Cloud collaborative traceability and correction method for power prediction deviation of photovoltaic power station

Abstract

The invention discloses a cloud collaborative traceability and correction method for power prediction deviation of a photovoltaic power station, and the method comprises the following steps: multi-source data collection and preprocessing: collecting the multi-source operation data of the photovoltaic power station in real time through a field edge node of the photovoltaic power station, performing abnormal value elimination, data normalization and timestamp alignment preprocessing on the acquired data to eliminate data noise and format difference; cloud collaborative architecture construction: constructing an edge node-regional cloud sub-platform-global cloud main platform three-level collaborative architecture, and realizing hierarchical collaboration of edge end real-time data processing, regional end aggregation analysis and global end unified modeling through the edge node-regional cloud sub-platform-global cloud main platform three-level collaborative architecture. The problems of high time delay of single cloud processing and insufficient computing power of an edge end are effectively solved, the data transmission and processing efficiency is remarkably improved, and it is ensured that a deviation event can be quickly captured and responded.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of deviation data processing, and particularly relates to a cloud collaborative traceability and correction method for power prediction deviation of a photovoltaic power station. BACKGROUND

[0002] At present, the power prediction deviation processing technology of a photovoltaic power station has become one of the key supporting technologies for efficient operation of the photovoltaic power station and stable dispatching of a power grid. In the prior art, data acquisition equipment is mainly used to obtain irradiance, environmental temperature, component operating state, historical power and power grid dispatching data of the photovoltaic power station, and the data is transmitted to a data processing platform for deviation analysis after preprocessing. The data processing platform mainly adopts a single cloud or local edge architecture, wherein the single cloud architecture is responsible for centralized storage of data and running of a prediction deviation traceability model, and the local edge architecture focuses on rapid processing and preliminary correction of on-site data. The traceability model is usually developed around a single factor, such as analysis of only data abnormality or model parameter drift, and the correction strategy mainly adopts fixed experience value adjustment or model parameter updating. Finally, the corrected prediction result is used for power generation amount statistics of the power station or simple power grid dispatching reference, and the overall technical system has been preliminarily applied in small and medium-sized photovoltaic power stations.

[0003] However, the prior art still has obvious limitations in actual application, and it is difficult to meet the efficient operation needs of a large-scale photovoltaic power station cluster. In terms of collaborative efficiency, the single cloud architecture has the problem of high data transmission delay, and the local edge architecture cannot complete complex traceability analysis due to limited computing power, resulting in lag in deviation event response; in terms of traceability accuracy, the single-factor traceability model cannot comprehensively cover multi-dimensional deviation sources such as data, model and environment, and it is difficult to quantify the contribution of each factor, so that the correction lacks scientific basis; in terms of correction strategy, the prior art mainly adopts a general adjustment method, does not accurately address the root cause of the deviation, and lacks a dynamic iteration mechanism, so it is difficult to stably control the deviation within the target threshold; in terms of application value, the correction result is mainly limited to internal management of a single power station, and is not deeply connected with the power grid dispatching system, so it cannot maximize the value; for a multi-power station cluster, the prior art needs to separately develop traceability and correction work, which has large repetitive workload, high operation and maintenance cost, and is difficult to improve the overall prediction accuracy of the region, and these deficiencies have restricted the further development of the power prediction deviation processing technology of the photovoltaic power station. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the application provides a cloud collaborative traceability and correction method for power prediction deviation of a photovoltaic power station, which solves the problems of low collaborative efficiency, inaccurate traceability, non-targeted correction and high operation and maintenance cost of multiple power stations in the prior art.

[0005] To achieve the above-mentioned purposes, the application provides the following technical solutions: A cloud collaborative traceability and correction method for power prediction deviation of a photovoltaic power station, comprising the following steps: S1. Multi-source data acquisition and preprocessing: through the edge node of the photovoltaic power station site, real-time acquisition of multi-source operation data of the photovoltaic power station, execution of outlier rejection, data normalization and timestamp alignment preprocessing on the collected data, elimination of data noise and format difference; S2. Cloud collaborative architecture building: a three-level collaborative architecture of edge node-regional cloud sub-platform-global cloud main platform is built, wherein: the edge node uploads the preprocessed data to the regional cloud sub-platform according to the preset period of the global cloud main platform; the regional cloud sub-platform gathers the preprocessed data of multiple photovoltaic power stations within its jurisdiction, performs local preliminary deviation calculation, and uploads the data and preliminary calculation results exceeding the preset preliminary deviation threshold of the global cloud main platform to the global cloud main platform; the global cloud main platform receives the uploaded data of the multi-regional cloud sub-platform, builds a unified deviation traceability and correction model library, and solves the problems of high processing delay of single cloud and insufficient edge computing power through hierarchical collaboration; S3. Multi-dimensional deviation traceability analysis: the global cloud main platform calls the preset multi-dimensional deviation traceability model to trace the uploaded threshold-exceeding deviation data, and the multi-dimensional deviation traceability model at least integrates: a data quality attribution module for identifying data implicit abnormalities and quantifying their contribution to deviation; a prediction model error attribution module for disassembling model input feature weight, structure and parameter drift deviation influence; an environmental interference attribution module for analyzing the interference coefficient and deviation contribution proportion of extreme environmental interference on prediction; S4. Dynamic correction and feedback iteration: the global cloud main platform generates a targeted correction strategy according to the multi-dimensional traceability result, and issues it to the edge node of the corresponding photovoltaic power station through the three-level collaborative architecture; the edge node executes the correction strategy, compares the corrected predicted power with the subsequent actual power, calculates the corrected deviation rate, and if the corrected deviation rate is still greater than the preset target threshold of the global cloud main platform, re-triggers the collaborative traceability and correction process of steps S2-S4 until the deviation rate is not greater than the target threshold; S5. Correction result application output: the global cloud main platform encapsulates the corrected power prediction result and the traceability analysis report into a standardized interface, and connects the photovoltaic power station operation management system and the power grid dispatching system; S6. Test Interface Deployment and Operation: A test interface based on a B / S architecture is built. This interface is divided into six functional modules: data acquisition and monitoring area, architecture operation status area, source tracing and analysis results area, correction strategy execution area, cross-power station collaboration area, and system configuration area. The data acquisition and monitoring area displays a list of multi-source data acquisitions and a visualization chart of preprocessing effects, supporting data completion operations. The architecture operation status area displays the three-level architecture in a topology diagram, showing key parameters such as data transmission latency and protocol operation status, and has a fault alarm function. The source tracing and analysis results area displays the contribution of deviation sources through a pie chart and provides a source tracing report export function. The correction strategy execution area compares power and deviation rate before and after correction using a dual-axis line chart, recording the feedback iteration process. The cross-power station collaboration area displays the regional power station deviation rate distribution through a heat map, providing common deviation analysis and unified correction strategy viewing functions. The system configuration area supports custom settings of key parameters, user permission management, and system log querying, realizing visualized monitoring, operation, and configuration of the entire collaborative source tracing and correction process. S7. Full-Process Data Analysis: In the data preprocessing stage, outlier judgment intervals are determined by calculating the data mean and standard deviation. Secondary verification distinguishes between true and false anomalies. The distribution characteristics and dispersion coefficients of the data before and after normalization are analyzed, and the timestamp alignment pass rate is verified to ensure the quality of preprocessed data. In the deviation tracing stage, algorithms such as chi-square test, sensitivity analysis, and regression analysis are used to quantify the contribution of each deviation source and verify the rationality of the attribution logic and calculation results. In the correction effect verification stage, the change and reduction of the deviation rate before and after correction are calculated. The t-test is used to verify the significance of the correction effect and analyze the stability of the deviation rate. For cross-power station collaboration scenarios, the changes in the deviation rate of each power station and the overall regional deviation rate before and after collaboration are calculated to assess the proportion of reduction in operation and maintenance costs. A time series prediction model is built based on multi-period data to analyze the confidence interval of deviation trend prediction, providing data support for early warning and early correction, and realizing scientific analysis and effect verification of data at each stage.

