A power energy intelligent scheduling method based on big data analysis

Abstract

The application discloses a kind of power energy wisdom scheduling method based on big data analysis, it is related to big data analysis technical field, including: generating trusted alignment power state frame;Generation scheduling form risk calibration table;Improved SOFTS model is constructed;Generation form enhancement scheduling state representation;Generation multi-source scheduling prediction result;Generation safety scheduling ability correlation table;Generation scheduling risk correction mark table;Generation wisdom scheduling strategy and write back result;Through execution result write back forms closed loop update.The application combines multi-source power data trusted alignment, form symbol risk calibration, improved SOFTS prediction, power flow safety projection, energy storage life constraint and risk stratification correction, completes the collaborative scheduling of load, new energy, energy storage, line and demand response, improves scheduling state identification precision, safety constraint adaptation capability, new energy consumption capacity and strategy execution stability.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a smart power energy dispatching method based on big data analytics. Background Technology

[0002] With the expansion of new energy grid connection, increased power load fluctuations, and greater access to energy storage resources, power dispatch is gradually shifting from manual, experience-based dispatch to data-driven dispatch. Existing power dispatch systems primarily rely on automated dispatch platforms, energy management systems, load forecasting models, and fixed rule bases to collect and analyze power generation operation data, new energy output data, energy storage status data, load consumption data, and grid operation data, and generate dispatch instructions based on the forecast results. While this approach can accomplish basic monitoring, plan allocation, and alarm response, it is insufficient in handling time differences, channel reliability differences, and topology node binding differences among multi-source data. Furthermore, the dispatch status representation is susceptible to data gaps, abnormal spikes, upload delays, and execution deviations.

[0003] Existing dispatch forecasting methods primarily focus on numerical load forecasting or renewable energy output forecasting, failing to adequately characterize the cross-channel coupling characteristics between load, renewable energy, energy storage, transmission lines, and demand response. This makes it difficult to identify the interconnected changes between supply-demand gaps, limited renewable energy absorption, insufficient energy storage support, and transmission line overruns. While some systems incorporate deep learning models, the forecast results lack morphological constraints, event time-delay correlations, and power flow safety projections, making it difficult to directly translate the predicted state into dispatch capabilities that meet constraints such as node power balance, transmission line capacity, voltage safety, regional exchange boundaries, and energy storage lifetime. Furthermore, the failure to effectively write dispatch execution feedback back into the forecasting and risk correction processes leads to problems such as delayed response, inaccurate risk stratification, and accumulated execution deviations in intelligent dispatch strategies.

[0004] Therefore, how to provide a smart power energy dispatching method based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a smart power energy dispatching method based on big data analysis. This invention integrates multi-source power data reliable alignment, morphological symbol aggregation approximation algorithm and improved SOFTS model to complete the joint prediction of load, new energy, energy storage and line status, safety capability assessment and risk correction dispatching. It has the advantages of high data reliability, fast dispatching response, strong safety constraints and high new energy absorption capacity.

[0006] A smart power energy dispatching method based on big data analysis according to an embodiment of the present invention includes: Collect power energy operation data and historical dispatch execution data, and perform preprocessing on the data to generate a reliable aligned power status frame; Based on the trusted aligned power state frame, a morphological symbol aggregation approximation algorithm with a cross-channel scheduling coupling symbol constraint mechanism is used to perform symbol calibration processing and generate a scheduling morphological risk calibration table. An improved SOFTS model is constructed, which includes a morphological symbol-guided fusion layer, a scheduling event causal gating layer, and a power grid flow projection layer. The input is a trusted aligned power state frame and a scheduling morphological risk calibration table, and the causal gating channel weights are generated. Input the causal gating channel weights and scheduling morphological risk calibration table into the morphological symbols to guide the fusion layer to perform core aggregation and state reallocation, and generate a morphologically enhanced scheduling state representation. The morphologically enhanced scheduling state representation is used as the input scheduling event. The causal gating layer performs joint prediction of the scheduling state to generate multi-source scheduling prediction results. The multi-source scheduling prediction results are input into the power flow projection layer to perform power flow safety projection, energy storage lifetime constraints and capacity calculation, and generate a safety scheduling capacity association table. Based on the security scheduling capability association table and the scheduling form risk calibration table, perform uncertainty risk hierarchical output and generate a scheduling risk correction mark table; A smart scheduling strategy is generated based on the security scheduling capability association table and the scheduling risk correction flag table, and the execution results of the smart scheduling strategy are written into the historical scheduling execution data.

[0007] Optionally, generating a trusted aligned power state frame includes: Collect power generation operation records, new energy output records, energy storage status records, load power consumption records, and power grid operation records to generate power energy operation data; Collect scheduling instruction numbers, execution times, planned adjustment amounts, actual response amounts, and execution deviation records to generate historical scheduling execution data; The power energy operation data and historical dispatch execution data are mapped, dimensionally unified, and sampled according to the dispatch time axis to generate raw power energy data. Perform missing data marking, anomaly marking, time alignment, channel credibility marking, topology node binding, and line topology marking on the raw power energy data to generate a trusted aligned power status frame carrying node binding relationships and line topology records.

[0008] Optionally, the generation of the scheduling pattern risk calibration table includes: Read the time-series status records of each power channel from the trusted aligned power status frame, perform multi-scale window segmentation according to channel type, scheduling cycle and node binding relationship, and generate channel scheduling window sequence; Perform morphological symbol mapping on the channel scheduling window sequence to generate a single-channel morphological symbol string carrying amplitude level, direction level and response stage markers; Based on the cross-channel scheduling coupling symbol constraint mechanism, the single-channel morphological symbol strings within the same scheduling window are constructed into a cross-channel symbol constraint graph according to the node binding relationship, generating a scheduling coupling symbol group; Read the execution deviation flag from the trusted aligned power state frame, perform risk level recalibration and symbol threshold calibration on the scheduling coupled symbol group based on the execution deviation flag, and generate coupled risk labels and calibration symbol thresholds; The scheduling morphology risk calibration table is generated by associating the channel scheduling window sequence, single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label, execution deviation mark and calibration symbol threshold.

[0009] Optionally, the generation of causal gating channel weights includes: An improved SOFTS model is established, connecting the morphological symbol guidance fusion layer, the scheduling event causal gating layer, and the power grid flow projection layer according to the data transmission sequence. Read the time-series status records of each power channel in the trusted aligned power state frame, and read the single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold in the scheduling morphology risk calibration table to generate model input records; Based on the model input records, channel representation encoding and event trigger matching are performed in the scheduling event causal gating layer to generate a scheduling event matching table; Based on the scheduling event matching table, gating weights are assigned to each power channel to generate causal gating channel weights. The improved SOFTS model was trained by reading trusted aligned power state frames, scheduling morphology risk calibration tables, and historical scheduling execution data. The trusted aligned power state frames and scheduling morphology risk calibration tables were input into the improved SOFTS model, and forward computation was performed to obtain multi-source scheduling prediction results, a safe scheduling capability association table, and a scheduling risk correction label table. The combination of scheduling state prediction error, event causality gating error, morphology symbol fusion error, power flow safety projection error, energy storage lifetime constraint error, scheduling risk level discrimination error, and scheduling correction triggering error was used as the joint optimization objective. The network parameters of the morphology symbol guidance fusion layer, the scheduling event causality gating layer, and the power flow projection layer were continuously optimized. When the change in the joint loss value in five consecutive training rounds was less than 0.001, the improved SOFTS model was considered to have completed convergence training.

[0010] Optionally, the improved SOFTS model includes: The SOFTS model retains the core aggregation structure of the sequence, the channel state redistribution structure, the time series prediction backbone and the prediction output structure, and makes structural modifications to the input expression, morphological symbol fusion, event causal gating and power flow security constraints in the context of intelligent power energy dispatching. A trusted aligned power state frame input is introduced at the model input end, transforming the ordinary multivariate time series input into a power dispatch state input form that carries channel type, scheduling cycle, node binding relationship, channel trustworthiness flag and execution deviation flag; A morphological symbol-guided fusion layer is added at the core aggregation position of the sequence, embedding the single-channel morphological symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold from the scheduling morphological risk calibration table into the channel fusion process; The channel weight allocation method is improved to form a scheduling event causal gating layer, and the ordinary channel weight allocation method is transformed into a causal gating allocation method under the constraint of the event time delay association table. A power flow projection layer is added to the back end of the prediction output. The multi-source scheduling prediction results are input into the power flow security projection process, so that node power balance, line capacity, voltage security, regional exchange boundary and energy storage lifetime constraints participate in the selection of safe and feasible scheduling states. The output organization method is improved by expanding the model output target to include causal gating channel weights, morphologically enhanced scheduling state representation, multi-source scheduling prediction results, and a correlation table of safe scheduling capabilities.

[0011] Optionally, the generation of morphologically enhanced scheduling state representation includes: Read the single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold from the scheduling morphology risk calibration table, input the morphology symbol guidance fusion layer, and generate morphology symbol index records according to the power channel number; Match the causal gating channel weights with the morphological symbol index records of the execution channel to generate morphological gating fusion weights carrying coupling risk constraints; Based on morphological gating fusion weights, weighted core aggregation is performed on the corresponding power channel states in the trusted aligned power state frames to generate scheduling core states. Based on the reallocation of the execution state of the scheduling core state and the scheduling coupled symbol group, a sub-channel enhanced state record is generated. By associating the enhanced state records of the associated channels, the core state of the scheduling, and the morphological gating fusion weights, a morphologically enhanced scheduling state representation is generated.

[0012] Optionally, generating multi-source scheduling prediction results includes: The reading morphology enhances the scheduling status representation. Load forecast representation, new energy forecast representation, energy storage forecast representation, line forecast representation, and demand response forecast representation are extracted according to power channel type and scheduling window number to generate a scheduling channel forecast representation set. Based on the scheduling channel prediction representation set, execution event time delay matching is performed to generate an event time delay association table; Generate event causal prediction weights by using a time-delay association table of related events and causal gating channel weights. Based on the event causal prediction weights, gated joint prediction is performed on the scheduling channel prediction representation set to generate power channel prediction records; Based on the power channel forecast records, supply and demand deviation calculations are performed according to the node binding relationship to generate node supply and demand deviation forecast records. The power channel forecast records and node supply and demand deviation forecast records are then linked to generate multi-source scheduling forecast results.

