Intelligent power dispatching optimization method and system based on multi-source data fusion

Abstract

The application provides a smart power dispatching optimization method and system based on multi-source data fusion, and belongs to the technical field of power system dispatching automation. The method comprises the following steps: through an end-side sensor and an intelligent terminal, multi-source data in a power system is synchronously collected to generate a multi-source heterogeneous data set; an edge computing node is used to perform real-time caching on the multi-source heterogeneous data set; according to the generated multi-source heterogeneous data set, a systematic pretreatment operation is performed to generate a pretreated data set; data fusion is more in-depth, a systematic multi-source data processing technology is adopted to complete time-space calibration, noise elimination and feature enhancement, the data collaborative value is fully tapped, the prediction error is greatly reduced, and accurate data support is provided for dispatching decision-making.

Description

Technical Field

[0001] This invention proposes a smart power dispatch optimization method and system based on multi-source data fusion, belonging to the field of power system dispatch automation technology. Background Technology

[0002] With the advancement of the "dual carbon" goals and the accelerated construction of new power systems, the power grid structure has undergone fundamental changes. A large-scale integration of diverse entities, such as high-proportion new energy sources (wind power and photovoltaics), distributed energy storage, virtual power plants, and flexible loads, has broken the traditional single operation mode of the power grid, which is "source follows load." This has led to the power grid exhibiting the core characteristics of "random fluctuations on both the source and load sides, increased multi-energy coupling, and dynamic and varied operating states." Currently, the core pain points facing power grid operation are becoming increasingly prominent: On the one hand, the phenomenon of data silos from multiple sources is serious. Data from power grid PMU phasor monitoring, renewable energy output time-series data, meteorological forecast data, equipment status monitoring data (vibration, temperature, insulation, etc.), load breakdown data, and power grid topology data are heterogeneous in source, format, and time and space, and suffer from problems such as data noise, missing data, and redundancy. The prediction error of a single data source generally exceeds 15%, directly resulting in a lack of accurate data support for dispatching decisions. On the other hand, traditional power dispatching relies on fixed experience models and static optimization strategies, which are difficult to adapt to the dynamic supply and demand balance under the high penetration of renewable energy. This results in problems such as insufficient dispatching accuracy, low renewable energy absorption rate, conservative power grid safety margin settings, and delayed response (the conventional dispatching update cycle is 15-30 minutes). This not only wastes clean energy but also increases the operating costs of the power grid, failing to meet the core dispatching requirements of "safety, efficiency, low carbon, and flexibility" of smart grids.

[0003] In the current field of power dispatching, the closest existing technical solutions to this invention mainly focus on the dispatching mode of "single data source driven + traditional algorithm optimization". The specific implementation paths can be divided into two categories: one is based on conventional thermal power and hydropower units, relying on historical load data and unit output parameters, combined with traditional heuristic algorithms such as linear programming and genetic algorithms to complete the short-term (within 24 hours) dispatch optimization of the power grid, focusing on solving the problems of power grid power balance and unit economic operation; the other is to initially introduce multi-source data splicing technology, simply superimpose new energy output data and load data, combine basic machine learning algorithms such as BP neural network and LSTM for power prediction, and then complete the dispatching scheme generation through conventional optimization algorithms.

[0004] While existing solutions improve the rationality of scheduling to some extent, they have significant limitations from a technical perspective and are clearly inferior to the advanced nature of this invention: First, at the data processing level, they only achieve shallow stitching of multi-source data, without systematic spatiotemporal calibration, noise removal, feature enhancement, and uncertainty quantification, failing to fully explore the synergistic value of multi-source data, and the prediction accuracy is insufficient to meet the needs of refined scheduling; Second, at the algorithm level, the prediction model has weak generalization ability and poor adaptability to random fluctuations in renewable energy output and extreme weather conditions, and the scheduling optimization algorithm has limited solution dimensions, failing to simultaneously consider multiple objectives such as grid security constraints, economic operating costs, renewable energy consumption, and carbon emission control; Third, at the architecture level, there is a lack of a real-time control architecture with cloud-edge-device collaboration, with data processing and scheduling decisions mostly concentrated in the cloud, resulting in high response latency and failing to achieve real-time linkage and dynamic control of end-side devices (distributed energy storage, flexible loads); Fourth, the level of intelligence is insufficient, failing to introduce cutting-edge technologies such as large-scale energy models and digital twins, failing to achieve self-learning, self-optimization, and closed-loop iteration of scheduling decisions, and failing to adapt to the complex and ever-changing operating scenarios of new power systems. Summary of the Invention

