Power system communication scheduling data interaction method and system

Abstract

The invention relates to the technical field of power dispatching, and discloses a power system communication dispatching data interaction method and system.The method comprises the steps that a power plant side multi-source data channel and a power grid side multi-source data channel are accessed, multi-source data are processed in a standardized mode through a unified point taking catalog and a versioning point table, and standardized data streams are obtained; constructing a partition isolation network-related data channel, and generating a scheduling interaction data packet and a session element record through encryption packaging and integrity verification; collecting the data into a data table, establishing an anchor point event dictionary and an event link identifier, constructing a process fragment packet and an index, and forming a network-related process fragment library; and calling the fragment library to construct a training sample, training the multi-task time sequence deep learning model and outputting an analysis result, organizing an intraday rolling load instruction based on the result, calculating an assessment electric quantity, and generating a load interaction execution result. The technical problems that link traceability, fragment reusability and load execution supporting by analysis of cross-source scheduling interaction data are difficult are solved.

Description

Technical Field

[0001] This invention relates to the field of power dispatching technology, specifically to a method and system for power system communication dispatching data interaction. Background Technology

[0002] With the continuous evolution of dispatch automation, plant-grid interaction, and AGC / primary frequency regulation control, online data acquisition of power plant-side distributed control system (DCS) operation data channels, thermal electronics station data channels, plant-grid interaction platform data channels, and grid-side remote terminal unit (RTU) data channels and grid-connected substation data channels has gradually become routine. Conventional methods typically achieve multi-source measurement field mapping and caliber unification through a unified point catalog and versioned point tables. Furthermore, standardized data streams are encrypted, integrity-verified, and session element-recorded on the isolated grid-connected data channels in dispatch zone 1 and dispatch zone 2 to support the transmission of dispatch interaction data packets and the execution of intraday rolling load commands.

[0003] Existing technologies still have shortcomings in terms of temporal consistency and traceable link characterization of cross-source data: On the one hand, changes in measurement caliber, sampling attributes, and point table versions of multi-source data channels are more likely to cause field consistency drift, making it difficult to obtain stable pairing evidence for scheduling instruction issuance events, scheduling receipt confirmation events, and DCS reception flag events under the same session number; on the other hand, the lack of a standardized accumulation mechanism for event link identification and process segment packets with anchor event dictionaries as the core makes it difficult to form reusable training samples for related segments of AGC adjustment process boundary events and primary frequency regulation action trigger events, thereby limiting the accuracy and replayability of the operational analysis results in supporting load interaction execution results and power generation plan assessment calculations. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the present invention aims to provide a power system communication and dispatch data interaction method and system to solve the technical problems of cross-source dispatch interaction data being difficult to trace, reuse, and analyze to support load execution.

[0005] This invention is achieved through the following technical solution: In a first aspect, the present invention provides a power system communication dispatch data interaction method, comprising: By connecting to multi-source data channels on the power plant side and the power grid side, and establishing a unified data point catalog and versioned data point table, the multi-source data is standardized to obtain a standardized data stream. Based on the standardized data stream, a partitioned and isolated network data channel is constructed. Through encryption encapsulation and integrity verification, scheduling interaction data packets and session element records are generated. The scheduling interaction data packets and session element records are imported into the data platform, an anchor event dictionary and event link identifier are established, process fragment packets and fragment indexes are constructed, and a network-related process fragment library is formed. The network-related process fragment library is called to construct training samples, a multi-task time-series deep learning model is trained, and the model is run to generate running analysis results. Based on the operational analysis results, daily rolling load instructions are organized and the power generation plan assessment electricity calculation is completed, generating load interaction execution results.

[0006] Preferably, the standardization process for multi-source data to generate a standardized data stream includes the following steps: Collect and register the channel identifiers and collection scope of the DCS operation data channels, thermal electronics station data channels, and power plant-grid interaction platform data channels on the power plant side, as well as the RTU data channels and grid-related substation data channels on the power grid side, to form a list of channels on the power plant side and the power grid side. Summarize the lists of channels on both sides, extract the key parameter names, measurement calibers and sampling attributes of each channel, generate a unified point sampling directory and establish a naming caliber for data items, and form a versioned point table; Based on the versioned point table, field mapping and consistency marking are performed on the data items sent from each channel to obtain a standardized data stream.

[0007] Preferably, generating scheduling interaction data packets and session element records specifically includes the following steps: Based on the data source markings of the standardized data flow, the data channel is divided into scheduling zone one and zone two channels and the partition relationship is recorded to generate a partition channel table; Configure independent identifiers and access control policies for each partition channel according to the partition channel table, solidify the isolation relationship, and build a partition-isolated network data channel. The standardized data stream carried by the isolated network data channel is encrypted and encapsulated to obtain the scheduling interaction data packet. At the same time, integrity verification is performed, and the session number, channel identifier, timestamp reference and data source mark are recorded to generate session element record.

[0008] Preferably, the establishment of an anchor event dictionary and event link identifiers includes the following steps: The scheduling interaction data packets and session element records are aggregated according to session number and channel identifier to form a session-related dataset; Extract elements of various scheduling-related events from the session association dataset, solidify event type names and judgment fields, and generate an anchor event dictionary; Based on the anchor event dictionary, the relationships between instruction issuance, receipt confirmation, and DCS reception flag events associated with the same session number are paired and recorded to generate alignment evidence records; Based on the alignment evidence record, extract key information such as unit identifier, dispatch master station identifier, and anchor event timestamp set, and combine them to generate event link identifier.

