AI-based energy data multi-source heterogeneous fusion method, system and equipment

Abstract

The invention relates to the technical field of energy data processing and artificial intelligence, and particularly discloses an AI-based energy data multi-source heterogeneous fusion method, system and equipment, and the method comprises the steps: screening a historical fusion error case, and judging a coupling reconstruction cause; dividing quantitative change / working condition reconstruction coupling reconstruction; determining a distribution offset comprehensive index critical value of offset reconstruction; pre-training a working condition reconstruction exclusive AI fusion model; the current fusion adaptation is adapted in real time and differentiated, fusion errors caused by coupling reconstruction are accurately recognized through closed-loop design of error screening, reconstruction classification, critical value determination, model pre-training and real-time adaptation, and a differentiated AI strategy is adopted for offset and working condition reconstruction, so that the accuracy of fusion is improved. The problems that in the prior art, dynamic changes of the coupling relation cannot be coped with, the fusion precision is low, and the stability is poor are solved, the precision and robustness of energy data multi-source heterogeneous fusion are remarkably improved, and the method adapts to complex and changeable energy system working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy data processing and artificial intelligence, in particular to an AI-based energy data multi-source heterogeneous fusion method, system and device. BACKGROUND

[0002] With the large-scale grid connection of new energy power generation (photovoltaic, wind power, etc.), the energy system presents the characteristics of multi-source heterogeneity, and energy data fusion has become the core support for realizing system optimization scheduling and stable operation. The existing energy data multi-source heterogeneous fusion method mainly depends on traditional machine learning models (such as support vector machine, ordinary neural network), adopts fixed fusion logic and model parameters, and does not fully consider the dynamic changes of feature coupling relationships in the energy system.

[0003] In actual operation, the feature coupling relationship of the energy system (such as the correlation between photovoltaic output and energy storage power, load and thermal power output) is easily affected by factors such as scheduling target switching, policy adjustment, equipment modification, and data distribution offset, and is reconfigured, which leads to mismatch between the logic of the original fusion model and the current system coupling logic, and further causes the fusion error to soar. The existing technology has the following defects:

[0004] It is unable to accurately distinguish the causes of fusion error, and it is easy to misjudge the error caused by coupling reconstruction as data noise or model aging, resulting in invalid processing strategy; the coupling reconstruction types are not subdivided, and a one-size-fits-all optimization strategy is adopted, which cannot adapt to the different characteristics of quantitative change (data gradual offset driven) and qualitative change (system event driven) reconstruction; there is a lack of scientific quantitative standard (such as distribution offset critical value) to guide real-time fusion optimization, and it is difficult to early warn the coupling reconstruction risk; the model training and working condition adaptation are disconnected, and when the working condition changes, the model needs to be retrained, which has a lagging response and cannot meet the real-time requirements of the energy system.

[0005] Therefore, the present application provides an AI-based energy data multi-source heterogeneous fusion method, system and device. SUMMARY

[0006] The purpose of the present application is to provide an AI-based energy data multi-source heterogeneous fusion method, system and device to solve the above background problems.

[0007] The purpose of the present application can be achieved by the following technical solution: an AI-based energy data multi-source heterogeneous fusion method, comprising the following steps:

[0008] Collecting historical multi-source heterogeneous energy data, screening error fusion with fusion error, analyzing the feature coupling relationship in the error fusion window, determining whether the fusion error is caused by coupling relationship reconstruction, if so, marking it as coupling reconstruction error fusion, and statistically analyzing the coupling reconstruction error fusion to determine whether there is a coupling reconstruction fusion error phenomenon;

[0009] If there is, calculate the distribution offset comprehensive index by feature distribution offset analysis, and combine the coupling reconstruction change feature analysis to fuse the coupling reconstruction error into offset reconstruction fusion and working condition reconstruction fusion;

[0010] For offset reconstruction fusion, correlation analysis is performed on feature distribution offset and coupling reconstruction to determine the critical value of distribution offset comprehensive index;

[0011] For working condition reconstruction fusion, the working condition changes leading to coupling relationship reconstruction are counted, and a dedicated fusion model is pre-trained based on historical steady-state data;

[0012] For the matching analysis of feature coupling relationship and working condition in current energy data fusion, if it is not matched, the coupling reconstruction type is identified, and the difference is called by combining the comparison analysis of the critical value of the distribution offset comprehensive index.

[0013] Further, the determination method of the coupling reconstruction error fusion is:

[0014] The root mean square error and error fluctuation coefficient of the multi-source heterogeneous energy data in the historical period are collected;

[0015] If one of the root mean square error or error fluctuation coefficient meets the requirements, there is a fusion error;

[0016] Divide the steady-state section and the fusion error section, the steady-state section is the window before the error fusion window, and the error fusion section is the window of the fusion error anomaly;

[0017] For any error fusion, filter out the feature pairs with coupling relationship, and mark the coupling type, including linear, nonlinear, and causal relationship;

[0018] For the feature pairs with linear relationship, calculate the Pearson correlation coefficient change rate of the steady-state section and the fusion error section;

[0019] For the feature pairs with nonlinear relationship, calculate the mutual information value change rate of the steady-state section and the fusion error section;

[0020] For the feature pairs with causal relationship, test the causal directionality between the features in the steady-state section and the fusion error section;

[0021] If the Pearson correlation coefficient change rate, the mutual information value change rate, and the causal directionality meet the requirements, there is a coupling relationship reconstruction;

[0022] Extract the coupling quantitative index reconstruction time point and the time point of the first error anomaly:

[0023] If the coupling quantitative index reconstruction time point is earlier than or synchronous with the time point of the first error anomaly, it meets the causal timing, and the error fusion is marked as coupling reconstruction error fusion.