[0005] This invention provides a smart power dispatch optimization method and system based on multi-source data fusion to solve the problems mentioned in the background section above: The present invention proposes a smart power dispatch optimization method based on multi-source data fusion, the method comprising: S1. Through end-side sensors and smart terminals, multi-source data in the power system are synchronously collected to generate a multi-source heterogeneous dataset; and edge computing nodes are used to cache the multi-source heterogeneous dataset in real time; based on the generated multi-source heterogeneous dataset, systematic preprocessing operations are performed to generate a preprocessed dataset. S2. Deeply fuse the generated preprocessed dataset, using a two-layer fusion strategy of feature-level fusion and decision-level fusion; dynamically allocate the decision weights of each data source in conjunction with the energy big data model, mine the collaborative value of multi-source data, output fused data, and generate a fused dataset. S3. Based on the generated fusion dataset, a prediction model is constructed using a large energy model and reinforcement learning algorithms to predict load demand and renewable energy output in real time for the next 15 minutes to 24 hours; and the prediction results are dynamically adjusted in conjunction with extreme weather warning data to generate an accurate load and renewable energy output prediction dataset. S4. Based on the generated accurate load and new energy output prediction dataset, set three core conditions; use quantum heuristic algorithm and multi-objective optimization algorithm to solve the scheduling scheme, determine the optimal scheduling combination, and generate the optimal scheduling scheme; S5. Based on the cloud-edge-device collaborative architecture, the generated optimal scheduling scheme is distributed to the edge nodes; the edge nodes, in conjunction with the real-time operating status of the device, fine-tune the optimal scheduling scheme and generate actual scheduling execution data. S6. Simulate the actual scheduling execution data generated by the digital twin and collect actual running data; compare the actual running data with the simulated data to determine the rationality of the scheduling scheme; use the self-learning function of the AI ​​agent to dynamically optimize the algorithm parameters and scheduling strategy to generate an optimized scheduling scheme; repeat the above steps S1 to S5, and continuously optimize the prediction model and adjust the scheduling scheme based on the newly collected data.

[0006] The present invention proposes a smart power dispatch optimization system based on multi-source data fusion, the system comprising: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0007] Beneficial effects of this invention: The data fusion is more in-depth, and a systematic multi-source data processing technology is adopted to complete spatiotemporal calibration, noise removal and feature enhancement, fully explore the collaborative value of data, significantly reduce prediction errors, and provide accurate data support for scheduling decisions; The algorithm is more adaptable, integrating cutting-edge algorithms such as energy big data models and reinforcement learning to construct a multi-objective scheduling optimization model, which can flexibly cope with new energy fluctuations and extreme operating conditions, and achieve multi-objective coordinated optimization such as safety, economy, and low carbon. The architecture is more real-time, building a cloud-edge-device collaborative control architecture to achieve real-time data collection on the device side, preprocessing on the device side, and global optimization in the cloud, thereby shortening the scheduling response delay and improving dynamic control capabilities. With a higher level of intelligence, it introduces technologies such as digital twins and AI agents to achieve self-learning, self-optimization and closed-loop iteration of scheduling decisions. It can adapt to the complex and ever-changing operating scenarios of new power systems, significantly improve the efficiency of power grid operation and the absorption rate of new energy sources, reduce operating costs, and fill existing technological gaps. Attached Figure Description

[0008] Figure 1 This is a diagram illustrating the steps of the method described in this invention. Detailed Implementation

