An AI-based power transaction position intelligent optimization management method

Abstract

The application relates to the technical field of power transaction data processing, and discloses an AI-based intelligent optimization management method for power transaction positions, which comprises the following steps: acquiring low-frequency contracts and high-frequency spot data, splitting the data into basic time series and binding power grid topology nodes, combining contract liquidity identifiers to establish a four-dimensional space-time position matrix; dynamically updating a power transfer distribution factor matrix for calculation, generating equivalent power flow occupation values of a restricted area; combining prediction confidence interval data to calculate a dynamic truncation threshold value; when the power flow occupation value breaks through the threshold value, extracting the ratio of the absolute value of the power transfer distribution factor to the contract liquidity identifier to generate a portfolio adjustment utility ratio composite index; generating a priority portfolio adjustment sequence in descending order according to the index, extracting front-end records to generate an optimization instruction and outputting the same. The application comprehensively considers physical adjustment sensitivity and financial resistance cost, can automatically screen out an optimal portfolio adjustment strategy, and effectively improves the accuracy and economy of power transaction position optimization.

Description

Technical Field

[0001] This invention relates to the field of power trading data processing technology, specifically to an AI-based intelligent optimization management method for power trading positions. Background Technology

[0002] Electricity trading position management is a key business link of the electricity market operation center. It mainly refers to the process by which market participants coordinate and control the risks of electricity contracts formed in various electricity markets. In the complex electricity market trading environment, spot market volume data has the characteristics of high-frequency fluctuations, while medium- and long-term bilateral contracts are in a low-frequency trading state. The trading center needs to integrate the contract data of the above different time scales, monitor the overall position load level formed by various market participants within a specific period, and prevent congestion in the physical power grid.

[0003] Existing power trading position control schemes rely on pre-set static grid models and fixed safe transmission limits. Operations personnel perform simple arithmetic summations on collected power trading data and then compare it to fixed limit parameters to determine if there is a risk of exceeding limits. Once a load exceedance is detected, the system initiates a reduction operation on the contracted power volume of each node according to a random allocation principle or a manually determined empirical ratio. During this process, the system allocates an equal number of computing threads to all nodes to perform Monte Carlo probability simulations, thereby outputting a risk assessment report. Traders manually input position adjustment data based on the assessment report and distribute it to each execution terminal.

[0004] Current processing methods simply sum the heterogeneous contracts across time scales, severing the constraints of physical network spatial location. They also ignore the financial resistance costs incurred by different nodes in adjusting their positions and the differences in their physical sensitivity to power flow regulation. This results in a rebalancing strategy that fails to balance economic costs with efficiency in mitigating power outages. Fixed safe transmission limits cannot adapt to network capacity fluctuations caused by predictive uncertainty. The system performs deep probabilistic inference calculations even for nodes in a safe state, consuming significant memory and CPU resources, leading to a substantial decrease in concurrent computing efficiency. The lack of a quantitative outage allocation model results in inaccurate power adjustment magnitudes, leading to either excessive intervention increasing costs or insufficient intervention causing physical outages. The reliance on manual intervention severs the data control link from risk prediction to automated execution.

[0005] Therefore, the purpose of this invention is to provide an AI-based intelligent optimization management method for power trading positions, in order to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an AI-based intelligent optimization management method for power trading positions. This method solves the problems in existing power trading position management, such as the inability of adjustment strategies to balance economic costs and efficiency in mitigating over-limit situations, the waste of system computing resources due to ineffective probability extrapolation, and the lack of a quantitative allocation model leading to inaccurate adjustment ranges and the inability to achieve automated execution.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] An AI-based intelligent optimization management method for power trading positions includes the following steps:

[0009] Obtain low-frequency business contract data and high-frequency spot market volume data, and split the low-frequency business contract data into basic time series.

[0010] By binding the power grid topology node identifier to the basic time series, calculating the contract liquidity identifier, and establishing a four-dimensional spatiotemporal position matrix;

[0011] Acquire a snapshot of the power grid topology and update the power transfer distribution factor matrix when the state changes;

[0012] The power transfer distribution factor matrix is ​​used to calculate the four-dimensional spatiotemporal storage matrix, generating a constrained region local storage cluster containing equivalent power flow occupancy values.

[0013] Obtain the predicted confidence interval data and calculate the dynamic truncation threshold;

[0014] The equivalent power flow occupancy value of the local position cluster in the restricted area is compared with the dynamic cutoff threshold. When the equivalent power flow occupancy value exceeds the dynamic cutoff threshold, the absolute value of the power transfer distribution factor in the power transfer distribution factor matrix and the contract liquidity identifier are extracted, and the ratio is calculated to generate a composite index of rebalancing utility ratio.

[0015] Based on the composite index of rebalancing efficiency ratio, the data records in the local position cluster of the restricted area are sorted in descending order to generate a priority rebalancing sequence. The data records at the front of the priority rebalancing sequence are extracted to generate optimization instructions and output.

[0016] Preferably, the steps of acquiring low-frequency business contract data and high-frequency spot market volume data, and then splitting the low-frequency business contract data into basic time series data, specifically include:

[0017] Extract the decomposition curve type identifier from low-frequency business contract data;

[0018] When the decomposition curve type is identified as a flat base load curve, the total contracted electricity volume in the low-frequency business contract data is proportionally distributed to each contracted time slice within the contract effective period.

[0019] When the decomposition curve type is identified as a peak-valley segmented curve, the parameters of the division intervals of peak period, stable period and low period defined in the transaction rules and the peak-valley electricity conversion ratio are read, and the total contracted electricity is unevenly distributed to the corresponding time intervals according to the peak-valley electricity conversion ratio.

[0020] When the decomposition curve type is identified as a custom historical fitting curve, the load shape coefficient matrix is ​​retrieved, and the load shape coefficient matrix is ​​used to perform a weighted calculation on the total contracted electricity volume.

[0021] The electricity values ​​allocated to the corresponding time intervals are converted into a basic time series with a fixed minimum unit.

[0022] Preferably, the steps for binding the basic time series data to power grid topology node identifiers, calculating contract liquidity identifiers, and establishing a four-dimensional spatiotemporal position matrix specifically include:

[0023] Extract the transaction entity code from the basic time series, retrieve the asset topology mapping table that records the mapping relationship between the transaction entity code and the grid topology node identifier, and bind the basic time series with the grid topology node identifier;

[0024] The default penalty rate, remaining time to delivery, and additional resistance constant contained in the original contract records corresponding to the basic time series are analyzed and weighted summed with preset static weight coefficients to generate a contract liquidity identifier.

[0025] Retrieve the expected electricity consumption value corresponding to the base time series;

[0026] A four-dimensional spatiotemporal position matrix is ​​constructed by combining the timestamp of the basic time series, the bound grid topology node identifier, the expected electricity value, and the calculated contract liquidity identifier.

[0027] Preferably, the steps of obtaining a snapshot of the power grid topology and updating the power transfer distribution factor matrix when the state changes specifically include:

[0028] Retrieve a snapshot of the historical topology cached in local memory;

[0029] The current power grid topology snapshot is compared with the historical topology snapshot item by item to identify whether there is topology status change data. The topology status change data includes the switching status reversal of physical lines or the relative change of branch reactance parameters exceeding the preset safety tolerance range.

[0030] When topology status change data is detected, a parameter recalculation trigger signal is generated;

[0031] Upon receiving the parameter recalculation trigger signal, the latest power grid topology snapshot data is read, the network parameters of the changed area are extracted, and the local power transfer distribution factor is calculated based on the DC power flow model to generate the latest power transfer distribution factor matrix.

