Self-updating new energy power prediction method and system

Abstract

The invention relates to the technical field of supervision control and data acquisition, in particular to a self-updating new energy power prediction method and system, and the method comprises the steps: carrying out the element space reconstruction of a new energy station, a meteorological station and a power grid topology of a target region, and determining a new energy spectrum; deploying a first updating area and a second prediction area, building a power predictor, and embedding the power predictor into a target area center controller; introducing a differential processing plug-in to establish communication interaction; and the transmission data triggers the power predictor to execute processing, power prediction data is determined, and discrete release management and control of the new energy station are carried out. The technical problem that a traditional data acquisition control system cannot provide comprehensive and accurate data support for new energy power prediction due to the fact that the traditional data acquisition control system is difficult to adapt to real-time interaction and dynamic processing requirements of multi-source data is solved, and real-time fusion and dynamic updating of the multi-source data are achieved; and comprehensive and accurate data support is provided for new energy power prediction, and the prediction precision and the management and control effect are improved.

Description

Technical Field

[0001] This invention relates to the field of monitoring and control and data acquisition technology, and in particular to a self-updating method and system for predicting new energy power. Background Technology

[0002] In renewable energy power forecasting, accurate power prediction is crucial for the stable operation of the power grid. Current technologies primarily rely on single data sources or static models, acquiring data from renewable energy plants and meteorological data using simple data acquisition equipment. While these methods have some application in steady-state grid environments or small-scale renewable energy integration scenarios, their limitations become apparent when applied to dynamic grid scenarios due to large-scale renewable energy grid integration and the increasing complexity of grid topologies.

[0003] Due to the complexity of power grid topology and the volatility of renewable energy output, traditional methods cannot effectively integrate multi-source heterogeneous data and lack dynamic update mechanisms. This results in incomplete data acquisition, poor model adaptability, and difficulty in meeting the accuracy requirements of power prediction under large-scale renewable energy grid integration. In particular, existing data acquisition and control systems struggle to achieve real-time interaction and dynamic processing of multi-source data such as meteorological and power grid topology data, failing to provide comprehensive and real-time data support for power prediction. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a self-updating method and system for predicting new energy power. This method and system can solve the problem that traditional data acquisition and control systems are unable to adapt to the real-time interaction and dynamic processing requirements of multi-source data, thus failing to provide comprehensive and accurate data support for predicting new energy power.

[0005] To achieve the above objectives, the present invention adopts the following specific technical solution:

[0006] The self-updating renewable energy power prediction method provided by this invention includes:

[0007] S1. For the new energy power stations, meteorological stations, and power grid topology deployed in the target area, perform spatial reconstruction of elements to determine the new energy map;

[0008] S2. Deploy the first update zone based on time-varying reset of the new energy map, deploy the second prediction zone based on power trend decision under time-varying state migration, build a power predictor and embed it in the target area central control.

[0009] S3. By introducing a differential processing plugin, establish communication interaction between new energy power plants, meteorological stations, power grid topology and power predictors under directional data loop;

[0010] S4. Using meteorological data and power distribution data as variable data, the auxiliary differential processing plug-in performs communication and transmission, triggers the power predictor to perform first-order update and second-order decision processing, determines the power prediction data, and performs discrete decentralization management of new energy power plants.

[0011] Further, in step S1, the element space is reconstructed to determine the new energy map, as follows:

[0012] S11. Using new energy power stations and meteorological stations as nodes and propagation relationships and power grids as edges, an initialization space is constructed. The propagation relationships are determined by introducing meteorological dynamics operators.

[0013] S12. By analogy with various meteorological dynamic operators, a set of codes is determined, and two sets of codes are determined based on the power grid distribution characteristics. The combination of the first set of codes and the two sets of codes is used as the feature code. The feature code has time-varying attention weights.

[0014] S13. Construct a new energy map based on the initialization space and feature encoding.

[0015] Furthermore, in step S2, the power predictor is constructed as follows:

[0016] S21. Based on the new energy map, perform inner loop training using time-varying reset based on feature coding to determine the first update zone. The inner loop is a data loop consisting of the meteorological station, the power grid topology, and the first update zone.

[0017] S22. Use the outer loop training based on the Markov state change of the update difference to determine the second prediction region. The outer loop is a data loop consisting of the first update region, the second prediction region and the new energy power station.

[0018] S23, cascaded with the first update region and the second prediction region, constitutes a power predictor.

[0019] Furthermore, in step S3, communication interaction is established between the new energy power plant, meteorological station, power grid topology, and power predictor under directional data loop, as detailed below:

[0020] S31. For new energy power plants, meteorological stations and power grid topology, data interfaces are opened on the output side of each source data;

[0021] S32. The power predictor is deployed in the central control of the target area. A multi-source data interface is opened in the power predictor, and communication and interaction between the data interface and the multi-source data interface are established.

[0022] Furthermore, the data interfaces opened on the output side of each source data are embedded with differential processing plugins, which determine the differential relaxation coefficient by balancing data value and data privacy.

[0023] Furthermore, in step S4, the auxiliary differential processing plugin performs communication interaction and transmission, as follows:

[0024] S41. Determine the first meteorological data based on the weather station, and perform data transformation through the differential processing plugin to determine the first differential meteorological data;

[0025] S42. Determine the power distribution demand based on the power grid topology, and perform data conversion through the differential processing plug-in to determine the first differential power distribution data;

[0026] S43. Transmit the first differential meteorological data and the first differential power data to the power predictor deployed in the target area.

[0027] Furthermore, in step S4, the power predictor is triggered to perform a first-order update, as follows:

[0028] S44. Based on the first update region, the attention weight of a group of coding parts is updated using the first differential meteorological data, and the attention weight of a second group of coding parts is updated using the first differential power distribution data to determine the updated map.

[0029] S45. Transfer the updated map to the second prediction area, and perform iterative storage of the updated map and the new energy map in the first update area.

[0030] Further, in step S4, the power predictor is triggered to perform second-order decision processing, as follows:

[0031] S46. For the updated map and the new energy map, measure the time-varying update difference, which is the migration of meteorological state and power distribution state.