[0009] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0010] One embodiment of the present invention, such asFigure 1 As shown, a smart power dispatch optimization method based on multi-source data fusion includes: S1. Simultaneously collect multi-source data from the power system via end-side sensors and intelligent terminals. This multi-source data includes PMU phasor monitoring data, renewable energy output time-series data, meteorological data, equipment status data, load breakdown data, and power grid topology data, generating a multi-source heterogeneous dataset. This dataset is then cached in real-time using edge computing nodes. Based on the generated multi-source heterogeneous dataset, systematic preprocessing operations are performed, including: spatiotemporal calibration based on timestamp synchronization and geographic coordinate matching to eliminate data spatiotemporal asynchrony; noise removal by using wavelet transform algorithms to remove abnormal interference signals; data completion based on an improved Bayesian algorithm combined with historical similar data; and redundancy removal by deleting duplicate and invalid data through correlation analysis. Finally, a preprocessed dataset is generated. S2. Deeply fuse the generated preprocessed dataset using a two-layer fusion strategy of feature-level fusion and decision-level fusion: extract core features from each data source through a convolutional neural network, construct a unified feature vector, and complete feature-level fusion; dynamically allocate decision weights for each data source using a large energy model, explore the collaborative value of multi-source data, output standardized, high-precision fused data, and reduce prediction error to below 8%; generate a fused dataset. S3. Based on the generated fusion dataset, a prediction model is constructed using a large energy model and reinforcement learning algorithms to predict load demand and renewable energy output in real time for the next 15 minutes to 24 hours; and the prediction results are dynamically adjusted in conjunction with extreme weather warning data to generate an accurate load and renewable energy output prediction dataset. S4. Based on the generated accurate load and renewable energy output forecast dataset, set three core conditions: power grid security constraints, economic constraints, and low-carbon constraints; use quantum heuristic algorithms and multi-objective optimization algorithms to solve the scheduling scheme, determine the optimal scheduling combination, and generate the optimal scheduling scheme. S5. Based on the cloud-edge-device collaborative architecture, the generated optimal scheduling scheme is distributed to the edge nodes; the edge nodes, in conjunction with the real-time operating status of the end-side devices, fine-tune the optimal scheduling scheme, control the charging and discharging of distributed energy storage, flexible load adjustment, and virtual power plant aggregation response, so as to realize the real-time implementation and execution of the scheduling scheme; and generate actual scheduling execution data. S6. Simulate the actual scheduling execution data generated by the digital twin and collect actual operation data; compare the actual operation data with the simulated data to determine the rationality of the scheduling scheme; use the self-learning function of the AI ​​agent to dynamically optimize the algorithm parameters and scheduling strategy to achieve closed-loop iteration and continuous optimization of scheduling decisions; generate the optimized scheduling scheme; repeat the above steps S1 to S5, continuously optimize the prediction model and adjust the scheduling scheme based on the newly collected data, and realize real-time dynamic scheduling and intelligent management of the power system.

[0011] The working principle and effects of the above technical solution are as follows: By synchronously acquiring and systematically preprocessing multi-source data, the spatiotemporal benchmarks of various types of data can be unified, data quality can be improved, missing data can be supplemented, and insufficient scheduling support caused by spatiotemporal asynchrony and abnormal interference of data can be avoided. Simultaneously, a two-layer data fusion strategy can fully explore the collaborative value of data, output high-precision unified data, and control prediction errors to a lower level, avoiding decision-making biases caused by single data support. Combining energy big data models and reinforcement learning to build a prediction system can improve the prediction accuracy of load and renewable energy output, adapting to the fluctuation characteristics of renewable energy. Solving scheduling schemes through multi-objective optimization algorithms can balance the requirements of grid safety and economy with low-carbon operation, improving the adaptability of scheduling schemes. Relying on a cloud-edge-device collaborative architecture to execute scheduling commands can shorten the control response time and improve the efficiency of distributed resource linkage. Utilizing digital twins and AI agents to complete closed-loop iteration can continuously optimize algorithm parameters and scheduling strategies, avoiding scheduling strategy rigidity and response lag, ensuring stable grid operation, improving renewable energy absorption, and reducing overall operating costs.

[0012] In one embodiment of the present invention, S1 includes: S11. Start-up end-side sensors and intelligent terminals synchronously collect PMU phasor monitoring data, new energy output time-series data, meteorological data, equipment status data, load breakdown data and power grid topology data in the power system to form a complete multi-source heterogeneous dataset; S12. Transmit the multi-source heterogeneous dataset to the edge computing node to complete the real-time caching and temporary storage of the data, ensuring that the data is not lost during the collection process; S13. Perform timestamp synchronization and geographic coordinate matching operations on multi-source heterogeneous datasets to unify the spatiotemporal benchmark of various types of data and eliminate the problem of spatiotemporal asynchrony of multi-source data. S14. Use wavelet transform algorithm to scan and identify abnormal interference signals in the data, complete noise removal, and improve data purity. S15. The improved Bayesian algorithm is used to fill in the missing data with historical similar data, and duplicate and invalid data are deleted through correlation analysis to generate a preprocessed dataset.

[0013] The working principle and effects of the above technical solution are as follows: By synchronously collecting various types of power operation data through edge sensors and intelligent terminals, it can comprehensively cover key dimensions of power grid operation, improving the comprehensiveness and completeness of data collection. Edge computing nodes perform real-time caching and temporary storage of collected data, reducing the probability of data loss and avoiding data gaps caused by collection interruptions. Timestamp synchronization and geographic coordinate matching operations are performed on the dataset to unify the spatiotemporal benchmark of the data and reduce calculation deviations caused by spatiotemporal asynchrony. Wavelet transform algorithms are used to remove abnormal interference signals, improving data purity and preventing noisy data from interfering with subsequent analysis and calculations. An improved Bayesian algorithm is used to fill in missing data, combined with correlation analysis to remove duplicate and invalid data, improving data quality and usability. The entire process ensures the integrity and validity of the original data and provides a stable and reliable foundation for subsequent data fusion and scheduling calculations, avoiding distorted scheduling decisions due to insufficient data quality.