[0032] Preferably, the step of calculating the four-dimensional spatiotemporal storage matrix using the power transfer distribution factor matrix to generate a constrained region local storage cluster containing equivalent power flow occupancy values ​​specifically includes:

[0033] The expected power values ​​of each node under the preset time slice are superimposed and converted under the grid spatial constraints. By judging whether the power transfer distribution factor in the power transfer distribution factor matrix is ​​greater than the preset correlation sensitivity threshold, the set of topological nodes that are electrically coupled with the restricted transmission section is determined.

[0034] The expected power value of each node is multiplied by the power transfer distribution factor and aggregated to obtain the equivalent power flow occupancy value generated by the aggregation of all relevant node positions on the restricted transmission section. The restricted transmission section index and the time slice index of the preset time slice are extracted.

[0035] Logically group the underlying records based on the restricted transmission section index and time slice index to generate a restricted area local warehouse cluster of structured data set.

[0036] Preferably, the steps of obtaining the predicted confidence interval data and calculating the dynamic truncation threshold specifically include:

[0037] Obtain the prediction confidence interval data for the corresponding time slice of the pre-prediction model, and separate the upper and lower confidence interval values ​​from the prediction confidence interval data;

[0038] Obtain the absolute difference between the upper limit and the lower limit of the confidence interval, and assign the absolute difference to the confidence interval width index.

[0039] Read the static restricted physical limit parameters recorded in the system configuration file, obtain the physical capacity extreme value corresponding to the static restricted physical limit parameter, and generate the dynamic truncation threshold by subtracting the product of the uncertainty penalty coefficient and the confidence interval width index from the physical capacity extreme value.

[0040] Preferably, after comparing the equivalent power flow occupancy value of the local warehouse cluster in the restricted area with the dynamic truncation threshold, the method further includes:

[0041] When the equivalent power flow occupancy value is less than the dynamic cutoff threshold, the conditional risk value is statically assigned to the local position cluster in the restricted area that has not exceeded the dynamic cutoff threshold, and the conditional risk value is assigned to a preset safety constant.

[0042] The Monte Carlo sampling calculation for physical topology nodes belonging to the restricted area local warehouse cluster that has not exceeded the dynamic truncation threshold is blocked, the system operation threads pre-allocated to the safe nodes are reclaimed, and the buffer blocks reserved for the safe nodes in memory are cleared.

[0043] Preferably, the steps of extracting the absolute value of the power transfer distribution factor and the contract liquidity identifier, and calculating the ratio to generate a composite index of rebalancing utility ratio specifically include:

[0044] Extract a list of pointers to the set of independent contract records contained within the local position cluster in the restricted area, and read the corresponding contract liquidity identifier and power transfer distribution factor;

[0045] The rebalancing utility ratio is calculated by dividing the absolute value of the power transfer distribution factor by the contract liquidity indicator.

[0046] The rebalancing utility ratio is added as an additional field to the four-dimensional spatiotemporal position matrix to generate a composite index of rebalancing utility ratio.

[0047] Preferably, the data records at the front end of the priority rebalancing sequence are extracted, including:

[0048] The over-limit power flow gap index is calculated by subtracting the dynamic cutoff threshold from the equivalent power flow occupancy value. The remaining over-limit power flow gap index is set, and the initial value of the remaining over-limit power flow gap index is set as the over-limit power flow gap index.

[0049] According to the priority relocation sequence, contract records are extracted one by one from top to bottom. The minimum value between the expected power volume and the product of the remaining over-limit power flow gap index divided by the absolute value of the power transfer distribution factor and the time slice duration conversion factor is selected as the target adjustment power volume.

[0050] The remaining overload power flow gap index is updated cyclically by using the power load after the target adjustment power conversion to offset the remaining overload power flow gap index until the remaining overload power flow gap index is less than or equal to zero or all contract records in the priority rebalancing sequence have been traversed.

[0051] Preferably, the step of extracting data records from the front end of the priority rebalancing sequence to generate optimization instructions and outputting them specifically includes:

[0052] The unique identifier of the contract record with a target adjustment amount greater than zero, the time slice identifier, the original expected amount of electricity, and the target adjustment amount are combined to generate a structured optimization instruction and send it to the power trading execution terminal.

[0053] This invention provides an AI-based intelligent optimization management method for power trading positions. It offers the following advantages:

[0054] 1. This invention obtains low-frequency business contracts and high-frequency spot market volume data across time scales by setting up a four-dimensional spatiotemporal position matrix, a dynamically updated power transfer distribution factor matrix, and a composite index of position adjustment utility ratio. These data are then uniformly split into basic time series bound to grid topology nodes. When the calculated equivalent power flow occupancy value exceeds the dynamic truncation threshold, the absolute value of the power transfer distribution factor and the ratio calculated by the contract liquidity identifier are extracted and sorted in descending order. This achieves accurate spatial mapping of transaction data to physical transmission sections. It comprehensively considers the sensitivity of grid power flow regulation and the financial resistance cost of position intervention, and can automatically select the priority position adjustment sequence with the highest efficiency and lowest economic cost in alleviating over-limit situations, thereby improving the accuracy and economy of power trading position optimization.

[0055] 2. This invention sets a dynamic truncation threshold based on the width of the predicted confidence interval and a calculation blocking mechanism for positions that have not exceeded the threshold. When the equivalent power flow occupancy value is less than the dynamic truncation threshold, it directly performs static assignment of conditional risk value to the corresponding restricted area local position cluster and intercepts Monte Carlo sampling calculations for the physical topology nodes within the restricted area local position cluster. It can dynamically adjust the boundary of the grid available capacity based on the uncertainty of the system prediction, effectively filter invalid probability inferences in a safe state, and promptly reclaim system computing threads and clear memory buffer blocks, thereby reducing the invalid occupation of the system's central processing unit and memory resources and improving the concurrent computing efficiency of risk control judgment and data processing.

[0056] 3. This invention sets up an iterative reduction operation logic based on a priority rebalancing sequence and a structured optimization instruction generation mechanism. After calculating the excess flow gap index, it extracts contract records one by one according to the descending queue of rebalancing utility ratio, selects the minimum value between the expected power value and the converted gap value as the target adjustment power value and performs cyclic reduction until the gap index returns to zero. This can accurately control the power adjustment range of a single contract, avoid extra costs caused by excessive intervention or excess due to insufficient intervention, and directly convert the optimization result into a structured instruction containing an identification code and the adjustment power value and send it to the terminal, thus opening up the data control link from risk deduction to automated execution. Attached Figure Description

[0057] Figure 1 This is a flowchart of the method of the present invention;

[0058] Figure 2 This is a system architecture diagram of the present invention;

[0059] Figure 3 This is a line graph showing the evolution of the equivalent power flow and dynamic truncation threshold of the restricted transmission section according to the present invention.

[0060] Figure 4This is a bar chart comparing the rebalancing costs and computation time of the intelligent optimization strategy of this invention with traditional strategies. Detailed Implementation

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0062] See attached document Figure 1 With appendix Figure 2 This invention includes an AI-based intelligent optimization management system for power trading positions. The system comprises a data alignment module, a storage engine, a topology aggregation module, a risk control judgment module, and an instruction generation module. The data alignment module connects to an external database to obtain contract data and performs format conversion.

[0063] The storage engine receives and records the processed data structure. The topology aggregation module reads the data from the storage engine and performs physical dimension calculations in conjunction with external parameters. The risk control judgment module controls the flow of the calculation thread based on external input parameters. The instruction generation module receives the dataset issued by the risk control judgment module, performs parameter calculations, and outputs instruction data packets.