[0032] S47. Determine the upper-level output data of the new energy power stations, wherein the upper-level output data is the power distribution status of each new energy power station before the update;

[0033] S48. Using the time-varying update difference as the state independent variable, perform knowledge deduction to determine the power prediction data, where the power prediction data includes the power data of each new energy power station.

[0034] Furthermore, in step S4, the decentralized management and control of new energy power stations is carried out, as detailed below:

[0035] Using the data interface of the new energy power station as the receiving end, the power prediction data is discretely transmitted.

[0036] Based on the installed capacity of each new energy power station, the received power prediction data is used to allocate equipment output and control the drive.

[0037] This invention provides a self-updating renewable energy power prediction system, which applies the above-mentioned self-updating renewable energy power prediction method and includes:

[0038] The new energy map acquisition module is used to reconstruct the spatial elements of new energy power plants, meteorological stations, and power grid topology deployed in the target area to determine the new energy map;

[0039] The power predictor building module is used to deploy the first update zone with time-varying reset based on the new energy map, deploy the second prediction zone with power trend decision under time-varying state migration, build the power predictor and embed it in the central control of the target area;

[0040] The communication interaction construction module is used to establish communication interaction between new energy power plants, meteorological stations, power grid topology and power predictors under directional data loop by introducing differential processing plug-in;

[0041] The discrete decentralization control execution module is used to transmit meteorological data and power distribution data as variables, assisting the differential processing plug-in in communication interaction, triggering the power predictor to perform first-order update and second-order decision processing, determining the power prediction data, and performing discrete decentralization control of new energy power plants.

[0042] The present invention can achieve the following technical effects:

[0043] This invention constructs a new energy map by spatially reconstructing elements such as new energy power plants, meteorological stations, and power grid topology. It then builds a power predictor using a first update zone with time-varying reset and a second prediction zone based on state-trending decision-making. A differential processing plugin enables directional cyclical interaction of multi-source data. Meteorological and distribution data drive the power predictor to perform first-order updates and second-order decisions, thereby determining power prediction data and enabling discrete decentralized control of new energy power plants. This solves the problem that traditional data acquisition and control systems struggle to adapt to real-time interaction and dynamic processing of multi-source data, achieving accurate prediction and control of new energy power. It provides comprehensive and accurate data support for new energy power prediction, improving prediction accuracy and control effectiveness. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the self-updating new energy power prediction method provided in an embodiment of the present invention.

[0045] Figure 2 This is a schematic diagram of the structure of a self-updating new energy power prediction system provided in an embodiment of the present invention. Detailed Implementation

[0046] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0048] This invention provides a self-updating method for predicting renewable energy power, the process of which is as follows: Figure 1 As shown, it includes the following steps:

[0049] S1. For the new energy power plants, meteorological stations, and power grid topology deployed in the target area, perform spatial reconstruction of elements to determine the new energy map.

[0050] In this embodiment of the invention, the new energy power station refers to the new energy power generation units such as wind farms and photovoltaic power stations deployed within the target area, serving as one of the core nodes in the construction of the new energy map. The power grid topology is the physical connection structure of the power grid within the target area, including the spatial layout and connection relationships of elements such as substations, transmission lines, and distribution networks. The new energy map is a digital model constructed using new energy power stations and meteorological stations as nodes, and propagation relationships and power grid topology as edges, forming an initial spatial structure.

[0051] For target areas, such as areas with concentrated access to new energy sources covered by the power grid, three key physical elements need to be deployed: new energy power plants (including wind farms, photovoltaic power plants, etc.), whose spatial distribution and installed capacity directly affect power output characteristics; meteorological stations, used to collect meteorological data such as wind speed, light intensity, temperature, and humidity in real time, and to quantify the propagation relationship of meteorological elements to new energy output by introducing meteorological dynamics operators; and power grid topology, including the connection structure of substations, transmission lines, and distribution networks, serving as the physical carrier of power transmission. These three elements achieve real-time data interaction through SCADA (Supervisory Control and Data Acquisition System, with a collection frequency of 5 minutes / time). New energy power plants and meteorological stations serve as map nodes, and the power grid topology and meteorological propagation relationship serve as connecting edges, providing a physical foundation for subsequent spatial reconstruction of elements and construction of the new energy map. Simultaneously, by embedding differential processing plugins in the data interface, the value utilization and privacy protection needs in the multi-source data acquisition process are balanced.

[0052] An initialization space is constructed using new energy power plants and meteorological stations as nodes and the propagation relationship and power grid as edges. The propagation relationship is determined by meteorological dynamics operators. A set of codes is generated by analogy with meteorological dynamics operators, and two sets of codes are determined by combining the power grid distribution characteristics. These are combined to form a feature code with time-varying attention weights. Then, a new energy map is constructed based on the initialization space and the feature code. The specific steps are shown in S11 to S13:

[0053] S11. Using new energy power stations and meteorological stations as nodes and propagation relationships and power grids as edges, an initialization space is constructed. The propagation relationships are determined by introducing meteorological dynamics operators.

[0054] S12. By analogy with various meteorological dynamic operators, a set of codes is determined, and two sets of codes are determined based on the power grid distribution characteristics. The combination of the first set of codes and the two sets of codes is used as the feature code. The feature code has time-varying attention weights.

[0055] S13. Construct a new energy map based on the initialization space and feature encoding.

[0056] By constructing an initial space and building a new energy map with new energy power plants and meteorological stations as nodes and propagation relationships and power grids as edges, the spatial reconstruction and feature fusion of multi-source elements in the target area are realized, providing a digital infrastructure for power prediction.

[0057] Meteorological dynamics operators are mathematical models or computational operators used to quantify the propagation patterns and influence mechanisms of meteorological elements (such as wind speed and light intensity) in space. Power grid distribution characteristics are a set of parameters reflecting the physical connection characteristics and operating status of the power grid during the distribution process. Time-varying attention weights are dynamic weight coefficients assigned to different feature components in feature encoding; their values ​​are adjusted according to changes in real-time information such as meteorological data and power distribution data.