[0006] Preferably, in step S1, the multi-source operating data includes real-time irradiance data, environmental and equipment temperature data, equipment operating status data, historical power data, and power grid dispatch constraint data.

[0007] Preferably, in step S1, the multi-source operation data also includes the geographic information data of the photovoltaic power station and the historical operation and maintenance data of the components, and the edge node realizes the localized deployment of the preprocessing algorithm through the edge computing chip.

[0008] Preferably, in step S2, in the three-level collaborative architecture, the edge nodes and the regional cloud sub-platform communicate using a low-latency protocol, and the regional cloud sub-platform and the global cloud main platform communicate using an encrypted transmission protocol; the regional cloud sub-platform sets a data caching threshold, and data exceeding the threshold is automatically archived to the distributed database of the global cloud main platform.

[0009] Preferably, in step S3, the multi-dimensional deviation tracing model also integrates a power grid dispatch constraint attribution module. This module quantifies the prediction deviation caused by changes in dispatch constraints by comparing the conflict periods between power grid dispatch constraint data and predicted power, and incorporates such deviations into the tracing results. The multi-dimensional deviation tracing model has a complete workflow and clear algorithmic support, as detailed below: Overall workflow of the multi-dimensional deviation tracing model: Data Input and Initialization: The global cloud main platform receives the over-threshold deviation data packets (including preprocessed multi-source data, preliminary deviation values, and data collection timestamps) uploaded by the regional cloud sub-platform. After performing format verification on the data (ensuring field integrity and data type matching), the traceability model parameters (such as data quality judgment threshold, model error analysis window duration, and environmental interference coefficient calculation base) are initialized. Parallel computation of modules: The data quality attribution module, the prediction model error attribution module, the environmental interference attribution module, and the power grid dispatch constraint attribution module are triggered to perform parallel operations. Each module performs targeted analysis on the target data based on a preset algorithm and outputs the original value of the deviation contribution of its respective dimension. Contribution normalization: The Min-Max normalization algorithm is used to map the original contribution values ​​of each module to the [0, 100%] interval, eliminating numerical bias caused by differences in the calculation dimensions of different modules. The formula is as follows: in, The contribution after normalization Output the original contribution score for the module. / The minimum / maximum contribution calculated historically for this module; Results integration and output: Summarize the normalized contributions of each module, generate a deviation source tracing result table (including deviation source type, contribution percentage, and key influencing parameters), and synchronously store it in the source tracing result sub-library of the global model library for subsequent correction strategy generation and use.

[0010] Preferably, in step S4, the targeted correction strategy includes: if the deviation mainly originates from data quality, triggering data completion and optimization algorithms to correct the input data of the edge nodes; if the deviation mainly originates from prediction model errors, updating the prediction model parameters of the corresponding power station in the global model library or pushing a new model adapted to the current operating conditions to the regional cloud sub-platform; if the deviation mainly originates from environmental interference or changes in power grid dispatch constraints, issuing corresponding compensation coefficients, which are then superimposed by the edge nodes in real-time prediction. The model determines the main source of the deviation using the following rules: Single-source dominance determination: Set a contribution threshold (default 50%, can be dynamically adjusted). If the contribution of a certain module exceeds the threshold, it is directly determined as the main source. Multi-source overlay determination: If the contribution of all modules is less than or equal to the threshold, the two largest contributing modules are selected. If the difference is greater than or equal to 20%, the higher one is considered the main source. If the difference is less than 20%, data quality / model error modules are selected first. Special adaptation: If the contribution of scheduling constraints is ≥40%, it is listed as a secondary primary source, and an adaptation strategy is generated synchronously.

[0011] Preferably, in step S4, the dynamic correction strategy also includes scenario-based correction templates. The global cloud main platform generates standardized correction templates in advance for typical scenarios based on historical traceability data. When the deviation data uploaded by the regional cloud sub-platform matches the typical scenario, the corresponding template is directly called to issue correction instructions.

[0012] Preferably, in step S5, the source analysis report includes the main sources of deviation, the contribution ratio of each source, the description of correction measures, and the deviation trend prediction. Based on the multi-cycle source analysis results and correction effects, the global cloud platform uses a time series prediction algorithm to predict the deviation change trend in the future preset period. If the predicted trend shows that the deviation will exceed the threshold again, the warning information will be pushed to the photovoltaic power station operation and maintenance terminal and the grid dispatch terminal in advance.