[0024] Further, the judgment method of whether there is a coupling reconstruction fusion error phenomenon is:

[0025] The proportion of coupling reconstruction error fusion in all error fusion is counted to obtain a coupling reconstruction error fusion proportion, and if the coupling reconstruction error fusion proportion meets the requirement, there is a coupling reconstruction fusion error phenomenon.

[0026] Further, the calculation process of the distribution offset comprehensive index is:

[0027] For any feature with a coupling relationship, the steady-state segment data and the error fusion window data are extracted, and the distribution offset comprehensive index is calculated, specifically:

[0028] The mean relative change rate = |offset after mean - offset before mean| / offset before mean, the variance change rate = offset after variance / offset before steady-state variance, and the value range compression / expansion rate = |offset before steady-state maximum - offset before steady-state minimum (offset after maximum - offset after minimum) - (offset before steady-state maximum - offset before steady-state minimum)|.

[0029] The mean relative change rate, the variance change rate, and the value range compression / expansion rate are added to obtain the distribution offset comprehensive index.

[0030] Further, the method of dividing the coupling reconstruction error fusion into offset reconstruction fusion and working condition reconstruction fusion is:

[0031] If the distribution offset comprehensive index meets the requirement, there is a quantitative change premise of data gradual offset.

[0032] The coupling reconstruction time point and the distribution offset standard time point are compared, and if the distribution offset standard time point is in front and the coupling reconstruction time point is in back, it meets the offset reconstruction characteristic, otherwise, it meets the working condition reconstruction characteristic.

[0033] It is checked whether there is a system event before the coupling reconstruction time point, and if there is, the coincidence degree of the event occurrence time and the reconstruction time is calculated: coincidence degree = 1 - |event time - reconstruction time| / 1 hour.

[0034] If there is no related system event, it meets the offset reconstruction characteristic, and if there is a related system event and the coincidence degree meets the requirement, it meets the working condition reconstruction characteristic.

[0035] If the time and event offset reconstruction characteristics are met at the same time, it is offset reconstruction fusion, and if the time and event working condition reconstruction characteristics are met at the same time, it is working condition reconstruction fusion.

[0036] Further, the calculation method of the offset comprehensive index critical value is:

[0037] The coupling reconstruction index of the offset reconstruction fusion is calculated:

[0038] Correlation coefficient change rate = |Reconstructed correlation coefficient - Pre-reconstruction steady-state correlation coefficient| / Pre-reconstruction steady-state correlation coefficient;

[0039] Logical direction reversal coefficient: coupling logic direction unchanged = 0, coupling logic direction reversed = 1;

[0040] Add the correlation coefficient change rate and the logical direction reversal coefficient to obtain the coupling reconstruction index;

[0041] Group the distribution offset comprehensive index 5% as an interval, covering the distribution offset comprehensive index range of all offset reconstruction fusion;

[0042] For each group of data, count the number of offset reconstruction cases in the index interval to obtain the number of samples in the group, count the number of cases triggering coupling reconstruction in the group to obtain the reconstruction occurrence number, and calculate the reconstruction occurrence probability by proportion between the reconstruction occurrence number and the number of samples in the group;

[0043] From the first group, calculate the deviation of the reconstruction occurrence probability of adjacent two groups in turn, and compare the reconstruction occurrence probability with the average value of the reconstruction occurrence probability of all previous groups after each calculation. If the reconstruction occurrence probability is greater than or equal to the average value of the reconstruction occurrence probability of all previous groups, it is a mutation interval, and the lower limit of the mutation interval is the distribution offset comprehensive index critical value.

[0044] Further, the process of pre-training the exclusive fusion model based on historical steady-state data is:

[0045] Obtain all historical working condition reconstruction fusion, and statistically analyze three dimensions of event type-coupling reconstruction logic-impact feature to form a working condition change-reconstruction logic mapping table;

[0046] Obtain the historical steady-state data corresponding to the working condition and the coupling reconstruction logic rule, and select the historical data after the working condition changes, the coupling reconstruction is stable, and there is no fusion error;

[0047] Select the domain self-adaptation + causal inference hybrid architecture, train the model for any type of working condition, and obtain the exclusive fusion model.

[0048] Further, the way to identify the coupling reconstruction type is:

[0049] For the current multi-source heterogeneous energy real-time data, calculate the coupling quantitative indicators: linear coupling (Pearson correlation coefficient), nonlinear coupling (mutual information), and causal coupling (Granger direction);

[0050] Numerical matching degree = 1 - |Current coupling index - Standard coupling index| / Standard coupling index;

[0051] Logical matching degree = 1 (current coupling direction / link is consistent with the standard), 0 (direction / link is reversed);

[0052] Comprehensive matching degree = numerical matching degree + logical matching degree, if the comprehensive matching degree does not meet the requirements, it is not matched;

[0053] Check the current system event log for whether there is a working condition change-reconfigure the working condition change event in the logical mapping table;

[0054] If there is a corresponding event, and the coincidence degree of the event occurrence time and the mismatching time meets the requirements, it is determined that the working condition reconfiguration is not matched;

[0055] If there is no corresponding event, calculate the distribution offset comprehensive index of the current feature, if the distribution offset comprehensive index is greater than or equal to the distribution offset comprehensive index threshold, it is an offset reconfiguration mismatch.