[0009] Preferably, the process fragment package and fragment index are constructed to form a network-related process fragment library, which specifically includes the following steps: Based on the event link identifier, locate the associated scheduling interaction data packet, extract the corresponding time series data range to generate the original fragment, and combine it with the alignment evidence record to obtain the aligned fragment; Based on the versioned point table, the alignment segment is subjected to measurement caliber normalization and field consistency adjustment to form a caliber normalized segment; The original fragments, aligned fragments, caliber-normalized fragments, and aligned evidence records are encapsulated into process fragment packages, which are then associated with event link identifiers to generate fragment package records. Key information is extracted from the fragment packet records to generate a fragment index, which is then combined with the fragment packet records to form a network-related process fragment library.

[0010] Preferably, the training samples are constructed by calling the network-related process fragment library, which specifically includes the following steps: Retrieve process segment packages associated with AGC regulation and primary frequency modulation action triggering events by segment index, collect them into a candidate segment set, and extract the normalized segment to generate candidate sample source; The candidate sample sources are grouped according to the unit identifier and instruction type to generate a grouped sample sequence, and the process curve category and anomaly cause label are labeled to form a training sample set.

[0011] Preferably, the training of a multi-task temporal deep learning model includes the following steps: According to the rule that the unit identifier and the instruction type are consistent, the training sample set is divided into a training subset and a validation subset to obtain the sample partitioning result; Based on the sample partitioning results, the multi-task temporal deep learning model is iteratively trained and its parameters are updated, while validation records are generated simultaneously using the validation subset. The trained model version identifier is fixed and the parameters are saved to obtain a multi-task temporal deep learning model.

[0012] Preferably, generating the analysis results includes the following steps: Retrieve the process segment package corresponding to the latest scheduling instruction from the network process segment library according to the event link identifier, and extract the normalized segment to generate a real-time segment sequence. The corresponding version of the multi-task temporal deep learning model is invoked to perform inference on the real-time segment sequence, and information such as network performance index prediction results and process curve categories are output to generate model inference results. The model inference results and key information from the session element records are combined to form the operational analysis results.

[0013] Preferably, generating the load interaction execution result specifically includes the following steps: The results of the operation analysis are analyzed, and information such as unit identification and grid-related performance index prediction results are extracted to form a unit operation summary. Combined with the intraday rolling load plan, a set of candidate load instructions is generated. Based on the unit operation summary, target load instructions are filtered, converted into a data structure available to the control side, and sent to the unit DCS to generate a load instruction issuance record. The rolling load plan and actual output data within the day are combined and aligned according to the unit identifier to generate plan execution aligned data; The power generation plan assessment data is calculated based on the alignment data of the plan execution, and the results are summarized with the load instruction issuance records to form the load interaction execution results.

[0014] Secondly, the present invention also provides a power system communication dispatch data interaction system, comprising: The multi-source access module is used to access multi-source data channels on the power plant side and the power grid side. By establishing a unified point acquisition directory and a versioned point table, the multi-source data is standardized to obtain a standardized data stream. The security encapsulation module is used to construct a partitioned and isolated network data channel based on the standardized data stream, and generate scheduling interaction data packets and session element records through encryption encapsulation and integrity verification. The segment management module is used to import the scheduling interaction data packets and session element records into the data platform, establish an anchor event dictionary and event link identifier, construct process segment packets and segment indexes, and form a network-related process segment library. The model training module is used to call the network process fragment library to construct training samples, train a multi-task time series deep learning model, and run the model to generate running analysis results. The scheduling execution module is used to organize intraday rolling load instructions and complete the calculation of power generation plan assessment based on the operation analysis results, and generate load interaction execution results.

[0015] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a power system communication dispatch data interaction method. It maps and marks key parameter names, measurement calibers, and sampling attributes using a unified point catalog and versioned point tables, standardizing the data flow to obtain stable point table version numbers and timestamp references, providing a verifiable data foundation for subsequent session element records. By using partitioned channel tables and partitioned isolation of grid-connected data channels, it achieves channel isolation between dispatch zone 1 and dispatch zone 2, and performs encryption encapsulation and integrity verification on dispatch interaction data packets, ensuring that session numbers, channel identifiers, timestamp references, and data source markers form a consistent session-related dataset. Anchor event dictionaries are used to solidify event type names and event judgment fields, generating aligned evidence records and event link identifiers. Process fragment packages are encapsulated as original fragments, aligned fragments, and caliber-normalized fragments, forming a fragment index, making the grid-connected process fragment library searchable, replayable, and reusable. A multi-task time-series deep learning model is trained and invoked, directly using the operational analysis results for intraday rolling load instruction screening, load instruction issuance record generation, and power generation plan assessment calculation, improving the interpretability and verification consistency of load interaction execution results. Attached Figure Description

[0016] Figure 1 This is a flowchart of the power system communication scheduling data interaction method in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the process of generating scheduling interaction data packets and session element records in an embodiment of the present invention. Figure 3 This is a flowchart for generating event chain identifiers in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the process of forming a network-related process fragment library in an embodiment of the present invention; Figure 5 This is a schematic diagram of the power system communication and dispatch data interaction system in an embodiment of the present invention; In the diagram: 1. Multi-source access module; 2. Security encapsulation module; 3. Fragment management module; 4. Model training module; 5. Scheduling and execution module. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] The purpose of this invention is to provide a method and system for communication and scheduling data interaction in power systems, so as to solve the technical problems of cross-source scheduling interaction data being difficult to trace, reuse, and analyze to support load execution.

[0020] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1 See Figure 1 In one embodiment of the present invention, a method for communication and dispatch data interaction in a power system is provided, comprising the following steps: S1. Access the multi-source data channels on the power plant side and the power grid side, and establish a unified point acquisition directory and versioned point table to generate standardized data streams.

[0021] S1.1. Collect and register the channel identifiers and collection ranges of the power plant-side DCS (Distributed Control System) operation data channels, thermal electronics station data channels, and power plant-grid interaction platform data channels. The collection range represents the set of collectable data items and the range of sampling attributes corresponding to the channel identifier. The set of collectable data items is obtained by reading the DCS point table, the thermal electronics station data interface list, and the power plant-grid interaction platform data interface list. The range of sampling attributes is determined by the sampling period, timestamp accuracy, and measurement caliber fields recorded in the point table and data interface list. Write the channel identifiers and collection ranges into the power plant-side channel entries and summarize them to form the power plant-side channel list.