[0013] Preferably, it also includes a cross-power station collaborative optimization step: the global cloud main platform performs correlation analysis on the traceability and correction data of multiple geographically adjacent photovoltaic power stations with similar operating conditions, identifies the common sources of deviation, generates a unified correction strategy across power stations, and simultaneously distributes it to the edge nodes of the corresponding multiple power stations.

[0014] Preferably, in step S1, outlier removal is based on statistical criteria, data normalization maps the data to a preset interval, and timestamp alignment ensures that the timestamps of data from different sources are consistent. In step S6, the test interface supports access from mainstream browsers such as Chrome and Firefox, adopts a responsive design to adapt to PC and tablet devices, and switches between modules through tabs. The data refresh frequency can be set to 1 second / time, 5 seconds / time or manual refresh. In step S7, outlier removal in the data preprocessing stage adopts the 3σ criterion, data normalization is mapped to the [0,1] interval by default, timestamps are uniformly converted to UTC+8 time zone and accurate to the second, the default significance level α=0.05 in the correction effect verification stage, the standard for judging the stability of the deviation rate is standard deviation ≤1%, the standard for judging the consistency of the deviation rate of regional power stations in the cross-power station collaboration scenario is coefficient of variation ≤0.2, and the default confidence level of the time series prediction model is 95%.

[0015] The technical effects and advantages of the cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviations of this invention are as follows: 1. This invention achieves layered collaboration of real-time data processing at the edge, aggregation and analysis at the regional end, and unified modeling at the global end through a three-level collaborative architecture of edge nodes, regional cloud sub-platforms, and global cloud main platform. It effectively solves the problems of high latency in single cloud processing and insufficient computing power at the edge, significantly improves data transmission and processing efficiency, and ensures that deviation events can be quickly captured and responded to.

[0016] 2. This invention relies on a multi-dimensional deviation tracing model, which integrates attribution analysis of multiple modules such as data quality, prediction model error, environmental interference, and power grid dispatch constraints. It can accurately locate the source of deviation and quantify the contribution of each factor from multiple levels such as data, model, environment, and dispatch, overcoming the one-sidedness of traditional single-factor tracing and providing a scientific basis for subsequent correction.

[0017] 3. The invention generates a targeted correction strategy based on the source tracing results, which can accurately implement measures according to the root causes of deviation (data quality, model error, environmental interference, etc.) and continuously optimize the correction effect by combining a feedback iteration mechanism; at the same time, it introduces scenario-based correction templates to shorten the strategy generation time, ensure that the power prediction deviation is stably controlled within the target threshold, and significantly improve the prediction accuracy.

[0018] 4. This invention connects the revised prediction results and source analysis report to the photovoltaic power plant operation management system and the power grid dispatching system through a standardized interface. This not only provides data support for power generation calculation and equipment operation and maintenance plan formulation, but also provides decision-making basis for power grid purchase and sale plan optimization and grid-connected power regulation, realizing the transformation of technological achievements into management and economic value.

[0019] 5. This invention generates a unified correction strategy for common deviation sources of geographically adjacent power plants with similar operating conditions through cross-power plant collaborative optimization. This reduces the workload of repeated modeling and individual correction, improves the overall prediction accuracy of power plants in the region, reduces operation and maintenance costs, and improves management efficiency.

[0020] 6. The newly added B / S architecture test interface of this invention realizes full-process visual monitoring through six major functional modules: real-time data acquisition status can be checked, timely warning of architecture operation failures, intuitive display of traceability results, dynamic tracking of the correction process, visualization of cross-power station collaboration effects, and flexible configuration of key parameters. It greatly reduces the technical application threshold, meets the operation and monitoring needs of users with different permissions (administrators, maintenance personnel, and viewers), and solves the problems of traditional methods such as "no visualization carrier, complex operation, and difficulty in fault location".

[0021] 7. The invention adds a full-process data analysis that covers four major stages: data preprocessing (outlier removal effect, normalization distribution, and timestamp alignment pass rate verification), deviation tracing (contribution calculation review and attribution logic verification), correction effect (deviation rate change, significance test, and stability analysis), and cross-power station collaboration (regional deviation rate consistency and cost-benefit assessment). Through statistical analysis and algorithm verification, it ensures the data quality and scientific nature of the technical logic in each stage, avoiding the shortcomings of traditional methods such as "correction without basis and difficulty in verifying effects," and further improving the reliability and reproducibility of the technical solution. Attached Figure Description

[0022] Figure 1 This is a flowchart of a cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviations proposed in this invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "includes..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] refer to Figure 1This invention provides a cloud-based collaborative tracing and correction method for power prediction deviations in photovoltaic power plants. It achieves data collaboration through a three-tiered architecture of "edge nodes - regional cloud sub-platforms - global cloud main platform," locates the source of deviations using a multi-dimensional tracing model, dynamically generates correction strategies, and achieves full-process visual monitoring through a testing interface. The effectiveness of each stage is verified through full-process data analysis, and finally, the results are output to the operation and scheduling system via a standardized interface. The following embodiments are based on actual operating data verification from a photovoltaic industrial park (containing 10 grid-connected photovoltaic power plants) in Northwest my country. Each power plant has an installed capacity of 50-100MW, uses centralized inverters and polycrystalline silicon modules, and has a historical average power prediction deviation rate of 8-12%.

[0026] Example 1 This embodiment provides a cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviations, used for the specific implementation of multi-source data acquisition and cloud-based collaborative architecture: Purpose of implementation: Verify the feasibility of deploying the three-tier collaborative architecture in practice, complete the accurate collection and preprocessing of multi-source data, and provide a high-quality data foundation for subsequent deviation tracing.