[0056] An AI-based multi-source heterogeneous fusion system for energy data, characterized by comprising the following modules:

[0057] Coupling reconfiguration determination module: collect historical multi-source heterogeneous energy data, screen error fusion with fusion error, analyze the coupling relationship of features in the error fusion window, determine whether the fusion error is caused by coupling relationship reconfiguration, if so, mark it as coupling reconfiguration error fusion, and perform statistical analysis on the coupling reconfiguration error fusion to determine whether there is a coupling reconfiguration fusion error phenomenon;

[0058] Reconfiguration fusion type division module: if there is, calculate the distribution offset comprehensive index through feature distribution offset analysis, and divide the coupling reconfiguration error fusion into offset reconfiguration fusion and working condition reconfiguration fusion combined with coupling reconfiguration change feature analysis;

[0059] Offset threshold calculation module: for offset reconfiguration fusion, associate analysis of feature distribution offset and coupling reconfiguration to determine the distribution offset comprehensive index threshold;

[0060] Special fusion model construction module: for working condition reconfiguration fusion, statistics of working condition changes leading to coupling relationship reconfiguration, and pre-training of special fusion model based on historical steady-state data;

[0061] Real-time matching and differential calling module: match and analyze the feature coupling relationship and working condition in the current energy data fusion, if not matched, identify the coupling reconfiguration type, and perform differential calling combined with comparison analysis of the distribution offset comprehensive index threshold.

[0062] An electronic device, characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus;

[0063] a memory for storing a computer program;

[0064] a processor for executing the program stored on the memory to implement the method steps of any one of claims 1-8.

[0065] The beneficial effects of the present application are as follows:

[0066] Accurate identification of coupling reconstruction error: through the triple mechanism of error screening + coupling reconstruction judgment + timing verification, effectively distinguish the error caused by coupling reconstruction and other causes (such as noise, model aging), avoid misjudgment, and improve the identification accuracy;

[0067] Subdivision of reconstruction type and accurate adaptation: for the first time, coupling reconstruction is divided into quantitative change / work condition reconstruction, and critical value early warning + weight optimization, exclusive model pre-training + rapid calling strategy are used accordingly, solving the drawbacks of traditional one-size-fits-all methods and improving the fusion accuracy;

[0068] Based on historical data and physical mechanism, the distribution deviation comprehensive index critical value is determined to realize early warning of coupling reconstruction risk, inhibit the positive feedback of deviation-reconstruction-more serious deviation, and improve the fusion stability;

[0069] Real-time response to working condition changes: pre-train exclusive AI model library, and quickly call the adaptive model when the working condition changes, without the need for retraining, shorten the response time, and meet the real-time requirements of energy systems;

[0070] Adopting domain self-adaptation + causal inference AI architecture, adapting to the characteristics of multi-source heterogeneous data, ensuring that the fusion results meet the physical rules of energy systems, and being applicable to various energy scenarios such as photovoltaic, wind power, and energy storage. BRIEF DESCRIPTION OF DRAWINGS

[0071] The present application will be further described below with reference to the accompanying drawings.

[0072] Figure 1 is a flowchart of the AI-based multi-source heterogeneous fusion method for energy data according to embodiment 1 of the present application;

[0073] Figure 2 is a logic judgment diagram for judging whether there is a coupling reconstruction fusion error phenomenon in embodiment 1 of the present application;

[0074] Figure 3 is a functional module diagram of the AI-based multi-source heterogeneous fusion system for energy data in embodiment 2 of the present application. DETAILED DESCRIPTION

[0075] In order to make the technical means, creative features, purposes and effects achieved by the present application easy to understand, the present application will be further described below with reference to the specific embodiments.

[0076] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, the AI-based multi-source heterogeneous fusion method for energy data specifically includes the following steps:

[0077] Step 1: Collect historical multi-source heterogeneous energy data, screen for error fusion with fusion errors, and determine whether the fusion error is caused by the reconstruction of coupling relationship by analyzing the characteristic coupling relationship within the error fusion window. If so, mark it as coupling reconstruction error fusion, and perform statistical analysis on coupling reconstruction error fusion to determine whether there is a coupling reconstruction fusion error phenomenon.

[0078] Please see Figure 2 As shown, in step one, the error fusion screening process includes:

[0079] Collect multi-source heterogeneous energy data within historical periods, and calculate the fusion error, including root mean square error and mean absolute error, for each fusion window in the historical data.

[0080] The error fluctuation coefficient is obtained by proportionally calculating the standard deviation of the error to the average error.

[0081] If the root mean square error is greater than the historical steady-state error, or the error fluctuation coefficient is greater than the preset fluctuation coefficient, then there is a fusion error.

[0082] It is understandable that the preset fluctuation coefficient is a critical value for determining whether the error is unstable. If it is exceeded, it indicates that the error fluctuates violently and there may be coupling reconstruction. It is set based on historical data statistics.

[0083] In step one, the process of determining whether the fusion error is caused by the reconstruction of coupling relationships includes:

[0084] The first point to clarify is that determining whether a coupling relationship exists during reconstruction involves:

[0085] Divide the data into a steady-state segment and a fusion error segment: the steady-state segment is the first 72 hours before the error fusion window (no error, steady-state data, fluctuation range <5%), and the error fusion segment is the window of abnormal fusion error.

[0086] For any error fusion:

[0087] Feature pairs with coupling relationships are selected and the coupling type is labeled, including linear, nonlinear, and causal relationships.