[0022] Collect and register the channel identifiers and collection ranges of the power grid-side RTU (Remote Terminal Unit) data channels and the data channels of the grid-connected substations. The collection range represents the set of collectable data items and the range of sampling attributes corresponding to the channel identifier. The set of collectable data items is obtained by reading the RTU point table and the grid-connected substation table. The range of sampling attributes is determined by the sampling period, timestamp accuracy, and measurement caliber fields recorded in the RTU point table and the grid-connected substation table. Write the channel identifiers and collection ranges into the power grid-side channel entries and summarize them to form a power grid-side channel list.

[0023] The power plant-side channel list and the power grid-side channel list are compiled, and the key parameter names, measurement calibers and sampling attributes corresponding to the channel identifiers are read one by one. The key parameter names are used to identify the measurement points, the measurement calibers are used to record the engineering meaning, measurement units and statistical calibers, and the sampling attributes are used to record the sampling period, timestamp accuracy and sampling method. The key parameter names, measurement calibers and sampling attributes are associated with the channel identifiers and written into the point entry, and then compiled into a unified point directory.

[0024] S1.2. Establish naming standards for each data item in the unified sampling directory. The naming standard for each data item should include at least the full name of the data item, the field name, the measurement unit identifier, and the standard version number. Write the naming standard for each data item into the point table entry and manage it according to the standard version number to form a versioned point table. The versioned point table is used to record the set of point table entries corresponding to the standard version number and retain the standard version number change record.

[0025] Based on the versioned point table, field mapping and consistency marking are performed on the data items sent by each channel corresponding to the power plant-side channel list and the power grid-side channel list. Field mapping is used to map the channel data item name to the field name in the versioned point table. Consistency marking is used to record the measurement caliber matching status and the sampling attribute matching status. The data items that have completed field mapping and consistency marking are organized into a standardized data stream according to the channel identifier and time sequence.

[0026] S2. Establish partitioned and isolated network data channels for standardized data streams, perform encryption encapsulation and integrity verification, and generate scheduling interaction data packets and session element records, such as... Figure 2 As shown.

[0027] S2.1. Read the data source marker from the standardized data stream. The data source marker is obtained by mapping the channel identifier recorded in the power plant-side channel list and the power grid-side channel list. According to the data source marker, divide the data channels corresponding to the standardized data stream into dispatch zone 1 channels and dispatch zone 2 channels. Write the correspondence between dispatch zone 1 channels, dispatch zone 2 channels and channel identifiers into the partition relationship record and summarize to form a partition channel table.

[0028] Read the partition channel table and configure independent channel identifiers and access control policies for the scheduling zone 1 channel and the scheduling zone 2 channel respectively. The access control policy records the binding relationship between the data source mark and the channel identifier that is allowed to be transmitted. The channel identifier and access control policy are written into the channel configuration record and the isolation relationship is fixed. The channel configuration records are summarized to form the partitioned isolated network data channel.

[0029] S2.2. Read the partitioned isolated network data channel and transmit the standardized data stream on the partitioned isolated network data channel. The standardized data stream is framed and encapsulated according to the channel identifier and a data source mark is attached to generate a scheduling interaction data packet. When performing integrity verification on the scheduling interaction data packet, first extract the frame header field, payload field and check field from the scheduling interaction data packet. The frame header field contains at least the channel identifier, data source mark and frame sequence number. The payload field carries the sequence of data items of the standardized data stream. The check field stores the check value calculated by the payload field. Then, perform hash digest calculation on the payload field to obtain the digest value and compare it with the check value in the check field to obtain the integrity verification result. Write the frame sequence number, digest value, check value, consistency comparison result and channel identifier of the scheduling interaction data packet into the integrity verification record. The integrity verification record is bound to the scheduling interaction data packet to form a verification record.

[0030] When reading the verification record, the channel identifier, frame number, and the first frame transmission start timestamp of the scheduling interaction data packet are extracted from the verification record in frame sequence order and merged to form the session start element. The first frame transmission start timestamp of the scheduling interaction data packet in the session start element is taken from the first frame timestamp field recorded in the frame header field of the scheduling interaction data packet. The channel identifier and the first frame transmission start timestamp of the scheduling interaction data packet are combined according to the preset splicing rules to generate the session number. The timestamp reference is determined by the timing unit and time zone identifier corresponding to the timestamp precision field recorded in the standardized data stream and written into the session element field. The session number, channel identifier, timestamp reference, data source mark, and integrity verification result in the verification record are written into the session element record, and the scheduling interaction data packet and session element record are output.

[0031] S3. Import the scheduling interaction data packets and session element records into the network source coordination data platform and establish an anchor event dictionary, generating event link identifiers, such as... Figure 3 As shown, process fragment packages and fragment indexes are constructed to form a network-related process fragment library, such as... Figure 4 As shown.

[0032] S3.1. Collect scheduling interaction data packets and session element records and write them into the network source coordination data platform. Merge the scheduling interaction data packets according to the session number and channel identifier and establish an association relationship with the session element records to form a session association dataset. The session association dataset retains the standardized data stream data item sequence corresponding to the session number, channel identifier, data source mark, point table version number, timestamp reference, scheduling interaction data packet frame sequence number and scheduling interaction data packet payload field.

