CN-121543832-B - Zero-carbon intelligent park cloud platform data optimization processing method and system

Abstract

The invention provides a zero-carbon intelligent park cloud platform data optimization processing method and system, and relates to the technical field of data processing, wherein the method comprises the following steps: step 1, constructing a multidimensional state space according to four high-dimensional state vectors, and respectively calculating covariance matrixes of the state vectors and the rest three state vectors by taking the state vectors of the photovoltaic grid-connected points in the multidimensional state space as references to obtain three independent characteristic subspaces, wherein the four high-dimensional state vectors comprise the photovoltaic grid-connected points, an energy storage access point, a main load point and a power grid interaction point. According to the invention, by constructing a multidimensional state space and a multi-level virtual energy guide graph and combining with the collaborative reconstruction of a power time sequence section, the time sequence prediction of an integrated attention mechanism and the joint iteration update, the net load power prediction precision is improved, the rationality of an energy storage charge and discharge instruction is optimized, the dynamic adaptation of the running state of an energy system of a zero-carbon intelligent park is ensured, and the accurate processing and the high-efficiency low-carbon scheduling of energy data are realized.

Inventors

• LI FEI

Assignees

• 远盈智慧能源有限公司

Dates

Publication Date

20260508

Application Date

20260115

Claims (9)

