the corresponding log data is located according to the task number, and the number of data interactions between the task and different devices during the execution process is counted one by one and organized in chronological order. During the statistical process, a time window of 60 seconds is set. This time window is determined based on the typical cycle of photovoltaic power station control command issuance and feedback to ensure that the collaboration frequency can truly reflect the collaboration intensity within a short period. Then, all interaction counts within the time window are accumulated to obtain the collaboration frequency value.

[0065] In the process of quantifying the cross-device collaboration frequency, the current collaboration frequency value is compared with the maximum number of collaborations recorded in the historical operation data. The maximum number of collaborations is obtained by statistically analyzing the operation data over 30 consecutive days, and the maximum value is selected as the benchmark. The current collaboration frequency is divided by this maximum value, and the result is converted into a ratio value between 0 and 1, thereby obtaining the cross-device collaboration frequency. This ratio is used to characterize the degree of dependence of the task on multi-device collaboration resources.

[0066] During edge load calculation, the number of associated tasks for each task is first extracted from the task association data. Specifically, this involves counting the number of task nodes that have a direct connection with the task. During the counting process, an upper limit of 20 is set. This upper limit is determined based on the historical statistical results of the system's concurrent task scale and the processing capacity of edge nodes. When the statistical result exceeds 20, it is processed as 20 to prevent abnormal data from affecting the load assessment. Subsequently, the cross-device collaboration frequency and the number of associated tasks are processed item by item. The edge load value is obtained by multiplying the collaboration frequency by the number of associated tasks. This value is used to reflect the comprehensive processing pressure on the edge side when executing the task.

[0067] During the cloud bandwidth determination process, edge load values ​​are divided into three intervals: a low load interval (less than 5), a medium load interval (between 5 and 10), and a high load interval (greater than 10). These interval divisions are based on statistical analysis of historical load changes and network resource usage. After the intervals are determined, a corresponding bandwidth requirement is assigned to each load interval: 10 Mbps for the low load interval, 50 Mbps for the medium load interval, and 100 Mbps for the high load interval. These bandwidth values ​​are determined based on the actual configuration capabilities of the communication network and the test results of equipment transmission requirements.

[0068] In the process of calculating device energy consumption, the execution time data of each task is first obtained, specifically by reading the start time and end time of the task and calculating the time difference between them in seconds. Then, the unit time energy consumption standard is determined according to the bandwidth range corresponding to the task, where low bandwidth corresponds to a unit time energy consumption of 1, medium bandwidth corresponds to 3, and high bandwidth corresponds to 6. This unit time energy consumption value is determined based on the power consumption test results of the device under different communication load conditions. The unit time energy consumption and the task execution time are accumulated item by item, that is, the energy consumption is accumulated according to the energy consumption corresponding to each second, so as to obtain the device energy consumption value of the task.

[0069] In the process of constructing the resource allocation priority queue, the tasks are first initially sorted according to their priority ranking to ensure that high-priority tasks are at the top. Then, under the same priority conditions, the tasks are sorted again according to their device energy consumption values ​​from smallest to largest, so that tasks with lower resource consumption are executed first. The resource allocation priority queue is formed through the above two-level sorting method, thereby taking into account both the urgency of tasks and the efficiency of resource utilization.

[0070] During the coordination cycle determination process, the coordination cycle requirement set in the system scheduling strategy is obtained. This cycle is set to 300 seconds, and this value is determined based on the comprehensive statistical results of photovoltaic power station scheduling response time, equipment control cycle, and communication delay. Then, the execution time of each task is accumulated one by one according to the resource allocation priority queue to obtain the cumulative execution time. When the cumulative execution time is less than or equal to 300 seconds, it is determined that the current queue meets the coordination cycle requirement. When the cumulative execution time exceeds 300 seconds, tasks are removed one by one starting from the end of the queue, and the cumulative execution time is recalculated after each removal until the cumulative execution time meets the cycle limit.