[0033] Read the session-associated dataset and extract event elements from events such as dispatch instruction issuance, dispatch receipt confirmation, DCS reception flag, key unit output change, AGC adjustment process boundary, and primary frequency regulation trigger. Each event element includes at least the session number, channel identifier, event occurrence timestamp, unit identifier, dispatch master station identifier, instruction type, and instruction sequence number. The event type name and event determination field are then merged to form an anchor event dictionary. The event determination field includes the field value conditions corresponding to the event type name and the position of the event occurrence timestamp field. The anchor event dictionary is used to locate the anchor event timestamp set within the session-associated dataset.

[0034] S3.2. Retrieve the anchor event dictionary to locate the scheduling instruction issuance event, scheduling receipt confirmation event, and DCS reception flag event associated with the same session number. Pair the scheduling instruction issuance event, scheduling receipt confirmation event, and DCS reception flag event according to the instruction sequence number and instruction type, and write them into the pairing relationship record. The pairing relationship record includes the session number, instruction type, instruction sequence number, timestamp of the scheduling instruction issuance event, timestamp of the scheduling receipt confirmation event, and timestamp of the DCS reception flag event. Summarize the pairing relationship records to form an alignment evidence record, and extract the set of anchor event timestamps under the same session number from the alignment evidence record to form an anchor event timestamp set record.

[0035] Retrieve alignment evidence records and extract the set of records containing unit identifier, dispatch master station identifier, instruction type, instruction sequence number, session number, point table version number, and anchor event timestamp. Organize unique combinations according to unit identifier, instruction type, and instruction sequence number to generate event link identifiers. Establish association relationships between event link identifiers and session numbers and channel identifiers and write them into the event link identifier record. The event link identifier record is used to locate the dispatch interaction data packets and anchor event timestamp sets corresponding to the same event link identifier.

[0036] S3.3. Retrieve the event link identifier record to locate the associated scheduling interaction data packet and retrieve the anchor event timestamp set record. Determine the continuous time series data range according to the anchor event timestamp set record, and extract the standardized data stream data item sequence corresponding to the continuous time series data range from the scheduling interaction data packet payload field to form the original fragment.

[0037] The alignment evidence record is retrieved, and the event timestamp, scheduling receipt confirmation event timestamp, and DCS receiving flag event timestamp are issued according to the scheduling instructions recorded in the alignment evidence record to determine the time offset relationship. The time offset relationship is then applied to the event occurrence timestamp of the original segment to obtain the aligned segment.

[0038] The versioned point table is retrieved, and the aligned segments are normalized for measurement caliber and field consistency according to the field names, measurement caliber fields, and sampling period fields of the versioned point table to obtain caliber-normalized segments. The original segments, aligned segments, caliber-normalized segments, and alignment evidence records are encapsulated to form process segment packages, and a relationship is established with the event link identifier to generate segment package records. The unit identifier, instruction type, instruction sequence number, session number, anchor event timestamp set, point table version number, and segment time range are extracted from the segment package records to generate segment indexes, and the segment indexes and segment package records are summarized to form a network-related process segment library.

[0039] S4. Call the network process fragment library to build training samples, train a multi-task time series deep learning model, run the multi-task time series deep learning model, and generate running analysis results.

[0040] S4.1. Based on the fragment index, retrieve the process fragment packages associated with the boundary events of the AGC adjustment process and aggregate them to form an AGC candidate fragment set. At the same time, based on the fragment index, retrieve the process fragment packages associated with the primary frequency regulation action trigger event and aggregate them to form a primary frequency regulation candidate fragment set. Merge the AGC candidate fragment set and the primary frequency regulation candidate fragment set to form a candidate fragment set. Extract the caliber-normalized fragments one by one from the candidate fragment set and retain the event link identifier, unit identifier, instruction type, instruction sequence number, session number, point table version number, anchor point event timestamp set and fragment time range to form a candidate sample source.

[0041] S4.2. Group the candidate sample sources according to the unit identifier and instruction type to form a grouped sample sequence. Each sample in the grouped sample sequence is expanded into a time-series sample matrix by the timestamp sequence and field name sequence of the normalized segment. Using the anchor event timestamp set and segment time range, locate the time step position corresponding to the anchor event timestamp in the timestamp sequence of the normalized segment and write it into the sample positioning field to form a key segment positioning label. The key segment positioning label includes the key segment start time step position and the key segment end time step position. Bind the process curve category label and the anomaly cause label to the grouped sample sequence according to the event link identifier to form a training sample set.

[0042] S4.3. Divide the training sample set into a training subset and a verification subset based on the rule that the unit identifier and instruction type are consistent, to form the sample partitioning result.

[0043] Extract field names from the versioned point table to form an input field name set. The input field name set and the output task name set are written into the training configuration record of the multi-task time series deep learning model. The output task name set includes network performance index prediction results, process curve categories, anomaly cause labels, and key segment location information.

[0044] Extract the normalized segment from each item in the training subset and locate the data item sequence corresponding to the field name according to the input field name set. Organize the sequence into a training window sequence according to the timestamp order and feed it into the multi-task time series deep learning model to perform forward computation to obtain the training output result. The training output result includes the prediction result of network performance index and the output probability of process curve category, and maintains a correspondence with the sample sequence number consistent with the training subset item set.

[0045] The sequence of data items whose field names correspond to the prediction results of network performance indicators is extracted from the normalized segment to form the true value sequence of network performance indicators. At the same time, the process curve category labels in the training subset item set are read to form the true value labels of process curve categories. The true value sequence of network performance indicators, the true value labels of process curve categories and the training output results are summarized to form the loss calculation dataset.

[0046] The mean squared error value is obtained by performing network performance index prediction error calculation on the loss calculation dataset, and the cross-entropy loss value is obtained by performing process curve category labeling loss calculation on the loss calculation dataset. The mean squared error value and the cross-entropy loss value are added together to obtain the training loss value.

[0047] Backpropagation is performed on the training loss value to obtain the gradient of the parameters of the multi-task temporal deep learning model. Gradient descent is then performed based on the parameter gradient to update the parameters of the multi-task temporal deep learning model and form a parameter update record. The parameter update record and the training loss value are written into the training record.