1. 1. The method for optimizing and processing the data of the cloud platform of the zero-carbon intelligent park is characterized by comprising the following steps of: Step 1, constructing a multi-dimensional state space according to four high-dimensional state vectors, and respectively calculating covariance matrixes of the three state vectors and the rest by taking the state vectors of the photovoltaic grid-connected points in the multi-dimensional state space as references to obtain three independent characteristic subspaces, wherein the four high-dimensional state vectors comprise the photovoltaic grid-connected points, an energy storage access point, a main load point and a power grid interaction point; Step 2, calculating generalized included angles between every two of three feature subspaces, dividing the multidimensional state space into a plurality of decision subzones according to the generalized included angles, setting projections of state vectors of three representative points of an energy storage access point, a main load point and a power grid interaction point in each decision subzone in the decision subzone, and connecting the representative points of each decision subzone with a reference state vector point to construct a multi-level virtual energy guide graph; Step 3, calculating a first path conduction coefficient of a link connecting the reference state vector point and the energy storage access point representative point, a second path conduction coefficient of a link connecting the reference state vector point and the main load point representative point and a third path conduction coefficient of a link connecting the reference state vector point and the power grid interaction point representative point based on the geometric relationship of each link in the multi-level virtual energy guide graph; continuously monitoring the values of the first path conductivity, the second path conductivity and the third path conductivity; when the instantaneous variation amplitude of the conduction coefficient of any path exceeds a preset abrupt change threshold, judging that abnormal abrupt change of the conduction coefficient occurs, and recording the occurrence time of the abnormal abrupt change; matching the occurrence time of the abnormal mutation with a time mark recorded in a switching event log, and if the matching is successful, carrying out data correction on a topological node corresponding to the path conduction coefficient with the abnormal mutation according to the value of the path conduction coefficient with the abnormal mutation in the virtual energy guide graph at the occurrence time, and collecting to obtain reconstructed power data after the correction is completed; Step 4, training a time sequence prediction model of an integrated attention mechanism by using the reconstructed power data and the meteorological time sequence section to obtain a future payload power prediction interval, and performing coverage verification on the prediction interval and the real-time power time sequence section to obtain an adjusted prediction interval; And step 5, determining a charging and discharging instruction of the energy storage unit based on the adjusted prediction interval, and performing feedback according to the instruction to perform joint iterative updating on the collaborative reconstruction process and the time sequence prediction model.
2. 2. The method for optimizing data of a zero-carbon intelligent park cloud platform according to claim 1, wherein the determining process of the four high-dimensional state vectors is as follows: Collecting active power time sequence sections of four topological nodes of a photovoltaic grid connection point, an energy storage access point, a main load point and a power grid interaction point through a cloud platform, collecting irradiance time sequence sections and temperature time sequence sections provided by meteorological monitoring equipment, and forming a meteorological time sequence section together, and collecting state conversion records generated by a static reactive generator as switching event logs; The method comprises the steps of carrying out time sequence alignment on four power time sequence sections and weather time sequence sections by adopting sliding time windows with the same time length, respectively calculating the mean value, variance and frequency domain energy characteristics of each power time sequence section as statistical characteristics and frequency spectrum characteristics of each power time sequence section in each aligned time window, extracting irradiance mean value and temperature mean value of weather time sequence sections in corresponding windows as weather characteristics, splicing and normalizing the power characteristics and weather characteristics of each topological node in each time window to form characteristic vectors of the corresponding topological node in the corresponding time window, and arranging the characteristic vectors of each topological node in all time windows in time sequence to form a high-dimensional state vector representing the running state of the corresponding topological node to obtain four high-dimensional state vectors.
3. 3. The method for optimizing data of the zero-carbon intelligent park cloud platform according to claim 2, wherein the step1 comprises the following steps: In the multidimensional state space, the high-dimensional state vector corresponding to the photovoltaic grid-connected point is used as a reference state vector, and covariance of the reference state vector and the high-dimensional state vector corresponding to the energy storage access point in the time dimension is calculated respectively to form a first covariance matrix; calculating covariance of the high-dimensional state vector corresponding to the reference state vector and the main load point in the time dimension to form a second covariance matrix; Calculating covariance of the reference state vector and the high-dimensional state vector corresponding to the power grid interaction point in the time dimension to form a third covariance matrix; Performing matrix decomposition on the first covariance matrix, extracting feature vectors corresponding to the k maximum feature values before presetting to form a first independent feature subspace representing the dynamic association relation between the photovoltaic grid-connected point and the energy storage access point, performing matrix decomposition on the second covariance matrix, extracting feature vectors corresponding to the L maximum feature values before presetting to form a second independent feature subspace representing the dynamic association relation between the photovoltaic grid-connected point and the main load point, performing matrix decomposition on the third covariance matrix, extracting feature vectors corresponding to the J maximum feature values before presetting to form a third independent feature subspace representing the dynamic association relation between the photovoltaic grid-connected point and the power grid interaction point, wherein k, L and J are minimum positive integers enabling the cumulative contribution rate of the feature values to be more than or equal to 90%.
4. 4. The method for optimizing data of the zero-carbon intelligent park cloud platform according to claim 3, wherein the step 2 comprises the following steps: Calculating the minimum included angle between the main component vector sets of the first independent feature subspace and the second independent feature subspace as a first generalized included angle, and calculating the minimum included angle between the main component vector sets of the first independent feature subspace and the third independent feature subspace as a second generalized included angle; The method comprises the steps of comparing a first generalized included angle, a second generalized included angle and a third generalized included angle with preset angle thresholds respectively, dividing a region which satisfies the condition that the first generalized included angle is larger than the angle threshold and the second generalized included angle and the third generalized included angle are smaller than the angle threshold in a multi-dimensional state space into a first type decision sub-region; For each divided decision sub-domain, orthogonally projecting high-dimensional state vectors of an energy storage access point, a main load point and a power grid interaction point to a main plane of the corresponding decision sub-domain to obtain three projection coordinate points, and setting the three projection coordinate points as three representative points in the corresponding decision sub-domain; In each decision sub-domain, taking a reference state vector point of the corresponding decision sub-domain as a starting point, and respectively connecting to three representative points of the corresponding decision sub-domain to form a local virtual energy conduction link in the corresponding decision sub-domain; the local virtual energy conduction links of all the decision subzones are combined to form a multi-level virtual energy guide graph reflecting the cross-region conduction relationship of energy in the multidimensional state space.