[0027] Implementation steps: Multi-source data acquisition deployment: Select Power Station A (60MW installed capacity) within the park and deploy an edge node on its inverter cabinet side. This node uses an ARM Cortex-A53 architecture edge computing chip (2TOPS computing power) to collect the following data in real time: Environmental data: direct / scattered irradiance (sampling frequency 1 time / minute, accuracy ±5%), ambient temperature (-40~80℃, accuracy ±0.5℃), module backsheet temperature (100 sampling points are set up, distributed in different areas of the module array); Equipment data: Inverter output power, efficiency (sampling frequency 1 time / 10 seconds), component attenuation rate (tested and calibrated monthly based on IV curve), combiner box on / off status (recorded as Boolean values). Historical and dispatch data: Forecasted and actual power from 9:00 to 15:00 daily for the past 3 years, and the daily grid connection limit (the limit for the morning peak from 8:00 to 11:00 is 90% of the installed capacity). Extended data: Power station latitude and longitude (N38°, E106°), altitude 1100m, no tall obstructions within 3km, and the last cleaning time of the modules was 1 month ago.

[0028] Data preprocessing operations: Outlier removal: The 3σ criterion is used to remove data in the irradiance that exceed the mean ± 3 times the standard deviation, such as the outlier value of "0 irradiance but 10MW power" that appeared on a certain day due to sensor failure. Data normalization: Mapping physical quantities such as temperature and irradiance to the [0,1] interval, for example, -40℃ corresponds to 0 and 80℃ corresponds to 1; Timestamp alignment: Unify all data to UTC+8 time zone, accurate to the second, and solve the timing misalignment problem caused by the clock deviation between the irradiance sensor and the inverter.

[0029] Three-tier collaborative architecture operation: Edge nodes upload pre-processed data to the regional cloud sub-platform (deployed in the provincial energy data center, using a 20-node server cluster) every 5 minutes. The regional cloud sub-platform calculates the initial deviation. When the absolute deviation between the real-time predicted power and the actual power is greater than 8%, the data (including anomaly markers and initial deviation values) is packaged and uploaded to the global cloud main platform (deployed on the national-level new energy cloud platform and using distributed storage HadoopHDFS) via the 5G low latency protocol (latency <100ms). The regional cloud sub-platform sets a 72-hour data caching threshold. Expired data is archived to the global cloud main platform (10PB storage capacity, supporting tens of millions of concurrent data accesses) via HTTPS+SM4 encryption protocol.

[0030] Test interface deployment and function verification: Interface Deployment: The test interface is developed based on a B / S architecture and deployed on a regional cloud sub-platform server (using Tomcat 9.0 as the web server and MySQL 8.0 to store interface configuration data). It supports access from Chrome 110+ and Firefox 109+ browsers and adopts a responsive design to adapt to 1920×1080 (PC) and 2560×1600 (tablet) resolutions. Module functionality verification: Data acquisition and monitoring area: Check the multi-source data acquisition list (confirm that the status labels such as "irradiance data - acquisition node 1 - 1 time / minute - normal" are accurate), compare the irradiance data fluctuation before and after preprocessing through a line graph (the data fluctuation amplitude is reduced by 60% after preprocessing), and trigger the LSTM data completion algorithm to complete 20 minutes of missing data (completion error 2.8%). Architecture Operation Status Area: View the status of edge nodes (blinking blue icons, indicating data transmission in progress), regional cloud sub-platforms (solid green icons), and global cloud main platform (solid orange icons) through the topology diagram. Real-time data transmission latency is displayed (currently 180ms, historical average 195ms). Simulate edge node offline (communication disconnection), trigger audible and visual alarms, and record fault logs ("2024-06-10 14:30: Edge node 1 of power station A is offline, it is recommended to check the communication link"). System Configuration Area: Adjust the "Preliminary Deviation Threshold" from 8% to 7%, the "Data Upload Cycle" from 5 minutes to 3 minutes, assign "Operations and Maintenance Personnel" permissions (only allowed to view data and perform corrections, prohibited from modifying parameters), and query the "Data Processing Log" for the past 24 hours (a total of 1200 preprocessing operations were recorded, with no abnormalities).

[0031] Implementation results: The data collection coverage reached 100%, and the data anomaly rate after preprocessing decreased from 12% to 1.5%; the data transmission latency of the three-level architecture was <200ms, and the data archiving efficiency of the regional sub-platform was improved by 40%; all modules of the test interface functioned normally, the data display was real-time and accurate (refresh delay <1 second), the fault alarm response time was <3 seconds, the parameter configuration and permission management were effective, and the needs of visual monitoring and operation were met.

[0032] Example 2 This embodiment provides a cloud-based collaborative method for tracing and correcting power prediction deviations in photovoltaic power plants, applicable to the specific application of multi-dimensional deviation tracing models: Purpose of implementation: By examining actual deviation events, we can verify the ability of the multi-dimensional source tracing model to accurately pinpoint the source of deviations and clarify the contribution ratio of each factor to the deviation.

[0033] Implementation steps: Deviation event triggering: The operating data of Power Station A from 12:00 to 13:00 on a certain day was selected. The predicted power during this period was 45MW, and the actual power was 32MW, with a deviation rate of 28.9%. This triggered a multi-dimensional traceability process. The global cloud main platform received the deviation data packet and initialized the parameters.

[0034] Multi-dimensional source tracing execution: Data quality attribution: Chi-square test analysis of the preprocessed data revealed a discontinuity in the irradiance data from 11:50 to 12:10 (sensor communication interrupted for 20 minutes, data jumped to 30% of the mean), and the contribution of this data anomaly to the bias was calculated to be 60%. Attribution of prediction model error: Using model analysis, it was found that the weight of the "component temperature" feature in the model (0.12) is lower than the actual impact weight (0.25), and the model training parameters are not adapted to the 3% decay rate of the component. The combined contribution of these two factors to the bias is 25%. Environmental interference attribution: Through equivalent circuit simulation of the components, it was found that from 12:30 to 12:45, the temporary parking of the inspection vehicle caused local shadows, resulting in a mismatch between the series and parallel connections of the components, with an interference coefficient of 0.15 and a corresponding deviation contribution of 15%. Attribution of grid dispatch constraints: Comparing the grid dispatch constraint data with the predicted power, it was found that the grid connection limit was not adjusted at 12:00 on the same day, and this factor contributed 0% to the deviation.

