Power resource deployment method and system for new energy access

Abstract

The application discloses a power resource deployment method and system for new energy access, and particularly relates to the field of new energy grid connection scheduling, and is used for solving the problem of frequency fluctuation and wind and light abandonment caused by flexible capacity configuration lag. The application is characterized in that the evolution track of wind and light prediction error is described in real time, and the evolution track is directly converted into climbing demand changing with time. A flexible capacity weighting coefficient generated by machine learning is introduced in the scheduling decision link, so that the available margin of energy storage and heat engine is immediately adjusted according to fluctuation intensity. The capacity configuration is then subjected to power flow sensitivity checking to exclude network bottlenecks, and market quotation is used to complete economic sorting and write the execution feedback back to the model, so as to continuously improve the judgment accuracy of the next round. The capacity scheduling is closely related to the instantaneous change without excessive risk, the passive abandonment limit is no longer in the sudden drop scene, the frequency offset is reduced, the wind and light abandonment is reduced to a negligible level, the scheduling cost is synchronously converged with the fine calling of flexible resources, and finally the stability, economy and green are considered under the high proportion of new energy grid connection.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new energy grid-connected scheduling, and more particularly, to a power resource deployment method and system for new energy access. BACKGROUND

[0002] After high-proportion wind and light power are connected to the grid, a real-time scheduling link often pre-sets an upward climbing ability curve, leaves a margin in a direct superposition manner of predicted net load and static uncertainty envelope. The dispatch center arranges start-stop of thermal power units and charging-discharging of energy storage according to the curve, and executes rollingly with peak-valley period switching as the rhythm. When wind and light output suffers from cloud blocking or wind speed instability, the prediction error rapidly accumulates along the time axis, the climbing demand surges but cannot trigger adaptive revision. The margin is overdrawn in a short time, and frequency control can only be recovered through power abandonment or power cut, causing yield shrinkage on the power generation side and power supply reliability fluctuation on the user side. The scenario reveals that the real-time climbing demand is set only by the prediction initial value, lacking closed-loop logic adjustment according to the actual error trajectory, which will reveal capacity gap when the fluctuation amplitude is amplified.

[0003] The disclosed patent (CN115936407A considers the power system optimization scheduling method and storage medium of new energy uncertainty) adopts a two-stage distribution robust model, fixes the flexible climbing demand in the first stage into the decision variable, and measures the output error with a single Wasserstein radius. This method assumes that the error distribution remains stable during the operation period, does not update the real-time measurement error and climbing demand, and does not set a margin to supplement the path. When a sudden drop event deviates from the prediction envelope, the start-stop rhythm of thermal power is still constrained by the first stage, the energy storage response window is passive misalignment, resulting in centralized issuance of frequency adjustment instructions and priority sorting conflict. The root cause of the problem is that the capacity configuration logic is static: the reserved climbing ability does not form a feedback channel with error evolution, the risk is hidden in the early stage of the decision chain, and only extreme fluctuations appear as high-cost emergency measures.

[0004] To solve the above problems, a technical solution is provided. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present application provide a power resource deployment method and system for new energy access, which depicts the evolution track of wind and light prediction error in real time and directly converts it into ramping demand changing with time, introduces the flexible capacity weighting coefficient generated by machine learning into the dispatching decision link, so that the available margin of energy storage and heat engine is immediately adjusted according to the fluctuation intensity; the capacity configuration then receives the power flow sensitivity check to exclude the network bottleneck, and then completes the economic sorting with market quotation and feeds back the execution to the model for continuous improvement of the judgment accuracy of the next round; the capacity scheduling closely follows the rapid changes without excessive risk, the passive abandonment of limit in sudden drop scenarios is no longer passive, the frequency deviation is reduced, the abandoned wind and light is reduced to a negligible level, the dispatching cost is converged with the fine calling of flexible resources, and finally the stability, economy and green are considered under the high proportion of new energy grid connection, so as to solve the problems raised in the above background art.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] The power resource deployment method for new energy access comprises the following steps:

[0008] S1: receiving the real-time measurement sequence and the same period prediction sequence of wind and light, calculating the difference to form the error track;

[0009] S2: calculating the slope and duration of the error track, and constructing the ramping demand prediction matrix;

[0010] S3: collecting the unit start-stop constraint, the energy storage state of charge and the demand response window, mapping the flexible capacity weighting coefficient through the preset adaptive kernel ridge regression model according to the average slope and duration in the ramping demand prediction matrix, the energy storage amplitude modulation potential and the unit load margin, generating a candidate capacity configuration vector and writing it into a dispatch candidate table;

[0011] S4: performing power flow sensitivity check on the capacity configuration vector in the dispatch candidate table, and writing it into the bottleneck-free buffer area after excluding the bottleneck triggering scheme;

[0012] S5: extracting the capacity configuration vector from the bottleneck-free buffer area, implementing economic sorting on the bottleneck-free vector based on the real-time market quotation sequence, outputting the optimal dispatching instruction and issuing it to the execution layer, and at the same time, feeding back the adaptive kernel ridge regression model for subsequent iteration.

[0013] In a preferred embodiment, step S1 comprises the following contents:

[0014] Real-time measurement data and contemporaneous prediction data of the wind-solar power plant are acquired through a supervisory control and data acquisition system of a power system, and time marks of the real-time measurement data and the contemporaneous prediction data are ensured to be aligned; for each pair of time-aligned real-time measurement data and contemporaneous prediction data, a difference between a real-time measurement power value and a contemporaneous prediction power value is calculated to generate an error value sequence; the error value sequence is arranged in time sequence to form an error trajectory.