[0088] Calculate the coupling quantization index for each feature pair in the steady-state segment and the fusion error segment, specifically including:

[0089] For the characteristic pair of linear coupling relationship, the linear coupling index is Pearson correlation coefficient, the Pearson correlation coefficient change rate of the characteristic pair of the steady state section and the fusion error section is calculated, and if the Pearson correlation coefficient change rate is greater than or equal to the preset change rate or the correlation direction is reversed, it indicates that the linear coupling relationship is reconstructed;

[0090] Wherein, the preset change rate is a critical value for determining whether the linear coupling is reconstructed, reflecting the significant degree of correlation coefficient change, and is set based on engineering experience;

[0091] For the characteristic pair of nonlinear coupling relationship, the nonlinear coupling index is mutual information, and the mutual information value change rate of the characteristic pair of the steady state section and the fusion error section is calculated, and if the mutual information value change rate exceeds 1 times, it indicates that the nonlinear coupling relationship is reconstructed;

[0092] Wherein, the mutual information value change rate 1 times threshold is a critical value for determining whether the nonlinear coupling is reconstructed, and the doubling of nonlinear correlation strength indicates reconstruction, which is a fixed threshold (industry consensus);

[0093] For the characteristic pair of causal coupling relationship, the causal coupling index is Granger causality test, and the causal directionality between characteristics is tested, and if the causal direction is reversed in the steady state section and the fusion error section, it indicates that the causal coupling is reconstructed;

[0094] Secondly, it is necessary to explain that the fusion error and the coupling relationship reconstruction are verified for time sequence consistency, specifically:

[0095] Extract the coupling quantization index reconstruction time point and the error rising time point:

[0096] The coupling quantization index reconstruction time point is the time point when the coupling relationship of the characteristic pair is first reconstructed, and the error rising time point is the time when the root mean square error first exceeds the abnormal threshold;

[0097] If the coupling quantization index reconstruction time point is earlier than or synchronous with the error rising time point, the causal time sequence is met, and the error fusion is marked as coupling reconstruction error fusion;

[0098] In step one, the process of judging whether there is coupling reconstruction error fusion phenomenon includes:

[0099] The proportion of coupling reconstruction error fusion in all error fusion is counted to obtain the coupling reconstruction error fusion proportion, and compared with the preset proportion, and if the coupling reconstruction error fusion proportion is greater than or equal to the preset proportion, there is coupling reconstruction error fusion phenomenon;

[0100] It can be understood that the preset proportion is a critical value for determining whether there is a systematic coupling reconstruction error phenomenon, avoiding interference of incidental cases, and is set based on engineering requirements;

[0101] It should be noted that the role of error fusion screening and coupling reconstruction inducement determination is to distinguish between incidental cases and systemic phenomena by statistical proportion, and to provide decision basis for subsequent classification processing;

[0102] Step two: if exists, calculate the distribution offset comprehensive index by feature distribution offset analysis, and combine the coupling reconstruction change feature analysis to divide the coupling reconstruction error fusion into offset reconstruction fusion and working condition reconstruction fusion;

[0103] In step two, the process of dividing the coupling reconstruction fusion into offset reconstruction fusion and working condition reconstruction fusion includes:

[0104] First, the feature is analyzed by feature distribution offset analysis to determine whether there is a quantitative change premise of data progressive offset, specifically:

[0105] For any coupling reconstruction error fusion:

[0106] For any feature with coupling relationship, extract the steady-state segment data and error fusion window data respectively, and calculate the distribution offset comprehensive index, specifically:

[0107] The mean relative change rate = |post-offset mean-pre-offset mean| / pre-offset mean, reflecting the change amplitude of feature center trend;

[0108] The variance change rate = post-offset variance / pre-offset steady-state variance, reflecting the change amplitude of feature dispersion degree;

[0109] The value range compression / expansion rate = |pre-offset steady-state maximum-pre-offset steady-state minimum(post-offset maximum-post-offset minimum)-(pre-offset steady-state maximum-pre-offset steady-state minimum)|, reflecting the change amplitude of feature fluctuation range;

[0110] If the variance change rate is greater than 5, it is calculated as 5, and if it is less than 5, it is calculated as the actual value;

[0111] The mean relative change rate, variance change rate, and value range compression / expansion rate are added to obtain the distribution offset comprehensive index;

[0112] Compare the distribution offset comprehensive index with the offset threshold value, if the distribution offset comprehensive index is greater than or equal to the offset threshold value, there is a quantitative change premise of data progressive offset;

[0113] Second, the coupling reconstruction change feature analysis is used to divide the offset reconstruction fusion and the working condition reconstruction fusion, specifically:

[0114] Two key time points are extracted, including:

[0115] Coupling reconstruction time point: the time when the feature meets the reconstruction standard for the first time on the coupling quantitative index;

[0116] Distribution offset compliance time point: the time when the distribution offset comprehensive index is first greater than or equal to the offset threshold value;

[0117] Timing verification is performed:

[0118] If the distribution offset compliance time point is earlier and the coupling reconstruction time point is later, it meets the offset reconstruction feature, otherwise, if the coupling reconstruction time point is earlier and the distribution offset compliance time point is later, it meets the working condition reconstruction feature;

[0119] Cause tracing analysis:

[0120] Whether there is a system event (dispatch target switching, policy adjustment, equipment modification, etc.) within 1 hour before and after the coupling reconstruction time point is found;

[0121] If there is, the coincidence degree between the event occurrence time and the reconstruction time is calculated: coincidence degree = 1- |event time-reconstruction time| / 1 hour;

[0122] If there is no related system event, it meets the offset reconstruction feature;

[0123] If there is a related system event, and the coincidence degree is greater than or equal to the preset coincidence degree, it meets the working condition reconstruction feature;

[0124] Reconstruction range and intensity analysis:

[0125] The number of feature pairs that meet the coupling reconstruction standard in the coupling reconstruction error fusion is counted;

[0126] The coupling index change rate is calculated: for linear coupling feature pairs, the difference between the correlation coefficients of the error segment and the steady state segment is calculated, and the change amplitude per hour is obtained;