[0035] End-to-end data analysis: Data preprocessing stage analysis: Outlier removal effect: Calculate the mean (850W / ㎡ before removal, 880W / ㎡ after removal) and standard deviation (120W / ㎡ before removal, 45W / ㎡ after removal) of the irradiance data before and after removal to confirm that the interference of outliers on the statistical characteristics of the data has been eliminated. Data normalization analysis: Plot histograms of ambient temperature before and after normalization (normalization shows a normal distribution, and the distribution shape remains unchanged after normalization, with the data concentrated in the interval [0.3, 0.7]), and calculate the coefficient of variation of the normalized data (0.22 ≤ 0.3, which meets the requirements). Timestamp alignment verification: Statistical analysis of the timestamp alignment pass rate of each data source during the period from 12:00 to 13:00 (60 timestamps in total, 60 successfully aligned, pass rate ≥ 99%, meeting the standard). Deviation tracing phase analysis: Contribution calculation verification: The original contribution of each attribution module (data quality 85, model error 35, environmental interference 20, scheduling constraint 0) was normalized using Min-Max (formula: normalized contribution = (original value - minimum value) / (maximum value - minimum value) × 100%). After verification, the contribution was consistent with the original results (60%, 25%, 15%, 0%). Attribution logic verification: After simulating "completing missing irradiance data", the deviation was recalculated (the predicted power after completion was 40MW, the actual power was 32MW, the deviation rate was 25%, which was 3.9% lower than the original deviation rate), verifying the actual impact of data quality on the deviation, which is consistent with the logic of the attribution results (60% contribution).

[0036] Test interface source tracing results display: View the results in the "Source Analysis Results" section of the test interface: The pie chart shows the contribution of each source of deviation (data quality 60%, model error 25%, environmental interference 15%, scheduling constraints 0%). When the mouse hovers over "data quality", a prompt appears: "Hidden anomaly: irradiance data interrupted for 20 minutes, contribution 60%". The traceability results details table displays information such as "Source of deviation: Data quality - Key parameters: Irradiance data continuity - Contribution 60% - Analysis algorithm: Chi-square test - Confidence level 98%". Click the "Export Report" button to generate an Excel-format traceability report (including raw data, calculation process, and results conclusions).

[0037] Implementation results: The system accurately identifies three types of deviation sources, with each factor's contribution measured in less than 3% of the quantified error, overcoming the limitations of traditional single-factor tracing. Full-process data analysis verifies that the preprocessed data quality meets standards (100% timestamp alignment pass rate, normalized coefficient of variation 0.22), and the tracing logic is rigorous (the reduction in deviation rate after data completion matches the attribution contribution). The test interface displays tracing results intuitively, and the report export function works correctly, providing a scientific basis for subsequent corrections.

[0038] Example 3 This embodiment provides a cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviations, including the specific process of dynamic correction and feedback iteration: Purpose of implementation: Verify the targeting of the dynamic correction strategy and the effectiveness of the feedback iteration mechanism. Verify the significance and stability of the correction effect through data analysis, and control the deviation rate within the target range.

[0039] Implementation steps: Targeted correction strategy generation and execution: Based on the source tracing results of Example 2, the global cloud main platform generates the following correction strategy: Data quality correction: Trigger the LSTM time series completion algorithm to complete 20 minutes of missing data based on the irradiance trend before 11:50 (completion error <3%). Model parameter update: Using online gradient descent, the component decay rate parameter was corrected from 0% to 3%, and the weight of the "component temperature" feature was adjusted to 0.25; Environmental interference compensation: A local shading compensation coefficient of 0.15 is issued, and the edge nodes are superimposed with compensation terms in real-time prediction (correction formula: prediction power = original prediction value × (1-0.15)).

[0040] Feedback and iterative verification: After the initial correction, monitoring data from 13:00 to 14:00 showed a predicted power of 34MW and an actual power of 32MW, with a deviation rate of 6.2% (still > 5% of the target threshold). The tracing process was retried, and it was found that the model did not adapt to the component efficiency degradation characteristics under high summer temperatures (above 35℃). The model structure was supplemented and corrected (by adding a high-temperature degradation factor). After the second correction, the data from 14:00 to 15:00 was monitored. The predicted power was 32.8MW, the actual power was 32MW, and the deviation rate was reduced to 3.8% (≤5%), at which point the iteration was terminated.

[0041] Correction effect data analysis: Analysis of deviation rate changes: The deviation rates before and after correction were calculated (original 28.9% → first time 6.2% → second time 3.8%), with reductions of 78.5% and 38.7% respectively, and an overall reduction of 86.8%. Significance test: A t-test (α=0.05) was performed on the six sets of deviation rate data (3.8%, 3.6%, 4.0%, 3.9%, 3.7%, 3.8%) within one hour after the second correction and compared with the data before correction. The calculated t-value was 8.23 ​​> (10) = 2.23, P < 0.05, verifying that the correction effect is statistically significant; Stability analysis: The standard deviation of the deviation rate after the second correction (0.15%≤1%) was calculated to confirm that the correction effect is stable.

[0042] Monitoring of the test interface correction process: View the test interface under "Correction Strategy Execution Area": The dual-axis line graph (horizontal axis time, left vertical axis power, right vertical axis deviation rate) shows that: before correction, the predicted power (45MW) and the actual power (32MW) deviated significantly. After the first correction, the fit improved (predicted 34MW vs. actual 32MW). After the second correction, they basically overlapped (predicted 32.8MW vs. actual 32MW). The red dashed line (target threshold 5%) shows that the deviation rate was always lower than the threshold after the second correction. The feedback iteration record list shows "Iteration 1 - Time 2024-06-10 15:00 - Correction measures: complete data + update parameters - Deviation rate 6.2% - Not up to standard; Iteration 2 - Time 2024-06-10 15:30 - Correction measures: supplement high temperature factor - Deviation rate 3.8% - Up to standard".

[0043] Implementation results: After two iterations, the deviation rate decreased from 28.9% to 3.8%, meeting the power grid dispatching requirements for prediction accuracy (≤5%). Data analysis verified that the correction effect was significant (P<0.05) and stable (standard deviation 0.15%). The response time of the correction strategy execution was <10 minutes. The test interface monitored the correction process in real time, and the iteration records were complete and traceable, avoiding the incompleteness of a single correction.