[0058] For large-scale integrated distributed base station scenarios within a region, firstly, with the new energy power stations and meteorological stations deployed in the target area as basic nodes, corresponding meteorological dynamic operators are introduced for meteorological elements such as wind speed, light intensity, and temperature gradient:

[0059] For wind speed propagation, the Navier-Stokes equations are selected based on fluid dynamics theory. A three-dimensional flow field computational domain is established using topographic data (such as mountain height and vegetation coverage) and initial wind speed data monitored by meteorological stations in the target area. The computational domain is discretized into grid cells, and boundary conditions (such as ground roughness and friction velocity) are substituted to solve the equations, obtaining the wind speed distribution field at different height levels. Furthermore, the wind speed attenuation coefficient during the propagation process from the meteorological station to the new energy power station is calculated; for example, the wind speed attenuates by 0.5 m / s per kilometer on flat terrain. Turbulence intensity parameters, such as the gradient of turbulence intensity decreasing with increasing height, are determined through turbulence spectrum analysis, ultimately quantifying the attenuation law and turbulence characteristics of wind speed during spatial propagation.

[0060] For solar radiation, the radiative transfer equation (RTE) is used to describe the photon transmission process in the atmosphere. Inputs include atmospheric optical properties (including aerosol optical depth (AOD), ozone concentration, and water vapor content) and raw solar intensity data from meteorological stations. By solving the RTE equation, considering atmospheric scattering and absorption effects such as the oxygen molecule absorption band and carbon dioxide absorption peak, and cloud cover attenuation coefficients (e.g., cumulus clouds attenuate direct sunlight by up to 40%), a solar intensity attenuation model is established from the meteorological monitoring point to the photovoltaic power station. The model needs to iteratively calculate the attenuation coefficients for different wavelengths of solar radiation (e.g., visible light 400-760 nm, near-infrared 760-1100 nm) and, combined with the tilt angle and azimuth angle of the photovoltaic panels, correct for the actual solar intensity distribution received by the power station.

[0061] Secondly, the meteorological propagation parameters calculated by the above operators, such as wind speed, wind speed attenuation coefficient, turbulence intensity parameter, cloud cover attenuation coefficient, and light intensity attenuation coefficient, are transformed into attribute values ​​of edges between nodes. Taking wind speed as an example, the wind speed influence weight coefficient between a certain meteorological station and surrounding stations is calculated based on the flow field model. This coefficient serves as the weight of the edge, representing the intensity of the meteorological station's wind speed influence on the surrounding stations. Similarly, the light intensity attenuation coefficient output by the light intensity attenuation model constitutes the propagation delay attribute of the edge. Finally, through the quantitative calculation of meteorological dynamics operators, the abstract meteorological propagation relationship is transformed into computable edge parameters in the initialization space, realizing the dynamic connection construction between new energy power stations and meteorological stations in the spatial dimension.

[0062] Simultaneously, the physical connection structure of the power grid within the target area is digitally abstracted: with substations as key nodes and transmission lines and distribution cables as connecting links, a graph model of the power grid topology is established based on electrical connection relationships. The parameters of each transmission line (such as line impedance, voltage level, and transmission capacity) are used as the attribute values ​​of the edges, and the power transmission direction and constraint relationship between substations are represented by the edge weights. For large-scale integrated distributed base station scenarios within a region, there are multiple independent power grid topologies, such as distribution networks at different voltage levels and dedicated power supply lines for new energy power plants. By extracting the boundary nodes of each independent topology, such as the step-up substations connecting new energy power plants and regional distribution stations, an interrelation matrix between topologies is established. That is, the edge structures of each independent topology are cascaded using the power flow parameters (such as injected power and voltage fluctuations) of the boundary nodes as the link. For example, the topology of a wind farm's dedicated transmission line is cascaded with the regional main power grid topology through the step-up substation node. Its edge attributes (line impedance) and the edge attributes (main line transmission capacity) of the main power grid form an association constraint. Finally, an initialization space containing the interrelation of multiple independent topologies is constructed. This space integrates the dispersed power grid structure into a unified topology network through edge cascading, enabling the power grid distribution characteristics to achieve cross-topology collaborative representation in multi-source data fusion. Compared with the traditional method of processing each topology independently, this architecture reduces data interaction links and significantly simplifies the complexity of element association.

[0063] Secondly, a set of codes is generated by analogy with the mathematical expressions of various meteorological dynamic operators (such as the coefficient matrix of the wind speed propagation equation) to characterize the dynamic characteristics of meteorological propagation; two sets of codes are constructed using characteristic parameters such as power grid distribution voltage level and load distribution to reflect power transmission constraints. An attention mechanism is used to assign time-varying weights to both sets of codes, with the weight update cycle consistent with the SCADA data acquisition frequency of 5 minutes per cycle. This allows the feature codes to dynamically focus on key influencing factors; for example, during strong winds, the weight of the wind speed propagation code is automatically increased to over 60%.

[0064] Finally, the feature codes are embedded in the nodes and edges of the initialization space as attribute values. That is, the node attributes include basic information such as the installed capacity of new energy power plants and meteorological station monitoring parameters, as well as the weight allocation of the feature codes. The edge attributes integrate meteorological propagation parameters, power grid distribution characteristics, and the dynamic weights of the corresponding codes, thereby building a new energy map that integrates multi-source information from meteorology, power grid, and power plants.

[0065] By constructing an initialization space with multiple interconnected topologies and generating feature codes with time-varying weights, the physical relationships and dynamic characteristics of new energy power stations, meteorological stations, and power grid topologies in the target area are transformed into a computable map, solving the problems of scattered multi-source data and insufficient correlation analysis in traditional technologies.

[0066] S2. Deploy the first update zone using time-varying reset based on the new energy map, and deploy the second prediction zone using power trend decision under time-varying state migration. Build a power predictor and embed it in the target area central control.