[0064] See attached document Figure 1 This invention provides an AI-based intelligent optimization management method for power trading positions, comprising the following steps:

[0065] S10, the data alignment module obtains low-frequency business contract data and high-frequency spot quantity data, and executes the time granularity alignment algorithm to split the low-frequency business contract data into a basic time series;

[0066] S20, the data alignment module binds the grid topology node identifier to the basic time series, calculates the contract liquidity identifier according to the rule base, and establishes a four-dimensional spatiotemporal position matrix in the storage engine that includes timestamps, topology node identifiers, expected electricity values ​​and contract liquidity identifiers;

[0067] S30, the topology aggregation module obtains a snapshot of the power grid topology and performs parameter updates to the power transfer distribution factor matrix when the topology snapshot status data changes;

[0068] S40, the topology aggregation module uses the currently effective power transfer distribution factor matrix to perform spatial aggregation calculation on the four-dimensional spatiotemporal position matrix, converting the transaction electricity on discrete topology nodes into the equivalent power flow occupancy value of the transmission section, and generating local position clusters in the restricted area in the storage engine;

[0069] S50, the risk control judgment module obtains the predicted confidence interval data, extracts the difference between the upper and lower limits as the confidence interval width index, and calculates the dynamic truncation threshold based on the confidence interval width index;

[0070] S60, the risk control judgment module compares the equivalent power flow occupancy value of the local position cluster in the restricted area with the dynamic cutoff threshold. When the equivalent power flow occupancy value is lower than the dynamic cutoff threshold, the risk sampling calculation process for the local position cluster in the restricted area is stopped.

[0071] S70, when the equivalent power flow occupancy value exceeds the dynamic cutoff threshold, the risk control judgment module transmits the data records inside the local position cluster in the restricted area to the instruction generation module. The instruction generation module extracts the absolute value of the power transfer distribution factor and the contract liquidity identifier, calculates the ratio, and generates a composite index of rebalancing utility ratio.

[0072] S80, the instruction generation module sorts the data records in the local position cluster of the restricted area in descending order according to the composite index of position adjustment efficiency ratio, extracts the data records at the front of the sorting queue, encapsulates them into intervention instructions containing node identifier parameters and target intervention power parameters, and executes and outputs them.

[0073] In step S10, the data alignment module acquires low-frequency business contract data and high-frequency spot quantity data, and executes a time granularity alignment algorithm to split the low-frequency business contract data into a basic time series. Step S10 further includes the following sub-steps:

[0074] S101, the data alignment module connects to an external trading database via an application programming interface (API), establishes a data transmission channel, and captures trading data. The external trading database stores low-frequency business contract data and high-frequency spot trading volume data. Low-frequency business contract data includes, but is not limited to, annual bilateral long-term contracts, monthly centralized bidding contracts, and weekly trading contracts. The raw time dimension of low-frequency business contract data spans a large range, appearing as natural months, natural weeks, or specific multi-day aggregation periods. High-frequency spot trading volume data includes day-ahead and real-time market trading records, with a time granularity of 15 minutes or 1 hour.

[0075] The data alignment module concurrently captures the above-mentioned multi-dimensional raw power records according to the set time window. The specific implementation process of data capture, network communication handshake and data encryption verification through the application programming interface can be implemented by those skilled in the art based on conventional computer network and database access technologies.

[0076] S102, for the acquired low-frequency business contract data, the data alignment module extracts the decomposition curve type identifier from each contract record and calls the preset numerical decomposition rule library based on the decomposition curve type identifier. Power generation and consumption entities agree on different power decomposition methods when signing medium- and long-term contracts. The numerical decomposition rule library includes flat-segment base load curve decomposition rules, peak-valley flat-segment curve decomposition rules, and custom historical fitting curve decomposition rules.

[0077] When the data alignment module determines that the decomposition curve type is a flat base load curve, it uses an arithmetic average to evenly distribute the total contracted electricity volume to each agreed time slice within the contract's effective period. When it is determined to be a peak-valley flat curve, the data alignment module reads the specific interval parameters for peak, stable, and low periods defined in the trading rules, extracts the preset peak-valley electricity conversion ratio, and distributes the total contracted electricity volume unevenly to the corresponding time segments based on the peak-valley electricity conversion ratio. When it is determined to be a custom historical fitting curve, the data alignment module retrieves the load shape coefficient matrix of the trading entity in the historical load sample library, uses the load shape coefficient matrix to perform a weighted calculation on the current total contracted electricity volume, and derives an electricity distribution value that conforms to the actual production behavior of the trading entity.

[0078] S103, the data alignment module executes a time granularity alignment algorithm to uniformly convert the electricity values ​​allocated to each time segment into a basic time series with 15 minutes as the fixed minimum unit. When the parsed original data granularity is greater than 15 minutes, the data alignment module triggers granularity refinement logic. For example, for daytime transaction data with an original time granularity of 1 hour, or for medium- and long-term contract data that has been decomposed to the hourly level in step S102, the data alignment module uses linear interpolation calculation or mean smoothing distribution calculation to convert the 1-hour level electricity values ​​into four consecutive 15-minute level electricity values.

[0079] After calculation and data format reconstruction, the data alignment module generates a basic time series in memory. Each element record in the basic time series has a 15-minute timestamp parameter and an expected delivery quantity parameter constrained by the timestamp parameter. All heterogeneous contract records across time scales are mapped to a unified time axis.

[0080] In step S20, the data alignment module binds the grid topology node identifier to the basic time series, calculates the contract liquidity identifier based on the rule base, and establishes a four-dimensional spatiotemporal position matrix in the storage engine that includes timestamps, topology node identifiers, expected electricity values, and contract liquidity identifiers. Step S20 further includes the following sub-steps:

[0081] S201, the data alignment module extracts the transaction entity codes from the basic time series. These codes include generator unit codes on the generation side or enterprise codes on the consumption side. The data alignment module sends a query request to the storage engine to retrieve the asset topology mapping table. This table records the corresponding mapping relationship between transaction entity codes and power grid physical topology bus node identifiers. Based on the asset topology mapping table, the data alignment module binds each record in the basic time series to a specific topology node identifier. After this binding operation, the records in the basic time series acquire physical network spatial location information.

[0082] S202, the data alignment module analyzes the commercial terms contained in the original contract records corresponding to the base time series, and calculates the contract liquidity identifier based on the rule base. The contract liquidity identifier quantifies the overall resistance cost faced when financially closing out, reverse hedging, or intervening in default on a given contract volume under the current market trading rules. The higher the value of the contract liquidity identifier, the higher the financial cost of adjusting the corresponding contract record. The commercial terms extracted by the data alignment module include the default penalty rate, the remaining time until delivery, and the additional resistance constant. The data alignment module calculates the contract liquidity identifier using the following formula: ;

[0083] In the formula, For the first Each contract record corresponds to a contract liquidity indicator; For the first The contract records the penalty rate for breach of contract, which is directly extracted from the breach of contract clause in the original contract and is a decimal between 0 and 1. For the first The contract records the remaining time from the physical delivery of the electricity at the current time, in hours, and the value range is a value greater than 0 (when the delivery is close, a very small positive threshold is set to prevent abnormal division by zero calculation). The additional resistance constant is a value set by market rules for different trading categories. The additional resistance constant is obtained by consulting the preset market trading rule configuration table, and its value range is preset to be between 0 and 10. The additional resistance constant value of long-term contracts is greater than that of short-term contracts. and The static weighting coefficients preset for the system.