[0044] Example 4 This embodiment provides a cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviations, used for application of correction results and trend early warning: Purpose of implementation: The study aims to verify the practical application value of the correction results in operation management and power grid dispatch, achieve early warning of deviation trends through time series analysis, and verify the ability of cross-power station collaborative optimization to solve common deviations.

[0045] Implementation steps: Application of the correction results: Connecting to the photovoltaic power plant operation and management system: Based on the corrected power data, the daily power generation deviation was calculated to be reduced by 12,000 kWh. Combined with the decrease in the proportion of deviation caused by sandstorms, the module cleaning plan for the next day was postponed by 1 day. Connecting to the power grid dispatch system: The corrected forecast data (error ±3%) is used to optimize the daily power purchase and sale plan, reducing curtailment of solar power by 0.8 million kWh and improving power plant revenue.

[0046] Deviation Trend Prediction and Early Warning: Based on the source data and correction effects of the past 10 days, the global cloud platform uses a time series prediction algorithm to predict the deviation trend for the next 7 days. It is expected that the deviation rate will rise to 7% (>5%) on the 5th day due to sandstorm weather. A "component cleaning preparation" warning is sent to the operation and maintenance end 12 hours in advance, and a "reserved power adjustment redundancy" warning is sent to the scheduling end. After following the warning instructions on the 5th day, the actual deviation rate was controlled at 4.5%.

[0047] Cross-power station collaborative optimization: Common Deviation Identification: Power plants B, C, and D within the park (geographically adjacent to power plant A, all using the same batch of polycrystalline silicon modules, with operating condition similarity > 85%) were selected. Analysis of winter (January) operating data revealed that when the ambient temperature was below -10℃, all three power plants exhibited increased deviation rates (9.5% for power plant B, 10.2% for power plant C, and 9.8% for power plant D). After correlation analysis on the global cloud platform, the root cause of the deviation was determined: the model did not include the low-temperature IV curve correction coefficient (the fill factor of the same batch of modules decreased by 8% at -10℃ compared to 25℃). Unified correction strategy generation: Based on common root causes, a cross-power station correction model is generated (low temperature compensation coefficient = 1 - 0.004 × (25 - ambient temperature), effective when ambient temperature < 0℃). Strategy distribution and execution: Select power stations B, C, and D through the "Cross-Power Station Collaboration Zone" test interface, click "Distribute Unified Strategy", and the edge nodes will receive the strategy and overlay the low temperature compensation item in the real-time prediction. Collaborative effect data analysis: Deviation rate comparison: Before collaboration, the average deviation rate of power plants B, C, and D was 9.8%, which decreased to 4.6% after collaboration (≤5% target threshold). Consistency analysis: The coefficient of variation of the deviation rate of the three power stations after coordination (standard deviation 0.3% / mean 4.6% ≈ 0.065 ≤ 0.2) was calculated to confirm that the prediction accuracy in the region tends to be consistent; Cost analysis: Cross-power station collaboration reduces the number of individual modeling steps by 3 (originally, one low-temperature correction model was needed for each of B, C, and D), shortens the operation and maintenance time from 3 days to 1 day, and reduces labor costs by 66.7%.

[0048] Cross-power station collaborative display on the test interface: View via the "Cross-Power Station Collaboration Zone" in the test interface: The heat map shows that before collaboration, the areas of power plants B, C, and D were dark in color (high deviation rate), and after collaboration, the color became lighter (low deviation rate). The bar chart compares the deviation rates before and after coordination (Power Plant B: 9.5% → 4.7%, Power Plant C: 10.2% → 4.5%, Power Plant D: 9.8% → 4.6%). Clicking the "Common Deviation Analysis" button displays "Common Root Cause: Low Temperature IV Curve Mismatch - Unified Strategy: Low Temperature Compensation Coefficient Formula - Synergistic Effect: Average Deviation Rate Reduced by 53.1%".

[0049] Implementation results: After the correction results were applied, the daily power generation of the power plant increased by 2.5%, and the curtailment rate of solar power on the grid decreased by 1.8%. The accuracy of deviation trend prediction reached 97.1% (actual 6.8% vs. predicted 7%), and early warnings were provided to avoid the risk of deviations exceeding the threshold. Cross-power plant collaboration reduced the regional average deviation rate from 9.8% to 4.6%, improved operation and maintenance efficiency by 30%, and reduced the modeling cost of a single power plant by 50%. The heat map and bar chart on the test interface intuitively displayed the collaborative effect, and the common deviation analysis function was clear, meeting the needs of multi-power plant cluster management.

[0050] Example 5 This embodiment provides a cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviations, used for cross-power plant collaborative optimization: Purpose of implementation: Verify the ability of cross-power station collaborative optimization to resolve common deviations and improve the overall prediction accuracy of multiple power stations in the region.

[0051] Implementation steps: Common deviation identification involved selecting power stations B, C, and D within the park (geographically adjacent and all using the same batch of modules). Analysis of their winter operating data revealed that when the ambient temperature was below -10℃, all three power stations exhibited excessively high prediction deviations, with deviation percentages exceeding 40%. A global cloud-based main platform correlation analysis of the three power stations determined that the root cause of the deviation was the model's failure to incorporate the low-temperature IV curve correction coefficient (a common characteristic of modules from the same batch).

[0052] Unified revision strategy formulation and distribution: Generate a cross-power station correction model: add a low-temperature compensation formula (compensation coefficient = 0.02 × (25℃ - ambient temperature)); Through a three-tier collaborative architecture, the correction model is synchronously distributed to the edge nodes of power plants B, C, and D, ensuring that each power plant executes the correction synchronously.

[0053] The effect of the correction was verified by monitoring the operation data of the three power stations after the correction, with a focus on the deviation rate during the period below -10℃.

[0054] Implementation results: The average deviation rate of power stations B, C, and D decreased from 9.2% to 4.8%, and the overall prediction accuracy in the region improved by 48%. Cross-power station collaboration reduces the workload of repetitive modeling, improves operation and maintenance efficiency by 30%, and reduces the modeling cost of a single power station by 50%.