[0127] If the number of reconstruction feature pairs is less than or equal to 2, and the change rate is less than or equal to 10% / hour, it meets the offset reconstruction feature;

[0128] If the number of reconstruction feature pairs is greater than or equal to 3, and the change rate is greater than or equal to 50% / hour, it meets the working condition reconstruction feature;

[0129] If the offset reconstruction features of timing, cause, reconstruction range and intensity are met at the same time, it is an offset reconstruction fusion;

[0130] If the working condition reconstruction features of timing, cause, reconstruction range and intensity are met at the same time, it is a working condition reconstruction fusion;

[0131] The role of coupling reconstruction error fusion classification is: through the triple verification of timing+cause+range / intensity, avoid single dimension misjudgment, and provide accurate case input for quantitative and qualitative changes;

[0132] Step three: for the offset reconstruction fusion, the correlation analysis of the feature distribution offset and the coupling reconstruction is performed to determine the distribution offset comprehensive index critical value;

[0133] In step three, the determination process of the distribution offset comprehensive index critical value includes:

[0134] Obtain the distribution offset comprehensive index of the offset reconstruction fusion;

[0135] Calculate the coupling reconstruction index of the offset reconstruction fusion:

[0136] Convert the qualitative change of the coupling relationship into a calculable index, and highlight the decisive role of the core logic direction reversal:

[0137] The correlation coefficient change rate = |reconstruction correlation coefficient - pre-reconstruction steady-state correlation coefficient| / pre-reconstruction steady-state correlation coefficient, reflecting the change amplitude of coupling strength;

[0138] The logic direction reversal coefficient: coupling logic direction unchanged = 0, coupling logic direction reversed = 1, reflecting the qualitative change of coupling logic;

[0139] If the correlation coefficient change rate is greater than 100%, it is uniformly calculated as 100%; if it is less than 100%, it is calculated according to the actual value;

[0140] Add the correlation coefficient change rate and the logic direction reversal coefficient to obtain the coupling reconstruction index;

[0141] Divide the group according to the distribution offset comprehensive index 5% as an interval, covering the distribution offset comprehensive index range of all offset reconstruction fusions;

[0142] For each group of data, the number of offset reconstruction cases in the index interval is counted to obtain the number of samples in the group, the number of cases triggering coupling reconstruction in the group is counted to obtain the reconstruction occurrence number, and the reconstruction occurrence probability is calculated by proportion between the reconstruction occurrence number and the number of samples in the group;

[0143] It should be noted that the index interval (i.e. the mutation interval) where the reconstruction occurrence probability jumps from a low probability to a high probability is found, and the lower limit of the interval is the critical value, which is:

[0144] Starting from the first group, the deviation of the reconstruction occurrence probability of adjacent two groups is calculated in turn, and after each calculation, it is compared with the mean value of the reconstruction occurrence probability of all previous groups. If the reconstruction occurrence probability is greater than or equal to the mean value of the reconstruction occurrence probability of all previous groups, it is the mutation interval, and the lower limit of the mutation interval is the distribution offset comprehensive index critical value;

[0145] It can be understood that the physical meaning of the distribution deviation comprehensive index critical value is that the quantitative demarcation point of the feature distribution deviation in the energy system triggering the coupling relationship reconstruction, which essentially reflects the maximum distribution deviation limit that the original coupling relationship can withstand under the current operating conditions of the energy system (such as device configuration, control strategy, and power grid structure);

[0146] It should be noted that the role of determining the distribution deviation comprehensive index critical value is to provide a quantitative standard for real-time early warning and to suppress the positive feedback of deviation-reconstruction-more serious deviation from the root;

[0147] Step four: for working condition reconstruction fusion, the working condition changes leading to coupling relationship reconstruction are counted, and a dedicated fusion model is pre-trained based on historical steady-state data;

[0148] In step four, the process of counting the working condition changes leading to coupling relationship reconstruction includes:

[0149] All historical working condition reconstruction fusions are obtained, and are counted according to three dimensions of event type-coupling reconstruction logic-impact feature to form a mapping table, for example:

[0150] | Working condition change type | Typical event example | Triggered coupling reconstruction logic | Impact core feature pair |;

[0151] | Scheduling target switching | Economic optimization-new energy consumption priority | Linear coupling direction reversal (positive correlation-negative correlation) | Photovoltaic-energy storage, load-thermal power |;

[0152] | Policy / Market mechanism adjustment | Benchmark price-present spot price mechanism start | Nonlinear coupling strength sudden increase (mutual information doubled) | Load-price, unit output-price |;

[0153] | Equipment structure modification | New large-scale energy storage power station connected to the grid | Causal coupling link added (photovoltaic-energy storage-tidal flow) | Photovoltaic-energy storage, energy storage-line tidal flow |;

[0154] Form a working condition coupling reconstruction: working condition change-reconstruction logic mapping table, which clearly defines the core impact feature pair and reconstruction rule corresponding to each working condition change;

[0155] In step four, the process of pre-training the corresponding dedicated fusion model based on historical steady-state data includes:

[0156] Obtain the historical steady-state data of the corresponding working condition and the coupling reconstruction logic rules (such as photovoltaic high-energy storage charging under the consumption priority working condition), and select the historical data after the working condition change, coupling reconstruction stability, and no fusion error;

[0157] Divide into training set (for model training) and test set (for precision verification) according to the ratio of 7:3;

[0158] Model architecture: Choose domain adaptation + causal inference hybrid architecture (adapt to multi-source heterogeneous data, ensure that the fusion result meets the coupling logic);