[0084] Static weighting coefficients are used to adjust the proportion of various cost factors. These coefficients are pre-set using expert scoring or the analytic hierarchy process (AHP) to meet specific requirements. The values ​​are all decimals between 0 and 1. The data alignment module converts the commercial terms into standardized numerical indicators.

[0085] In S203, the data alignment module combines the timestamp of the base time series, the bound topology node identifier, the expected electricity value, and the calculated contract liquidity identifier to generate a structured data tuple. Since market participants face pricing uncertainty in the spot bidding process, the expected electricity value can include a definite bid electricity value or the extreme values ​​of the electricity confidence interval output by the preceding model based on historical bidding behavior.

[0086] The data alignment module continuously writes structured data tuples into the storage engine. The storage engine aggregates the data tuples and constructs a four-dimensional spatiotemporal position matrix. The four-dimensional spatiotemporal position matrix records the mapping distribution of discrete trading behaviors on time slices and spatial nodes. For the table structure establishment and index construction process of the underlying relational database or time series database, those skilled in the art can implement it based on conventional database development techniques.

[0087] In step S30, the topology aggregation module acquires a grid topology snapshot, and performs parameter updates to the power transfer distribution factor matrix when the topology snapshot status data changes. Step S30 further includes the following sub-steps:

[0088] S301, the topology aggregation module establishes a data communication connection with the external power grid dispatching system and obtains a power grid topology snapshot according to a preset polling period, which is set to 5 minutes or 15 minutes. The power grid topology snapshot records the current operating status data of key nodes, transmission lines, and circuit breakers within the power grid. The operating status data includes line connectivity identifiers and branch reactance parameters. For the specific implementation process of accessing the remote database and retrieving status data via a communication protocol, those skilled in the art can use conventional network communication technologies.

[0089] S302, the topology aggregation module caches historical topology snapshots obtained in the previous polling cycle in local memory. The topology aggregation module compares the currently obtained power grid topology snapshot with the historical topology snapshots item by item for node and branch data to identify whether there is topology status change data. Topology status change data specifically refers to the switching state reversal caused by the commissioning, shutdown, or maintenance of physical lines, or the relative change of branch reactance parameters exceeding the preset safety tolerance range.

[0090] The safety tolerance range is obtained by consulting the system configuration file, and the value range is set to 3% to 5%. When topology status change data is detected, the topology aggregation module generates a parameter recalculation trigger signal and sends the parameter recalculation trigger signal containing the changed node and line information to the computing service process. When no topology status change data is detected, the topology aggregation module maintains the physical parameters in the current memory unchanged and waits for the next polling cycle.

[0091] S303: After receiving the parameter recalculation trigger signal, the topology aggregation module reads the latest power grid topology snapshot data, extracts the network parameters of the changed area, and calculates the local power transfer distribution factor of the affected area based on the DC power flow model. The power transfer distribution factor is used to characterize the degree of influence of a node in the power grid injecting or extracting a unit of active power on the power flow distribution on a specific transmission section, and is expressed as a percentage value.

[0092] After the calculation is completed, the topology aggregation module generates the latest power transfer distribution factor matrix. The topology aggregation module performs a memory replacement operation, unloading the old power transfer distribution factor matrix from memory and loading the latest power transfer distribution factor matrix into the system's memory database. The calculation process of deriving the node admittance matrix and solving the power transfer distribution factor based on the DC power flow model can be implemented by those skilled in the art based on conventional power system analysis theory.

[0093] The topology aggregation module performs spatial aggregation calculations on the four-dimensional spatiotemporal position matrix using the currently effective power transfer distribution factor matrix, converting the traded electricity on discrete topology nodes into equivalent power flow occupancy values ​​of transmission sections, and generating local position clusters in restricted areas in the storage engine. Step S40 further includes the following sub-steps:

[0094] S401, the topology aggregation module retrieves the four-dimensional spatiotemporal warehouse matrix from the storage engine, and simultaneously retrieves the power transfer distribution factor matrix that was loaded into the memory database in the preceding steps and is currently in effect. The topology aggregation module establishes a cross-index retrieval table for the timestamp parameter, the topology node identifier parameter, and the power transfer distribution factor matrix. For the specific implementation process of multi-table join queries through the database indexing mechanism, those skilled in the art can use conventional database structured query language statements or memory key-value pair retrieval techniques. The data retrieval process is a well-known technique in the field and will not be described in detail here.

[0095] S402, the topology aggregation module performs superposition and conversion of the expected power values ​​of each node under the preset time slice under the grid spatial constraints. The spatial aggregation calculation is achieved through the following formula:

[0096] ;

[0097] In the formula, The time slice index corresponds to the 15-minute timestamp in the base time series. Its value ranges from 1 to 96, and it is determined by dividing a natural day into 96 equal parts. This is the restricted transmission section index. The restricted transmission section index is determined by the transmission channel number defined by the external power grid dispatching system, and the value range is determined according to the actual number of restricted power grid sections configured. This is the physical topology node index, which corresponds to the bus node identifier of the power grid topology. The value range is determined based on the total number of bus nodes contained in the actual power grid. For the first Index of contract records; To be compatible with restricted power transmission sections The set of topological nodes with electrical coupling associations is determined by whether the power transfer distribution factor is greater than a preset association sensitivity threshold. The association sensitivity threshold is preset by the system configuration file and is a floating-point number between 0.01 and 0.05. To belong to the topology node A collection of independent contract records; For topology nodes Next The contract record is in a time slice The expected electricity consumption value, in megawatt-hours; To the topology node To restricted transmission sections The power transfer distribution factor, which is a floating-point number ranging from -1 to 1, is obtained through the aforementioned power transfer distribution factor matrix update process. For time slices At that time, all relevant node positions were located at the restricted transmission section. The equivalent power flow occupancy value generated by the aggregation is expressed in megawatts. Through the above calculations, the electricity traded on each physical topology node is quantitatively converted into the active power flow value on the corresponding restricted transmission section.

[0098] In step S403, the topology aggregation module sends the calculated equivalent power flow occupancy value back to the storage engine. The storage engine logically groups the underlying records based on the restricted transmission section index and the time slice index, generating restricted area local storage clusters within the storage engine. These restricted area local storage clusters are structured data sets, and their storage fields include the restricted transmission section identifier, timestamp, equivalent power flow occupancy value, and a list of pointers to the sets of independent contract records that generate active power contributions.

[0099] By generating local position clusters in restricted areas, the data processing system completes the spatial aggregation of trading position data to physically restricted sections, providing input for threshold comparison in the risk control judgment module. For the specific implementation of the underlying database aggregating and grouping data according to fields and dividing data subsets, those skilled in the art can use conventional database cluster partitioning or table partitioning techniques.

[0100] In step S50, the risk control judgment module acquires the predicted confidence interval data, extracts the difference between the upper and lower limits as the confidence interval width index, and calculates the dynamic truncation threshold based on the confidence interval width index. Step S50 further includes the following sub-steps:

[0101] S501, the risk control judgment module obtains the upper and lower limits of the confidence interval for the corresponding time slice of the preceding prediction model. The risk control judgment module establishes a data connection interface with the preceding prediction model and retrieves the predicted confidence interval data output by the preceding prediction model through this interface. The preceding prediction model can be a load prediction model based on a long short-term memory network. The predicted confidence interval data records the net load fluctuation prediction results for a specific restricted transmission section within a future time period. The AI ​​features are mainly reflected in the step of the prediction model outputting the confidence interval.