[0067] In this embodiment, time-varying reset refers to the process of periodically updating the node and edge attributes of the new energy graph through a real-time data-driven dynamic adjustment mechanism. Time-varying state transition refers to predicting and analyzing the dynamic trend of new energy power generation by modeling the transition law of system state over time. The target area control unit is the core control unit for deploying the power predictor, and its functions include real-time data acquisition, multi-source information fusion, and decision command issuance.

[0068] Optionally, based on the new energy map, the first update region is determined through inner-loop training (the weather station, power grid topology, and the first update region constitute a data loop) using time-varying reset based on feature encoding. Then, the second prediction region is determined through outer-loop training (the first update region, the second prediction region, and the new energy power station constitute a data loop) using Markov state-change based on update differences. The two are cascaded to form a power predictor. Finally, the power predictor is built and embedded in the target area's central control system, with specific steps shown in S21-S23.

[0069] S21. Based on the new energy map, perform inner loop training using time-varying reset based on feature coding to determine the first update zone. The inner loop is a data loop consisting of the meteorological station, the power grid topology, and the first update zone.

[0070] S22. Use the outer loop training based on the Markov state change of the update difference to determine the second prediction region. The outer loop is a data loop consisting of the first update region, the second prediction region and the new energy power station.

[0071] S23, cascaded with the first update region and the second prediction region, constitutes a power predictor.

[0072] Real-time calibration of the new energy map was achieved through a dynamic weight update mechanism. By combining the LSTM-HMM cascade model and ensemble forecasting technology, the mean absolute error of traditional power prediction schemes was reduced, and the prediction accuracy under extreme weather scenarios was improved. Finally, a power prediction infrastructure with dynamic adaptability was constructed.

[0073] First, internal circulation training was conducted based on the new energy map: data such as wind speed and solar radiation from weather stations and power grid topology distribution parameters were collected at 5-minute intervals, and the weights of each indicator were calculated using the coefficient of variation method.

[0074] First, the mean of each indicator (such as wind speed, light intensity, power distribution parameters, etc.) is calculated to reflect the average level of the data. Second, the standard deviation of each indicator is calculated to measure the dispersion of the data. Next, the standard deviation of each indicator is divided by its mean to obtain the coefficient of variation, eliminating the influence of dimensions and reflecting the relative dispersion of the indicators. Finally, the coefficient of variation of each indicator is normalized, that is, the coefficient of variation of each indicator is divided by the sum of the coefficients of variation of all indicators to obtain the weight of each indicator. For example, if the mean of the wind speed data is 10 m / s and the standard deviation is 2.2 m / s, then the coefficient of variation is 0.22. Assuming that the sum of the coefficients of variation of other indicators such as light intensity is 0.28, then the weight of wind speed is 0.22 / (0.22+0.28) = 0.44, thereby realizing the time-varying reset of feature encoding.

[0075] The meteorological station, power grid topology, and the first update zone form an inner loop of data closure. After each data update, differential processing is used to transform the input meteorological data and power distribution data. For example, wind speed data is converted into a differential sequence, and then the weight of the feature encoding is updated through an attention mechanism. For example, the weight of wind speed-related encoding is increased to 0.6 under strong wind weather, forming a dynamically optimized first update zone.

[0076] Next, the second prediction region is determined using outer loop training: a Markov state transition model is constructed based on the update difference. First, the state space of the renewable energy power station is defined, such as dividing it into discrete states like normal output, overload, and low output. Second, the update map output from the first update region (including feature encoding weights, time-varying attention parameters, etc.) and the historical output data of the renewable energy power station (such as the power distribution sequence of the past hour and day) are used as inputs to extract the time-varying update difference (i.e., the migration between meteorological and power distribution states) as the state independent variable. Then, by statistically analyzing the transition frequency between each state in the historical data, the state transition probability is calculated using Bayesian estimation or maximum likelihood method to construct a transition probability matrix. For example, the probability of a power station transitioning from normal output to overload is calculated to be 0.12, and the probability of transitioning from normal output to low output is 0.08, etc. The final output is a matrix containing the transition probabilities between each state, providing probabilistic model support for subsequent power trend decisions based on the update difference.

[0077] In the outer loop, the first update zone, the second prediction zone and the new energy power station form a data loop. The model is trained using historical power deviation data. When an update difference is detected, such as when the meteorological state migration exceeds the threshold of 15%, the Markov model is triggered to update the state transition matrix, thereby realizing dynamic prediction of power trend changes.

[0078] Finally, the first update zone and the second prediction zone are cascaded to form a power predictor integrating dynamic updating and trend prediction functions, which is then embedded in the central control unit of the target area. The central control system receives multi-source data in real time through the SCADA interface. After receiving differential meteorological data (such as the first differential wind speed data) and differential power distribution data, the first update zone first completes feature encoding optimization, and then the second prediction zone outputs the power prediction value based on the Markov state trend. For example, the power prediction range for a certain wind farm in the next hour is [8.2MW, 10.5MW], and the peak value at 90% confidence level is 9.3MW.

[0079] By constructing a dual closed-loop training mechanism of inner loop time-varying reset and outer loop Markov state trend change, combined with feature encoding dynamic optimization and multi-source data closed-loop interaction, the average absolute error of traditional power prediction schemes is reduced, and the prediction accuracy under extreme weather scenarios is improved. Finally, through dual-zone cascade architecture and embedded deployment, high-precision real-time prediction and dynamic control of new energy power are realized.

[0080] S3. By introducing a differential processing plugin, establish communication interaction between new energy power plants, meteorological stations, power grid topology and power predictors under directional data loop.

[0081] The differential processing plugin is a functional module embedded in the data interfaces of the data output side of each source data in new energy power plants, meteorological stations, and power grid topologies.

[0082] For new energy power plants, meteorological stations, and power grid topology, data interfaces are opened on the output side of each source data. Power predictors are deployed in the target area central control and multi-source data interfaces are opened therein. Communication and interaction between the data interfaces and multi-source data interfaces are established. The specific steps are shown in S31 to S32:

[0083] S31. For new energy power plants, meteorological stations and power grid topology, data interfaces are opened on the output side of each source data;

[0084] S32. The power predictor is deployed in the central control of the target area. A multi-source data interface is opened in the power predictor, and communication and interaction between the data interface and the multi-source data interface are established.