[0013] Optionally, the generation of the security scheduling capability association table includes: The power flow projection layer reads power channel prediction records and node supply and demand deviation prediction records from multi-source scheduling prediction results, and reads node binding relationships and line topology records from trusted aligned power state frames to generate power flow projection input records. Input the power flow projection input record into the power grid power flow projection layer, perform node power balance verification, line capacity verification, voltage safety verification and regional exchange boundary verification, and generate power flow safety verification record; Based on the power channel prediction records in the multi-source scheduling prediction results, extract the energy storage status prediction records, read the energy storage operation boundary records in the reliable aligned power status frames, perform energy storage lifetime constraint verification, and generate energy storage dispatchable boundary records. Based on the power flow safety verification record and the energy storage schedulable boundary record, a safe and feasible state projection is performed to generate a safe and feasible schedulable state; Based on the calculation of power generation regulation capacity, new energy absorption capacity, energy storage charging and discharging capacity, load response capacity and line transmission margin in the safe and feasible dispatch state, an energy unit dispatchable capacity matrix is ​​generated. By associating the safe and feasible scheduling status, the energy unit scheduling capability matrix, the energy storage scheduling boundary record, the node binding relationship, and the power flow safety verification record, a safe scheduling capability association table is generated.

[0014] Optionally, generating the scheduling risk correction flag table includes: Read the safe and feasible scheduling status, energy unit scheduling capability matrix, power flow safety verification record, and energy storage scheduling boundary record from the safe scheduling capability association table; Power flow safety margin is generated based on power flow safety verification records, energy storage lifetime constraint margin is generated based on energy storage schedulable boundary records, and scheduling priority is generated based on energy unit schedulable capability matrix. Safety capability risk records are generated by associating safe and feasible scheduling status. Read the coupling risk label, calibration symbol threshold and execution deviation mark from the scheduling morphological risk calibration table to generate a morphological risk calibration record; Associate security capability risk records and morphological risk calibration records, perform layered output of uncertainty risks, and generate scheduling risk levels; Based on the scheduling risk level and scheduling priority, a correction trigger flag is generated, and a risk correction association record is established according to the energy unit number and scheduling window number; Associate scheduling risk levels, correction trigger flags, risk correction association records, and execution deviation flags to generate a scheduling risk correction flag table.

[0015] Optionally, the generation of intelligent scheduling strategies based on the security scheduling capability association table and the scheduling risk correction flag table includes: Read the safe and feasible scheduling status, energy unit scheduling capability matrix, energy storage scheduling boundary record and power flow security verification record from the safe scheduling capability association table, and generate the strategy capability input record. Read the scheduling risk level, scheduling correction trigger flag, and risk correction association record from the scheduling risk correction flag table to generate a strategy risk input record; Associate the policy capability input record and policy risk input record, and generate a combination of scheduling instructions according to the energy unit number and scheduling window number; Intelligent scheduling strategies are generated based on combinations of scheduling instructions, and the scheduling instruction number, execution time, planned adjustment amount, actual response amount and execution deviation record corresponding to the intelligent scheduling strategy are collected to generate strategy execution results; Write the strategy execution results into the historical scheduling execution data.

[0016] The beneficial effects of this invention are: This invention proposes a smart power energy dispatching method based on big data analysis. By collecting power energy operation data and historical dispatch execution data, it performs field mapping, unit unification, time alignment, channel credibility marking, topology node binding, and line topology marking to generate a reliable aligned power status frame. This effectively reduces the impact of missing, abnormal, delayed, and asynchronous multi-source power data on dispatching decisions. Furthermore, by introducing a morphological symbol aggregation approximation algorithm with a cross-channel dispatching coupled symbol constraint mechanism, it performs morphological symbol mapping, coupled constraint mapping, and execution deviation calibration for loads, new energy sources, energy storage, lines, and demand response channels. This improves the accuracy of identifying dispatching risks such as supply-demand gaps, limited new energy absorption, insufficient energy storage support, and line over-limits.

[0017] This invention constructs an improved SOFTS model comprising a morphological symbol-guided fusion layer, a scheduling event causal gating layer, and a power flow projection layer. It inputs a scheduling morphological risk calibration table and a reliable aligned power state frame into the model to complete morphologically enhanced state expression, event time-delay correlation prediction, power flow security projection, and energy storage lifetime constraint verification. Compared to single load forecasting or fixed-rule scheduling methods, this invention can generate multi-source scheduling prediction results, a secure scheduling capability correlation table, and a scheduling risk correction label table. This improves the adaptability of scheduling strategies to node power balance, line capacity, voltage security, regional exchange boundaries, and energy storage operation boundaries, offering advantages such as high data reliability, accurate risk stratification, timely scheduling response, strong renewable energy absorption capacity, and high grid operation security. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a smart power energy dispatching method based on big data analysis proposed in this invention; Figure 2 This is a schematic diagram of the improved SOFTS model in the intelligent power energy dispatching method based on big data analysis proposed in this invention. Figure 3 This is a schematic diagram illustrating the process of generating the safety dispatch capability association table and the dispatch risk correction mark table in the intelligent power energy dispatching method based on big data analysis proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 , Figure 2 and Figure 3 A smart power energy dispatching method based on big data analysis includes: Collect power energy operation data and historical dispatch execution data, and perform preprocessing on the data to generate a reliable aligned power status frame; Based on the trusted aligned power state frame, a morphological symbol aggregation approximation algorithm with a cross-channel scheduling coupling symbol constraint mechanism is used to perform symbol calibration processing and generate a scheduling morphological risk calibration table. An improved SOFTS model is constructed, which includes a morphological symbol-guided fusion layer, a scheduling event causal gating layer, and a power grid flow projection layer. The input is a trusted aligned power state frame and a scheduling morphological risk calibration table, and the causal gating channel weights are generated. Input the causal gating channel weights and scheduling morphological risk calibration table into the morphological symbols to guide the fusion layer to perform core aggregation and state reallocation, and generate a morphologically enhanced scheduling state representation. The morphologically enhanced scheduling state representation is used as the input scheduling event. The causal gating layer performs joint prediction of the scheduling state to generate multi-source scheduling prediction results. The multi-source scheduling prediction results are input into the power flow projection layer to perform power flow safety projection, energy storage lifetime constraints and capacity calculation, and generate a safety scheduling capacity association table. Based on the security scheduling capability association table and the scheduling form risk calibration table, perform uncertainty risk hierarchical output and generate a scheduling risk correction mark table; A smart scheduling strategy is generated based on the security scheduling capability association table and the scheduling risk correction flag table, and the execution results of the smart scheduling strategy are written into the historical scheduling execution data.

[0021] In this embodiment, generating a trusted aligned power state frame includes: Collect power generation operation records, new energy output records, energy storage status records, load power consumption records, and power grid operation records to generate power energy operation data; Collect scheduling instruction numbers, execution times, planned adjustment amounts, actual response amounts, and execution deviation records to generate historical scheduling execution data; The power energy operation data and historical dispatch execution data are mapped, dimensionally unified, and sampled according to the dispatch time axis to generate raw power energy data. The raw power energy data is subjected to missing data marking, anomaly marking, time alignment, channel credibility marking, topology node binding, and line topology marking to generate a trusted aligned power status frame carrying node binding relationships and line topology records. The trusted aligned power status frame includes scheduling time axis number, sampling time, power channel number, channel type, generation operation status field, new energy output status field, energy storage status field, load power consumption status field, grid operation status field, historical scheduling execution field, demand response status field, missing data marking, anomaly marking, sampling marking, channel credibility marking, node binding relationship, line topology record, and execution deviation marking.