[0102] The risk control judgment module extracts fields from the predicted confidence interval data, separating the upper and lower limits of the confidence interval. The upper and lower limits of the confidence interval represent the numerical boundaries that the predicted object will reach under a preset confidence level, which is set to 95%. For the specific implementation process of retrieving external service data through the application programming interface and extracting fields from structured messages, those skilled in the art can use conventional data crawling and text parsing techniques.

[0103] In S502, the risk control judgment module uses the difference between the upper and lower limits to characterize the prediction uncertainty and establish the confidence interval width index. The risk control judgment module extracts the upper limit and lower limit of the confidence interval for the same restricted transmission section under the same time slice, performs arithmetic subtraction, and obtains the absolute difference between the upper limit and lower limit of the confidence interval. The risk control judgment module assigns the absolute difference to the confidence interval width index. The confidence interval width index is a quantitative value, and the value of the confidence interval width index is positively correlated with the deviation range of the output result of the previous prediction model. A larger confidence interval width index value indicates an increase in the operational uncertainty faced by the underlying physical power grid.

[0104] S503, the risk control judgment module combines the physical capacity extreme value and the confidence interval width index to perform an inverse proportional function calculation to generate a dynamic truncation threshold. The risk control judgment module reads the static restricted physical limit parameters recorded in the system configuration file and calculates the dynamic truncation threshold using the following formula:

[0105] ;

[0106] In the formula, Index of restricted transmission sections In time slice index The dynamic cutoff threshold is given below, in MW. Index of restricted transmission sections; Index for time slices; Index of restricted transmission sections The corresponding physical capacity extreme value, in MW; Index of restricted transmission sections In time slice index The confidence interval width index, in MW; This represents the uncertainty penalty coefficient.

[0107] The restricted transmission section index is determined by the transmission channel number defined by the external power grid dispatching system, and its value range is determined according to the actual number of restricted power grid sections configured; the time slice index corresponds to the 15-minute timestamp in the basic time series, and is determined by dividing a natural day into 96 equal parts, with a value range of positive integers between 1 and 96; the physical capacity extreme value is preset by the power grid dispatching rules and stored in the system configuration file, and the value range of the physical capacity extreme value is determined by the stability limit of the actual power grid lines; the confidence interval width index is obtained by calculating the absolute difference between the upper limit and the lower limit of the confidence interval.

[0108] The uncertainty penalty coefficient is used to adjust the degree to which the prediction uncertainty index reduces the extreme value of physical capacity. The uncertainty penalty coefficient is determined through multiple linear regression analysis of historical operating data, and its value is set to a floating-point number between 0.1 and 0.5. Through formula calculation, the risk control judgment module numerically binds the reserved power grid safety margin with the prediction uncertainty. When the prediction error amplifies, it automatically lowers the benchmark for determining available capacity and outputs a dynamic truncation threshold to the memory database.

[0109] In step S60, the risk control judgment module blocks invalid calculations by setting calculation boundary conditions to avoid invalid calculations occupying system computing resources. Step S60 further includes the following sub-steps:

[0110] S601, the risk control judgment module retrieves the restricted area local position clusters generated in the aforementioned steps from the storage engine, extracts the equivalent power flow occupancy value from the restricted area local position clusters, and simultaneously retrieves the dynamic truncation threshold calculated in the aforementioned steps. The risk control judgment module compares the equivalent power flow occupancy value derived from each restricted area local position cluster with the dynamic truncation threshold to generate a risk trigger status value. The numerical judgment logic is implemented through the following conditional judgment formula:

[0111] ;

[0112] In the formula, Index of restricted transmission sections In time slice index The risk trigger state value below; For time slices At that time, all relevant node positions were located at the restricted transmission section. The equivalent power flow occupancy generated by the aggregation, in MW; Index of restricted transmission sections In time slice index The dynamic truncation threshold is given below, in MW.

[0113] The restricted transmission section index is determined by the transmission channel number defined by the external power grid dispatching system, and its value range is determined based on the actual number of restricted power grid sections configured; time slice index. The corresponding 15-minute timestamp in the basic time series is determined by dividing a natural day into 96 equal parts, and the value ranges from 1 to 96 as a positive integer.

[0114] In time slice index At that time, the equivalent power flow occupancy value generated by the aggregation of all relevant node positions on the restricted transmission section is obtained through the aforementioned spatial aggregation calculation; the dynamic truncation threshold of the restricted transmission section index under the time slice index is generated through the aforementioned inverse proportional function calculation; the restricted transmission section index under the time slice index The risk trigger status value below is used to indicate whether the current set of nodes is within the physical security boundary, when in the time slice index. When the equivalent power flow occupancy value generated by the aggregation of all relevant node positions on the restricted transmission section is less than the dynamic truncation threshold of the restricted transmission section index under the time slice index, the risk trigger status value of the restricted transmission section index under the time slice index is assigned to 0. When the equivalent power flow occupancy value generated by the aggregation of all relevant node positions on the restricted transmission section is greater than or equal to the dynamic truncation threshold of the restricted transmission section index under the time slice index, the risk triggering state value of the restricted transmission section index under the time slice index is assigned to 1.

[0115] In step S602, when the risk trigger status value of the restricted transmission section index under the time slice index is 0, it indicates that the local position cluster in the restricted area has not exceeded the dynamic truncation threshold. The risk control judgment module performs static assignment of conditional risk value for the local position cluster in the restricted area that has not exceeded the dynamic truncation threshold. Conditional risk value is used to quantify the potential power over-limit gap under extreme scenarios. Static assignment of conditional risk value means directly assigning the conditional risk value to a preset safety constant, instead of performing probabilistic deep calculations on the underlying data. The preset safety constant is uniformly set to 0MW. The risk control judgment module writes the static assignment result of conditional risk value as the underlying parameter to the storage engine. The storage engine updates the data record of the corresponding local position cluster in the restricted area, completing the data writing based on the safety status.

[0116] S603, when the risk trigger state value of the restricted transmission section index under the time slice index is 0, the risk control judgment module blocks the Monte Carlo sampling allocation for safe nodes to release system computing threads and memory usage. Safe nodes refer to physical topology nodes belonging to the local warehouse clusters in restricted areas that have not exceeded the dynamic truncation threshold. Monte Carlo sampling simulates the probability distribution of the power grid operation state by generating a large number of random load scenarios. The Monte Carlo sampling calculation process will occupy a large number of central processing unit calculation cycles and storage space. The risk control judgment module sends an interrupt command to the system's operation scheduler to intercept the entry of local warehouse clusters in restricted areas that have not exceeded the dynamic truncation threshold into the Monte Carlo sampling calculation queue. After receiving the interrupt command, the operation scheduler reclaims the system computing threads pre-allocated to safe nodes and clears the buffer blocks reserved for safe nodes in memory.

[0117] The recycled system computing threads re-enter the system's available resource pool and are subsequently allocated to local warehouse clusters in restricted areas where the risk trigger status value of the restricted transmission section index is 1 under the time slice index for computation. For the specific implementation of multi-threaded concurrency control and memory fragmentation reclamation, those skilled in the art can use conventional operating system resource scheduling techniques.

[0118] In step S70, the instruction generation module receives the data records issued by the risk control judgment module, combines network topology parameters and contract resistance parameters to generate a position adjustment utility ratio, establishes the correlation between physical indicators and financial indicators, and provides data basis for position adjustment. Step S70 further includes the following sub-steps:

[0119] S701, the instruction generation module receives the restricted area local position clusters with a risk trigger status value of 1 under the time slice index from the restricted transmission section index transmitted by the risk control judgment module. The instruction generation module extracts a pointer list of the independent contract record sets contained within the corresponding restricted area local position clusters. The instruction generation module reads the underlying four-dimensional spatiotemporal position matrix through the pointer list of independent contract record sets to obtain the first... Each contract record corresponds to a contract liquidity identifier, and information from the topology nodes. The power transfer distribution factor to the restricted transmission section.