[0085] First, for source data terminals such as SCADA systems of new energy power plants (such as photovoltaic power plants and wind farms), monitoring equipment of meteorological stations, and smart meters of power grid topology, standardized data interface modules are deployed at the hardware level. These modules support Modbus TCP / IP and OPCUA protocols, and firmware upgrades ensure that the output data of each device conforms to a unified format specification.

[0086] Secondly, the power predictor is embedded in the target area's central control system, such as an industrial server cluster in a power grid dispatch center. A multi-source data interface engine is developed within the predictor's software architecture. This engine supports simultaneous access to over 500 channels of meteorological data (such as wind speed and solar radiation), power grid topology parameters (such as line impedance and load distribution), and power station output data. By establishing an interface mapping table, such as associating meteorological station IDs with power station geographical locations, real-time routing and protocol conversion of multi-source data are achieved. For example, the IEC61850 protocol data of the power grid topology is automatically converted to the JSON format required by the predictor, with processing latency controlled within 10ms.

[0087] Ultimately, through the standardization of source data interfaces and the collaboration of the predictor's multi-source interface engine, an end-to-end directional data loop is constructed: real-time data from new energy power plants, meteorological stations, and power grid topology are transmitted to the central control unit via output interfaces at 50ms intervals. The predictor parses and feeds back the processing results in real time through the multi-source interface, forming a closed-loop interaction of data acquisition, prediction, and control, providing a high-speed and reliable data transmission channel for the real-time updating and accurate decision-making of the power predictor.

[0088] Through the collaborative design of standardized source data interfaces and multi-source interface engines, the effect of building low-latency, highly reliable directional data loop communication and interaction is achieved, solving the problems of low efficiency and protocol incompatibility in traditional multi-source data transmission.

[0089] Specifically, differential processing plugins are embedded in the data interfaces opened on the output side of each source data. The differential relaxation coefficients are determined by balancing data value and data privacy, as detailed below:

[0090] The data interfaces on the output side of each source data are embedded with differential processing plugins, in which the differential relaxation coefficient is determined by balancing data value and data privacy.

[0091] Balancing data value involves balancing the data's utility in power prediction with privacy protection needs during data interaction. This is achieved by adjusting differential relaxation coefficients, ensuring data availability while preventing the leakage of sensitive information. The differential relaxation coefficient is a key parameter embedded in the differential processing plugins of each source data output interface, used to balance data value and privacy.

[0092] Specifically, traditional solutions that directly transmit raw meteorological and distribution data lack effective protection for the privacy of data from multiple sites, potentially leading to the leakage of sensitive information such as grid topology and power plant output strategies. For example, leaked raw power data from a wind farm could be maliciously exploited, causing grid instability. To address this issue, differential processing plugins are embedded in the data interfaces on the output sides of each source data point. The following steps achieve a balance between privacy protection and data value:

[0093] First, a data sensitivity assessment model is established: meteorological data (such as wind speed and light intensity) and power distribution data (such as load distribution and line impedance) are classified and labeled. Information entropy is used to calculate the privacy risk value of each field. For example, the entropy value of substation coordinates in the power grid topology is 3.2 bits, which is considered highly sensitive data, while the entropy value of conventional wind speed data is 1.8 bits, which is considered low-sensitivity data. Based on sensitivity, the data is divided into three levels: high, medium, and low, corresponding to different privacy protection thresholds.

[0094] Secondly, a dynamic optimization algorithm for the differential relaxation coefficient is designed: A Pareto optimal model is constructed with privacy protection strength (ε value) and data availability (F1 score) as optimization objectives. The range of coefficient values ​​is determined through cross-validation: For example, when processing highly sensitive data, the relaxation coefficient is adjusted to 0.8, at which point the added Laplace noise intensity is 0.5. Although this leads to a 7% decrease in data accuracy, the privacy protection level is improved by 40%. When processing low-sensitivity data, the coefficient is reduced to 0.3, the noise intensity is 0.2, and the data accuracy decreases by only 2%, meeting the power prediction requirements.

[0095] Finally, an adaptive noise injection module is integrated into the plugin: it dynamically calls the corresponding coefficient based on the real-time data sensitivity to perform differential conversion on the first meteorological data and the power distribution demand data. For example, when the power distribution demand data of the power grid topology is detected to contain highly sensitive load distribution information, differential processing with a coefficient of 0.8 is automatically enabled to generate the first differential power distribution data. While ensuring that the data cannot restore the original load details, it retains the trend characteristics required for power prediction, such as load fluctuation cycles.

[0096] By embedding a differential processing plugin with dynamic relaxation coefficients into the data interface, a quantitative balance between privacy protection and power prediction accuracy in multi-site data interaction is achieved, effectively resolving the contradiction between data sharing and privacy security in traditional solutions.

[0097] S4. Using meteorological data and power distribution data as variable data, the auxiliary differential processing plug-in performs communication and transmission, triggers the power predictor to perform first-order update and second-order decision processing, determines the power prediction data, and performs discrete decentralization management of new energy power plants.

[0098] First, the auxiliary differential processing module performs communication and transmission, that is, it determines the first meteorological data based on the weather station and the power distribution demand based on the power grid topology. After being converted into the first differential meteorological data and the first differential power distribution data by the differential processing module, they are transmitted to the power predictor deployed in the central control of the target area. The specific steps are shown in S41 to S43:

[0099] S41. Determine the first meteorological data based on the weather station, and perform data transformation through the differential processing plugin to determine the first differential meteorological data;

[0100] S42. Determine the power distribution demand based on the power grid topology, and perform data conversion through the differential processing plug-in to determine the first differential power distribution data;

[0101] S43. Transmit the first differential meteorological data and the first differential power data to the power predictor deployed in the target area.