[0022] In this embodiment, generating the scheduling morphology risk calibration table includes: Read the time-series status records of each power channel from the trusted aligned power state frame, perform multi-scale window segmentation according to channel type, scheduling cycle, and node binding relationship, and generate a channel scheduling window sequence, where: Read the time-series status records of each power channel from the trusted aligned power state frame, and perform multi-scale window segmentation according to channel type, scheduling cycle, and node binding relationship, specifically: Read the scheduling time axis number, sampling time, power channel number, channel type, generation operation status field, new energy output status field, energy storage status field, load power consumption status field, power grid operation status field, and node binding relationship from the trusted aligned power status frame; According to the power channel number and sampling time, the power generation operation status field, new energy output status field, energy storage status field, load power consumption status field and power grid operation status field are sorted to generate time-series status records for each power channel; According to the channel type, the time sequence status records of each power channel are divided into power generation channel, new energy channel, energy storage channel, load channel and line channel; The execution window for the time-series status record of each power channel is segmented using the scheduling cycle, half of the scheduling cycle, and two consecutive scheduling cycles as the basic scale, short scale, and long scale, respectively. The segmented window fragments are bound to the corresponding nodes according to the node binding relationship, and the scheduling time axis number, power channel number, channel type, window scale marker, window start and end time and state value sequence within the window are associated to generate a channel scheduling window sequence. Perform morphological symbol mapping on the channel scheduling window sequence to generate a single-channel morphological symbol string carrying amplitude level, direction level, and response stage markers, where: Perform morphological symbol mapping on the channel scheduling window sequence, specifically as follows: Read the power channel number, channel type, window scale marker, window start and end time, and window status value sequence from the channel scheduling window sequence; Calculate the window mean, window maximum, window minimum, window first value, and window last value of the sequence of state values ​​within the window; The difference between the maximum and minimum values ​​of the window is divided by the absolute value of the window mean to generate the amplitude change rate. When the amplitude change rate is less than or equal to 2%, it is marked as low amplitude level; when the amplitude change rate is greater than 2% and less than or equal to 5%, it is marked as medium-low amplitude level; when the amplitude change rate is greater than 5% and less than or equal to 10%, it is marked as medium-high amplitude level; when the amplitude change rate is greater than 10%, it is marked as high amplitude level. The first value of the window is subtracted from the last value of the window, and then divided by the absolute value of the first value of the window to generate the rate of change between the first and last values. When the absolute value of the first and last change rate is less than or equal to 1%, it is marked as a stable direction level; when the first and last change rate is greater than 1%, it is marked as an upward direction level; when the first and last change rate is less than -1%, it is marked as a downward direction level. Based on the window start time being located in the first 1 / 3, middle 1 / 3, and last 1 / 3 of the scheduling cycle, generate start response phase markers, continuous response phase markers, and termination response phase markers respectively. Associate the power channel number, channel type, window scale marker, amplitude level, direction level, and response stage marker to generate a single channel morphology symbol string; Based on the cross-channel scheduling coupling symbol constraint mechanism, single-channel morphological symbol strings within the same scheduling window are used to construct a cross-channel symbol constraint graph according to node binding relationships, generating a scheduling coupling symbol group, where: The cross-channel scheduling coupling symbol constraint mechanism is as follows: Read the power channel number, channel type, window scale marker, amplitude level, direction level, and response stage marker from the single-channel morphology symbol string; Read node binding relationships, line topology records, and scheduling timeline numbers from trusted aligned power status frames; Based on the scheduling time axis number, window scale mark and node binding relationship, the single channel form symbol string corresponding to the load channel, new energy channel, energy storage channel, line channel and demand response channel belonging to the same node within the same scheduling window is established as a symbol node; Establish a supply-demand coupling edge based on load channels and new energy channels; Establish an energy storage support edge based on the energy storage channel and the load channel; Establish a consumption and transmission edge based on new energy channels and transmission lines; Establish a response peak-shaving edge based on the demand response channel and the load channel; Power transfer edges between adjacent nodes are established based on line topology records; By associating symbol nodes, supply and demand coupling edges, energy storage support edges, absorption and transmission edges, response peak shaving edges, and power transfer edges, a cross-channel symbol constraint graph is generated. Extract symbol combinations that simultaneously satisfy high amplitude level, rising direction level, insufficient energy storage support, increased line load, and insufficient demand response from the cross-channel symbol constraint diagram to generate scheduling coupling symbol groups; Execution deviation markers are read from trusted aligned power state frames. Based on these execution deviation markers, risk level recalibration and symbol threshold calibration are performed on the scheduling coupled symbol group, generating coupled risk labels and calibration symbol thresholds, where: Based on the execution deviation label, the risk level recalibration and symbol threshold calibration of the scheduling coupled symbol group are performed as follows: Read the execution deviation flag from the trusted aligned power state frame, and read the scheduling instruction number, execution time, planned adjustment amount and actual response amount from the historical scheduling execution field in the trusted aligned power state frame; Read the scheduling window number, node number, power channel number, amplitude level, direction level, and response stage marker from the scheduling coupling symbol group; The execution deviation markers are matched with the scheduling coupling symbol group according to the execution time, scheduling window number, node number, and power channel number; The execution deviation rate is generated by dividing the absolute value of the difference between the actual response and the planned adjustment by the absolute value of the planned adjustment. When the execution deviation rate is less than or equal to 5%, a low-risk coupling label is generated; When the execution deviation rate is greater than 5% and less than or equal to 10%, a medium-risk coupling label is generated; When the execution deviation rate is greater than 10%, a high-risk coupling label is generated; The average execution deviation rate is calculated based on the same power channel number, the same amplitude level, and the same coupled risk label. Multiply the original amplitude level boundary by 0.7 to generate the original boundary retention value; multiply the average execution deviation rate by 0.3 to generate the deviation feedback correction value; add the original boundary retention value and the deviation feedback correction value to generate the calibration symbol threshold. Low-risk coupling labels, medium-risk coupling labels, and high-risk coupling labels are grouped into coupling risk labels; Establish a correspondence between the coupling risk label and the calibration symbol threshold according to the power channel number, amplitude level, and coupling risk label; The scheduling morphology risk calibration table is generated by associating the channel scheduling window sequence, single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label, execution deviation mark and calibration symbol threshold.

[0023] In this embodiment, the generation of causal gating channel weights includes: An improved SOFTS model is established, connecting the morphological symbol guidance fusion layer, the scheduling event causal gating layer, and the power grid flow projection layer according to the data transmission sequence. Read the time-series state records of each power channel in the trusted aligned power state frame, and read the single-channel morphological symbol string, scheduling coupling symbol group, coupling risk label, and calibration symbol threshold from the scheduling morphological risk calibration table to generate the model input record, wherein: Model input records, specifically including: The scheduling time axis number, sampling time, power channel number, channel type, generation operation status field, new energy output status field, energy storage status field, load power consumption status field, power grid operation status field, channel credibility flag, node binding relationship, line topology record, and execution deviation flag are all from the trusted aligned power status frame. The channel scheduling window sequence, single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label, and calibration symbol threshold from the scheduling morphology risk calibration table; Based on the scheduling time axis number, power channel number, channel type and node binding relationship, a correspondence is established between the status field in the trusted aligned power status frame and the morphological calibration data in the scheduling morphological risk calibration table to generate model input records; Based on the model input records, channel representation encoding and event trigger matching are performed in the scheduling event causal gating layer to generate a scheduling event matching table, where: The event causal gating layer for scheduling includes: Dispatch representation reading end: Reads the model input record and extracts the generation channel representation, new energy channel representation, energy storage channel representation, load channel representation and line channel representation according to the power channel number and channel type; Event Delay Matching Area: Based on the scheduling channel prediction representation set, identify the lag response relationship between different channels and generate an event delay association table; Causal weight generation channel: The weights of the time-delay association table of related events and the causal gating channel are used to generate causal prediction weights for events; Prediction record output: Based on the event causal prediction weights, gated joint prediction is performed on the scheduling channel prediction representation set, and the power channel prediction record is output. Generate a scheduling event matching table, specifically as follows: Through the scheduling representation reading end in the scheduling event causal gating layer, the scheduling time axis number, sampling time, power channel number, channel type, power generation operation status field, new energy output status field, energy storage status field, load power consumption status field, power grid operation status field, node binding relationship, line topology record, single channel morphological symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold are read from the model input record; According to the power channel number, channel type and sampling time, the power generation operation status field, new energy output status field, energy storage status field, load power consumption status field and power grid operation status field are encoded to generate power generation channel representation, new energy channel representation, energy storage channel representation, load channel representation and line channel representation; Embed the single-channel morphological symbol string, coupled risk label, and calibration symbol threshold into the corresponding channel representation to generate a morphological risk constraint channel representation. By using the event time delay matching zone, sampling points with amplitude levels of medium to low amplitude levels and direction levels of rising or falling direction levels are extracted from the morphological risk constraint channel representation to generate channel event points. Based on the node binding relationship, load channel event points, new energy channel event points, energy storage channel event points, line channel event points and power generation channel event points within the same node are grouped into the same node event set; The event points of the load channel and the new energy channel are used as trigger event points, and the event points of the energy storage channel, the line channel, and the power generation channel are used as response event points. Within the same node event set, the sampling time of the response event point is subtracted from the sampling time of the trigger event point to generate the event delay value; When the event delay value is greater than or equal to 0 and less than or equal to 1 scheduling cycle, the trigger event point and the response event point will be established as a candidate delay event pair; According to the line topology record, the same event time delay value calculation is performed on the line channel event points and power generation channel event points between adjacent nodes to generate cross-node candidate time delay event pairs; The amplitude level, direction level, coupling risk label and event time delay value in the candidate time delay event pairs are matched and verified, and the event pairs corresponding to load increase and energy storage discharge, load increase and line load increase, new energy output decrease and power generation regulation enhancement, and new energy output decrease and energy storage discharge are retained. The event pairs, node numbers, power channel numbers, sampling times, event time delay values, and coupling risk labels are associated and retained to generate a scheduling event matching table; Based on the scheduling event matching table, gating weights are assigned to each power channel to generate causal gating channel weights, where: Gating weights are assigned to each power channel based on the scheduling event matching table, specifically as follows: By generating the causal weight in the causal gating layer of the scheduling event, the node number, power channel number, scheduling window number, event delay value, event matching count, and coupling risk label are read from the scheduling event matching table; Risk coefficients are set according to the coupling risk labels: 0.3 for low-risk coupling labels, 0.6 for medium-risk coupling labels, and 1.0 for high-risk coupling labels. The delay coefficient is set according to the event delay value. When the event delay value is less than or equal to 1 / 3 of the scheduling cycle, it corresponds to 1.0. When the event delay value is greater than 1 / 3 of the scheduling cycle and less than or equal to 2 / 3 of the scheduling cycle, it corresponds to 0.7. When the event delay value is greater than 2 / 3 of the scheduling cycle and less than or equal to 1 scheduling cycle, it corresponds to 0.4. The channel gating score is generated by multiplying the risk coefficient, the time delay coefficient, and the number of event matches. Divide the channel gating score of each power channel within the same scheduling window by the sum of all channel gating scores to generate causal gating channel weights. The improved SOFTS model is trained by reading trusted aligned power state frames, scheduling morphology risk calibration tables, and historical scheduling execution data. The trusted aligned power state frames and scheduling morphology risk calibration tables are input into the improved SOFTS model, and forward computation yields multi-source scheduling prediction results, a safe scheduling capability correlation table, and a scheduling risk correction label table. A combination of scheduling state prediction error, event causality gating error, morphology symbol fusion error, power flow safety projection error, energy storage lifetime constraint error, scheduling risk level discrimination error, and scheduling correction triggering error is used as the joint optimization objective. The network parameters of the morphology symbol guidance fusion layer, the scheduling event causality gating layer, and the power flow projection layer are continuously optimized. When the change in the joint loss value over five consecutive training rounds is less than 0.001, the improved SOFTS model is considered to have completed convergence training. The improved SOFTS model is trained as follows: Read the scheduling timeline number, sampling time, power channel number, channel type, generation operation status field, new energy output status field, energy storage status field, load power consumption status field, power grid operation status field, node binding relationship, line topology record and execution deviation flag from the trusted aligned power status frame; Read the channel scheduling window sequence, single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold from the scheduling morphology risk calibration table; Read the scheduling instruction number, execution time, planned adjustment amount, actual response amount and execution deviation record from historical scheduling execution data, and generate the actual scheduling risk level and actual correction trigger mark based on the execution deviation record; The trusted aligned power state frame and the scheduling morphology risk calibration table are input into the improved SOFTS model. The morphology-enhanced scheduling state representation is generated by the state frame reading end, morphology symbol index area, coupled risk gating channel and core state write-back end in the fusion layer through morphology symbol guidance. The causal gating channel weights and multi-source scheduling prediction results are generated by the scheduling representation reading end, event time delay matching area, causal weight generation channel and prediction record output end in the causal gating layer of the scheduling event; A safe dispatch capability association table is generated through the power flow constraint reading end, safe projection calculation area, energy storage boundary verification channel and capacity matrix output end in the power grid power flow projection layer, and a dispatch risk correction mark table is also generated. The difference between the multi-source scheduling prediction result and the actual state value of the corresponding future scheduling window is calculated. The average value of the squared difference is then used to obtain the scheduling state prediction error. The difference between the causal gating channel weight and the channel response ratio obtained based on the actual response quantity statistics is used to obtain the average value of the squared difference, and the event causal gating error is obtained. The difference between the morphological level corresponding to the morphological enhancement scheduling state characterization and the coupling risk label is calculated. The average value of the squared level difference is then obtained to obtain the morphological symbol fusion error. The power flow safety projection result, energy storage schedulable boundary record, scheduling risk level and scheduling correction trigger mark corresponding to the safety scheduling capability association table are subtracted from the line topology record, energy storage status field, real scheduling risk level and actual correction trigger mark, respectively, to generate power flow safety projection error, energy storage lifetime constraint error, scheduling risk level discrimination error and scheduling correction trigger error. The errors are accumulated according to preset weights to generate a joint loss value. The gradient value corresponding to the joint loss value is multiplied by the learning rate of 0.001 and the network parameters are updated until the change of the joint loss value in 5 consecutive training rounds is less than 0.001. The improved SOFTS model is then considered to have completed convergence training.