[0055] Comparative Example 1 This comparison provides a traditional approach, including the following: It adopts a local correction mode for a single power plant, without a cloud-based collaborative architecture, and only performs linear correction based on historical power deviations, without involving multi-dimensional deviation tracing.

[0056] Test scenario: The same deviation event occurred at Power Plant A from 12:00 to 13:00 as in Examples 1-3 (predicted power 45MW, actual power 32MW, deviation rate 28.9%).

[0057] Implementation steps: Traditional methods only compare historical data from the same period to discover the superficial phenomenon of "predicted values ​​being higher than actual values"; Using an empirical correction method, 5MW was directly subtracted from the predicted value to obtain a corrected predicted power of 40MW; Monitor the corrected data from 13:00 to 14:00 and calculate the deviation rate.

[0058] For a comparison of the implementation results, please see Table 1.

[0059] Table 1 compares the parameters of Comparative Example 1 with those of the present invention.

[0060] Compared with Examples 1-5 and Comparative Example 1, the core difference between Examples 1-5 and Comparative Example 1 (traditional single power plant local correction method) lies in the technical architecture and source-tracing correction logic. Through multi-dimensional comparison, the technical advantages and practical value of the present invention can be clearly demonstrated.

[0061] From a data foundation perspective, Example 1 establishes a three-tier collaborative architecture of "edge node - regional cloud sub-platform - global cloud main platform". It achieves accurate collection of multi-source data (environment, equipment, scheduling, etc.) through edge computing chips. After preprocessing, the data anomaly rate is reduced from 12% to 1.5%, and the transmission latency is <200ms. In contrast, Comparative Example 1 has no collaborative architecture and only collects local power data. The data anomaly rate is as high as 10% or more. Moreover, due to the lack of preprocessing such as timestamp alignment, timing misalignment problems often occur, which poses a hidden danger for subsequent correction.

[0062] In the deviation localization stage, Example 2 uses a multi-dimensional source tracing model (data quality, model error, environmental interference, etc.) to accurately quantify the contribution of each deviation source (error < 3%). For example, in a certain deviation event, it successfully identifies three root causes: data anomaly (60%), model drift (25%), and local shadow (15%). In contrast, Comparative Example 1 can only find the surface phenomenon of "deviation between predicted value and actual value" and cannot locate the root cause, resulting in a lack of targeted correction.

[0063] In terms of correction effect, the dynamic correction and feedback iteration mechanism of Example 3 has significant advantages: in a certain deviation event, after two iterations, the deviation rate dropped from 28.9% to 3.8% (≤5% target threshold), and the correction response time was <10 minutes; in Comparative Example 1, the empirical correction of "reducing fixed value" was adopted, and the deviation rate after correction was still 15.6%, and repeated adjustments were required when the environment changed, resulting in extremely poor adaptability.

[0064] At the application expansion level, Example 4 integrates the correction results with the operation and dispatch system, which increases the daily power generation of the power plant by 2.5% and reduces the curtailment rate of solar power by 1.8%. At the same time, it avoids deviation risks in advance through trend prediction (accuracy of 90%). In contrast, the correction results of Example 1 are only used for local power adjustment, have no value for extended application, and have no early warning capability. They are prone to causing dispatch violations due to deviations exceeding the threshold.

[0065] For multi-power station scenarios, the cross-power station collaborative optimization in Example 5 can identify common deviations (such as the low-temperature characteristics of components in the same batch) and generate a unified correction strategy, which reduces the average deviation rate of regional power stations from 9.2% to 4.8% and improves operation and maintenance efficiency by 30%. In contrast, Example 1 requires independent correction for each station, which involves a large amount of repetitive work and makes it difficult to reduce the overall deviation rate of the region. The operation and maintenance cost is 50% higher than that of this invention.

[0066] In summary, Comparative Example 1, lacking a collaborative architecture, multi-dimensional tracing, and iterative mechanisms, can only achieve "surface correction." In contrast, this invention, through full-process collaboration and precise tracing, achieves "root cause correction + value expansion," reducing the average deviation rate of regional power plants from 10.5% to 4.2%, saving approximately 800,000 yuan in annual operation and maintenance costs. It is comprehensively superior to traditional methods and has extremely strong engineering application value.