5. 5. The method for optimizing data of the zero-carbon intelligent park cloud platform according to claim 4, wherein the step 4 comprises: Organizing the reconstructed power data into a power training sequence according to time sequence, organizing the synchronous meteorological time sequence section into a meteorological training sequence, and training a time sequence prediction model integrating an attention mechanism by using the power training sequence and the meteorological training sequence to obtain a trained time sequence prediction model; the time sequence prediction model after training outputs net load power predicted values of future continuous time points and probability distribution parameters corresponding to each predicted value according to the input power history sequence and weather history sequence; calculating a confidence upper bound and a confidence lower bound of each future time point predicted value based on the probability distribution parameters, and forming a future payload power predicted interval by the confidence upper bound and the confidence lower bound; collecting real-time power time sequence sections actually generated in a period covered by a future payload power prediction interval, counting the number of data points in the real-time power time sequence sections falling into a corresponding time prediction interval, and calculating to obtain an actual coverage proportion; And comparing the actual coverage ratio with a preset expected coverage ratio, if the actual coverage ratio is continuously lower than the expected coverage ratio, carrying out back propagation correction on probability distribution parameters output by the time sequence prediction model according to historical prediction error data in a set time period, outputting new probability distribution parameters, and generating a new confidence upper bound and a confidence lower bound to form an adjusted prediction interval.
6. 6. The method for optimizing data on a cloud platform of a zero-carbon intelligent park according to claim 5, wherein the training of the time sequence prediction model to output the net load power predicted value and the probability distribution parameter corresponding to each predicted value at the future continuous time point according to the input power history sequence and the weather history sequence comprises: the time sequence prediction model after training carries out time sequence feature coding on the input power history sequence and the weather history sequence respectively to obtain a power feature sequence and a weather feature sequence, and self-attention weights of time step feature vectors in the power feature sequence and cross-attention weights between the power feature sequence and the weather feature sequence are calculated by using an attention mechanism in the time sequence prediction model after training; The power characteristic sequence and the weather characteristic sequence are subjected to weighted fusion based on the self-attention weight and the cross-attention weight to form fusion characteristics; And the trained time sequence prediction model generates the net load power predicted value of each continuous time point in the future and probability distribution parameters representing the uncertainty of each predicted value in parallel according to the fusion characteristics through an output layer.
7. 7. The method for optimizing data of the zero-carbon intelligent park cloud platform according to claim 6, wherein the step 5 comprises: solving by a robust optimization scheduling algorithm based on the adjusted prediction interval to obtain a charge and discharge power instruction sequence of the energy storage unit in a future scheduling period; after executing the charge and discharge power instruction sequence, acquiring actual power data of four topological nodes, calculating to obtain an actual net load power curve, and calculating the difference between the actual net load power curve and the median sequence in the adjusted prediction interval to form a power deviation sequence; Integrating the power deviation sequence, the corresponding meteorological data, the switching event log and the abnormal mutation information of the conduction coefficient to form a closed loop feedback data set, identifying a time period when the deviation degree exceeds a preset deviation threshold value based on the statistical distribution characteristics of the power deviation in the closed loop feedback data set, extracting the abnormal mutation record of the related conduction coefficient in the time period, carrying out parameter adjustment on the mutation threshold value used for judging the abnormal mutation of the conduction coefficient according to the mutation record, taking historical power data, meteorological data and the corresponding actual net load power data in the closed loop feedback data set as new training samples, inputting a time sequence prediction model to execute training so as to update the network parameters of the time sequence prediction model, and periodically executing the mutation threshold value parameter adjustment and the training of the time sequence prediction model to complete the joint iteration updating of the collaborative reconstruction process and the time sequence prediction model.
8. 8. The method for optimizing data of the cloud platform of the zero-carbon intelligent park according to claim 7, wherein the method is characterized in that based on the adjusted prediction interval, the method is used for solving through a robust optimization scheduling algorithm to obtain a charge and discharge power instruction sequence of the energy storage unit in a future scheduling period, and the method comprises the following steps: The method comprises the steps of extracting the numerical values of a confidence upper bound and a confidence lower bound of each time point in an adjusted prediction interval as a deterministic boundary constraint of the net load power at the corresponding moment, and constructing a robust optimization scheduling model by taking the sum of the total power purchase cost and the equipment loss cost of a park for minimizing a future scheduling period as an objective function; Taking rated power, charging and discharging efficiency, current state of charge and allowable range of the state of charge of the energy storage unit as operation constraint of a robust optimization scheduling model, and solving the robust optimization scheduling model to obtain a set of charging and discharging power values of the energy storage unit which meet all deterministic boundary constraint and operation constraint; and arranging the charge and discharge power value sets of the energy storage units in time sequence to generate a charge and discharge power instruction sequence of the energy storage units in a future scheduling period.
9. 9. A zero-carbon smart park cloud platform data optimization processing system implementing the method of any one of claims 1 to 8, comprising: The space construction module is used for constructing a multidimensional state space according to four high-dimensional state vectors, and respectively calculating covariance matrixes of the space construction module and the three remaining state vectors by taking the state vectors of the photovoltaic grid-connected points in the multidimensional state space as references so as to obtain three independent characteristic subspaces, wherein the four high-dimensional state vectors comprise the photovoltaic grid-connected points, the energy storage access points, the main load points and the power grid interaction points; The guide diagram construction module is used for calculating generalized included angles between every two of three characteristic subspaces, dividing the multidimensional state space into a plurality of decision subzones according to the generalized included angles, setting projections of state vectors of three representative points of an energy storage access point, a main load point and a power grid interaction point in each decision subzone in the decision subzone, and connecting the representative points of the decision subzones with reference state vector points to construct a multi-level virtual energy guide diagram; The abnormality detection module is used for analyzing path conduction coefficients of all levels in the virtual energy guide graph, and when detecting that the abnormal mutation of the conduction coefficient of any level is matched with the switching event log, the abnormality detection module performs collaborative reconstruction on the power time sequence section of the corresponding event to generate reconstructed power data; the interval verification module is used for training a time sequence prediction model of an integrated attention mechanism by using the reconstructed power data and the meteorological time sequence section to obtain a future payload power prediction interval, and performing coverage verification on the prediction interval and the real-time power time sequence section to obtain an adjusted prediction interval; and the iteration optimization module is used for determining the charge and discharge instruction of the energy storage unit based on the adjusted prediction interval, and carrying out feedback on the collaborative reconstruction process and the time sequence prediction model according to the instruction execution.