[0120] S702, the instruction generation module performs the relocation utility ratio calculation. The relocation utility ratio measures the efficiency of adjusting the electricity volume of a single contract in alleviating power flow over-limit at a specific restricted transmission section. The instruction generation module calculates the relocation utility ratio using the following formula:

[0121] ;

[0122] In the formula, For the first Contract records relative to restricted transmission section index The efficiency ratio of portfolio rebalancing; To the topology node To restricted transmission sections The power transfer distribution factor; For the first Each contract record corresponds to a contract liquidity indicator; Index of restricted transmission sections; For physical topology node indexes.

[0123] No. The rebalancing utility ratio of each contract record relative to the restricted transmission section index is a dimensionless floating-point value. The larger the rebalancing utility ratio, the more effective the adjustment. The higher the ratio of the physical network security benefits obtained from the contract record to the financial resistance costs incurred; the power transfer distribution factor from the topology node to the restricted transmission section is a floating-point number ranging from -1 to 1, obtained through the aforementioned power transfer distribution factor matrix update process. The formula takes the absolute value of the power transfer distribution factor because both forward and reverse high-sensitivity nodes can provide highly efficient over-limit mitigation capabilities when adjusting the corresponding positions; the first The contract liquidity identifier corresponding to each contract record is a standardized numerical indicator, obtained through the aforementioned commercial terms element conversion process; the restricted transmission section index is determined by the transmission channel number defined by the external power grid dispatching system, and the value range is determined according to the actual number of restricted power grid sections configured; the physical topology node index corresponds to the power grid topology bus node identifier, and the value range is determined according to the total number of bus nodes included in the actual power grid.

[0124] S703, the instruction generation module will calculate the first... The rebalancing utility ratio of each contract record relative to the restricted transmission section index is added as an additional field to the four-dimensional spatiotemporal position matrix, generating a composite index containing the rebalancing utility ratio. The instruction generation module, for restricted transmission section indexes with a risk trigger status value of 1, associates the first local position cluster in the restricted area with the rebalancing utility ratio based on its numerical value. The contract records are sorted in descending order. After the descending sorting operation, the instruction generation module generates a priority rebalancing sequence in memory. The priority rebalancing sequence places the contract record with the highest rebalancing utility ratio at the top of the sequence, which will be the primary processing object for subsequent position adjustment algorithms. For the specific implementation of adding fields to the database to build composite indexes and sorting the dataset in memory, those skilled in the art can use conventional database operations or data structure sorting algorithms.

[0125] In step S80, the instruction generation module performs a quantitative allocation calculation of the excess gap based on the priority rebalancing sequence to generate control instructions. Step S80 further includes the following sub-steps:

[0126] S801, the instruction generation module retrieves the risk triggering state value of the restricted transmission section index under the time slice index. When the risk triggering state value of the restricted transmission section index under the time slice index is 1, the instruction generation module calculates the over-limit power flow gap index. The calculation formula is as follows:

[0127] ;

[0128] In the formula, Index of restricted transmission sections In time slice index The power flow gap indicator below the limit, in MW; For time slices At that time, all relevant node positions were located at the restricted transmission section. The equivalent power flow occupancy generated by the aggregation, in MW; Index of restricted transmission sections In time slice index The dynamic truncation threshold is given below, in MW.

[0129] The restricted transmission section index is determined by the transmission channel number defined by the external power grid dispatching system, and its value range is determined according to the actual number of restricted power grid sections configured. The time slice index corresponds to the 15-minute timestamp in the basic time series, which is determined by dividing a natural day into 96 equal parts, and its value range is a positive integer between 1 and 96. During time slicing, the equivalent power flow occupancy value generated by the aggregation of all relevant node positions on the restricted transmission section is obtained through the aforementioned spatial aggregation calculation. The dynamic truncation threshold of the restricted transmission section index under the time slice index is generated through the aforementioned inverse proportional function calculation.

[0130] The over-limit power flow gap index under the time slice index of the restricted transmission section index is used to quantify the active power value of the current physical section that exceeds the safety boundary. When the risk trigger state value of the restricted transmission section index under the time slice index is 1, the over-limit power flow gap index under the time slice index of the restricted transmission section index is a floating-point number greater than 0.

[0131] S802, the instruction generation module performs an iterative reduction operation based on the priority rebalancing sequence generated in the preceding steps. The instruction generation module extracts the first instruction from the priority rebalancing sequence from top to bottom. For each contract record, the target adjustment electricity volume for a single contract is calculated using the following formula:

[0132] ;

[0133] In the formula, For the first The contract record is in the time slice index. The target adjusted electricity volume is expressed in MWh. For topology nodes Next The contract record is in a time slice The expected power consumption value, in MWh; The remaining overload power flow gap is expressed in MW. To the topology node To restricted transmission sections The power transfer distribution factor; This is the time slice duration conversion factor, in hours; This is a function that takes the minimum value.

[0134] No. The contract record is in a time slice The target adjustment power under the index is the power reduction value determined based on the physical safety boundary, and the value is a floating-point number greater than or equal to 0; the first topology node... The expected electricity value of the contract recorded in the time slice is extracted from the aforementioned four-dimensional spatiotemporal position matrix; the initial value of the remaining over-limit power flow gap index is equal to the restricted transmission section index in the time slice. The out-of-limit power flow gap indicator under the index, the instruction generation module for each time targeting the first After the target adjustment electricity volume is calculated in this contract record, the remaining over-limit power flow gap index is reduced using the power load converted from the target adjustment electricity volume. The reduction and update formula is as follows:

[0135] ;

[0136] in, The remaining overload power flow gap is expressed in MW. This is the time slice duration conversion factor, in hours.

[0137] The instruction generation module updates the values ​​cyclically until the remaining overload power flow gap index is less than or equal to 0 or all contract records in the priority rebalancing sequence have been traversed; the power transfer distribution factor from the topology node to the restricted transmission section is a floating-point number between -1 and 1; the time slice duration conversion factor is a fixed value of 0.25, corresponding to a time span of 15 minutes; the physical topology node index corresponds to the grid topology bus node identifier.

[0138] S803, the instruction generation module stops iterative optimization calculation when the remaining over-limit power flow gap index is less than or equal to 0 or when all records in the priority rebalancing sequence have been traversed. The instruction generation module will then calculate the target adjustment power greater than 0. The field attributes of the contract record are combined with the target adjustment amount to generate a structured optimization instruction. The storage fields of the optimization instruction include the unique identifier of the contract record, the time slice identifier, the original expected amount of electricity, and the target adjustment amount of electricity. The instruction generation module sends the optimization instruction to the power trading execution terminal.

[0139] After the power trading execution terminal parses the optimization instruction, it performs a numerical rewriting operation on the underlying transaction record pointed to by the optimization instruction. As for the specific implementation process of encapsulating structured data into communication messages and sending them to external systems, those skilled in the art can use conventional network socket programming and message queue technology to implement it. The instruction message encapsulation and transmission process is a well-known technology in this field and will not be described in detail here.