[0067] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0069] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviations, characterized in that, Includes the following steps: S1. Multi-source data acquisition and preprocessing: Real-time acquisition of multi-source operation data of photovoltaic power station through edge nodes on the photovoltaic power station site, and preprocessing of the acquired data including outlier removal, data normalization and timestamp alignment to eliminate data noise and format differences; S2. Cloud-based Collaborative Architecture Construction: A three-tiered collaborative architecture is constructed, consisting of edge nodes, regional cloud sub-platforms, and a global cloud main platform. Edge nodes upload pre-processed data to regional cloud sub-platforms according to a preset cycle on the global cloud main platform. Regional cloud sub-platforms aggregate pre-processed data from multiple photovoltaic power plants within their jurisdiction, perform local preliminary deviation calculations, and upload data exceeding the preset preliminary deviation threshold and preliminary calculation results to the global cloud main platform. The global cloud main platform receives uploaded data from multiple regional cloud sub-platforms, constructs a unified deviation tracing and correction model library, and addresses the issues of high latency in single-cloud processing and insufficient computing power at the edge through layered collaboration. S3. Multi-dimensional Deviation Source Tracing Analysis: The global cloud main platform calls the preset multi-dimensional deviation source tracing model to trace the source of uploaded deviation data exceeding the threshold. The multi-dimensional deviation source tracing model integrates at least: a data quality attribution module, used to identify latent anomalies in the data and quantify their contribution to the deviation; a prediction model error attribution module, used to decompose the deviation impact of model input feature weights, structure, and parameter drift; and an environmental interference attribution module, used to analyze the interference coefficient and deviation contribution ratio of extreme environmental interference to the prediction. S4. Dynamic Correction and Feedback Iteration: The global cloud main platform generates targeted correction strategies based on multi-dimensional traceability results and distributes them to the edge nodes of the corresponding photovoltaic power station through a three-level collaborative architecture. After the edge nodes execute the correction strategies, they compare the corrected predicted power with the subsequent actual power and calculate the corrected deviation rate. If the corrected deviation rate is still greater than the preset target threshold of the global cloud main platform, the collaborative traceability and correction process of steps S2-S4 is re-triggered until the deviation rate is no greater than the target threshold. S5. Correction Result Application Output: The global cloud main platform encapsulates the corrected power prediction results and source analysis report into a standardized interface, which is then connected to the photovoltaic power plant operation and management system and the power grid dispatch system. S6. Test Interface Deployment and Operation: A test interface based on a B / S architecture is built. This interface is divided into six functional modules: data acquisition and monitoring area, architecture operation status area, source tracing and analysis results area, correction strategy execution area, cross-power station collaboration area, and system configuration area. The data acquisition and monitoring area displays a list of multi-source data acquisitions and a visualization chart of preprocessing effects, supporting data completion operations. The architecture operation status area displays the three-level architecture in a topology diagram, showing key parameters such as data transmission latency and protocol operation status, and has a fault alarm function. The source tracing and analysis results area displays the contribution of deviation sources through a pie chart and provides a source tracing report export function. The correction strategy execution area compares power and deviation rate before and after correction using a dual-axis line chart, recording the feedback iteration process. The cross-power station collaboration area displays the regional power station deviation rate distribution through a heat map, providing common deviation analysis and unified correction strategy viewing functions. The system configuration area supports custom settings of key parameters, user permission management, and system log querying, realizing visualized monitoring, operation, and configuration of the entire collaborative source tracing and correction process. S7. Full-Process Data Analysis: In the data preprocessing stage, outlier judgment intervals are determined by calculating the data mean and standard deviation. Secondary verification distinguishes between true and false anomalies. The distribution characteristics and dispersion coefficients of the data before and after normalization are analyzed, and the timestamp alignment pass rate is verified to ensure the quality of preprocessed data. In the deviation tracing stage, algorithms such as chi-square test, sensitivity analysis, and regression analysis are used to quantify the contribution of each deviation source and verify the rationality of the attribution logic and calculation results. In the correction effect verification stage, the change and reduction of the deviation rate before and after correction are calculated. The t-test is used to verify the significance of the correction effect and analyze the stability of the deviation rate. For cross-power station collaboration scenarios, the changes in the deviation rate of each power station and the overall regional deviation rate before and after collaboration are calculated to assess the proportion of reduction in operation and maintenance costs. A time series prediction model is built based on multi-period data to analyze the confidence interval of deviation trend prediction, providing data support for early warning and early correction, and realizing scientific analysis and effect verification of data at each stage.

2. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S1, the multi-source operational data includes real-time irradiance data, environmental and equipment temperature data, equipment operating status data, historical power data, and power grid dispatch constraint data.

3. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S1, the multi-source operation data also includes the geographic information data of the photovoltaic power station and the historical operation and maintenance data of the components. The edge node realizes the local deployment of the preprocessing algorithm through the edge computing chip.

4. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S2, in the three-level collaborative architecture, the edge nodes and the regional cloud sub-platform communicate using a low-latency protocol, and the regional cloud sub-platform communicates with the global cloud main platform using an encrypted transmission protocol. The regional cloud sub-platform sets a data caching threshold, and data exceeding the threshold is automatically archived to the distributed database of the global cloud main platform.

5. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S3, the multi-dimensional deviation tracing model also integrates a power grid dispatch constraint attribution module. This module quantifies the prediction deviation caused by changes in dispatch constraints by comparing the conflict periods between power grid dispatch constraint data and predicted power, and incorporates such deviations into the tracing results.

6. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S4, the targeted correction strategies include: if the deviation mainly comes from data quality, triggering data completion and optimization algorithms to correct the input data of edge nodes; if the deviation mainly comes from prediction model errors, updating the prediction model parameters of the corresponding power station in the global model library or pushing a new model adapted to the current operating conditions to the regional cloud sub-platform; if the deviation mainly comes from environmental interference or changes in power grid dispatch constraints, issuing corresponding compensation coefficients respectively, and having the edge nodes add compensation terms in real-time prediction.

7. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S4, the dynamic correction strategy also includes scenario-based correction templates. The global cloud main platform generates standardized correction templates in advance for typical scenarios based on historical traceability data. When the deviation data uploaded by the regional cloud sub-platform matches the typical scenario, the corresponding template is directly called to issue correction instructions.

8. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S5, the source analysis report includes the main sources of deviation, the contribution ratio of each source, the explanation of the correction measures, and the deviation trend prediction. Based on the multi-cycle source analysis results and correction effects, the global cloud platform uses a time series prediction algorithm to predict the deviation change trend in the future preset period. If the predicted trend shows that the deviation will exceed the threshold again, the early warning information will be pushed to the photovoltaic power station operation and maintenance terminal and the grid dispatch terminal in advance.

9. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, It also includes cross-power station collaborative optimization steps: the global cloud main platform will conduct correlation analysis on the traceability and correction data of multiple geographically adjacent photovoltaic power stations with similar operating conditions, identify common deviation sources, generate a unified correction strategy across power stations, and simultaneously distribute it to the edge nodes of the corresponding multiple power stations.

10. The cloud-based collaborative tracing and correction method for photovoltaic power plant power prediction deviation as described in claim 1, characterized in that, In step S1, outlier removal is based on statistical criteria, data normalization maps the data to a preset range, and timestamp alignment ensures that the timestamps of data from different sources are consistent. In step S6, the test interface supports access from mainstream browsers such as Chrome and Firefox, adopts a responsive design to adapt to PC and tablet devices, and switches between modules through tabs. The data refresh frequency can be set to 1 second / time, 5 seconds / time or manual refresh. In step S7, outlier removal in the data preprocessing stage adopts the 3σ criterion, data normalization is mapped to the [0,1] interval by default, timestamps are uniformly converted to UTC+8 time zone and accurate to the second, the default significance level α=0.05 in the correction effect verification stage, the standard for judging the stability of the deviation rate is standard deviation ≤1%, the standard for judging the consistency of the deviation rate of regional power stations in the cross-power station collaboration scenario is coefficient of variation ≤0.2, and the default confidence level of the time series prediction model is 95%.

Images