[0102] First, real-time meteorological data (such as wind speed and light intensity) from weather stations and power distribution demand data (such as load distribution and line impedance) based on the power grid topology are collected. Taking a 200MW wind farm as an example, its deployed weather stations collect wind speed data every 5 minutes, and the power grid SCADA system simultaneously collects real-time load demand data from the distribution network to form the raw data set.

[0103] Secondly, the two types of raw data are transformed by the differential processing plugin embedded in the data interface: the differential relaxation coefficient is dynamically adjusted according to the data sensitivity, with a value range of 0.1-0.8. For highly sensitive power distribution demand data (such as substation coordinates), a coefficient of 0.8 is used, and Laplace noise with an intensity of 0.5 is added to generate the first differential power distribution data; for low-sensitivity meteorological data (such as conventional wind speed), a coefficient of 0.3 and a noise intensity of 0.2 are used to obtain the first differential meteorological data.

[0104] Finally, the two types of differential data are transmitted to the power predictor deployed in the target area central control via a standardized data interface at a period of 50ms. The Modbus TCP / IP protocol is used to realize data encapsulation and transmission, and the communication latency is controlled below 80ms with a packet loss rate of less than 1%, ensuring the real-time performance and reliability of data interaction.

[0105] By dynamically reducing noise and standardizing the transmission of the raw data according to its sensitivity level through a differential processing plugin, the availability of the data required for power prediction is maintained while protecting data privacy, thus resolving the contradiction between data sharing and privacy security in traditional technologies.

[0106] Next, the power predictor is triggered to perform a first-order update. Based on the first update region, the attention weights of the first set of codes and the second set of codes are updated using the first differential meteorological data and the first differential power distribution data, respectively, to determine the updated map and transfer it to the second prediction region. At the same time, the updated map and the new energy map are iteratively stored in the first update region. The specific steps are shown in S44 to S45:

[0107] S44. Based on the first update region, the attention weight of a group of coding parts is updated using the first differential meteorological data, and the attention weight of a second group of coding parts is updated using the first differential power distribution data to determine the updated map.

[0108] S45. Transfer the updated map to the second prediction area, and perform iterative storage of the updated map and the new energy map in the first update area.

[0109] First, based on the feature coding structure of the first update region, attention weights are updated for a group of coding components using the first differential meteorological data (such as differential wind speed and differential light intensity). Specifically, the dynamic weights of the meteorological data are calculated using the coefficient of variation method (the process is the same as step A210, and will not be elaborated further here due to space limitations in the manual). For example, the coefficient of variation of the measured wind speed of a certain wind farm is 0.22, and the corresponding weight is adjusted to 0.458, an increase of 30% compared to before the update, significantly improving the model's sensitivity to strong winds. Simultaneously, the weights of the second group of coding components are optimized using the first differential power distribution data (such as differential load demand and differential line impedance). For example, differential data on grid load fluctuations increases the weight of the power distribution feature from 0.3 to 0.5, strengthening the influence of grid constraints.

[0110] Secondly, after the aforementioned weight update, the latest meteorological propagation relationships (such as changes in the radius of influence of wind speed) and power grid distribution characteristics (such as dynamic adjustment of line impedance) are integrated into the node and edge attributes of the graph to form an updated graph, which improves the correlation between meteorological factors and power output.

[0111] Subsequently, the updated spectrum is transferred to the second prediction region to provide dynamic data support for subsequent power trend analysis; at the same time, the updated spectrum and the new energy spectrum are iteratively stored in the first update region, and the sliding window algorithm is used to retain the spectrum versions of the most recent 100 updates, so as to realize the fusion storage of historical features and real-time status.

[0112] By using differential data-driven feature encoding weight dynamic updates and a dual-zone map iteration mechanism, the power prediction model can adapt to changes in meteorological and power grid conditions in real time, thereby improving the dynamic response capability of the prediction system.

[0113] Simultaneously, the power predictor is triggered to perform second-order decision processing. This requires measuring the time-varying update difference between the updated map and the new energy map, i.e., the migration amount between meteorological and power distribution states, to determine the upper-level output data of the power distribution state of each new energy power station before the update. Then, using the time-varying update difference as the state independent variable, knowledge deduction is performed to determine the power prediction data containing the power data of each power station. The specific steps are shown in S46 to S48:

[0114] S46. For the updated map and the new energy map, measure the time-varying update difference, which is the migration of meteorological state and power distribution state.

[0115] S47. Determine the upper-level output data of the new energy power stations, wherein the upper-level output data is the power distribution status of each new energy power station before the update;

[0116] S48. Using the time-varying update difference as the state independent variable, perform knowledge deduction to determine the power prediction data, where the power prediction data includes the power data of each new energy power station.

[0117] First, for the updated map and the new energy map, the time-varying update difference is measured using the Euclidean distance algorithm. This difference is quantified as the migration of meteorological conditions (such as wind speed and solar intensity) and power distribution conditions (such as load distribution and line impedance). Taking a wind farm as an example, when the measured wind speed suddenly increases from 8 m / s to 12 m / s, the meteorological condition migration is calculated to be 4.2 units, and when the grid load increases from 50 MW to 75 MW, the power distribution condition migration is 25 MW. After normalization, the comprehensive update difference is 0.68.

[0118] Secondly, determine the upper-level output data of the new energy power plants, that is, the power distribution status of each plant before the update. Collect historical power data for one hour through the SCADA system. For example, the output of a photovoltaic power plant before the update is 150MW, accounting for 75% of its installed capacity. The data of each plant constitutes an initial power distribution vector [150, 200, 180]MW, corresponding to three plants, which serves as the benchmark value for subsequent decisions.

[0119] Finally, using the time-varying update difference as the state variable, Bayesian inference and a Markov state transition model are employed for knowledge derivation. The model construction process is the same as step A220, and will not be elaborated further here due to space limitations in the manual. A state transition probability matrix is ​​established. For example, when the meteorological state transition exceeds 0.5, the probability of a wind farm's power increase is 0.72. Combined with the upper-level power output data, power prediction data is obtained through matrix operations.