Description

Zero-carbon intelligent park cloud platform data optimization processing method and system Technical Field The invention relates to the technical field of data processing, in particular to a zero-carbon intelligent park cloud platform data optimization processing method and system. Background In the operation management of a zero-carbon intelligent park, a cloud platform is generally required to collect and analyze power data from a plurality of nodes such as photovoltaic power generation, an energy storage system, a main load, a power grid connection point and the like so as to realize net load prediction and energy storage optimization scheduling, the prior art generally carries out preprocessing on collected time sequence data through a data cleaning, filtering or simple statistical interpolation method so as to eliminate noise or abnormal values, and then trains a time sequence prediction model based on the processed historical data, however, such methods may face certain limitations when processing data mutation caused by switching operation of specific equipment in the park. For example, when a Static Var Generator (SVG) in a park is put into or cut off due to voltage regulation, a plurality of data sequences such as output of a photovoltaic inverter, power consumption of a main load, power of a power grid interaction point and the like are affected simultaneously in a short time, and when a peak abnormality occurs in power data of a certain node (such as the power grid interaction point) is detected, local smoothing or substitution is performed only on the data sequence of the node according to data of a front time window and a rear time window of the peak abnormality, which may not fully consider the SVG switching event as a system event, how to synchronously change dynamic association relations among multiple node operation states such as photovoltaic, load, energy storage and the like, and sometimes, due to neglecting changes of the multi-node coupling relation under the events, the real power flow state of the system is difficult to accurately restore only based on reconstruction processing performed by single-point data. The possible deviation between the reconstructed data and the real physical association, if directly used for subsequent model training, may affect the reliability of the prediction of the payload interval to a certain extent, thereby bringing uncertainty risk to the energy storage scheduling decision based on the prediction. Disclosure of Invention The invention aims to solve the technical problem of providing a zero-carbon intelligent park cloud platform data optimization processing method and system, and the stability of park energy operation is enhanced. In order to solve the technical problems, the technical scheme of the invention is as follows: in a first aspect, a method for optimizing data of a cloud platform of a zero-carbon intelligent park, the method comprising: Step 1, constructing a multi-dimensional state space according to four high-dimensional state vectors, and respectively calculating covariance matrixes of the three state vectors and the rest by taking the state vectors of the photovoltaic grid-connected points in the multi-dimensional state space as references to obtain three independent characteristic subspaces, wherein the four high-dimensional state vectors comprise the photovoltaic grid-connected points, an energy storage access point, a main load point and a power grid interaction point; Step 2, calculating generalized included angles between every two of three feature subspaces, dividing the multidimensional state space into a plurality of decision subzones according to the generalized included angles, setting projections of state vectors of three representative points of an energy storage access point, a main load point and a power grid interaction point in each decision subzone in the decision subzone, and connecting the representative points of each decision subzone with a reference state vector point to construct a multi-level virtual energy guide graph; Step 3, analyzing path conduction coefficients of all levels in the virtual energy guide graph, and when abnormal abrupt change of the conduction coefficient of any level is detected to be matched with a switching event log, performing cooperative reconstruction on a power time sequence section of a corresponding event to generate reconstructed power data; Step 4, training a time sequence prediction model of an integrated attention mechanism by using the reconstructed power data and the meteorological time sequence section to obtain a future payload power prediction interval, and performing coverage verification on the prediction interval and the real-time power time sequence section to obtain an adjusted prediction interval; And step 5, determining a charging and discharging instruction of the energy storage unit based on the adjusted prediction interval, and performing feedback according to the