[0140] Specific application examples:

[0141] To verify the effectiveness of the AI-based intelligent optimization management method for power trading positions proposed in this invention in actual power grid operation, addressing issues such as power flow exceeding transmission line limits due to high-frequency spot trading coupled with fluctuations in renewable energy output, and the excessively high intervention costs resulting from traditional position adjustment methods failing to consider both physical network and financial constraints, this embodiment is based on a provincial power grid during the summer peak season (with a set time slice index). This section provides a detailed explanation of the actual spot trading and operational monitoring scenarios from 14:00 to 14:15.

[0142] This embodiment is combined with the appendix Figure 3 and attached Figure 4 The data is presented and attached. Figure 3 and attached Figure 4 The data in this document are all comparative test data of the indicator results calculated in real time by the system in the in-memory database and the traditional proportional reduction strategy.

[0143] Before executing the location optimization task, the system obtains the underlying power grid parameters and the preceding AI prediction data. In this embodiment, the transmission channel L-01 defined by the power grid dispatching system is the current restricted transmission section (restricted transmission section index). ).

[0144] The physical capacity limit of the restricted transmission section L-01 preset in the system configuration file The risk control judgment module establishes a data connection with the pre-prediction model based on a Long Short-Term Memory (LSTM) network and retrieves time slices. The predicted confidence interval data is used to extract the upper and lower limits of the confidence interval, and then arithmetic subtraction is performed to obtain the confidence interval width index. The system extracts historical operational data to determine the uncertainty penalty coefficient. .

[0145] The risk control judgment module combines the physical capacity extreme value and the confidence interval width index to perform an inverse proportional function calculation, based on the formula. The dynamic cutoff threshold for this period is calculated. This indicates that after taking into account the prediction error of the AI ​​model output, the system automatically lowered the safety judgment benchmark from 800MW to 740MW, reserving a safety margin for the power grid.

[0146] The data alignment module has parsed the heterogeneous contracts into 15-minute basic time series and constructed a four-dimensional spatiotemporal location matrix in the storage engine. The topology aggregation module utilizes the currently active power transfer distribution factor matrix, according to the formula... Perform spatial aggregation calculations.

[0147] Combined with appendix Figure 3 As shown in the line chart, the horizontal axis represents the actual time corresponding to the time slice index (13:00 to 15:00), and the vertical axis represents the active power (MW). The gray dashed line represents the dynamic cutoff threshold, and the black solid line represents the equivalent power flow occupancy value. During the period from 13:00 to 13:45, the solid line is stably below the dashed line, and the system is in a safe range without exceeding the limits. When the system reaches 14:00 (slice...), the line cutoff threshold is reached. At that time, due to the surge in high-frequency spot trading volume at multiple nodes, the equivalent power flow occupancy value was calculated through spatial aggregation. .

[0148] The risk control judgment module compares the equivalent power flow occupancy value (770MW) with the dynamic cutoff threshold (740MW). Since 770 ≥ 740, the risk control judgment module, based on the conditional judgment formula, determines the risk trigger state value of the restricted transmission section L-01 under slice 56. When the value is set to 1, the risk control judgment module blocks the Monte Carlo sampling allocation for safe nodes that have not exceeded the limit in order to release the system's computing thread, and transmits the data records inside the local position cluster in the restricted area to the instruction generation module.

[0149] To minimize financial resistance costs while eliminating overruns, the instruction generation module performs composite index sorting and quantitative allocation calculations.

[0150] Reading the underlying four-dimensional spatiotemporal position matrix, the restricted region's local position cluster mainly contains elements belonging to node A (index). ) and node B (index) Two independent contract records, node A is associated with contract record 1 ( ): Expected power consumption The penalty rate for breach of contract is extracted from the original breach of contract clause. Time remaining until delivery Consult the rule configuration table to obtain the additional resistance constant. The power transfer distribution factor from node A to section L-01 .

[0151] Node B associated contract record 2 ( ): Expected power consumption Default penalty rate Time remaining until delivery Additional resistance constant The power transfer distribution factor from node B to section L-01 The system presets static weight coefficients. , .

[0152] Calculate the contract liquidity identifier based on the rule base:

[0153] Contract 1: .

[0154] Contract 2: .

[0155] Calculate the portfolio adjustment efficiency ratio:

[0156] Contract 1: .

[0157] Contract 2: .

[0158] The instruction generation module sorts restricted transmission sections with a risk trigger status value of 1 in descending order based on their relocation utility ratio. Generate a priority rebalancing sequence containing Contract 1 to Contract 2, and place Contract 1 as the first priority.

[0159] Calculate the initial overload power flow gap index The initial value of the remaining over-limit power flow gap indicator Extract the duration conversion factor for the time slice corresponding to 15 minutes. Extract Contract 1 from the top of the sequence:

[0160] Calculate target adjustment power Execute the reversal update formula: If the remaining out-of-limit power flow gap indicator is greater than 0, continue extracting Contract 2:

[0161] Calculate target adjustment power Execute the reversal update formula: At this point, the remaining overload power flow gap index is less than or equal to 0. The instruction generation module stops iterative optimization calculation and encapsulates the target adjustment electricity of 100MWh for Contract 1 and the target adjustment electricity of 133.33MWh for Contract 2 into a structured optimization instruction and sends it to the power trading execution terminal to perform the value rewriting operation.

[0162] To verify the specific effects of the technical solution of this application, the rebalancing strategy based on dynamic topology association output by this system was compared with the traditional proportional reduction strategy that only considers the amount of electricity and does not distinguish the liquidity of contracts in an engineering comparison test.

[0163] Combined with appendix Figure 4 As can be seen from the comparison bar chart, the horizontal axis represents different evaluation dimensions, namely the comprehensive rebalancing cost (ten thousand yuan) and the calculation time for eliminating the limit (seconds). The light gray bars represent the traditional proportional reduction strategy, while the dark gray bars represent the intelligent optimization strategy of this invention. In terms of the comprehensive rebalancing cost indicator, the traditional strategy ignores the differences in contract period and default rate and forcibly apportions the cost based on the declared volume, resulting in high financial penalties. The comprehensive rebalancing cost reaches 420,000 yuan (light gray bars).

[0164] This invention, through portfolio adjustment utility ratio calculation, prioritizes intervention for contracts with high physical contribution and low default resistance constant. The dark gray bars show that the overall portfolio adjustment cost plummeted to 180,000 yuan, a reduction of 57.1%, establishing an effective correlation between physical and financial indicators. Regarding the calculation time for exceeding limits, the traditional model performs redundant probability calculations on all network nodes. The risk control judgment module of this invention pre-sets a dynamic truncation threshold, accurately blocking invalid calculations and reclaiming system memory. This invention compresses the calculation time from 8.5 seconds to 1.2 seconds.

[0165] Summary of application examples:

[0166] This embodiment fully verifies the system efficiency of the AI-based intelligent optimization management method for power trading positions in actual operation. The system uses the predicted confidence interval output by the LSTM pre-model to correct physical extreme values ​​and generate a dynamic truncation threshold to block the calculation of safe nodes. When faced with a certain power flow limit exceedance, the instruction generation module introduces a position adjustment utility ratio index that includes power transfer distribution factors and commercial terms. This not only accurately smooths out the equivalent power flow gap in the physical dimension, but also avoids the default risk of high liquidity resistance contracts in the financial dimension, thus balancing the safety of power system operation and the economic benefits of market transactions.