[0048] The set of validation subset items is used to generate a validation window sequence according to the training window sequence construction rules. This sequence is then fed into a multi-task temporal deep learning model to perform forward computation and obtain the validation output. The validation output maintains a consistent correspondence with the sample sequence number of the set of validation subset items.

[0049] The validation subset of entries is read to extract the true value sequence of network performance indicators and the true value label of process curve category. These are then summarized with the validation output results to form a validation loss calculation dataset. The validation loss calculation dataset is then subjected to mean squared error calculation and cross-entropy loss calculation to obtain the validation loss value.

[0050] The validation accuracy value is obtained by comparing the output probability of the process curve category in the validation output with the true value annotation of the process curve category. The validation loss value and validation accuracy value corresponding to each iteration are written into the validation record. The validation record is associated with the model version identifier to generate a multi-task temporal deep learning model.

[0051] When calculating the prediction error of network performance indicators for the training subset of entries, the mean squared error expression is defined as follows: ; in, This indicates the prediction error of network performance indicators. This represents the number of samples in the training subset's entry set that participated in the computation. Indicates the sample number. This represents the true value of the network performance metric labeled on the training sample set. This represents the predicted network performance index output by the multi-task temporal deep learning model.

[0052] When calculating the process curve category labeling loss for the training subset of entries, the cross-entropy loss expression is defined as follows: ; in, Indicates the loss of the process curve category label. This represents the number of samples in the training subset's entry set that participated in the computation. This indicates the number of categories labeled for the process curve. Indicates that the training sample set is in the sample Category The labeled probability values ​​are as follows: This indicates that the multi-task temporal deep learning model is in the sample Category The output probability value is . Represents the natural logarithm operation.

[0053] S4.4. Retrieve the process segment package corresponding to the latest scheduling instruction from the network process segment library according to the event link identifier, and extract the caliber-normalized segment to form a real-time segment sequence. The real-time segment sequence retains the event link identifier, unit identifier, instruction type, instruction sequence number, session number, channel identifier, timestamp reference, point table version number, and integrity verification result. Read the set of input field names of the multi-task time series deep learning model corresponding to the model version identifier, and locate the data item sequence by field name in the caliber-normalized segment of the real-time segment sequence. Concatenate the data item sequences corresponding to each field name in timestamp order to form an inference sample matrix, and perform sampling period consistency sorting on the inference sample matrix to obtain an aligned inference sample matrix.

[0054] The aligned inference sample matrix is ​​divided into inference window sequences according to the sequence length segmentation rules recorded in the multi-task temporal deep learning model. Each inference window sequence is then fed into the multi-task temporal deep learning model for forward computation. Forward computation includes calculating the hidden state of each time step sequentially and outputting the task output vector for the last time step. The task output vector is parsed into network performance index prediction results, process curve category output probabilities, anomaly cause label output probabilities, and key segment location information. The process curve category output probabilities and anomaly cause label output probabilities are determined by the category name and label name corresponding to the highest probability. The key segment location information is determined by the start and end time step positions in the task output vector. The network performance index prediction results, process curve categories, anomaly cause labels, and key segment location information corresponding to the inference window sequences are summarized to form the model inference results. The model inference results are then combined with the channel identifier, timestamp baseline, and integrity verification results in the session element records to form the operational analysis results.

[0055] S5. Based on the operational analysis results, organize intraday rolling load instructions and complete the calculation of the power generation plan assessment electricity, and generate load interaction execution results.

[0056] S5.1. Parse the operation analysis results and extract the unit identifier, grid-related performance index prediction results and key segment location information. Summarize the grid-related performance index prediction results according to the unit identifier and associate them with the key segment location information to form a unit operation summary. Obtain the daily rolling load plan and extract the planned time period, planned output and dispatch load instruction items corresponding to the unit identifier in the daily rolling load plan. Sort them according to the planned time period to form a candidate load instruction set.

[0057] S5.2. Read the unit operation summary and match the unit identifier and planned time period in the candidate load instruction set, select the dispatch load instruction entry corresponding to the next time moment, and generate the target load instruction; convert the dispatch load instruction entry of the target load instruction into a data structure that can be used by the control side and send it to the unit DCS operation data channel, record the unit identifier, dispatch load instruction entry, sending timestamp, channel identifier and session number to form a load instruction issuance record.

[0058] S5.3. Collect the planned output data from the daily rolling load plan and the actual output data from the unit DCS operation data channel. Align and merge the planned output and actual output data according to the unit identifier and generate planned execution aligned data according to the planned time period. The planned execution aligned data includes the unit identifier, planned time period, planned output time series and actual output time series.

[0059] S5.4. Read the planned execution alignment data, and perform power generation plan assessment calculation on the planned output time series and the actual output time series to form the assessment calculation results. Summarize the assessment calculation results and load instruction issuance records to form the load interaction execution results.

[0060] The expression for calculating the power generation plan assessment volume is as follows: ; in, This represents the cumulative deviation of the assessed power generation from the calculated assessed power generation plan. This indicates the number of time points involved in the calculation within the planned execution aligned data. Indicates the time point sequence number. This indicates that the planned execution alignment data is at a specific time point. The corresponding planned output value, This indicates that the planned execution alignment data is at a specific time point. The corresponding actual output value, This indicates the time interval between adjacent time points, used to convert the output value into the cumulative amount of electricity.

[0061] To further verify the technical solution of the present invention, experimental simulation data of the power system communication scheduling data interaction method are provided.

[0062] Three grid-connected generating units within the same provincial power grid dispatch area were selected as test subjects, namely “300MW-1 unit”, “300MW-2 unit” and “600MW-1 unit”. The test window covers a continuous 24-hour rolling plan cycle and includes multiple AGC adjustment process boundary events and one frequency regulation action trigger event.