[0159] Input features: Core features related to reconstruction (such as priority operating conditions, input photovoltaic output, energy storage SOC, and load data);

[0160] For any type of working condition, use the training set to train the model to obtain a dedicated fusion model;

[0161] Store by working condition type-model version-applicable feature pair-precision index classification to obtain a dedicated fusion model;

[0162] Step five: Match and analyze the feature coupling relationship and working condition in the current energy data fusion. If it does not match, identify the coupling reconstruction type, and combine the distribution offset comprehensive index critical value comparison analysis to perform differential calling;

[0163] In step five, the process of matching and analyzing the feature coupling relationship and working condition in the current energy data fusion includes:

[0164] For the current multi-source heterogeneous energy real-time data, calculate the coupling quantitative indicators: linear coupling (Pearson correlation coefficient), nonlinear coupling (mutual information), and causal coupling (Granger direction);

[0165] Numerical matching degree = 1 - |current coupling index - standard coupling index| / standard coupling index;

[0166] Logical matching degree = 1 (current coupling direction / link is consistent with the standard), 0 (direction / link is reversed);

[0167] Comprehensive matching degree = numerical matching degree + logical matching degree. Compare the comprehensive matching degree with the preset matching degree. If the comprehensive matching degree is less than the preset matching degree, it does not match;

[0168] In step five, the process of determining the coupling reconstruction type corresponding to the mismatch includes:

[0169] Check the current system event log to see if there is a working condition change event in the working condition change-reconstruction logic mapping table;

[0170] If there is a corresponding event, and the event occurrence time and the time coincidence degree of the comprehensive matching degree being less than the preset matching degree is greater than or equal to the preset coincidence degree, it is determined that the working condition reconstruction does not match;

[0171] According to the current working condition label, match the corresponding dedicated model from the dedicated fusion model library to perform multi-source heterogeneous energy data fusion;

[0172] If there is no corresponding event, the distribution offset comprehensive index of the current feature is calculated and compared with the distribution offset comprehensive index threshold value;

[0173] If the distribution offset comprehensive index is less than the distribution offset comprehensive index threshold value, the fusion weight of the feature pair is adjusted to strengthen the contribution of stable features and weaken the interference of fluctuating features, so as to quickly recover the coupling matching;

[0174] If the distribution offset comprehensive index is greater than or equal to the distribution offset comprehensive index threshold value:

[0175] First, the sliding window dynamic normalization (basic de-biasing) is explained, which is specifically:

[0176] The statistical distribution parameters (mean, variance) of the features are updated in real time to align the first-order distribution difference (center value, dispersion) between the current and reference working conditions, which is specifically:

[0177] Stable data (excluding abnormal values) in the last 1 hour (4 15-minute data windows) is selected, and the mean and variance in each data window are updated every 1 data window;

[0178] The normalization is performed according to the formula: current feature standardized value = (current original value - window mean) / window standard deviation;

[0179] Second, the domain adaptive algorithm (high-order alignment, for features with large fluctuations) is explained:

[0180] For features with a fluctuation coefficient > 0.5, the feature covariance matrix of the reference working condition is extracted, the covariance matrix of the current window feature is calculated, and the matrix transformation is performed to make the current covariance matrix consistent with the reference covariance matrix;

[0181] For feature pairs with coupling quantitative indicators (correlation coefficient, mutual information) close to the standard value after de-biasing, the weight is adjusted upward, and for feature pairs still with slight deviation, the current weight is maintained.

[0182] The technical scheme and advantages of the embodiments of the present application are as follows: historical multi-source heterogeneous energy data is collected, error fusion with existing fusion errors is screened, feature coupling relationship in the error fusion window is analyzed, it is determined whether the fusion error is caused by coupling relationship reconstruction, if so, the coupling reconstruction error fusion is marked, statistical analysis is performed on the coupling reconstruction error fusion, and it is determined whether there is a coupling reconstruction fusion error phenomenon; if so, a distribution offset comprehensive index is calculated through feature distribution offset analysis, and the coupling reconstruction error fusion is divided into offset reconstruction fusion and working condition reconstruction fusion through coupling reconstruction change feature analysis; for the offset reconstruction fusion, correlation analysis is performed on the feature distribution offset and the coupling reconstruction, and a distribution offset comprehensive index threshold is determined; for the working condition reconstruction fusion, working condition changes causing the coupling relationship reconstruction are counted, and a dedicated fusion model is pre-trained based on historical steady-state data; matching analysis is performed on the feature coupling relationship and the working condition in the current energy data fusion, if not matching, the coupling reconstruction type is identified, and differential calling is performed through comparison analysis of the distribution offset comprehensive index threshold. The present application screens historical fusion error cases and determines coupling reconstruction causes, divides coupling reconstruction of quantity change / working condition reconstruction, determines the distribution offset comprehensive index threshold of offset reconstruction, pre-trains a working condition reconstruction dedicated AI fusion model, and performs real-time adaptation and differential processing on the current fusion, through the closed-loop design of error screening-reconstruction classification-threshold determination-model pre-training-real-time adaptation, the fusion error caused by coupling reconstruction is accurately identified, differential AI strategies are adopted for quantity change and working condition reconstruction, the problems that the prior art cannot cope with dynamic changes of coupling relationship, fusion precision is low, and stability is poor are solved, the precision and robustness of multi-source heterogeneous fusion of energy data are significantly improved, and the complex and variable working conditions of the energy system are adapted.