[0024] In this embodiment, the improved SOFTS model includes: The SOFTS model retains the core aggregation structure of the sequence, the channel state redistribution structure, the time series prediction backbone and the prediction output structure, and makes structural modifications to the input expression, morphological symbol fusion, event causal gating and power flow security constraints in the context of intelligent power energy dispatching. A trusted aligned power state frame input is introduced at the model input end, transforming the ordinary multivariate time series input into a power dispatch state input form that carries channel type, scheduling cycle, node binding relationship, channel trustworthiness flag and execution deviation flag; A morphological symbol-guided fusion layer is added at the core aggregation position of the sequence, embedding the single-channel morphological symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold from the scheduling morphological risk calibration table into the channel fusion process; The channel weight allocation method is improved to form a scheduling event causal gating layer, and the ordinary channel weight allocation method is transformed into a causal gating allocation method under the constraint of the event time delay association table. A power flow projection layer is added to the back end of the prediction output. The multi-source scheduling prediction results are input into the power flow security projection process, so that node power balance, line capacity, voltage security, regional exchange boundary and energy storage lifetime constraints participate in the selection of safe and feasible scheduling states. The output organization method is improved by expanding the model output target to include causal gating channel weights, morphologically enhanced scheduling state representation, multi-source scheduling prediction results, and a correlation table of safe scheduling capabilities.

[0025] In this embodiment, the generation of morphologically enhanced scheduling state representation includes: Read the single-channel morphological symbol string, scheduling coupling symbol group, coupling risk label, and calibration symbol threshold from the scheduling morphological risk calibration table, input them into the morphological symbol guidance fusion layer, and generate morphological symbol index records according to the power channel number, where: The morphological symbol-guided fusion layer specifically includes: Status frame reader: Reads the power channel number, channel type, sampling time, channel confidence flag, and node binding relationship from the trusted aligned power status frame; Morphological symbol index area: Read the single-channel morphological symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold from the scheduling morphological risk calibration table, and establish morphological symbol index records according to the power channel number; Coupling risk gating channel: Match the weights of causal gating channels with the morphological symbol index records, and adjust the fusion strength of different power channels according to the coupling risk labels; Core state write-back end: Performs core aggregation and state reallocation on the adjusted power channel state, and enhances the representation of scheduling state in the output form; Generate morphological symbol index records according to the power channel number, specifically as follows: The morphological symbol index area in the fusion layer is guided by morphological symbols, and the single-channel morphological symbol string, scheduling coupling symbol group, coupling risk label, execution deviation mark and calibration symbol threshold are read from the scheduling morphological risk calibration table. The single-channel morphological symbol strings are indexed and sorted according to the power channel number, channel type, dispatch window number, and node number; Write the scheduling coupling symbol group, coupling risk label, and calibration symbol threshold into the index position under the corresponding power channel number; Associate the power channel number, channel type, dispatch window number, node number, single channel morphological symbol string, dispatch coupling symbol group, coupling risk label, and calibration symbol threshold to generate a morphological symbol index record; The causal gating channel weights are matched with the morphological symbol index records to generate morphological gating fusion weights carrying coupling risk constraints, where: Generate morphological gating fusion weights carrying coupling risk constraints, specifically as follows: The coupling risk gating channel in the fusion layer is guided by morphological symbols, the weight of the causal gating channel and the morphological symbol index record are read, and the channel credibility tag is read from the trusted aligned power state frame. The causal gating channel weights are matched with the morphological symbol index records according to the power channel number, dispatch window number, and node number; The morphological risk adjustment coefficient is set according to the coupling risk label, with 0.3 corresponding to low-risk coupling label, 0.6 corresponding to medium-risk coupling label, and 1.0 corresponding to high-risk coupling label; The channel fusion score is generated by multiplying the causal gating channel weight, the morphological risk adjustment coefficient, and the channel credibility label. Divide the channel fusion score of each power channel within the same scheduling window by the sum of all channel fusion scores to generate the morphological gating fusion weight; Based on morphological gating fusion weights, a weighted core aggregation is performed on the corresponding power channel states in the trusted aligned power state frames to generate the scheduling core state, where: Perform weighted core aggregation on the corresponding power channel states in the trusted aligned power state frames, specifically as follows: By guiding the state frame reading end in the fusion layer with morphological symbols, the power channel number, channel type, sampling time, power generation operation status field, new energy output status field, energy storage status field, load power consumption status field, power grid operation status field, and node binding relationship are read from the trusted aligned power state frame. Extract the status of the corresponding power channel according to the power channel number and sampling time; By coupling the risk-gated channel, the power channel state and the morphological gating fusion weight are multiplied together to generate a weighted channel state; The weighted channel states are summed according to the same node number and the same scheduling window number to generate the scheduling core state; Based on the reallocation of the scheduling core state and the scheduling coupled symbol group execution state, a channel-specific enhanced state record is generated, in which: The reallocation of execution states based on the scheduling core state and the scheduling coupled symbol group is as follows: The core state write-back endpoint in the fusion layer is guided by morphological symbols to read the scheduling core state, scheduling coupling symbol group, and morphological symbol index record. Read the power channel number, amplitude level, and direction level of each power channel under the same scheduling window and the same node number from the scheduling coupling symbol group; Read the coupling risk labels corresponding to the power channel number, dispatch window number, and node number from the morphological symbol index record; The amplitude coefficient is set according to the amplitude level: 0.25 for low amplitude level, 0.50 for medium-low amplitude level, 0.75 for medium-high amplitude level, and 1.00 for high amplitude level. Set the direction coefficient according to the direction level: 0.30 for a steady direction level, 1.00 for an ascending direction level, and 0.70 for a descending direction level. Risk coefficients are set according to the coupling risk labels: 0.30 for low-risk coupling labels, 0.60 for medium-risk coupling labels, and 1.00 for high-risk coupling labels. The amplitude coefficient, direction coefficient, and risk coefficient corresponding to the same power channel are multiplied to generate a channel redistribution score; Divide the channel redistribution score of each power channel under the same scheduling window and the same node number by the sum of all channel redistribution scores to generate the status redistribution ratio; The scheduling core status is written back to the corresponding power channel according to the status redistribution ratio, generating the power channel status after writing back. Associate the power channel status, scheduling window number, node number, power channel number, status redistribution ratio, and coupling risk label after the write-back to generate sub-channel enhanced status records; By associating the channel-specific enhanced state records, scheduling core states, and morphological gating fusion weights, a morphological enhanced scheduling state representation is generated, wherein: Generate morphological enhancements to the scheduling state representation, specifically: Read the enhanced status records of the sub-channels, schedule the core status, and the morphological gating fusion weights; Establish a correspondence between the sub-channel enhanced status records and the core dispatch status according to the dispatch window number, node number, and power channel number; Write the morphological gating fusion weights into the enhanced state positions of the corresponding power channels; Associate the enhanced status records of the associated channels, the core status of the scheduling, the morphological gating fusion weight, the node number, and the scheduling window number to generate a morphological enhanced scheduling status representation.