[0071] During the resource scheduling sequence generation process, the priority queue of resource allocation that meets the coordination cycle requirements is expanded item by item in the current order to form a task execution sequence. Based on the location of the device to which the task belongs and the resource type, the tasks are allocated to the edge or the cloud for execution. Among them, tasks with high real-time requirements and low edge load are prioritized for execution on the edge, while tasks with high computational complexity or involving multi-device coordination are allocated to the cloud for execution, thus forming a cloud-edge collaborative resource scheduling sequence to achieve reasonable resource allocation and energy efficiency optimization during task execution.

[0072] S6 includes obtaining the resource scheduling sequence and task set to generate an initial task execution order list; selecting priority tasks from the initial task execution order list and sending them to the edge computing unit, which then monitors the priority tasks in real time to generate monitoring data; parsing the monitoring data to extract power generation change feature values, and calculating the expected recovery time if the power generation change feature values ​​are abnormal; uploading the power generation change feature values ​​and the expected recovery time to the cloud for global optimization to obtain the final hierarchical processing sequence to achieve intelligent optimization of the overall energy efficiency of the photovoltaic power station.

[0073] In one possible implementation, for the dynamic optimization process of task execution order, the resource scheduling sequence and the current task set are first obtained, and the two are matched item by item. Specifically, the task identifiers are read one by one according to the task order in the resource scheduling sequence, the corresponding task records are retrieved in the task set, and the matching accuracy is ensured by double verification through task number and timestamp. After the matching is completed, the order in the resource scheduling sequence is directly converted into the initial task execution order list, thereby forming a task sequence structure with a clear execution order relationship.

[0074] After the initial task execution order list is generated, the priority value of each task in the list is read and re-sorted according to the value from high to low. Then, the priority filtering threshold is set to the top 30 range. This threshold is determined based on the statistical results of the system's response time to critical tasks under high load conditions, that is, by analyzing the contribution ratio of the top 30 tasks to the overall power generation efficiency in historical operation. During the filtering process, tasks within this range are marked as priority tasks one by one, and a task list to be sent is generated in order. Then, the tasks are sent to the edge computing unit one by one through the scheduling communication interface, along with the task execution time, device number and control parameter information, to ensure that the tasks can be executed accurately.

[0075] After receiving the task, the edge computing unit starts the real-time monitoring process to continuously collect the power generation during the task execution. Specifically, the output power of the corresponding device is collected point by point according to a 1-second sampling period. This sampling period is determined based on the statistical results of the photovoltaic module power change response speed to ensure that rapid fluctuation information can be captured. During the collection process, the continuously sampled data is arranged in chronological order to form a monitoring data sequence.

[0076] In the process of calculating the characteristic value of power generation change, the power values ​​of adjacent time points in the monitoring data sequence are compared one by one, the power difference between each adjacent time point is calculated, and the continuous differences are accumulated to obtain the total power change in the current time period. Then, the total change is divided by the number of sampling points to obtain the average change amplitude, thus forming the characteristic value of power generation change, which is used to reflect the degree of power fluctuation. In the process of anomaly judgment, the characteristic value is compared with the change range under historical stable operating conditions. The upper limit of this range is set to 10 of the rated power, and this value is determined based on the statistical upper limit of normal fluctuation amplitude in long-term operating data. When the characteristic value exceeds this range, it is judged as an abnormal state.

[0077] During the calculation of expected recovery time, when an abnormal state is determined, records consistent with the current abnormal type are first screened from the historical operation database. The screening criteria include the power drop range, equipment type, and environmental condition range. The screening time range is set to the most recent 30 days, which is determined based on the requirements of environmental condition consistency and data validity. Recovery time data is extracted from each record in the screening results, and all recovery times are accumulated and divided by the number of records to obtain the average recovery time. This average value is used as the expected recovery time value.