[0014] In one embodiment of the present invention, S13 includes: Traverse all data items in the multi-source heterogeneous dataset, extract the collection time identifier corresponding to each data item, and establish a unified time benchmark; Using a unified time benchmark as a reference, time alignment is performed on all data items to smooth out differences in data acquisition timing. Analyze the geographic distribution information corresponding to each data item in the multi-source heterogeneous dataset and establish a unified spatial benchmark; use the unified spatial benchmark as a reference to perform location matching operation on all data items and regulate the spatial distribution of the data. By achieving dual alignment in both time and space dimensions, a data set with consistent spatiotemporal benchmarks is formed, eliminating the problem of spatiotemporal asynchrony among multi-source data.

[0015] The working principle and effects of the above technical solution are as follows: By extracting the collection time identifiers of all data items and establishing a unified time benchmark, the data collection sequence can be standardized, improving the consistency of the time dimension. Data alignment using a unified time benchmark smooths out differences in collection sequence, reduces calculation errors caused by time misalignment, and avoids the impact of time sequence chaos on subsequent data processing. Analyzing the geographical distribution information of the data and establishing a unified spatial benchmark can regulate the spatial attribution of various types of data, enhancing the uniformity of the spatial dimension. Performing location matching using a unified spatial benchmark regulates the spatial distribution of data, reducing analytical biases caused by spatial information mismatch. After dual alignment in time and space, a data set with a consistent benchmark is formed, completely eliminating the problem of spatiotemporal asynchrony between multi-source data. This provides a precise spatiotemporal basis for subsequent data fusion and scheduling calculations, while preventing spatiotemporal information disorder from causing scheduling decisions to deviate from the actual operating state, thus improving the overall reliability of data processing.

[0016] In one embodiment of the present invention, S2 includes: S21. Retrieve the preprocessed dataset, start the convolutional neural network to extract the core features of each data source, integrate and construct a unified feature vector, and complete feature-level fusion. S22. Access the energy big data model to dynamically allocate the decision weights of each data source in the scheduling scenario, and strengthen the contribution ratio of high-value data. S23. Combining feature-level fusion results with dynamically allocated decision weights, multi-source data decision-level fusion is completed, fully releasing the collaborative value of data; S24. Standardize the fused data to output high-precision, uniformly formatted data, controlling the prediction error to below 8%; solidify the standardized fusion results to form a fused dataset that can be directly used for prediction calculations.

[0017] The working principle and effects of the above technical solution are as follows: By extracting core features from multi-source data and constructing a unified feature vector through convolutional neural networks, the integration of data features can be improved, and the interference of redundant features on subsequent calculations can be reduced. The dynamic allocation of decision weights for data sources in the energy big data model can amplify the role of high-value data and reduce computational interference from low-value data. Combining feature-level and decision-level dual-layer fusion methods can fully release the collaborative value of multi-source data and enhance data support capabilities. Standardizing the fused data can output high-precision data in a unified format, keeping prediction errors at a lower level and avoiding computational deviations caused by inconsistent data formats. Solidifying the fusion results into a dedicated dataset can improve the efficiency of the prediction process. The entire process maximizes the utilization of data value and provides stable and accurate data support for subsequent load and renewable energy output predictions, avoiding insufficient scheduling decision accuracy due to inadequate data fusion.

[0018] In one embodiment of the present invention, step S22 includes: Load the underlying computing framework of the large energy model, import the preprocessed dataset and complete the format adaptation of the model input layer, converting multi-type heterogeneous data into a unified input format that the model can recognize, ensuring the stable operation of subsequent computing processes; Iterates through all data source types in the preprocessed dataset, fully analyzes the numerical distribution and variation patterns of each type of data, records the fluctuation characteristics and correlations of data under different operating conditions, and provides a basis for weight allocation. The energy big data model is used to quantify the influence of various data sources in power dispatch scenarios. Combined with the impact of real-time power grid operation status assessment data on dispatch results, a complete quantitative assessment result is formed. Based on the quantitative calculation results, corresponding scheduling decision weights are assigned to each data source, and the weight values ​​are dynamically adjusted to make the weight distribution fit the current power grid operation environment and scheduling needs; the contribution ratio of high-value data in the decision-making system is increased, the interference of low-value data on the overall calculation is reduced, and core data dominates the scheduling decision-making process.