[0183] Embodiment 2: Please refer to Figure 3 The AI-based multi-source heterogeneous energy data fusion system according to the embodiments of the present application includes the following modules:

[0184] The coupling reconstruction determination module collects historical multi-source heterogeneous energy data, screens error fusion with existing fusion errors, analyzes the feature coupling relationship in the error fusion window, determines whether the fusion error is caused by coupling relationship reconstruction, and if so, marks the coupling reconstruction error fusion, and performs statistical analysis on the coupling reconstruction error fusion to determine whether there is a coupling reconstruction fusion error phenomenon;

[0185] The reconstruction fusion type division module divides the coupling reconstruction error fusion into offset reconstruction fusion and working condition reconstruction fusion through feature distribution offset analysis and coupling reconstruction change feature analysis if there is a coupling reconstruction fusion error phenomenon;

[0186] The offset threshold calculation module determines the distribution offset comprehensive index threshold through correlation analysis of the feature distribution offset and the coupling reconstruction for the offset reconstruction fusion;

[0187] Special fusion model construction module: for working condition reconstruction fusion, the working condition change leading to coupling relationship reconstruction is counted, and a special fusion model is pre-trained based on historical steady-state data;

[0188] Real-time matching and differential calling module: matching analysis is performed on the feature coupling relationship and working condition in the current energy data fusion, if not matching, the coupling reconstruction type is identified, combined with the comparison analysis of the distribution offset comprehensive index critical value, differential calling is executed.

[0189] In embodiment 3, the electronic device further comprises:

[0190] The processor, the communication interface, the memory and the communication bus complete mutual communication through the communication bus;

[0191] The memory is used for storing a computer program;

[0192] The processor is used for executing the program stored on the memory, and realizes the steps of the display control method based on the information publishing screen.

[0193] The above describes one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the scope of the present application.

Claims

1. An AI-based method for multi-source heterogeneous fusion of energy data, characterized by: Includes the following steps: Historical multi-source heterogeneous energy data are collected, and error fusion with fusion error is screened. By analyzing the characteristic coupling relationship within the error fusion window, it is determined whether the fusion error is caused by the reconstruction of the coupling relationship. If so, it is marked as coupling reconstruction error fusion. Statistical analysis is performed on the coupling reconstruction error fusion to determine whether there is a coupling reconstruction fusion error phenomenon. If it exists, calculate the comprehensive index of distribution offset through characteristic distribution offset analysis, and combine it with the coupling reconstruction change characteristic analysis to divide the coupling reconstruction error fusion into offset reconstruction fusion and working condition reconstruction fusion. For offset reconstruction fusion, correlation analysis is performed on feature distribution offset and coupled reconstruction to determine the critical value of the comprehensive distribution offset index; For the fusion of working condition reconfiguration, the working condition changes that lead to the reconfiguration of coupling relationships are statistically analyzed, and a dedicated fusion model is pre-trained based on historical steady-state data; The system performs a matching analysis between the feature coupling relationship and the operating conditions in the current energy data fusion. If there is a mismatch, it identifies the coupling reconstruction type and performs a differentiated call based on the comparison analysis of the critical value of the distribution offset comprehensive index.

2. The AI-based multi-source heterogeneous fusion method for energy data according to claim 1, characterized in that: The determination method for the coupling reconstruction error fusion is as follows: The root mean square error and error fluctuation coefficient of multi-source heterogeneous energy data were collected over a historical period. If either the root mean square error or the error fluctuation coefficient meets the requirements, then a fusion error exists. The system is divided into a steady-state segment and a fusion error segment. The steady-state segment is the window before the error fusion window, and the error fusion segment is the window where the fusion error is abnormal. For any error fusion, feature pairs with coupling relationships are selected and the coupling type is marked, including linear, nonlinear, and causal relationships; For the feature pairs with linear relationships, calculate the rate of change of the Pearson correlation coefficient between the steady-state segment and the fusion error segment; For feature pairs with nonlinear relationships, calculate the rate of change of mutual information values ​​between the steady-state segment and the fusion error segment; For feature pairs with causal relationships, examine the causal directionality between features in the steady-state segment and the fusion error segment; If the rate of change of Pearson correlation coefficient, the rate of change of mutual information value, and the causal directionality meet the requirements, then there is a reconstruction of the coupling relationship; Extracting the time point for reconstructing coupled quantization metrics and the time point of the first error anomaly: If the reconstruction time of the coupled quantitative index is earlier than or synchronous with the time of the first error anomaly, it conforms to the causal time sequence, and the error fusion is marked as coupled reconstruction error fusion.

3. The AI-based multi-source heterogeneous fusion method for energy data according to claim 2, characterized in that: The method for determining whether there is a coupling reconstruction fusion error is as follows: The proportion of coupled reconstruction error fusion in all error fusions is statistically analyzed to obtain the proportion of coupled reconstruction error fusion. If the proportion of coupled reconstruction error fusion meets the requirements, then the phenomenon of coupled reconstruction fusion error exists.

4. The AI-based multi-source heterogeneous fusion method for energy data according to claim 1, characterized in that: The calculation process for the distribution offset composite index is as follows: For any feature with a coupling relationship, extract the steady-state segment data and the error fusion window data respectively, and calculate the distribution offset comprehensive index, specifically: The relative rate of change of the mean = |mean after shift - mean before shift| / mean before shift; the variance change factor = variance after shift / steady-state variance before shift; the range compression / expansion rate = | steady-state maximum value before shift - steady-state minimum value before shift (maximum value after shift - minimum value after shift) - (steady-state maximum value before shift - steady-state minimum value before shift)|; The distribution shift index is obtained by adding the relative rate of change of the mean, the factor of change of the variance, and the rate of compression / expansion of the range.