[0026] In this embodiment, generating multi-source scheduling prediction results includes: The system reads the enhanced scheduling status representation and extracts load forecasting representation, renewable energy forecasting representation, energy storage forecasting representation, line forecasting representation, and demand response forecasting representation according to power channel type and scheduling window number, generating a scheduling channel forecasting representation set, in which: Generate a scheduling channel prediction representation set, specifically as follows: By reading the scheduling representation in the causal gating layer of the scheduling event, the sub-channel enhanced status record, scheduling core status, morphological gating fusion weight, node number, scheduling window number, power channel number and channel type in the morphological enhanced scheduling status representation are read. According to the channel type, the channel enhancement status records are divided into load enhancement status, new energy enhancement status, energy storage enhancement status, line enhancement status and demand response enhancement status; The load enhancement state, renewable energy enhancement state, energy storage enhancement state, line enhancement state and demand response enhancement state are associated with the scheduling core state and the morphological gating fusion weight, respectively, to generate load forecasting representation, renewable energy forecasting representation, energy storage forecasting representation, line forecasting representation and demand response forecasting representation. The associated load forecasting representation, new energy forecasting representation, energy storage forecasting representation, line forecasting representation, demand response forecasting representation, node number and scheduling window number are used to generate a scheduling channel forecasting representation set; Based on the scheduling channel prediction representation set, execution event time delay matching is performed to generate an event time delay association table, where: Based on the scheduling channel prediction representation set, the execution event time delay matching is performed as follows: By using the event time delay matching area in the scheduling event causal gating layer, the node number, scheduling window number, load forecasting representation, new energy forecasting representation, energy storage forecasting representation, line forecasting representation, and demand response forecasting representation are read from the scheduling channel forecasting representation set. Extract load rise locations from the load forecasting representation and extract renewable energy output decline locations from the renewable energy forecasting representation. Mark the load rise locations and renewable energy output decline locations as trigger event points. Extract the locations of enhanced energy storage discharge from the energy storage prediction characterization, extract the locations of increased line load from the line prediction characterization, and extract the locations of enhanced demand response from the demand response prediction characterization. Mark the locations of enhanced energy storage discharge, increased line load, and enhanced demand response as response event points. Triggering event points and response event points are paired according to the same node number and the same scheduling window number; Subtract the sampling time corresponding to the trigger event point from the sampling time corresponding to the response event point to generate the event delay value; When the event delay value is greater than or equal to 0 and less than or equal to 1 scheduling cycle, the corresponding trigger event point and response event point are retained. Generate an event delay association table by associating node number, scheduling window number, trigger event point, response event point, event delay value, and power channel number; The event causal prediction weights are generated by using a time-delay association table and causal gating channel weights, where: The weights for generating causal predictions for events are as follows: By generating a causal weight in the causal gating layer of the scheduling event, the node number, scheduling window number, power channel number, event delay value, and trigger event point in the event delay association table are read. Read the power channel number, dispatch window number, and channel weight value from the causal gating channel weight; Match the event time delay association table with the causal gating channel weights according to the node number, scheduling window number, and power channel number; When the event delay value is less than or equal to 1 / 3 of the scheduling period, the delay decay coefficient is set to 1.0; When the event delay value is greater than 1 / 3 of the scheduling period and less than or equal to 2 / 3 of the scheduling period, the delay attenuation coefficient is set to 0.7. When the event delay value is greater than 2 / 3 of the scheduling cycle but less than or equal to 1 scheduling cycle, the delay attenuation coefficient is set to 0.4. Multiply the channel weight value by the time delay decay coefficient to generate the event causal prediction weight; Gated joint prediction is performed on the scheduling channel prediction representation set based on event causal prediction weights to generate power channel prediction records, where: Gated joint prediction is performed on the scheduling channel prediction representation set based on event causal prediction weights, specifically as follows: By reading the load forecast representation, new energy forecast representation, energy storage forecast representation, line forecast representation, demand response forecast representation, node number, dispatch window number and power channel number from the forecast record output terminal in the dispatch event causal gating layer; Read the node number, scheduling window number, power channel number, and event causal prediction weight value from the event causal prediction weight; Write the event causal prediction weight value into the corresponding prediction representation position according to the node number, scheduling window number, and power channel number; The load forecasting representation, renewable energy forecasting representation, energy storage forecasting representation, transmission line forecasting representation, and demand response forecasting representation are multiplied by the corresponding event causal forecasting weight values ​​to generate weighted load forecasting representation, weighted renewable energy forecasting representation, weighted energy storage forecasting representation, weighted transmission line forecasting representation, and weighted demand response forecasting representation. The characteristic mean values ​​of the weighted load forecast, weighted new energy forecast, weighted energy storage forecast, weighted line forecast, and weighted demand response forecast within the same scheduling window are calculated respectively, and the load forecast, new energy output forecast, energy storage charging and discharging forecast, line load forecast, and demand response forecast are generated. By associating load forecasts, renewable energy output forecasts, energy storage charging and discharging forecasts, line load forecasts, demand response forecasts, node numbers, dispatch window numbers, and power channel numbers, a power channel forecast record is generated. Based on power channel forecast records, supply and demand deviation calculations are performed according to node binding relationships, generating node supply and demand deviation forecast records. These records are then linked to generate multi-source scheduling forecast results, including: The multi-source scheduling prediction results are generated as follows: Read the load forecast, renewable energy output forecast, energy storage charging and discharging forecast, line load forecast, demand response forecast, node number, dispatch window number and power channel number from the power channel forecast record; Read node binding relationships and line topology records from trusted aligned power status frames; Based on the node binding relationship, the load forecast, new energy output forecast, energy storage charging and discharging forecast, line load forecast, and demand response forecast under the same node number and the same scheduling window number are aggregated. Based on the positive and negative directions of the energy storage charging and discharging prediction values, the energy storage charging and discharging prediction values ​​are divided into energy storage discharging prediction values ​​and energy storage charging prediction values. Based on the line connection direction in the line topology record, the line load prediction value is divided into line input prediction value and line output prediction value; The predicted values ​​of new energy output, energy storage discharge, and line input are added together to generate the predicted value of node supply. The nodal demand forecast is generated by subtracting the demand response forecast from the load forecast, energy storage charging forecast, and line output forecast. Subtract the node supply forecast from the node demand forecast to generate the node supply-demand deviation forecast. Associate node number, scheduling window number, node supply forecast, node demand forecast, node supply-demand deviation forecast, and line load forecast to generate a node supply-demand deviation forecast record; By associating power channel forecast records and node supply-demand deviation forecast records, multi-source scheduling forecast results are generated.