[0019] The working principle and effects of the above technical solution are as follows: Data format adaptation is achieved through the energy big data model framework, unifying the input format of various types of heterogeneous data, improving the stability of model calculations, and avoiding process interruptions caused by format incompatibility. By traversing and analyzing the distribution and fluctuation characteristics of all data sources, the inherent patterns of the data can be fully grasped, enhancing the rationality of weight allocation. Quantitative calculation of the intensity of data impact can accurately reflect the actual impact of various data on scheduling, reducing biases caused by subjective allocation. Dynamically adjusting decision weights based on real-time status allows the weight distribution to conform to the current operating conditions of the power grid, improving the accuracy of scheduling guidance. Increasing the proportion of high-value data and suppressing interference from low-value data allows core data to dominate the calculation process. The entire process maximizes the supporting role of effective data while avoiding interference from invalid information in scheduling judgments, reducing the risk of decision-making errors, and providing a reliable guarantee for subsequent data fusion and scheduling optimization.

[0020] In one embodiment of the present invention, step S3 includes: S31. Load the fusion dataset, build a prediction framework that integrates the energy big model and reinforcement learning algorithm, and complete the model parameter initialization; S32. The driving prediction model performs time-by-time simulation of the power grid load demand for the next 15 minutes to 24 hours and outputs the load demand prediction results. S33, the driving prediction model extrapolates the output of new energy sources from the next 15 minutes to 24 hours in a time-by-time manner and outputs the prediction results of new energy output; S34. Access extreme weather warning data to make real-time corrections to load demand and renewable energy output forecasts, reducing forecast deviations; integrate the corrected load and renewable energy output results to generate an accurate load and renewable energy output forecast dataset.

[0021] The working principle and effects of the above technical solution are as follows: By building a prediction framework and completing initialization through a large energy model and reinforcement learning algorithms, the response speed of prediction calculations can be improved, and the stability of the model's operating state can be maintained. Time-by-time simulation of grid load demand can fully cover short, medium, and long-term operating cycles, improving the comprehensiveness of load forecasting. Time-by-time simulation of renewable energy output can accurately capture output fluctuation patterns, enhancing the accuracy of renewable energy forecasts. Real-time correction of prediction results by incorporating extreme weather warning data can reduce prediction deviations and mitigate prediction inaccuracies caused by sudden weather changes. Integrating the corrected results to form an accurate prediction dataset can provide a reliable reference for dispatching scheme formulation. The entire process balances prediction duration and accuracy while avoiding prediction distortion under extreme conditions, preventing the disconnect between dispatching arrangements and actual operation due to prediction deviations.

[0022] In one embodiment of the present invention, step S4 includes: S41. Based on accurate load and new energy output forecast datasets, define the operating boundaries of power grid security constraints, economic constraints, and low-carbon constraints.

[0023] S42. Import quantum-inspired algorithms to construct the computational kernel, and combine them with multi-objective optimization algorithms to build a scheduling scheme solution system; S43. Input the constraints and prediction data into the solution system, traverse all feasible scheduling combinations and complete the calculation iteration; compare the comprehensive performance of each scheduling combination under the constraints, and select the scheduling combination with the highest fitness. S44. Solidify the execution logic and parameter configuration of the optimal scheduling combination, and generate an optimal scheduling scheme that can be deployed and executed.

[0024] The working principle and effects of the above technical solution are as follows: Defining triple constraint boundaries using accurate prediction data stabilizes the scope of power grid operation and control, improves the compliance of dispatching schemes, and prevents exceeding safe and economical operating limits. A solution system constructed using quantum-inspired algorithms combined with multi-objective optimization algorithms improves computational efficiency and solution accuracy, enhancing the ability to find optimal solutions in complex scenarios. Importing constraints and data into the system for iterative traversal and comparison screening comprehensively covers feasible dispatching paths, reducing the possibility of missing optimal solutions and avoiding overall imbalance caused by single-dimensional optimization. Solidifying the logic and parameters of the optimal combination and generating an executable solution simplifies the subsequent deployment process and improves the coherence of dispatching execution. The entire process balances the triple objectives of safety, economy, and low carbon emissions, while avoiding mismatches between dispatching schemes and actual operating conditions, reducing power grid operation risks and resource waste.

[0025] In one embodiment of the present invention, S43 includes: The power grid security constraints, economic constraints, low-carbon constraints, and load renewable energy output forecast data are synchronously imported into the dispatch scheme solution system to complete the full loading of all calculation elements. The scheduling scheme solution system performs a combined traversal of all adjustable resources in the power system, covers all feasible scheduling paths, and generates a complete set of scheduling combinations. The solution system performs round-by-round calculations on the complete set of scheduling combinations, updating the operating parameters and constraint matching degree of each scheduling combination; During the calculation iteration process, the operating status of each scheduling combination under multiple constraints is continuously compared, and the comprehensive score of each indicator is calculated. The optimal scheduling combination that best suits the current operating state of the power grid is obtained by extracting the option with the highest comprehensive score from all scheduling combinations.