[0120] By quantifying the time-varying update difference and combining historical power output data with Markov state inference, the system achieves accurate prediction of power trends under dynamic changes in weather and power grid conditions, reduces prediction errors in sudden change scenarios, and improves the real-time adaptability of new energy power prediction.

[0121] Finally, discrete control of new energy power plants is implemented. Using the data interfaces of the new energy power plants as receiving ends, power prediction data is discretely transmitted and distributed. Based on the installed capacity of each power plant, the received data is used for equipment output allocation and drive control, as detailed below:

[0122] Using the data interface of the new energy power station as the receiving end, the power prediction data is discretely transmitted.

[0123] Based on the installed capacity of each new energy power station, the received power prediction data is used to allocate equipment output and control the drive.

[0124] Discrete transmission is a process in which the power prediction data is divided into independent data frames and pushed for transmission, with the data interface of the new energy power station as the receiving end.

[0125] First, using the data interface of the renewable energy power plants as the receiving end, standardized communication protocols (such as Modbus TCP / IP) are employed to discretely transmit power prediction data. By deploying edge computing nodes at each power plant, the power prediction data generated by the central control system (e.g., a total power prediction of 500MW for a certain area in the next hour) is divided into independent data frames and pushed to the data interface of each power plant at a 50ms cycle. The transmission delay is controlled within 80ms, and the data integrity reaches 99.9%.

[0126] Secondly, based on the installed capacity of each new energy power station, such as new energy power station A with an installed capacity of 150MW and new energy power station B with an installed capacity of 250MW, the equipment output allocation coefficient is calculated proportionally. Specifically, when the total installed capacity is 400MW, the allocation coefficient of power station A is 150 / 400 = 0.375, corresponding to 37.5% of the predicted power allocation data, i.e., 187.5MW. This is then transmitted to inverters, converters, and other equipment through equipment drive protocols (such as IEC61850) to achieve output control of each piece of equipment.

[0127] By using a discrete data transmission and installed capacity allocation mechanism, precise control of the power output of new energy power plant equipment is achieved, reducing power allocation errors in traditional schemes and improving the overall operating efficiency and grid absorption capacity of new energy power generation systems.

[0128] This invention determines a new energy map by spatially reconstructing elements such as new energy power plants, meteorological stations, and power grid topology deployed in a target area. Based on time-varying reset and power trend decision-making under time-varying state migration of the new energy map, a power predictor is built and embedded in the central control of the target area. A differential processing plugin is introduced to establish communication interaction under directional data loop, with meteorological and distribution data assisting in transmission. This triggers the power predictor to perform first-order updates and second-order decision processing, determining the power prediction data and performing discrete decentralized control based on the installed capacity of each new energy power plant. This accurately predicts new energy power and enables refined control of power plant equipment output, significantly improving the accuracy of new energy power prediction and the grid's absorption capacity. It achieves real-time fusion and dynamic updating of multi-source data, providing comprehensive and accurate data support for new energy power prediction and enhancing prediction accuracy and control effectiveness.

[0129] This invention also provides a self-updating renewable energy power prediction system. Applying the above-described self-updating renewable energy power prediction method, the system structure is as follows: Figure 2 As shown, it includes:

[0130] The new energy map acquisition module is used to reconstruct the spatial elements of new energy power plants, meteorological stations, and power grid topology deployed in the target area to determine the new energy map.

[0131] The power predictor building module is used to deploy the first update zone with time-varying reset based on the new energy map, deploy the second prediction zone with power trend decision under time-varying state migration, build the power predictor and embed it in the central control of the target area;

[0132] The communication interaction construction module is used to establish communication interaction between new energy power plants, meteorological stations, power grid topology and power predictors under directional data loop by introducing differential processing plug-in;

[0133] The discrete decentralization control execution module is used to transmit meteorological data and power distribution data as variables, assisting the differential processing plug-in in communication interaction, triggering the power predictor to perform first-order update and second-order decision processing, determining the power prediction data, and performing discrete decentralization control of new energy power plants.

[0134] The new energy map acquisition module is used to perform the following steps:

[0135] An initialization space is constructed using new energy power plants and meteorological stations as nodes and propagation relationships and the power grid as edges. The propagation relationships are determined by introducing meteorological dynamic operators. A set of codes is determined by analogy with each meteorological dynamic operator, and two sets of codes are determined by the power grid distribution characteristics. The first set of codes and the second set of codes are combined as feature codes, where the feature codes have time-varying attention weights. Based on the initialization space and feature codes, a new energy map is constructed.

[0136] The power predictor building block is used to perform the following steps:

[0137] Based on the renewable energy map, an inner loop training method based on feature coding is used to determine the first update region. The inner loop consists of a data loop composed of a weather station, a power grid topology, and the first update region. An outer loop training method based on the Markov state change based on the update difference is used to determine the second prediction region. The outer loop consists of a data loop composed of the first update region, the second prediction region, and renewable energy power plants. The first update region and the second prediction region are cascaded to form a power predictor.

[0138] The communication interaction building module is used to perform the following steps:

[0139] For new energy power plants, meteorological stations and power grid topology, data interfaces are opened on the output side of each source data; power predictors are deployed in the central control of the target area, and multi-source data interfaces are opened in the power predictors to establish communication interaction between the data interfaces and the multi-source data interfaces.

[0140] The data interfaces on the output side of each source data are embedded with differential processing plugins, in which the differential relaxation coefficient is determined by balancing data value and data privacy.

[0141] The discrete decentralization control execution module is used to perform the following steps:

[0142] First meteorological data based on the weather station is determined, and the data is converted through a differential processing module to determine first differential meteorological data; power distribution demand based on the power grid topology is determined, and the data is converted through a differential processing module to determine first differential power distribution data; the first differential meteorological data and the first differential power distribution data are transmitted to a power predictor deployed in the target area.

[0143] Based on the first update region, attention weights are updated for one group of coding parts using the first differential meteorological data, and attention weights are updated for two groups of coding parts using the first differential power distribution data to determine the updated map; the updated map is then transferred to the second prediction region, and iterative storage of the updated map and the new energy map is performed in the first update region.