[0063] Channel identifiers are collected from the DCS (Distributed Control System) operation data channels, thermal electronics station data channels, and power plant-grid interaction platform data channels on the power plant side. The set of collectable data items, sampling period, timestamp accuracy, and measurement caliber fields are read from the DCS point table, thermal electronics station data interface list, and power plant-grid interaction platform data interface list to form a power plant-side channel list. Simultaneously, channel identifiers are collected from the RTU (Remote Terminal Unit) data channels and grid-connected substation data channels on the grid side. The set of collectable data items and sampling attribute range are read from the RTU point table and grid-connected substation site table to form a grid-side channel list. Subsequently, the power plant-side and grid-side channel lists are summarized, and the key parameter names, measurement calibers, and sampling attributes corresponding to each channel identifier are read, written into the sampling entries, and summarized into a unified sampling directory. Based on the unified sampling directory, a naming caliber is established for each data item, including at least the full name of the data item, field name, measurement unit identifier, and caliber version number. Versioned point tables are formed according to caliber version numbers, and caliber version number changes are recorded.

[0064] In the implementation phase, based on the versioned point table, field mapping and consistency marking are performed on the data items transmitted through each channel. Channel data item names are mapped to field names in the versioned point table, and the measurement caliber matching status and sampling attribute matching status are recorded in the consistency marking. Standardized data streams are organized according to channel identifiers and time sequence. Based on the data source markings of the standardized data streams, the data channels corresponding to the standardized data streams are divided into scheduling zone 1 channels and scheduling zone 2 channels, and written into the partitioned channel table. Independent channel identifiers and access control policies are configured and isolation relationships are solidified, forming partitioned isolated network-connected data channels. When the standardized data stream is transmitted on the partitioned isolated network-connected data channels, it is framed and encapsulated according to the channel identifier to generate scheduling interaction data packets. Hash digest calculation is performed on the payload field and compared with the consistency of the verification field to generate an integrity verification record, which is then bound to the scheduling interaction data packet to form a verification record. The first frame timestamp and channel identifier are extracted from the verification record according to the frame sequence number and written into the session element record.

[0065] The scheduling interaction data packets and session element records are written into the network source coordination data platform, and aggregated according to session number and channel identifier to form a session association dataset. From the session association dataset, event elements of scheduling instruction issuance events, scheduling receipt confirmation events, DCS reception flag events, key unit output change events, AGC adjustment process boundary events, and primary frequency regulation action trigger events are extracted, and event type names and event judgment fields are fixed to generate an anchor event dictionary. Using the anchor event dictionary, scheduling instruction issuance events, scheduling receipt confirmation events, and DCS reception flag events within the same session number are paired to generate alignment evidence records, and the unit identifier, scheduling master station identifier, instruction type, instruction sequence number, session number, point table version number, and anchor event timestamp set are extracted and combined to generate an event link identifier. Subsequently, based on the event link identifier, the associated scheduling interaction data packet is located and the continuous time series data range corresponding to the anchor event timestamp set is extracted to generate the original fragment. Based on the alignment evidence record execution time benchmark, the alignment fragment is uniformly formed. Based on the versioned point table, the measurement caliber is normalized and the field consistency is sorted to form the caliber normalization fragment. The original fragment, alignment fragment, caliber normalization fragment and alignment evidence record are encapsulated to form the process fragment package and generate the fragment index. The summaries form the network-related process fragment library.

[0066] Finally, process segment packages associated with AGC regulation process boundary events and primary frequency regulation trigger events are retrieved according to the segment index, and normalized segments are extracted to form candidate sample sources. Training sample sets are formed by grouping by unit identifier and instruction type and labeling process curve categories and anomaly cause tags. Training subset and validation subset sets are divided according to the rule of consistency between unit identifier and instruction type. Iterative training of the multi-task time-series deep learning model is performed based on the training configuration records, and validation records are generated on the validation subset set. During real-time operation, process segment packages corresponding to the latest dispatch instruction are retrieved according to the event link identifier, and normalized segments are extracted to form real-time segment sequences. The set of input field names is determined based on the model version identifier, and an aligned inference sample matrix is ​​constructed. The sequence is divided into inference window sequences according to sequence length, and forward calculation is performed window by window to output grid-related performance index prediction results, process curve categories, anomaly cause tags, and key segment location information. These are then summarized to form operational analysis results. The operational analysis results are used for subsequent intraday rolling load instruction organization and power generation plan assessment calculations.

[0067] The details are shown in Table 1 below: Table 1 Comparative Test Data of Key Indicators for Data Interaction Between Generating Units and the Grid

[0068] A quantitative comparison is given from five links: "consistent data standards", "reliable cross-regional transmission", "traceable event links", "searchable and replayable fragments", and "inference results can support scheduling execution and assessment".

[0069] First, the "field mapping consistency marking pass rate (%)" reached 99.4%~99.7% in all three units, while the comparative method only achieved 91.7%~93.1%. Simultaneously, the "standardized data flow missing point percentage (%)" significantly decreased to 0.3%~0.6%, while the comparative method maintained 4.2%~5.1%. This difference corresponds to the solidification effect of the unified sampling directory and versioned point table on the full name of data items, field names, measurement unit identifiers, and caliber version numbers. Field mapping and consistency marking explicitly record the measurement caliber matching status and sampling attribute matching status, enabling cross-source data to complete caliber convergence and missing point identification before entering the subsequent partitioned isolated grid-connected data channels. This reduces the data incomparability issues caused by "same name, different meaning" and "different name, same meaning" between the dispatching side and the power plant side.