[0026] The working principle and effects of the above technical solution are as follows: Simultaneous import of multiple constraints and prediction data to load computational elements ensures a complete and unified computational foundation, improves the reliability of subsequent solutions, and avoids solution bias caused by missing data. Performing a combined traversal of all adjustable resources covers all feasible scheduling paths, reducing the possibility of overlooking high-quality scheduling solutions. Iteratively updating the operating parameters and constraint matching degree of the scheduling combinations in each round continuously improves the fit between the solution and actual operating conditions, enhancing the rigor of the calculation process. Continuously comparing the operating status of each combination and calculating a comprehensive score objectively reflects the overall advantages and disadvantages of different solutions, avoiding selection errors caused by biased evaluations. Extracting the combination with the highest score as the optimal solution accurately matches the current operating state of the power grid. The entire process ensures both the comprehensiveness and fairness of the scheduling solution and avoids scheduling decisions biased towards a single objective, improving the stability and economy of power grid operation.

[0027] In one embodiment of the present invention, step S5 includes: S51. Through the cloud-edge-device collaborative architecture, the optimal scheduling scheme is distributed to each edge node to complete the full-domain push of the scheme; S52. Side nodes collect real-time data on the operating status of end-side devices and make local fine-tuning of the optimal scheduling scheme based on on-site conditions; S53. Side nodes issue execution commands to end-side devices to control distributed energy storage to complete charging and discharging actions; side nodes guide flexible loads to participate in regulation and drive the virtual power plant to complete aggregation response actions; S54. Record the entire instruction execution process and equipment feedback information to generate complete actual scheduling execution data.

[0028] The working principle and effects of the above technical solution are as follows: By using a cloud-edge-device collaborative architecture to push scheduling plans across the entire domain, the speed of command transmission can be accelerated, the scheduling coverage can be improved, and the problem of local nodes being unable to receive scheduling commands can be avoided. Edge nodes can fine-tune the plan based on field conditions, improving the adaptability of scheduling execution and reducing execution deviations caused by mismatches between the global plan and field conditions. Edge nodes can directly control energy storage to regulate loads and drive virtual power plant responses, shortening control response time, enhancing the collaborative linkage effect of distributed resources, and preventing resource regulation lags from affecting grid balance. The entire execution process and equipment feedback are recorded and complete data is generated, retaining full-process execution information and providing a reliable basis for subsequent review and optimization. The entire process ensures that scheduling commands are implemented quickly and that field execution is more flexible and smooth, reducing the grid operation risks caused by scheduling execution disconnects.

[0029] In one embodiment of the present invention, step S6 includes: S61. Load actual scheduling execution data, drive the power grid digital twin to carry out full-process simulation operation of scheduling, and output simulation operation data; S62. Compare the actual operating data with the simulated operating data item by item to evaluate the effectiveness and rationality of the scheduling plan. S63. Start the AI ​​agent, automatically adjust the algorithm parameters and update the core logic of the scheduling strategy based on the data comparison results; complete the algorithm and strategy iterative optimization, and generate a more adaptable optimized scheduling scheme. S64. Based on the optimized scheduling scheme, the S1 to S5 processes are executed cyclically to support the continuous real-time dynamic scheduling and intelligent management of the power system.

[0030] The working principle and effects of the above technical solution are as follows: By loading actual execution data into a digital twin to conduct full-process simulation, the scheduling operation scenario can be completely reproduced, improving the realism of the solution evaluation and avoiding interference with the normal operation of the power grid caused by offline testing. Item-by-item comparison of actual and simulated operation data can intuitively reflect the execution effect of the scheduling solution, enhance the accuracy of problem identification, and reduce deviation omissions. The AI ​​agent automatically adjusts parameters and updates strategies based on the comparison results, enabling rapid optimization iteration, improving the solution's adaptability, and preventing strategies from becoming rigid and unable to keep up with changes in the power grid. Continuously executing the optimized solution throughout the entire process forms a continuously upgraded scheduling closed loop, enhancing the system's intelligence level. The entire process can complete solution verification without affecting power grid operation, and allow the scheduling strategy to continuously evolve, reducing the risk of decision lag in long-term operation.