[0144] For the updated map and the new energy map, the time-varying update difference is measured, where the time-varying update difference is the migration amount of meteorological state and power distribution state; the upper-level output data of the new energy power stations is determined, where the upper-level output data is the power distribution state of each new energy power station before the update; and the power prediction data is determined by knowledge deduction using the time-varying update difference as the state independent variable, where the power prediction data includes the power data of each new energy power station.

[0145] Using the data interface of the new energy power station as the receiving end, the power prediction data is distributed and transmitted discretely; based on the installed capacity of each new energy power station, the received power prediction data is used for equipment output allocation and drive control.

[0146] The self-updating new energy power prediction system provided in this embodiment of the invention can execute the self-updating new energy power prediction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0147] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0148] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

[0149] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A self-updating method for predicting renewable energy power, characterized in that, include: S1. For the new energy power stations, meteorological stations, and power grid topology deployed in the target area, perform spatial reconstruction of elements to determine the new energy map; S2. Deploy the first update region based on the time-varying reset of the new energy map, deploy the second prediction region based on the power trend decision under the time-varying state migration, build a power predictor and embed it in the central control of the target region. S3. By introducing a differential processing plugin, establish communication interaction between the new energy power station, meteorological station, power grid topology and the power predictor under directional data loop; S4. Using meteorological data and power distribution data as variable data, the differential processing plug-in is assisted in communication and transmission, triggering the power predictor to perform first-order update and second-order decision processing, determining the power prediction data, and performing discrete decentralization management of new energy power plants.

2. The self-updating new energy power prediction method according to claim 1, characterized in that, In step S1, the element space is reconstructed to determine the new energy map, as detailed below: S11. Using new energy power stations and meteorological stations as nodes and propagation relationships and power grids as edges, an initialization space is constructed. The propagation relationships are determined by introducing meteorological dynamics operators. S12. By analogy with various meteorological dynamic operators, a set of codes is determined, and two sets of codes are determined based on the power grid distribution characteristics. The first set of codes and the two sets of codes are combined as feature codes, and the feature codes have time-varying attention weights. S13. Construct a new energy map based on the initialization space and the feature encoding.

3. The self-updating new energy power prediction method according to claim 1, characterized in that, In step S2, the power predictor is built as follows: S21. Based on the new energy map, perform inner loop training using time-varying reset based on feature coding to determine the first update zone. The inner loop is a data loop consisting of the meteorological station, the power grid topology, and the first update zone. S22. Use the outer loop training based on the Markov state change of the update difference to determine the second prediction region. The outer loop is a data loop consisting of the first update region, the second prediction region and the new energy power station. S23. The first update region and the second prediction region are cascaded to form the power predictor.

4. The self-updating new energy power prediction method according to claim 1, characterized in that, In step S3, a directional data loop communication interaction is established between the new energy power station, meteorological station, power grid topology, and the power predictor, as follows: S31. For the aforementioned new energy power stations, meteorological stations and power grid topology, data interfaces are opened on the output side of each source data; S32. The power predictor is deployed in the target area for centralized control. A multi-source data interface is opened in the power predictor, and communication interaction is established between the data interface and the multi-source data interface.

5. The self-updating new energy power prediction method according to claim 4, characterized in that, The data interfaces on the output side of each source data are embedded with differential processing plugins, which determine the differential relaxation coefficient by balancing data value and data privacy.

6. The self-updating new energy power prediction method according to claim 1, characterized in that, In step S4, the differential processing plugin is assisted in communication and data transmission, as follows: S41. Determine the first meteorological data based on the weather station, and perform data transformation through the differential processing plugin to determine the first differential meteorological data; S42. Determine the power distribution demand based on the power grid topology, and perform data conversion through the differential processing plug-in to determine the first differential power distribution data; S43. The first differential meteorological data and the first differential power data are transmitted to the power predictor deployed in the central control of the target area.

7. The self-updating new energy power prediction method according to claim 6, characterized in that, In step S4, the power predictor is triggered to perform a first-order update, as follows: S44. Based on the first update region, the attention weight of one group of coding parts is updated using the first differential meteorological data, and the attention weight of two groups of coding parts is updated using the first differential power distribution data to determine the updated map. S45. Transfer the updated map to the second prediction region, and perform iterative storage of the updated map and the new energy map in the first update region.

8. The self-updating new energy power prediction method according to claim 7, characterized in that, In step S4, the power predictor is triggered to perform second-order decision processing, as follows: S46. For the updated map and the new energy map, measure the time-varying update difference, where the time-varying update difference is the migration amount between meteorological state and power distribution state; S47. Determine the upper-level output data of the new energy power stations, wherein the upper-level output data is the power distribution status of each new energy power station before the update; S48. Using the time-varying update difference as a state independent variable, perform knowledge deduction to determine the power prediction data, wherein the power prediction data includes the power data of each new energy power station.

9. The self-updating new energy power prediction method according to claim 8, characterized in that, In step S4, the decentralized management and control of new energy power stations is carried out, as detailed below: The power prediction data is discretely transmitted by using the data interface of the new energy power station as the receiving end; Based on the installed capacity of each new energy power station, the received power prediction data is used to allocate equipment output and control the drive.

10. A self-updating renewable energy power prediction system, employing the self-updating renewable energy power prediction method according to any one of claims 1 to 9, characterized in that, include: The new energy map acquisition module is used to reconstruct the spatial elements of new energy power plants, meteorological stations, and power grid topology deployed in the target area to determine the new energy map; The power predictor building module is used to deploy the first update zone with time-varying reset based on the new energy map, deploy the second prediction zone with power trend decision under time-varying state migration, build the power predictor and embed it in the central control of the target area. The communication interaction construction module is used to establish communication interaction between the new energy power station, meteorological station, power grid topology and the power predictor under directional data loop by introducing a differential processing plugin; The discrete decentralization control execution module is used to use meteorological data and power distribution data as variable data, assist the differential processing plug-in in communication and transmission, trigger the power predictor to perform first-order update and second-order decision processing, determine the power prediction data, and perform discrete decentralization control of new energy power plants.

Images