[0070] Secondly, the "failure rate (‰) of scheduling interaction data packet integrity verification" decreased to 0.10‰~0.15‰ across the three units, compared to 0.98‰~1.22‰ in the comparison method; the "success rate (%) of session number reconstruction increased to 99.5%~99.8%, compared to 93.1%~94.4% in the comparison method. These differences are directly related to the channel identifiers of the partitioned isolated network data channels, the solidification of isolation relationships through access control policies, and the consistency comparison of hash digests performed on the payload fields after frame encapsulation. Integrity verification records and verification records provide verifiable frame sequence numbers and first frame timestamps for session number generation, reducing the impact of out-of-order transmission, missing frames, and replays on the consistency of session element records during cross-regional transmission.

[0071] Third, the "anchor event pairing success rate (%)" in this invention reaches 98.4%~99.2%, while the comparative method is only 84.6%~85.1%. At the same time, the "event link identifier generation delay (ms)" is shortened to 160~210ms, while the comparative method is 820~910ms. The anchor event dictionary solidifies the event type name and event judgment field, so that dispatch instruction issuance events, dispatch receipt confirmation events and DCS reception flag events can be paired according to instruction type and instruction sequence number within the same session number to form an alignment evidence record. The event link identifier is composed of the unit identifier, dispatch master station identifier, session number, point table version number and anchor event timestamp set, which has a stable positioning key value, avoiding mismatch and omission caused by relying solely on time window matching.

[0072] Fourth, the average time taken to retrieve fragment indexes (ms) is 90-110ms in this invention, compared to 395-455ms in the comparative method. This indicates that the structured organization of the fragment index in terms of unit identifier, instruction type, instruction sequence number, session number, and fragment time range reduces the cost of replaying and reanalyzing the fragment library in the network process.

[0073] Finally, regarding model-related indicators, the "mean absolute error of grid performance indicator prediction (MW)" decreased from 9.1~10.1MW to 4.3~5.2MW, the "key segment positioning deviation (seconds)" decreased from 4.5~5.2 seconds to 1.4~1.8 seconds, and the "hit rate of next moment dispatch load command (%)" increased to 95.4%~96.9%. On the assessment side, this is reflected in the "cumulative value of power generation plan assessment deviation E_ (MWh)" decreasing from 26.9~30.2MWh to 11.6~13.8MWh.

[0074] The above results indicate that the parallel encapsulation of original segments, aligned segments, and caliber-normalized segments in the process segment package provides stronger comparability and label consistency for the training sample set. The multi-task temporal deep learning model jointly outputs network performance index prediction results, process curve categories, anomaly cause labels, and key segment location information within the same inference window. This is beneficial for considering both prediction bias and key segment location information during the load instruction screening stage, thereby improving the matching degree of intraday rolling load instructions and reducing the cumulative value of assessment bias.

[0075] Example 2 according to Figure 5 As shown, this embodiment also provides a power system communication dispatch data interaction system, including: Multi-source access module 1 is used to access multi-source data channels on the power plant side and the power grid side. By establishing a unified point acquisition directory and a versioned point table, the multi-source data is standardized to obtain a standardized data stream. Security encapsulation module 2 is used to construct a partitioned and isolated network data channel based on the standardized data stream, and generate scheduling interaction data packets and session element records through encryption encapsulation and integrity verification processing; The segment management module 3 is used to import the scheduling interaction data packets and session element records into the data platform, establish an anchor event dictionary and event link identifier, construct process segment packets and segment indexes, and form a network-related process segment library. Model training module 4 is used to call the network-related process fragment library to construct training samples, train a multi-task time series deep learning model, and run the model to generate running analysis results. The scheduling execution module 5 is used to organize intraday rolling load instructions and complete the calculation of power generation plan assessment based on the operation analysis results, and generate load interaction execution results.

[0076] Example 3 This embodiment also provides a computer device applicable to the power system communication dispatch data interaction method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the power system communication dispatch data interaction method proposed in the above embodiment. The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0077] Example 4 This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the power system communication dispatch data interaction method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0078] In summary, this invention maps and marks key parameter names, measurement calibers, and sampling attributes using a unified point catalog and versioned point tables, standardizing data flow to obtain stable point table version numbers and timestamp benchmarks, providing a verifiable data foundation for subsequent session element records. It achieves channel isolation between scheduling zone 1 and scheduling zone 2 channels through partitioned channel tables and partitioned isolation of network-connected data channels, and performs encryption encapsulation and integrity verification on scheduling interaction data packets, ensuring that session numbers, channel identifiers, timestamp benchmarks, and data source markers form a consistent session-related dataset. It uses an anchor event dictionary to solidify event type names and event judgment fields, generating aligned evidence records and event link identifiers. Process fragment packages are encapsulated with original fragments, aligned fragments, and caliber-normalized fragments to form a fragment index, making the network-connected process fragment library searchable, replayable, and reusable. Finally, it trains and calls a multi-task time-series deep learning model, directly using the operational analysis results for intraday rolling load instruction screening, load instruction issuance record generation, and power generation plan assessment calculation, improving the interpretability and consistency of load interaction execution results.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for data interaction in power system communication dispatch, characterized in that, include: By connecting to multi-source data channels on the power plant side and the power grid side, and establishing a unified data point catalog and versioned data point table, the multi-source data is standardized to obtain a standardized data stream. Based on the standardized data stream, a partitioned and isolated network data channel is constructed. Through encryption encapsulation and integrity verification, scheduling interaction data packets and session element records are generated. The scheduling interaction data packets and session element records are imported into the data platform, an anchor event dictionary and event link identifier are established, process fragment packets and fragment indexes are constructed, and a network-related process fragment library is formed. The network-related process fragment library is called to construct training samples, a multi-task time-series deep learning model is trained, and the model is run to generate running analysis results. Based on the operational analysis results, daily rolling load instructions are organized and the power generation plan assessment electricity calculation is completed, generating load interaction execution results.