[0031] In one embodiment of the present invention, S63 includes: Start the AI ​​agent and import the comparison results between actual and simulated running data to complete the full reading of data characteristics and deviation information; The AI ​​agent performs in-depth analysis of the deviation distribution and changing trends in the comparison results to identify areas for optimization in the algorithm and strategy. The AI ​​agent adjusts the internal parameters of the algorithm based on the deviation characteristics obtained from the analysis, and corrects the deviation propagation path in the calculation process; the core logic of the AI ​​agent updates the scheduling strategy and optimizes the allocation method of scheduling resources and the control response rules. The AI ​​agent integrates and optimizes algorithm parameters and scheduling strategies to generate a more adaptable optimized scheduling scheme.

[0032] The working principle and effects of the above technical solution are as follows: By fully reading the data comparison results and deviation information through an AI agent, abnormal characteristics in scheduling operations can be accurately captured, improving the direction of subsequent optimization and avoiding a lack of basis for optimization work. In-depth analysis of deviation distribution and trends can quickly pinpoint weaknesses in algorithms and strategies, reducing the time spent on problem investigation. Adjusting algorithm parameters and correcting deviation propagation paths based on deviation characteristics can reduce calculation errors at the source, lowering the possibility of continuous expansion of scheduling deviations. Updating the core logic of the scheduling strategy and optimizing resource allocation and response rules can improve the strategy's adaptability to dynamic changes in the power grid, avoiding execution lag caused by strategy rigidity. Integrating and optimizing content to generate a more adaptable scheduling scheme allows subsequent scheduling execution to better fit the on-site conditions. The entire process enables the optimization process to proceed autonomously and efficiently, continuously improving the quality of scheduling decisions and preventing the accumulation of deviations from affecting the safe and stable operation of the power grid.

[0033] One embodiment of the present invention provides a smart power dispatch optimization system based on multi-source data fusion, the system comprising: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0034] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A smart power dispatch optimization method based on multi-source data fusion, characterized in that, The method includes: S1. Through end-side sensors and smart terminals, multi-source data in the power system are synchronously collected to generate a multi-source heterogeneous dataset; and edge computing nodes are used to cache the multi-source heterogeneous dataset in real time; based on the generated multi-source heterogeneous dataset, systematic preprocessing operations are performed to generate a preprocessed dataset. S2. Deeply fuse the generated preprocessed dataset, using a two-layer fusion strategy of feature-level fusion and decision-level fusion; dynamically allocate the decision weights of each data source in conjunction with the energy big data model, mine the collaborative value of multi-source data, output fused data, and generate a fused dataset. S3. Based on the generated fusion dataset, a prediction model is constructed using a large energy model and reinforcement learning algorithms to predict load demand and renewable energy output in real time for the next 15 minutes to 24 hours; and the prediction results are dynamically adjusted in conjunction with extreme weather warning data to generate an accurate load and renewable energy output prediction dataset. S4. Based on the generated accurate load and new energy output prediction dataset, set three core conditions; use quantum heuristic algorithm and multi-objective optimization algorithm to solve the scheduling scheme, determine the optimal scheduling combination, and generate the optimal scheduling scheme; S5. Based on the cloud-edge-device collaborative architecture, the generated optimal scheduling scheme is distributed to the edge nodes; the edge nodes, in conjunction with the real-time operating status of the device, fine-tune the optimal scheduling scheme and generate actual scheduling execution data. S6. Simulate the actual scheduling execution data generated by the digital twin and collect actual running data; compare the actual running data with the simulated data to determine the rationality of the scheduling scheme; use the self-learning function of the AI ​​agent to dynamically optimize the algorithm parameters and scheduling strategy to generate an optimized scheduling scheme; repeat the above steps S1 to S5, and continuously optimize the prediction model and adjust the scheduling scheme based on the newly collected data.

2. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 1, characterized in that, S1 includes: S11. Start-up end-side sensors and intelligent terminals synchronously collect PMU phasor monitoring data, new energy output time-series data, meteorological data, equipment status data, load breakdown data and power grid topology data in the power system to form a complete multi-source heterogeneous dataset; S12. Transmit the multi-source heterogeneous dataset to the edge computing node to complete the real-time caching and temporary storage of the data; S13. Perform timestamp synchronization and geographic coordinate matching operations on multi-source heterogeneous datasets to unify the spatiotemporal benchmark of various types of data and eliminate the problem of spatiotemporal asynchrony of multi-source data. S14. Use wavelet transform algorithm to scan and identify abnormal interference signals in the data, complete noise removal, and improve data purity. S15. The improved Bayesian algorithm is used to fill in the missing data with historical similar data, and duplicate and invalid data are deleted through correlation analysis to generate a preprocessed dataset.

3. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 2, characterized in that, S13 includes: Traverse all data items in the multi-source heterogeneous dataset, extract the collection time identifier corresponding to each data item, and establish a unified time benchmark; Using a unified time benchmark as a reference, time alignment is performed on all data items to smooth out differences in data acquisition timing. Analyze the geographic distribution information corresponding to each data item in the multi-source heterogeneous dataset and establish a unified spatial benchmark; use the unified spatial benchmark as a reference to perform location matching operation on all data items and regulate the spatial distribution of the data. By achieving dual alignment in both time and space dimensions, a data set with consistent spatiotemporal benchmarks is formed, eliminating the problem of spatiotemporal asynchrony among multi-source data.

4. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 1, characterized in that, The S2 includes: S21. Retrieve the preprocessed dataset, start the convolutional neural network to extract the core features of each data source, integrate and construct a unified feature vector, and complete feature-level fusion. S22. Access the large energy model and dynamically allocate the decision weights of each data source in the scheduling scenario; S23. Combine the feature-level fusion results with dynamically allocated decision weights to complete the decision-level fusion of multi-source data; S24. Standardize the fused data to output high-precision, uniformly formatted data, controlling the prediction error to below 8%; solidify the standardized fusion results to form a fused dataset that can be directly used for prediction calculations.

5. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 4, characterized in that, S22 includes: Load the underlying computing framework of the large energy model, import the preprocessed dataset and complete the format adaptation of the model input layer, converting multi-type heterogeneous data into a unified input format that the model can recognize; Iterate through all data source types in the preprocessed dataset, fully analyze the numerical distribution and variation patterns of each type of data, and record the fluctuation characteristics and correlations of the data under different operating conditions. The energy big data model is used to quantify the influence of various data sources in power dispatch scenarios. Combined with the impact of real-time power grid operation status assessment data on dispatch results, a complete quantitative assessment result is formed. Based on the quantitative calculation results, corresponding scheduling decision weights are assigned to each data source, and the weight values ​​are dynamically adjusted to make the weight distribution fit the current power grid operating environment and scheduling needs.

6. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 1, characterized in that, The S3 includes: S31. Load the fusion dataset, build a prediction framework that integrates the energy big model and reinforcement learning algorithm, and complete the model parameter initialization; S32. The driving prediction model performs time-by-time simulation of the power grid load demand for the next 15 minutes to 24 hours and outputs the load demand prediction results. S33, the driving prediction model extrapolates the output of new energy sources from the next 15 minutes to 24 hours in a time-by-time manner and outputs the prediction results of new energy output; S34. Access extreme weather warning data to make real-time corrections to load demand and renewable energy output forecasts, thereby reducing forecast deviations; integrate the corrected load and renewable energy output results to generate a load and renewable energy output forecast dataset.

7. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 1, characterized in that, The S4 includes: S41. Based on the load and new energy output forecast dataset, define the operating boundaries of power grid security constraints, economic constraints, and low-carbon constraints. S42. Import quantum-inspired algorithms to construct the computational kernel, and combine them with multi-objective optimization algorithms to build a scheduling scheme solution system; S43. Input the constraints and prediction data into the solution system, traverse all feasible scheduling combinations and complete the calculation iteration; compare the comprehensive performance of each scheduling combination under the constraints, and select the scheduling combination with the highest fitness. S44. Solidify the execution logic and parameter configuration of the optimal scheduling combination, and generate an optimal scheduling scheme that can be deployed and executed.

8. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 1, characterized in that, The S5 includes: S51. Through the cloud-edge-device collaborative architecture, the optimal scheduling scheme is distributed to each edge node to complete the full-domain push of the scheme; S52. Side nodes collect real-time data on the operating status of end-side devices and make local fine-tuning of the optimal scheduling scheme based on on-site conditions; S53. Side nodes issue execution commands to end-side devices to control distributed energy storage to complete charging and discharging actions; side nodes guide flexible loads to participate in regulation and drive the virtual power plant to complete aggregation response actions; S54. Record the entire instruction execution process and equipment feedback information to generate complete actual scheduling execution data.

9. The intelligent power dispatch optimization method based on multi-source data fusion according to claim 1, characterized in that, The S6 includes: S61. Load actual scheduling execution data, drive the power grid digital twin to carry out full-process simulation operation of scheduling, and output simulation operation data; S62. Compare the actual operating data with the simulated operating data item by item to evaluate the effectiveness and rationality of the scheduling plan. S63. Start the AI ​​agent, automatically adjust the algorithm parameters and update the core logic of the scheduling strategy based on the data comparison results; complete the algorithm and strategy iterative optimization, and generate a more adaptable optimized scheduling scheme. S64. Based on the optimized scheduling scheme, the S1 to S5 processes are executed cyclically to support the continuous real-time dynamic scheduling and intelligent management of the power system.

10. A smart power dispatching and optimization system based on multi-source data fusion, characterized in that, The system includes: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 9.

Images