Claims

1. An AI-based intelligent optimization management method for power trading positions, characterized in that, Includes the following steps: Obtain low-frequency business contract data and high-frequency spot market volume data, and then split the low-frequency business contract data into a basic time series. Bind the power grid topology node identifier to the basic time series, calculate the contract liquidity identifier, and establish a four-dimensional spatiotemporal position matrix; Acquire a snapshot of the power grid topology and update the power transfer distribution factor matrix when the state changes; The power transfer distribution factor matrix is ​​used to perform calculations on the four-dimensional spatiotemporal storage matrix to generate a constrained area local storage cluster containing equivalent power flow occupancy values; Obtain the predicted confidence interval data and calculate the dynamic truncation threshold; The equivalent power flow occupancy value of the local position cluster in the restricted area is compared with the dynamic cutoff threshold. When the equivalent power flow occupancy value exceeds the dynamic cutoff threshold, the absolute value of the power transfer distribution factor in the power transfer distribution factor matrix is ​​extracted and compared with the contract liquidity identifier. The ratio is calculated to generate a composite index of rebalancing utility ratio. Based on the composite index of rebalancing utility ratio, the data records in the local position cluster of the restricted area are sorted in descending order to generate a priority rebalancing sequence. The data records at the front of the priority rebalancing sequence are extracted to generate an optimization instruction and output.

2. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, The steps of acquiring low-frequency business contract data and high-frequency spot market volume data, and then splitting the low-frequency business contract data into a basic time series, specifically include: Extract the decomposition curve type identifier from the low-frequency business contract data; When the decomposition curve type is identified as a flat base load curve, the total contracted electricity volume in the low-frequency business contract data is proportionally distributed to each contracted time slice within the contract effective period. When the decomposition curve type is identified as a peak-valley segmented curve, the parameters for dividing the peak period, stable period and low period as defined in the transaction rules, as well as the peak-valley electricity conversion ratio, are read. Based on the peak-valley electricity conversion ratio, the total contracted electricity is unevenly distributed to the corresponding time segments. When the decomposition curve type is identified as a custom historical fitting curve, the load shape coefficient matrix is ​​retrieved, and the load shape coefficient matrix is ​​used to perform a weighted calculation on the total contracted electricity volume. The electricity values ​​allocated to the corresponding time intervals are converted into the basic time series with a fixed minimum unit.

3. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, The specific steps for binding the basic time series with power grid topology node identifiers, calculating contract liquidity identifiers, and establishing a four-dimensional spatiotemporal position matrix include: Extract the transaction entity code attached to the basic time series, retrieve the asset topology mapping table that records the mapping relationship between the transaction entity code and the power grid topology node identifier, and bind the basic time series with the power grid topology node identifier; The default penalty rate, remaining time to delivery, and additional resistance constant contained in the original contract records corresponding to the basic time series are analyzed and weighted summed with preset static weight coefficients to generate the contract liquidity identifier. Retrieve the expected electricity value corresponding to the base time series; The four-dimensional spatiotemporal position matrix is ​​constructed by combining the timestamp of the basic time series, the bound grid topology node identifier, the expected electricity value, and the calculated contract liquidity identifier.

4. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, The steps of obtaining a snapshot of the power grid topology and updating the power transfer distribution factor matrix when the state changes specifically include: Retrieve a snapshot of the historical topology cached in local memory; The current power grid topology snapshot is compared with the historical topology snapshot item by item to identify whether there is topology status change data. The topology status change data includes the switching status reversal of physical lines or the relative change of branch reactance parameters exceeding the preset safety tolerance range. Upon detecting the topology state change data, a parameter recalculation trigger signal is generated; Upon receiving the parameter recalculation trigger signal, the latest power grid topology snapshot data is read, the network parameters of the changed area are extracted, and the local power transfer distribution factor is calculated based on the DC power flow model to generate the latest power transfer distribution factor matrix.

5. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, The specific steps of using the power transfer distribution factor matrix to calculate the four-dimensional spatiotemporal storage matrix and generating a constrained region local storage cluster containing equivalent power flow occupancy values ​​include: The expected power values ​​of each node under the preset time slice are superimposed and converted under the grid spatial constraints. By judging whether the power transfer distribution factor in the power transfer distribution factor matrix is ​​greater than the preset correlation sensitivity threshold, the set of topological nodes that are electrically coupled with the restricted transmission section is determined. The expected power value of each node is multiplied by the power transfer distribution factor and aggregated to obtain the equivalent power flow occupancy value generated by the aggregation of all relevant node positions on the restricted transmission section. The restricted transmission section index and the time slice index of the preset time slice are then extracted. Logically group the underlying records according to the restricted transmission section index and the time slice index to generate the restricted area local warehouse cluster of the structured data set.

6. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, The steps of obtaining the predicted confidence interval data and calculating the dynamic truncation threshold specifically include: Obtain the prediction confidence interval data for the time slice corresponding to the pre-prediction model, and separate the upper confidence interval value and the lower confidence interval value from the prediction confidence interval data; Obtain the absolute difference between the upper limit of the confidence interval and the lower limit of the confidence interval, and assign the absolute difference to the confidence interval width index; Read the static restricted physical limit parameters recorded in the system configuration file, obtain the physical capacity extreme value corresponding to the static restricted physical limit parameters, and generate the dynamic truncation threshold by subtracting the product of the uncertainty penalty coefficient and the confidence interval width index from the physical capacity extreme value.

7. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, After comparing the equivalent power flow occupancy value of the local storage cluster in the restricted area with the dynamic truncation threshold, the method further includes: When the equivalent power flow occupancy value is less than the dynamic cutoff threshold, a static assignment of conditional risk value is performed on the local position clusters in the restricted area that have not exceeded the dynamic cutoff threshold, and the conditional risk value is assigned to a preset safety constant. Monte Carlo sampling calculations are blocked for physical topology nodes belonging to the restricted area local warehouse clusters that have not exceeded the dynamic truncation threshold. System operation threads pre-allocated to safe nodes are reclaimed, and buffer blocks reserved for safe nodes in memory are cleared.

8. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, The step of extracting the absolute value of the power transfer distribution factor and calculating the ratio with the contract liquidity identifier to generate a composite index for rebalancing utility ratio specifically includes: Extract a list of pointers to the set of independent contract records contained within the local position cluster in the restricted area, and read the corresponding contract liquidity identifier and the power transfer distribution factor; The rebalancing utility ratio is calculated by dividing the absolute value of the power transfer distribution factor by the contract liquidity identifier. The rebalancing utility ratio is added as an additional field to the four-dimensional spatiotemporal position matrix to generate the rebalancing utility ratio composite index.

9. The AI-based intelligent optimization management method for power trading positions according to claim 1, characterized in that, The extraction of the data records at the front end of the priority rebalancing sequence includes: The over-limit power flow gap index is calculated by subtracting the dynamic cutoff threshold from the equivalent power flow occupancy value, the remaining over-limit power flow gap index is set, and the initial value of the remaining over-limit power flow gap index is set as the over-limit power flow gap index. According to the priority relocation sequence, contract records are extracted one by one from top to bottom. The minimum value between the expected power volume value and the product of the remaining over-limit power flow gap index divided by the absolute value of the power transfer distribution factor and the time slice duration conversion factor is selected as the target adjustment power volume. The remaining overload power flow gap index is reduced by the power load after the target power adjustment conversion and updated cyclically until the remaining overload power flow gap index is less than or equal to zero or all contract records in the priority rebalancing sequence have been traversed.

10. The AI-based intelligent optimization management method for power trading positions according to claim 9, characterized in that, The step of extracting the data records at the front end of the priority rebalancing sequence to generate optimization instructions and outputting them specifically includes: The unique identifier of the contract record with the target adjustment amount greater than zero, the time slice identifier, the original expected amount value, and the target adjustment amount are combined to generate the structured optimization instruction and sent to the power trading execution terminal.

Images