[0078] During the data upload process, the characteristic values ​​of power generation change and the expected recovery time are sent to the cloud system through the communication module. The transmitted data includes task identifiers and time information to ensure that the data can be accurately matched with the corresponding tasks in the cloud. A sequential sending mechanism is adopted during the upload process to ensure complete data transmission.

[0079] During the cloud-based global optimization process, the received data is uniformly organized. First, it is categorized according to task identifiers, and the characteristic values ​​corresponding to each task are combined with the recovery time. Then, the power generation change characteristic values ​​of all tasks are sorted to obtain the impact ranking result. At the same time, the expected recovery time is sorted to obtain the recovery difficulty ranking result. In the comprehensive ranking process, the power generation change characteristic values ​​and the expected recovery time are weighted. The weight of the power generation change characteristic values ​​is set to 0.6, which is determined based on the direct impact of power fluctuations on power generation efficiency. The weight of the expected recovery time is set to 0.4, which is determined based on the continuous impact of the recovery process on the system. The comprehensive evaluation value of each task is obtained by accumulating the results item by item.

[0080] During the final hierarchical processing sequence generation process, the comprehensive evaluation values ​​of all tasks are sorted and divided into three level intervals according to their numerical values. When the comprehensive evaluation value is greater than 0.7, it is classified as a high processing level. This threshold is determined based on the statistical distribution of tasks that have a significant impact on the overall power generation efficiency in historical operation. When the comprehensive evaluation value is between 0.4 and 0.7, it is classified as a medium processing level. This interval is determined based on the normal fluctuation range of the system. When the comprehensive evaluation value is less than 0.4, it is classified as a low processing level. Subsequently, the task execution order is rearranged according to the level order to form the final hierarchical processing sequence. This sequence is used as the basis for scheduling execution, thereby achieving continuous optimization of the overall energy efficiency of the photovoltaic power station.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent optimization of photovoltaic power plant energy efficiency based on cloud-edge collaboration, characterized in that, include: S1. Through the cloud-edge collaboration architecture, the description information of all tasks to be processed is collected from the edge of the photovoltaic power station and aggregated in the cloud. The tasks include equipment operation adjustment tasks, fault handling tasks and energy efficiency optimization tasks. The task dependency chain length, cross-device or cross-system collaboration frequency and data sharing requirements are analyzed in multiple dimensions to construct a preliminary evaluation matrix of task association strength values ​​and obtain the association strength distribution between tasks. S2. Based on the distribution of task association intensity, analyze the probability of task conflict and the progress synchronization requirements. Combined with information transmission delay and feedback loop count, generate a cross-influence map between tasks and determine the task association intensity level. S3. If the task association intensity level is higher than the preset threshold, the urgency level of the task will be determined by collecting data on the fault response time and power loss duration of the photovoltaic power station, combined with the range of equipment affecting power generation efficiency and the frequency of fault occurrence. S4. Based on the urgency level, extract key business parameters and adjustment resource demand information that affect the energy efficiency of photovoltaic power plants, integrate background data on task source type and expected recovery time, generate a comprehensive task priority score, and obtain a task priority ranking for energy efficiency optimization. S5. By prioritizing tasks, and combining the real-time processing capabilities of the edge side with the global scheduling capabilities of the cloud, analyze the frequency of cross-device collaboration and the number of associated tasks for priority tasks, construct a resource allocation priority queue, and determine the resource scheduling sequence for cloud-edge collaboration.

2. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 1, characterized in that: S1 includes: Acquire real-time operational data from the edge of the photovoltaic power station and aggregate it in the cloud. Use semantic parsing technology to process the real-time operational data to generate standardized task descriptions. Identify the predecessor and successor relationships between task nodes based on standardized task descriptions, calculate the dependency chain length, and count the collaboration frequency and data sharing volume. The dependency chain length, collaboration frequency, and data sharing amount are used as multi-dimensional feature inputs for numerical weighted calculation to construct a preliminary evaluation matrix. Numerical density analysis was performed on the preliminary evaluation matrix to obtain the distribution of inter-task correlation strength.

3. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 1, characterized in that: S2 includes: Obtain the raw execution time sequence and node dependency data from the task management database, calculate the topological path length and coupling depth between nodes in the directed acyclic graph, and obtain the task association strength. The probability of conflict is determined by multi-dimensional weighted calculation based on the task association strength and the real-time resource occupancy rate, and by logical mapping using the communication frequency per unit time. If the probability of conflict is higher than the preset risk threshold, the information transmission delay and feedback loop count in the underlying protocol stack are extracted, and the buffer time in the task chain is dynamically corrected through a time delay compensation algorithm. The progress synchronization model is input with buffer time as a spatiotemporal constraint. The logical topology and interference paths between tasks are mapped in a multidimensional vector space to generate a cross-association graph. Based on the edge weight distribution and node clustering characteristics in the cross-association graph, the task association strength level is divided by clustering analysis algorithm.

4. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 1, characterized in that: S3 includes: Obtain the task association strength value. If the task association strength value is higher than the preset threshold, collect the fault response time limit value and the power loss duration value. The range of damaged equipment is located based on the fault response time limit value and the power loss duration value, and the proportion of affected installed capacity corresponding to the range of damaged equipment is calculated. Based on the frequency of failures within the scope of damaged equipment, and combined with the proportion of affected installed capacity, a quantitative index of urgency is constructed. The urgency level of a task is determined based on the numerical value of its urgency quantification index.

5. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 1, characterized in that: S4 includes: Acquire monitoring data and extract component status; Determine the deviation value based on the component status; If the deviation value exceeds the limit, an initial urgency level is generated; Adjust the gap based on the initial emergency quantification; Determine whether the energy storage margin meets the regulation gap to generate an intervention signal.

6. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 5, characterized in that: S4 further includes: Regarding the intervention signal matching recovery time; Calculate the loss rate using recovery time and adjustment gap; A comprehensive score is generated by weighting the loss rate; The priority ranking of tasks for energy efficiency optimization is obtained based on the comprehensive score.

7. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 1, characterized in that: S5 includes: Extract the collaboration frequency of priority tasks, and quantify the cross-device collaboration frequency based on the collaboration frequency; The edge load is obtained by matching the number of associated devices based on cross-device collaborative frequency, and the cloud bandwidth is obtained based on the edge load.

8. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 7, characterized in that: The S5 also includes: The device energy consumption is obtained based on the cloud bandwidth, and the device energy consumption is input into the resource allocation model to build a resource allocation priority queue; If the resource allocation priority queue meets the coordination cycle requirements, then the resource scheduling sequence for cloud-edge coordination is determined.

9. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 1, characterized in that, It also includes S6, which dynamically updates the execution order of photovoltaic power plant energy efficiency optimization tasks according to the resource scheduling sequence, monitors priority tasks in real time at the edge, and performs global optimization and adjustment in the cloud based on changes in power generation and expected recovery time to obtain the final hierarchical processing sequence, so as to achieve intelligent optimization of the overall energy efficiency of the photovoltaic power plant, specifically including: Obtain the resource scheduling sequence and task set to generate an initial task execution order list; Priority tasks are selected from the initial task execution order list and sent to the edge computing unit, which then monitors the priority tasks in real time and generates monitoring data.

10. The intelligent energy efficiency optimization method for photovoltaic power plants based on cloud-edge collaboration according to claim 9, characterized in that: S6 further includes: Analyze the monitoring data to extract characteristic values ​​of power generation changes, and if the characteristic values ​​of power generation changes are abnormal, calculate the expected recovery time. The characteristic values ​​of power generation change and expected recovery time are uploaded to the cloud for global optimization, resulting in a final hierarchical processing sequence to achieve intelligent optimization of the overall energy efficiency of the photovoltaic power station.

Images