[0027] In this embodiment, generating the security scheduling capability association table includes: The power flow projection layer reads power channel prediction records and node supply-demand deviation prediction records from multi-source scheduling prediction results, and reads node binding relationships and line topology records from trusted aligned power state frames to generate power flow projection input records, including: The power grid flow projection layer specifically includes: Power flow constraint reading end: Reads power channel prediction records and node supply and demand deviation prediction records from the multi-source scheduling prediction results, and reads node binding relationships and line topology records from the trusted aligned power state frames; Safety projection calculation area: Based on node power balance, line capacity, voltage safety and regional switching boundary, perform power flow safety projection to generate a safe and feasible scheduling state; Energy storage boundary verification channel: Extract the energy storage state prediction field from the power channel prediction record, combine it with the energy storage state field in the trusted aligned power state frame to perform energy storage lifetime constraint verification, and generate energy storage dispatchable boundary record; Capability matrix output: Calculate the energy unit dispatchable capability matrix based on the safe and feasible dispatchable state and energy storage dispatchable boundary record, and generate a safe dispatchable capability association table; The trend projection input record is as follows: Through the power flow constraint reading terminal in the power flow projection layer, power channel prediction records and node supply and demand deviation prediction records are read from the multi-source scheduling prediction results. Read the load forecast, renewable energy output forecast, energy storage charging and discharging forecast, line load forecast, demand response forecast, node number, dispatch window number and power channel number from the power channel forecast record; Read the node supply forecast value, node demand forecast value, and node supply-demand deviation forecast value from the node supply-demand deviation forecast record; Read node binding relationships and line topology records from trusted aligned power status frames; Based on the node number, scheduling window number, and power channel number, establish a correspondence between the power channel forecast record, node supply and demand deviation forecast record, node binding relationship, and line topology record, and generate power flow projection input record; The power flow projection input record is input into the power flow projection layer of the power grid to perform node power balance verification, line capacity verification, voltage safety verification, and regional exchange boundary verification, generating a power flow safety verification record, in which: Generate power flow security verification records, specifically as follows: Through the safe projection calculation area in the power grid power flow projection layer, read the node number, scheduling window number, node supply forecast value, node demand forecast value, node supply and demand deviation forecast value, line load forecast value, node binding relationship and line topology record from the power flow projection input record; Read the line capacity limit, node rated voltage, voltage safety limit, voltage safety lower limit, line impedance parameters, and area exchange boundary from the line topology record; The node power balance difference is generated by subtracting the node demand forecast from the node supply forecast. The node power balance deviation rate is generated by dividing the absolute value of the node power balance difference by the absolute value of the node demand forecast. When the node power balance deviation rate is less than or equal to 5%, node power balance is generated by marking. When the node power balance deviation rate is greater than 5%, a node power balance over-limit flag is generated; Divide the predicted line load by the upper limit of line capacity to generate the line capacity utilization rate. When the line capacity utilization rate is less than or equal to 100%, the generated line capacity is marked. When the line capacity utilization rate is greater than 100%, a line capacity over-limit flag is generated; Multiply the predicted supply-demand deviation at the node by the line impedance parameter to generate the estimated node voltage offset. The node voltage prediction value is generated by adding the node rated voltage to the node voltage offset estimate; When the predicted node voltage value is greater than or equal to the lower voltage safety limit and less than or equal to the upper voltage safety limit, a voltage safety pass flag is generated; When the predicted node voltage is less than the lower voltage safety limit or greater than the upper voltage safety limit, a voltage safety limit violation flag is generated. Based on the regional connection relationships in the line topology record, the cross-regional line load prediction values ​​are accumulated to generate regional exchange prediction values. The predicted area swap values ​​are compared with the area swap boundaries to generate area swap boundary verification markers. The associated node power balance pass mark, node power balance over-limit mark, line capacity pass mark, line capacity over-limit mark, voltage safety pass mark, voltage safety over-limit mark, and area exchange boundary verification mark are used to generate a power flow safety verification record; Based on the power channel prediction records in the multi-source scheduling prediction results, energy storage state prediction records are extracted. Energy storage operation boundary records are read from the trusted aligned power state frames. Energy storage lifetime constraint verification is performed to generate energy storage dispatchable boundary records, where: Generate a schedulable boundary record for energy storage, specifically as follows: Through the energy storage boundary verification channel in the power grid power flow projection layer, the predicted values ​​of energy storage charging and discharging, node number, scheduling window number and power channel number are extracted from the power channel prediction record in the multi-source scheduling prediction results. Read the energy storage state of charge, energy storage health status, energy storage temperature, energy storage cycle count, and energy storage available capacity from the energy storage state field in the trusted aligned power state frame; Values ​​greater than 0 in the predicted energy storage charge and discharge values ​​are marked as predicted energy storage discharge values, and values ​​less than 0 in the predicted energy storage charge and discharge values ​​are marked as predicted energy storage charge values. Set 20% as the lower limit of energy storage charge state and 90% as the upper limit of energy storage charge state; When the energy storage state of charge is less than or equal to 20%, the upper limit of discharge power is set to 0. When the energy storage state of charge is greater than or equal to 90%, the upper limit of charging power is set to 0. When the energy storage state of charge is greater than 20% and less than 90%, the upper limit of charging power and the upper limit of discharging power are calculated based on the available energy storage capacity and the scheduling cycle. When the energy storage health status is greater than or equal to 90%, the energy storage temperature is between 15℃ and 35℃, and the energy storage cycle count is less than or equal to 3000, the life protection factor is set to 1.0. When the energy storage health status is greater than or equal to 80% and less than 90%, the energy storage temperature is between 10℃ and 45℃, and the energy storage cycle count is less than or equal to 5000, the life protection factor is set to 0.7. When the energy storage health status is less than 80%, the energy storage temperature is below 10℃, the energy storage temperature is above 45℃, and the energy storage cycle count is greater than 5000, the life protection factor is set to 0.4. Multiply the upper limit of charging power and the upper limit of discharging power by the life protection factor respectively to generate the energy storage operation boundary record; When the predicted energy storage charge and discharge values ​​are within the energy storage operation boundary record, the predicted energy storage charge and discharge values ​​are retained. When the predicted value of energy storage charging and discharging exceeds the energy storage operation boundary record, the predicted value of energy storage charging and discharging will be clipped to the nearest boundary value of the energy storage operation boundary record. By associating the retained energy storage charge and discharge prediction values, the trimmed energy storage charge and discharge prediction values, the energy storage operation boundary record, the node number, the scheduling window number, and the power channel number, a dispatchable energy storage boundary record is generated. Based on the power flow safety verification record and the energy storage schedulable boundary record, a safe and feasible state projection is performed to generate a safe and feasible scheduling state, wherein: The process of generating a safe and feasible scheduling state is as follows: Through the safety projection calculation area in the power flow projection layer, read the node number, scheduling window number, node power balance difference, line capacity occupancy rate, node voltage prediction value, regional exchange prediction value, node power balance over-limit mark, line capacity over-limit mark, voltage safety over-limit mark and regional exchange boundary verification mark in the power flow safety verification record; Read the trimmed energy storage charge and discharge prediction values ​​and energy storage operation boundary records from the energy storage dispatchable boundary records; Establish a constraint priority sequence based on node power balance constraints, line capacity constraints, voltage safety constraints, regional exchange boundary constraints, and energy storage operation boundary constraints; Write the node power balance difference, line capacity utilization rate, node voltage prediction, regional exchange prediction, and pruned energy storage charging and discharging prediction into the constraint state vector of the same scheduling window. Based on the out-of-limit marker, the out-of-limit component is extracted from the constraint state vector to generate the scheduling out-of-limit vector; According to the constraint priority sequence, the node power balance difference is projected to 0, the line capacity utilization rate is projected to within 100%, the node voltage prediction value is projected to between the upper and lower voltage safety limits, the regional exchange prediction value is projected to within the regional exchange boundary, and the pruned energy storage charging and discharging prediction value is projected to within the energy storage operation boundary record. Write the projected constraint state vector back to the corresponding node number and scheduling window number to generate a safe and feasible scheduling state. Based on the calculation of power generation regulation capacity, renewable energy absorption capacity, energy storage charging and discharging capacity, load response capacity, and line transmission margin under safe and feasible dispatch conditions, an energy unit dispatchable capacity matrix is ​​generated, in which: The energy unit dispatchability matrix is ​​generated as follows: Through the output terminal of the capacity matrix in the power grid power flow projection layer, read the node number, scheduling window number, projected node supply and demand deviation prediction value, projected line load prediction value and safe and feasible flag in the safe and feasible scheduling status. Read the predicted values ​​of new energy output and demand response from the power channel prediction records in the multi-source dispatch prediction results; Read the upper limit of power generation output, the lower limit of power generation output, and the current power generation output from the power generation operation status field in the trusted aligned power status frame; Subtracting the current power generation output from the upper limit of power generation output generates the power generation upward adjustment capacity; subtracting the lower limit of power generation output from the current power generation output generates the power generation downward adjustment capacity; combining the power generation upward adjustment capacity and the power generation downward adjustment capacity generates the power generation regulation capacity. Read the line capacity limit from the power flow safety verification record, subtract the projected line load prediction value from the line capacity limit, and generate the remaining transmission capacity of the line. The smaller of the predicted output of new energy sources and the remaining transmission capacity of the lines is taken as the new energy absorption capacity. Read the energy storage operation boundary record and the pruned energy storage charge and discharge prediction values ​​from the energy storage dispatchable boundary record; Subtract the clipped energy storage charge and discharge prediction values ​​from the upper boundary of the energy storage operation boundary record to generate the energy storage discharge capacity. Subtract the clipped energy storage charge and discharge prediction values ​​from the lower boundary of the energy storage operation boundary record to generate the energy storage charging capacity. Combine the energy storage discharge capacity and the energy storage charging capacity to generate the energy storage charge and discharge capacity. The smaller of the absolute values ​​of the demand response forecast and the projected nodal supply-demand deviation forecast is taken as the load response capacity. The remaining transmission capacity of the line is used as the line transmission margin. Based on node number, dispatch window number, and power channel number, the power generation regulation capacity, new energy absorption capacity, energy storage charging and discharging capacity, load response capacity, and line transmission margin are associated to generate an energy unit dispatchable capacity matrix; By associating the safe and feasible scheduling status, the energy unit scheduling capability matrix, the energy storage scheduling boundary record, the node binding relationship, and the power flow safety verification record, a safe scheduling capability association table is generated.

[0028] In this embodiment, generating the scheduling risk correction flag table includes: Read the safe and feasible scheduling status, energy unit scheduling capability matrix, power flow safety verification record, and energy storage scheduling boundary record from the safe scheduling capability association table; Power flow safety margin is generated based on power flow safety verification records, energy storage lifetime constraint margin is generated based on energy storage schedulable boundary records, and scheduling priority is generated based on the energy unit schedulable capability matrix. A safety capability risk record is then generated by associating the safe and feasible scheduling status, where: The associated security feasible scheduling state generates a security capability risk record, specifically: Read the node number, scheduling window number, node power balance deviation rate, line capacity utilization rate, node voltage prediction value, voltage safety upper limit, voltage safety lower limit, area exchange prediction value, and area exchange boundary from the power flow safety verification record. Subtract the node power balance deviation rate from 5% to generate the node power balance margin. Subtract the line capacity utilization rate from 100% to generate the line capacity margin. Subtract the voltage safety lower limit from the node voltage prediction value to generate the voltage lower limit margin; The voltage safety upper limit is subtracted from the node voltage prediction value to generate the voltage upper limit margin; The smaller value between the lower voltage margin and the upper voltage margin is used to generate the voltage safety margin; The area exchange margin is generated by subtracting the absolute value of the area exchange forecast from the area exchange boundary. Associating node power balance margin, line capacity margin, voltage safety margin, and regional switching margin, a power flow safety margin is generated; Read the energy storage operation boundary record and the pruned energy storage charge and discharge prediction values ​​from the energy storage dispatchable boundary record; The energy storage discharge margin is generated by subtracting the clipped energy storage charge and discharge prediction values ​​from the upper boundary of the energy storage operation boundary record. The energy storage charging margin is generated by subtracting the lower boundary of the energy storage operation boundary record from the clipped energy storage charging and discharging prediction values. The smaller value between the energy storage discharge margin and the energy storage charge margin is used to generate the energy storage lifetime constraint margin. The power generation regulation capacity, new energy consumption capacity, energy storage charging and discharging capacity, load response capacity, and line transmission margin are read from the energy unit dispatchability matrix. Arrange scheduling priorities by sorting energy units from largest to smallest scheduling capacity; Associate the safe and feasible scheduling status, power flow safety margin, energy storage lifetime constraint margin, scheduling priority, node number and scheduling window number to generate a safety capability risk record; Read the coupling risk label, calibration symbol threshold and execution deviation mark from the scheduling morphological risk calibration table to generate a morphological risk calibration record; Associate security capability risk records and morphological risk calibration records, perform uncertainty risk layering output, and generate scheduling risk levels, including: The uncertainty risk stratification output is implemented as follows: Read the node number, scheduling window number, power flow safety margin, energy storage lifetime constraint margin, and scheduling priority from the safety capability risk record; Read the coupling risk label, calibration symbol threshold, and execution deviation marker from the morphological risk calibration record; Match the security capability risk records with the morphological risk calibration records according to the node number and scheduling window number; The smaller value between the power flow safety margin and the energy storage lifetime constraint margin is used to generate the comprehensive safety margin; When the overall safety margin is greater than 10%, a safety margin score of 0.2 is generated. When the overall safety margin is greater than 5% and less than or equal to 10%, a safety margin score of 0.5 is generated. When the overall safety margin is less than or equal to 5%, a safety margin score of 1.0 is generated. The morphological risk score is set according to the coupling risk label, with 0.3 corresponding to low risk coupling label, 0.6 corresponding to medium risk coupling label, and 1.0 corresponding to high risk coupling label; When an execution deviation is marked as present, an execution deviation score of 1.0 is generated. When the execution deviation is marked as no deviation, an execution deviation score of 0.2 is generated; The safety margin score, morphological risk score, and execution deviation score are multiplied by 0.4 and then summed to generate a comprehensive scheduling risk score. When the comprehensive scheduling risk score is less than or equal to 0.4, a low scheduling risk level is generated; When the comprehensive scheduling risk score is greater than 0.4 and less than or equal to 0.7, a medium scheduling risk level is generated. When the comprehensive scheduling risk score is greater than 0.7, a high scheduling risk level is generated; Low scheduling risk level, medium scheduling risk level and high scheduling risk level are merged into scheduling risk level; Based on the scheduling risk level and scheduling priority, a correction trigger flag is generated, and a risk correction association record is established according to the energy unit number and scheduling window number; Associate scheduling risk levels, correction trigger flags, risk correction association records, and execution deviation flags to generate a scheduling risk correction flag table.