2. The power system communication dispatch data interaction method according to claim 1, characterized in that, Standardizing multi-source data to generate a standardized data stream involves the following steps: Collect and register the channel identifiers and collection scope of the DCS operation data channels, thermal electronics station data channels, and power plant-grid interaction platform data channels on the power plant side, as well as the RTU data channels and grid-related substation data channels on the power grid side, to form a list of channels on the power plant side and the power grid side. Summarize the lists of channels on both sides, extract the key parameter names, measurement calibers and sampling attributes of each channel, generate a unified point sampling directory and establish a naming caliber for data items, and form a versioned point table; Based on the versioned point table, field mapping and consistency marking are performed on the data items sent from each channel to obtain a standardized data stream.

3. The power system communication dispatch data interaction method according to claim 1, characterized in that, Generating scheduling interaction data packets and session element records includes the following steps: Based on the data source markings of the standardized data flow, the data channel is divided into scheduling zone one and zone two channels and the partition relationship is recorded to generate a partition channel table; Configure independent identifiers and access control policies for each partition channel according to the partition channel table, solidify the isolation relationship, and build a partition-isolated network data channel. The standardized data stream carried by the partitioned isolated network data channel is encrypted and encapsulated to obtain the scheduling interaction data packet. At the same time, integrity verification is performed, and the session number, channel identifier, timestamp reference and data source mark are recorded to generate session element record.

4. The power system communication dispatch data interaction method according to claim 1, characterized in that, Establishing an anchor event dictionary and event chain identifiers includes the following steps: The scheduling interaction data packets and session element records are aggregated according to session number and channel identifier to form a session-related dataset; Extract elements of various scheduling-related events from the session association dataset, solidify event type names and judgment fields, and generate an anchor event dictionary; Based on the anchor event dictionary, the relationships between instruction issuance, receipt confirmation, and DCS reception flag events associated with the same session number are paired and recorded to generate alignment evidence records; Based on the alignment evidence record, extract key information such as unit identifier, dispatch master station identifier, and anchor event timestamp set, and combine them to generate event link identifier.

5. A power system communication dispatch data interaction method according to claim 1, characterized in that, The process fragment package and fragment index are constructed to form a network-related process fragment library, which includes the following steps: Based on the event link identifier, locate the associated scheduling interaction data packet, extract the corresponding time series data range to generate the original fragment, and combine it with the alignment evidence record to obtain the aligned fragment; Based on the versioned point table, the alignment segment is subjected to measurement caliber normalization and field consistency adjustment to form a caliber normalized segment; The original fragments, aligned fragments, caliber-normalized fragments, and aligned evidence records are encapsulated into process fragment packages, which are then associated with event link identifiers to generate fragment package records. Key information is extracted from the fragment packet records to generate a fragment index, which is then combined with the fragment packet records to form a network-related process fragment library.

6. The power system communication dispatch data interaction method according to claim 1, characterized in that, The process of constructing training samples by calling the network-related process fragment library includes the following steps: Retrieve process segment packages associated with AGC regulation and primary frequency modulation action triggering events by segment index, collect them into a candidate segment set, and extract the normalized segment to generate candidate sample source; The candidate sample sources are grouped according to the unit identifier and instruction type to generate a grouped sample sequence, and the process curve category and anomaly cause label are labeled to form a training sample set.

7. A power system communication dispatch data interaction method according to claim 1, characterized in that, The training of a multi-task temporal deep learning model includes the following steps: According to the rule that the unit identifier and the instruction type are consistent, the training sample set is divided into a training subset and a validation subset to obtain the sample partitioning result; Based on the sample partitioning results, the multi-task temporal deep learning model is iteratively trained and its parameters are updated, while validation records are generated simultaneously using the validation subset. The trained model version identifier is fixed and the parameters are saved to obtain a multi-task temporal deep learning model.

8. A power system communication dispatch data interaction method according to claim 1, characterized in that, Generating the analysis results involves the following steps: Retrieve the process segment package corresponding to the latest scheduling instruction from the network process segment library according to the event link identifier, and extract the normalized segment to generate a real-time segment sequence. The corresponding version of the multi-task temporal deep learning model is invoked to perform inference on the real-time segment sequence, and information such as network performance index prediction results and process curve categories are output to generate model inference results. The model inference results and key information from the session element records are combined to form the operational analysis results.

9. A power system communication dispatch data interaction method according to claim 1, characterized in that, Generating the load interaction execution results includes the following steps: The results of the operation analysis are analyzed, and information such as unit identification and grid-related performance index prediction results are extracted to form a unit operation summary. Combined with the intraday rolling load plan, a set of candidate load instructions is generated. Based on the unit operation summary, target load instructions are filtered, converted into a data structure available to the control side, and sent to the unit DCS to generate a load instruction issuance record. The rolling load plan and actual output data within the day are combined and aligned according to the unit identifier to generate plan execution aligned data; The power generation plan assessment data is calculated based on the alignment data of the plan execution, and the results are summarized with the load instruction issuance records to form the load interaction execution results.

10. A power system communication dispatch data interaction system, characterized in that, include: The multi-source access module is used to access multi-source data channels on the power plant side and the power grid side. By establishing a unified point acquisition directory and a versioned point table, the multi-source data is standardized to obtain a standardized data stream. The security encapsulation module is used to construct a partitioned and isolated network data channel based on the standardized data stream, and generate scheduling interaction data packets and session element records through encryption encapsulation and integrity verification. The segment management module is used to import the scheduling interaction data packets and session element records into the data platform, establish an anchor event dictionary and event link identifier, construct process segment packets and segment indexes, and form a network-related process segment library. The model training module is used to call the network process fragment library to construct training samples, train a multi-task time series deep learning model, and run the model to generate running analysis results. The scheduling execution module is used to organize intraday rolling load instructions and complete the calculation of power generation plan assessment based on the operation analysis results, and generate load interaction execution results.

Images