5. The AI-based multi-source heterogeneous fusion method for energy data according to claim 4, characterized in that: The method for dividing the coupling reconstruction error fusion into offset reconstruction fusion and working condition reconstruction fusion is as follows: If the distribution offset composite index meets the requirements, then there is a quantitative premise for gradual data offset. Compare the coupling reconstruction time point with the distribution offset compliance time point. If the distribution offset compliance time point is earlier and the coupling reconstruction time point is later, it conforms to the offset reconstruction characteristics. Otherwise, it conforms to the working condition reconstruction characteristics. Check if there are any system events before the coupling reconfiguration time point. If so, calculate the degree of coincidence between the event occurrence time and the reconfiguration time: degree of coincidence = 1 - |event time - reconfiguration time| / 1 hour; If there are no relevant system events, it meets the offset reconstruction characteristics; if there are relevant system events and the consistency meets the requirements, it meets the working condition reconstruction characteristics. If both time and event offset reconstruction characteristics are met, it is called offset reconstruction fusion; if both time and event condition reconstruction characteristics are met, it is called condition reconstruction fusion.

6. The AI-based multi-source heterogeneous fusion method for energy data according to claim 1, characterized in that: The critical value of the offset composite index is calculated as follows: Calculate the coupling reconstruction index for offset reconstruction fusion: Rate of change of correlation coefficient = |Correlation coefficient after reconstruction - Steady-state correlation coefficient before reconstruction| / Steady-state correlation coefficient before reconstruction; Logic direction reversal coefficient: Unchanged coupled logic direction = 0, Reversed coupled logic direction = 1; The coupling reconstruction index is obtained by adding the correlation coefficient change rate to the logical direction reversal coefficient; The distribution offset composite index is divided into groups based on a 5% interval, covering the range of the distribution offset composite index for all offset reconstruction and fusion. For each set of data, the number of offset reconstruction cases within the statistical index interval is used to obtain the number of samples in the set, the number of cases triggering coupled reconstruction within the set is used to obtain the number of reconstructions, and the ratio of the number of reconstructions to the number of samples in the set is used to calculate the probability of reconstruction. Starting from the first group, the deviation of the reconstruction probability between adjacent groups is calculated sequentially. After each calculation, it is compared with the mean reconstruction probability of all previous groups. If the reconstruction probability is greater than or equal to the mean reconstruction probability of all previous groups, it is a mutation interval. The lower limit of the mutation interval is the critical value of the distribution offset comprehensive index.

7. The AI-based multi-source heterogeneous fusion method for energy data according to claim 1, characterized in that: The process of pre-training the dedicated fusion model based on historical steady-state data is as follows: Acquire all historical working conditions for reconstruction and integration, and statistically analyze them in three dimensions: event type, coupled reconstruction logic, and impact characteristics to form a mapping table of working condition changes and reconstruction logic. Obtain historical steady-state data and coupling reconstruction logic rules for the corresponding operating conditions, and select historical data that are stable after the operating conditions change and the coupling reconstruction is stable with no fusion error; A hybrid architecture combining domain adaptation and causal inference is adopted. For any type of working condition, the model is trained to obtain a dedicated fusion model.

8. The AI-based multi-source heterogeneous fusion method for energy data according to claim 1, characterized in that: The method for identifying the coupling reconstruction type is as follows: For current real-time data of multi-source heterogeneous energy, calculate the coupling quantification index: linear coupling (Pearson correlation coefficient), nonlinear coupling (mutual information), and causal coupling (Granger direction). Numerical matching degree = 1 - |current coupling index - standard coupling index| / standard coupling index; Logical matching degree = 1 (current coupling direction / link is consistent with the standard), 0 (direction / link is reversed); Overall matching degree = numerical matching degree + logical matching degree. If the overall matching degree does not meet the requirements, then there is no match. Check the current system event log to see if there are any operating condition change events in the operating condition change-refactoring logic mapping table; If a corresponding event exists, and the coincidence between the event occurrence time and the mismatch time meets the requirements, then it is determined to be a mismatch in the working condition reconfiguration. If no corresponding event exists, calculate the distribution offset composite index of the current feature. If the distribution offset composite index is greater than or equal to the distribution offset composite index threshold, then the offset reconstruction is mismatched.

9. An AI-based multi-source heterogeneous fusion system for energy data, characterized in that: Includes the following modules: Coupling Reconstruction Judgment Module: Collects historical multi-source heterogeneous energy data, filters out error fusion with fusion error, analyzes the characteristic coupling relationship within the error fusion window to determine whether the fusion error is caused by the reconstruction of the coupling relationship. If so, it is marked as coupling reconstruction error fusion, and statistical analysis is performed on the coupling reconstruction error fusion to determine whether there is a coupling reconstruction fusion error phenomenon. Reconstruction fusion type classification module: If it exists, calculate the comprehensive index of distribution offset through feature distribution offset analysis, and combine it with the coupling reconstruction change feature analysis to classify the coupling reconstruction error fusion into offset reconstruction fusion and working condition reconstruction fusion. Offset Critical Value Calculation Module: For offset reconstruction fusion, the module performs correlation analysis between feature distribution offset and coupled reconstruction to determine the critical value of the comprehensive distribution offset index. Dedicated fusion model construction module: For working condition reconstruction fusion, the working condition changes that lead to the reconstruction of coupling relationship are statistically analyzed, and a dedicated fusion model is pre-trained based on historical steady-state data; Real-time matching and differentiated invocation module: Performs matching analysis on the feature coupling relationship and operating conditions in the current energy data fusion. If there is no match, it identifies the coupling reconstruction type, and performs differentiated invocation based on the comparison analysis of the critical value of the distribution offset comprehensive index.

10. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method described in any one of claims 1-8.

Images