[0029] In this embodiment, the generation of intelligent scheduling strategies based on the security scheduling capability association table and the scheduling risk correction flag table includes: Read the safe and feasible scheduling status, energy unit scheduling capability matrix, energy storage scheduling boundary record and power flow security verification record from the safe scheduling capability association table, and generate the strategy capability input record. Read the scheduling risk level, scheduling correction trigger flag, and risk correction association record from the scheduling risk correction flag table to generate a strategy risk input record; Associated strategy capability input records and strategy risk input records, and generate scheduling instruction combinations according to energy unit number and scheduling window number, wherein: The scheduling instruction combination is generated according to the energy unit number and the scheduling window number, specifically as follows: Read the safe and feasible scheduling status, energy unit scheduling capability matrix, energy storage scheduling boundary record, power flow safety verification record, energy unit number and scheduling window number from the strategy capability input record; Read the scheduling risk level, scheduling correction trigger flag, and risk correction association record from the strategy risk input record; Match the strategy capability input records with the strategy risk input records according to the energy unit number and the dispatch window number; When the scheduling risk level is high and the scheduling correction trigger mark is triggered, the energy storage charging and discharging capacity, load response capacity and line transmission margin are read first to generate energy storage support instructions, load response instructions and line transfer instructions. When the dispatch risk level is medium, read the generation regulation capacity, new energy consumption capacity and energy storage charging and discharging capacity, and generate generation regulation instructions, new energy consumption instructions and energy storage charging and discharging instructions. When the scheduling risk level is low, read the renewable energy consumption capacity and line transmission margin, and generate renewable energy priority consumption instructions and regional power exchange instructions. Each dispatch instruction is sorted according to dispatch window number, energy unit number, instruction type, target adjustment amount, execution time, and risk correction association record to generate a dispatch instruction combination; A smart scheduling strategy is generated based on the combination of scheduling instructions. The system collects the corresponding scheduling instruction number, execution time, planned adjustment amount, actual response amount, and execution deviation record to generate the strategy execution result, including: Intelligent scheduling strategies are generated based on combinations of scheduling instructions, specifically as follows: Read the scheduling window number, energy unit number, instruction type, target adjustment amount, execution time, and risk correction association record from the scheduling instruction combination; Sort the scheduling instruction combinations by time according to the scheduling window number; Bind the scheduling instruction combinations within the same scheduling window to the energy unit number; According to the type of instruction, the dispatch instructions are divided into power generation regulation instructions, new energy consumption instructions, energy storage charging and discharging instructions, load response instructions, and regional power exchange instructions. When the risk correction association record corresponds to a high dispatch risk level, the energy storage charging and discharging command and the load response command will be arranged before the generation regulation command; When the risk correction association record corresponds to the dispatch risk level, the power generation regulation command, energy storage charging and discharging command and new energy consumption command will be arranged from largest to smallest according to the target regulation amount; When the risk correction association record corresponds to a low dispatch risk level, the new energy consumption instruction and the regional power exchange instruction will be arranged before the power generation regulation instruction. By associating scheduling window number, energy unit number, instruction type, target adjustment amount, execution time, and instruction order, a smart scheduling strategy is generated. Write the strategy execution results into the historical scheduling execution data.

[0030] Example 1: This example is applied to the inspection of surface cracks in underwater engineering lining plates. During the inspection, the underwater camera continuously moves along the surface of the lining plate, collecting a batch of lining plate image data. The image data contains a total of 1280 images, with a resolution of 1024×1024 pixels per image. Due to the underwater acquisition environment, the images exhibit significant water scattering noise, local reflections, grayscale attenuation, and edge blurring. The average grayscale contrast of the original images is 0.21. Manual sampling revealed that the crack widths are mainly distributed between 1 and 6 pixels, and approximately 37% of the crack areas show signs of breakage, occlusion, or edge adhesion. Traditional manual review methods require examining each image individually, which can easily miss small cracks; ordinary segmentation networks, under water noise interference, are prone to misjudging sediment edges and shadow textures as cracks, making it difficult to stably output continuous crack contours.

[0031] In this embodiment, the system first performs normalization processing on 1280 original images, scaling pixel values ​​to the 0–1 range, and uses a 5×5 median filter window to remove high-frequency noise. The average signal-to-noise ratio of the images before processing is 14.2dB, which is improved to 18.7dB after processing, and the average grayscale contrast ratio is improved from 0.21 to 0.34. The preprocessed images are input into the encoder for feature extraction. The convolutional feature extraction unit uses three scale convolution kernels: 3×3, 5×5, and 7×7, to extract multi-scale responses for fine crack edges, local textures, and wider damaged areas, generating 32-channel, 64-channel, and 96-channel feature maps, respectively. Subsequently, the images enter the depthwise separable convolution unit to decouple and compress the channel features, mapping the features of different scales to a unified 128-dimensional channel space, forming an initial spatial feature map with a size of 256×256×128.

[0032] In the sequence modeling stage, the system unfolds the initial spatial feature map row by row according to the horizontal serialization scanning strategy, converting 256×256 spatial locations into a sequence of length 65536, with each sequence location corresponding to a 128-dimensional feature vector. The system then unfolds column by column according to the vertical serialization scanning strategy to form the second sequence feature representation. A bidirectional cross-Seg-LSTM module aligns the spatial coordinates of the two sequences, pairing and fusing the positional features in the horizontal sequence with the features corresponding to the same spatial coordinates in the vertical sequence. Taking the sequence location corresponding to spatial coordinates (120, 87) as an example, the system extracts the current crack edge features from the horizontal sequence and reads the contextual features corresponding to the same coordinates from the vertical sequence, inputting both into the cross-gating calculation process. The mean of the cross-correlation vector at this location is 0.43, and the variance is 0.12, significantly higher than the mean of 0.18 in the background region, indicating that the crack structure exhibits a stable response in both the horizontal and vertical sequences.

[0033] During the forward sequence processing, the system continuously reads the historical state of the previous sequence position and combines it with cross-branch information in the shared matrix memory to generate the current cross-correlation vector. At sequence position 12000, the average cross-gating value is 0.67, indicating that the current state is more biased towards the new input features of the crack region. After fusion, the average response value of the cross-fused state vector of the crack region reaches 0.81, while the average response value of the background region is 0.29. After completing the forward traversal, the system performs reverse sequence processing to supplement the contextual relationships on both sides of the crack break position, increasing the information entropy of the final enhanced sequence feature representation from 3.12 to 4.87, and significantly enhancing the feature continuity of the crack break region.

[0034] The system then performs deserialization reconstruction on the enhanced sequence feature representation. During reconstruction, sequence positions are mapped back to two-dimensional spatial coordinates one by one. Sequence position 35000 is mapped back to spatial coordinates (137, 88), and the corresponding enhanced feature is filled into the corresponding position in the two-dimensional feature map. After all sequence positions are mapped, a transitional two-dimensional feature map with a size of 256×256×256 is formed. The system compresses the number of channels from 256 to 128 through pointwise convolution to generate an enhanced two-dimensional feature map, and adds it element-wise to the initial spatial feature map to form a global context enhanced feature map. After stacking, the average feature intensity of the crack region is increased by 27%, the background deposition texture response is reduced by 18%, and the feature discrimination between crack and non-crack regions is improved from 0.52 to 0.76.

[0035] In the decoding stage, the system performs the first resolution restoration. The edge confidence prediction head weights the enhanced feature map, with an average edge weight of 0.83 for the crack region and an average weight of 0.21 for the background region. After the weighted features are input into the PixelShuffle module, the output size is increased to 1024×1024×32. In the second resolution restoration stage, edge prediction is performed again, increasing the average weight of the crack region to 0.89. After channel expansion and PixelShuffle processing, the output size is 2048×2048×16. In the third resolution restoration stage, instead of increasing the spatial size, the crack edges are refined, resulting in a 41% increase in crack edge gradient and an increase in the integrity rate of fractured crack connections from 71.6% to 88.9%.

[0036] After entering the segmentation prediction stage, the crack segmentation prediction head performs classification mapping on each pixel location. At the crack location (512, 768), the original response value is 2.17, and after normalization mapping, the crack probability is 0.897; at the background location (200, 300), the probability value is 0.083. The system performs binarization processing based on a 0.5 threshold to obtain a crack binarized segmentation mask. 8423 crack pixels were identified in a single sample image, accounting for 0.40% of the total pixels. In the connected component labeling stage, the image was traversed using an eight-connectivity rule, identifying 12 independent crack connected components, with the largest crack connected component containing 3261 pixels. After six consecutive iterations of morphological thinning processing, the crack width was thinned to a single-pixel skeleton, reducing the total number of crack pixels from 8423 to 5127, while maintaining crack connectivity. When the system calculated the parameters for the third connected region, the maximum Euclidean distance was 312 pixels, and the total number of pixels was 1450. The calculated crack length was 312 pixels and the average width was 4.65 pixels. The final detection report generated 12 crack records, each containing the spatial distribution range, crack length, crack width, and connected region number.

[0037] To verify the effectiveness of this embodiment, the system compares the method of this invention with the traditional U-Net method, using 4000 training samples and 1000 test samples for each. The test results show that the traditional U-Net achieves a crack pixel accuracy of 89.3%, a recall of 82.6%, an intersection-over-union ratio of 74.8%, an average Dice coefficient of 85.6%, a fine crack false negative rate of 16.9%, and an average processing time of 0.42 seconds per image. The method of this invention achieves a crack pixel accuracy of 94.7%, a recall of 91.8%, an intersection-over-union ratio of 84.9%, an average Dice coefficient of 91.7%, a fine crack false negative rate reduced to 6.3%, and an average processing time of 0.36 seconds per image. For crack samples with breaks or occlusions, the connected component recovery accuracy of the method of this invention reaches 88.9%, an improvement of 17.3% compared to the traditional U-Net. The average error in crack length and the average error in crack width in the inspection report are 3.8%, which meets the application requirements for underwater lining plate crack inspection, including the location of small cracks, restoration of continuous cracks, and parameterized output. As can be seen from the above simulation process, this invention can improve crack feature extraction, edge restoration, and inspection report generation efficiency under conditions of water scattering noise, low contrast, and crack fracture obscuring.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart power energy dispatching method based on big data analysis, characterized in that, include: Collect power energy operation data and historical dispatch execution data, and perform preprocessing on the data to generate a reliable aligned power status frame; Based on the trusted aligned power state frame, a morphological symbol aggregation approximation algorithm with a cross-channel scheduling coupling symbol constraint mechanism is used to perform symbol calibration processing and generate a scheduling morphological risk calibration table. An improved SOFTS model is constructed, which includes a morphological symbol-guided fusion layer, a scheduling event causal gating layer, and a power grid flow projection layer. The input is a trusted aligned power state frame and a scheduling morphological risk calibration table, and the causal gating channel weights are generated. Input the causal gating channel weights and scheduling morphological risk calibration table into the morphological symbols to guide the fusion layer to perform core aggregation and state reallocation, and generate a morphologically enhanced scheduling state representation. The morphologically enhanced scheduling state representation is used as the input scheduling event. The causal gating layer performs joint prediction of the scheduling state to generate multi-source scheduling prediction results. The multi-source scheduling prediction results are input into the power flow projection layer to perform power flow safety projection, energy storage lifetime constraints and capacity calculation, and generate a safety scheduling capacity association table. Based on the security scheduling capability association table and the scheduling form risk calibration table, perform uncertainty risk hierarchical output and generate a scheduling risk correction mark table; A smart scheduling strategy is generated based on the security scheduling capability association table and the scheduling risk correction flag table, and the execution results of the smart scheduling strategy are written into the historical scheduling execution data.

2. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generation of the trusted aligned power state frame includes: Collect power generation operation records, new energy output records, energy storage status records, load power consumption records, and power grid operation records to generate power energy operation data; Collect scheduling instruction numbers, execution times, planned adjustment amounts, actual response amounts, and execution deviation records to generate historical scheduling execution data; The power energy operation data and historical dispatch execution data are mapped, dimensionally unified, and sampled according to the dispatch time axis to generate raw power energy data. Perform missing data marking, anomaly marking, time alignment, channel credibility marking, topology node binding, and line topology marking on the raw power energy data to generate a trusted aligned power status frame carrying node binding relationships and line topology records.

3. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generated scheduling morphology risk calibration table includes: Read the time-series status records of each power channel from the trusted aligned power status frame, perform multi-scale window segmentation according to channel type, scheduling cycle and node binding relationship, and generate channel scheduling window sequence; Perform morphological symbol mapping on the channel scheduling window sequence to generate a single-channel morphological symbol string carrying amplitude level, direction level and response stage markers; Based on the cross-channel scheduling coupling symbol constraint mechanism, the single-channel morphological symbol strings within the same scheduling window are constructed into a cross-channel symbol constraint graph according to the node binding relationship, generating a scheduling coupling symbol group; Read the execution deviation flag from the trusted aligned power state frame, perform risk level recalibration and symbol threshold calibration on the scheduling coupled symbol group based on the execution deviation flag, and generate coupled risk labels and calibration symbol thresholds; The scheduling morphology risk calibration table is generated by associating the channel scheduling window sequence, single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label, execution deviation mark and calibration symbol threshold.

4. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generated causal gating channel weights include: An improved SOFTS model is established, connecting the morphological symbol guidance fusion layer, the scheduling event causal gating layer, and the power grid flow projection layer according to the data transmission sequence. Read the time-series status records of each power channel in the trusted aligned power state frame, and read the single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold in the scheduling morphology risk calibration table to generate model input records; Based on the model input records, channel representation encoding and event trigger matching are performed in the scheduling event causal gating layer to generate a scheduling event matching table; Based on the scheduling event matching table, gating weights are assigned to each power channel to generate causal gating channel weights. The improved SOFTS model was trained by reading trusted aligned power state frames, scheduling morphology risk calibration tables, and historical scheduling execution data. The trusted aligned power state frames and scheduling morphology risk calibration tables were input into the improved SOFTS model, and forward computation was performed to obtain multi-source scheduling prediction results, a safe scheduling capability association table, and a scheduling risk correction label table. The combination of scheduling state prediction error, event causality gating error, morphology symbol fusion error, power flow safety projection error, energy storage lifetime constraint error, scheduling risk level discrimination error, and scheduling correction triggering error was used as the joint optimization objective. The network parameters of the morphology symbol guidance fusion layer, the scheduling event causality gating layer, and the power flow projection layer were continuously optimized. When the change in the joint loss value in five consecutive training rounds was less than 0.001, the improved SOFTS model was considered to have completed convergence training.

5. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The improved SOFTS model includes: The SOFTS model retains the core aggregation structure of the sequence, the channel state redistribution structure, the time series prediction backbone and the prediction output structure, and makes structural modifications to the input expression, morphological symbol fusion, event causal gating and power flow security constraints in the context of intelligent power energy dispatching. A trusted aligned power state frame input is introduced at the model input end, transforming the ordinary multivariate time series input into a power dispatch state input form that carries channel type, scheduling cycle, node binding relationship, channel trustworthiness flag and execution deviation flag; A morphological symbol-guided fusion layer is added at the core aggregation position of the sequence, embedding the single-channel morphological symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold from the scheduling morphological risk calibration table into the channel fusion process; The channel weight allocation method is improved to form a scheduling event causal gating layer, and the ordinary channel weight allocation method is transformed into a causal gating allocation method under the constraint of the event time delay association table. A power flow projection layer is added to the back end of the prediction output. The multi-source scheduling prediction results are input into the power flow security projection process, so that node power balance, line capacity, voltage security, regional exchange boundary and energy storage lifetime constraints participate in the selection of safe and feasible scheduling states. The output organization method is improved by expanding the model output target to include causal gating channel weights, morphologically enhanced scheduling state representation, multi-source scheduling prediction results, and a correlation table of safe scheduling capabilities.

6. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generated morphological enhancement scheduling state representation includes: Read the single-channel morphology symbol string, scheduling coupling symbol group, coupling risk label and calibration symbol threshold from the scheduling morphology risk calibration table, input the morphology symbol guidance fusion layer, and generate morphology symbol index records according to the power channel number; Match the causal gating channel weights with the morphological symbol index records of the execution channel to generate morphological gating fusion weights carrying coupling risk constraints; Based on morphological gating fusion weights, weighted core aggregation is performed on the corresponding power channel states in the trusted aligned power state frames to generate scheduling core states. Based on the reallocation of the execution state of the scheduling core state and the scheduling coupled symbol group, a sub-channel enhanced state record is generated. By associating the enhanced state records of the associated channels, the core state of the scheduling, and the morphological gating fusion weights, a morphologically enhanced scheduling state representation is generated.

7. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generation of multi-source scheduling prediction results includes: The reading morphology enhances the scheduling status representation. Load forecast representation, new energy forecast representation, energy storage forecast representation, line forecast representation, and demand response forecast representation are extracted according to power channel type and scheduling window number to generate a scheduling channel forecast representation set. Based on the scheduling channel prediction representation set, execution event time delay matching is performed to generate an event time delay association table; Generate event causal prediction weights by using a time-delay association table of related events and causal gating channel weights. Based on the event causal prediction weights, gated joint prediction is performed on the scheduling channel prediction representation set to generate power channel prediction records; Based on the power channel forecast records, supply and demand deviation calculations are performed according to the node binding relationship to generate node supply and demand deviation forecast records. The power channel forecast records and node supply and demand deviation forecast records are then linked to generate multi-source scheduling forecast results.

8. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generated security scheduling capability association table includes: The power flow projection layer reads power channel prediction records and node supply and demand deviation prediction records from multi-source scheduling prediction results, and reads node binding relationships and line topology records from trusted aligned power state frames to generate power flow projection input records. Input the power flow projection input record into the power grid power flow projection layer, perform node power balance verification, line capacity verification, voltage safety verification and regional exchange boundary verification, and generate power flow safety verification record; Based on the power channel prediction records in the multi-source scheduling prediction results, extract the energy storage status prediction records, read the energy storage operation boundary records in the reliable aligned power status frames, perform energy storage lifetime constraint verification, and generate energy storage dispatchable boundary records. Based on the power flow safety verification record and the energy storage schedulable boundary record, a safe and feasible state projection is performed to generate a safe and feasible schedulable state; Based on the calculation of power generation regulation capacity, new energy absorption capacity, energy storage charging and discharging capacity, load response capacity and line transmission margin in the safe and feasible dispatch state, an energy unit dispatchable capacity matrix is ​​generated. By associating the safe and feasible scheduling status, the energy unit scheduling capability matrix, the energy storage scheduling boundary record, the node binding relationship, and the power flow safety verification record, a safe scheduling capability association table is generated.

9. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generation of the scheduling risk correction flag table includes: Read the safe and feasible scheduling status, energy unit scheduling capability matrix, power flow safety verification record, and energy storage scheduling boundary record from the safe scheduling capability association table; Power flow safety margin is generated based on power flow safety verification records, energy storage lifetime constraint margin is generated based on energy storage schedulable boundary records, and scheduling priority is generated based on energy unit schedulable capability matrix. Safety capability risk records are generated by associating safe and feasible scheduling status. Read the coupling risk label, calibration symbol threshold and execution deviation mark from the scheduling morphological risk calibration table to generate a morphological risk calibration record; Associate security capability risk records and morphological risk calibration records, perform layered output of uncertainty risks, and generate scheduling risk levels; Based on the scheduling risk level and scheduling priority, a correction trigger flag is generated, and a risk correction association record is established according to the energy unit number and scheduling window number; Associate scheduling risk levels, correction trigger flags, risk correction association records, and execution deviation flags to generate a scheduling risk correction flag table.

10. The intelligent power energy dispatching method based on big data analysis according to claim 1, characterized in that, The generation of intelligent scheduling strategies based on the security scheduling capability association table and the scheduling risk correction flag table includes: Read the safe and feasible scheduling status, energy unit scheduling capability matrix, energy storage scheduling boundary record and power flow security verification record from the safe scheduling capability association table, and generate the strategy capability input record. Read the scheduling risk level, scheduling correction trigger flag, and risk correction association record from the scheduling risk correction flag table to generate a strategy risk input record; Associate the policy capability input record and policy risk input record, and generate a combination of scheduling instructions according to the energy unit number and scheduling window number; Intelligent scheduling strategies are generated based on combinations of scheduling instructions, and the scheduling instruction number, execution time, planned adjustment amount, actual response amount and execution deviation record corresponding to the intelligent scheduling strategy are collected to generate strategy execution results; Write the strategy execution results into the historical scheduling execution data.

Images