[0015] In a preferred embodiment, step S2 includes the following contents:

[0016] A slope calculation is performed on the error trajectory, a ratio of error values of adjacent time points to time intervals is used, and a smooth slope sequence is generated through a three-point sliding window method; a duration calculation is performed on the error trajectory, a backtracking method is used to determine a duration of an error state of each time point, and a duration sequence is generated; the smooth slope sequence and the duration sequence are combined in time sequence into a ramping demand prediction matrix, each row of the matrix corresponds to a time point, and includes a smooth slope value and a duration value of the time point.

[0017] In a preferred embodiment, step S3 includes the following contents:

[0018] Start-up and shutdown time limits of thermal power units, ramping rate limits, state of charge of energy storage units, available time windows and maximum response power of demand response resources are acquired from a power market management database, a supervisory control and data acquisition (SCADA) database and a demand side management database through a real-time data interface, and the data are classified and stored into a resource state database.

[0019] In a preferred embodiment, step S3 further includes the following contents:

[0020] Based on data in the resource state database, a residual charge-discharge power difference of each energy storage unit is calculated, and the power difference is divided by a base load power to obtain a normalized power difference; the normalized power differences of all energy storage units are summed and multiplied by a dispatch window length to obtain an energy storage amplitude release index; a ramping rate difference and a weighted ramping rate difference of each thermal power unit are calculated, and a unit load response margin index is further calculated, which is obtained by dividing the weighted ramping rate difference of all thermal power units by a total maximum allowed ramping rate.

[0021] In a preferred embodiment, step S3 further includes the following contents:

[0022] The energy storage amplitude release index, the unit load response margin index, an average slope and a duration in the ramping demand prediction matrix are input into a pre-trained adaptive kernel ridge regression model to generate a flexible capacity weighting coefficient, and the initial candidate capacity configuration vector is adjusted by using the flexible capacity weighting coefficient to obtain an adjusted candidate capacity configuration vector, which is written into a dispatch candidate table.

[0023] In a preferred embodiment, step S4 includes the following:

[0024] Extracting the power flow sensitivity matrix and the maximum allowable power flow of the critical branches from the power network database; for each capacity configuration vector in the dispatch candidate table, calculating the power flow increment of all critical branches, which is calculated by multiplying the capacity adjustment amount of each node in the capacity configuration vector with the sensitivity value of the corresponding node pair to the specific branch in the power flow sensitivity matrix and summing up; dividing the power flow increment by the maximum allowable power flow of the critical branches to calculate the power flow sensitivity.

[0025] In a preferred embodiment, step S4 further includes the following:

[0026] According to the predetermined power flow sensitivity threshold, checking the power flow sensitivity value of each capacity configuration vector to all critical branches, if the power flow sensitivity value of any critical branch is greater than the power flow sensitivity threshold, marking the corresponding capacity configuration vector as having a bottleneck, if the power flow sensitivity value of all critical branches is less than or equal to the power flow sensitivity threshold, marking as no bottleneck; filtering out the capacity configuration vectors marked as no bottleneck from the dispatch candidate table, writing these capacity configuration vectors into the bottleneck-free buffer area, and appending the generation time metadata.

[0027] In a preferred embodiment, step S5 includes the following:

[0028] Obtaining the real-time market quotation sequence from the power market database, and extracting the capacity configuration vectors from the bottleneck-free buffer area; for each capacity configuration vector, identifying its resource type, extracting the corresponding real-time market quotation, calculating the dispatch cost, and appending the dispatch cost to the record of the capacity configuration vector; sorting the dispatch costs of the capacity configuration vectors in ascending order, selecting the capacity configuration vector with the lowest dispatch cost as the optimal dispatch instruction, and formatting it before issuing it to the execution layer; collecting the feedback data after execution, storing it in the feedback database, and updating the parameters of the adaptive kernel ridge regression model using the feedback data according to the preset period to optimize subsequent dispatch.

[0029] The power resource deployment system for new energy access includes:

[0030] Difference track generation module: receiving wind and light real-time measurement sequence and contemporaneous prediction sequence, calculating difference to form error track;

[0031] Demand shaping module: calculating slope and duration of error track, constructing ramping demand prediction matrix;

[0032] Flexible mapping module: The unit start-stop constraint, the energy storage state of charge and the demand response window are collected, the energy storage amplitude modulation potential, the unit load margin and the average slope and duration in the climbing demand prediction matrix are mapped to generate a flexible capacity weighting coefficient through a preset adaptive kernel ridge regression model, and a candidate capacity configuration vector is generated according to the flexible capacity weighting coefficient and written into a scheduling candidate table;

[0033] Bottleneck screening module: The capacity configuration vector in the scheduling candidate table is subjected to a power flow sensitivity check, and the bottleneck scheme is removed to write into a bottleneck-free buffer area;

[0034] Economic dispatching module: The capacity configuration vector is extracted from the bottleneck-free buffer area, the bottleneck-free vector is subjected to economic sorting based on a real-time market quotation sequence, and the optimal dispatching instruction is output and executed in the execution layer, and the adaptive kernel ridge regression model is fed back and written for subsequent iteration.

[0035] The technical effects and advantages of the power resource deployment method and system for new energy access of the present application are as follows:

[0036] The present application depicts the evolution trajectory of wind and light prediction error in real time and directly converts it into a climbing demand varying with time, introduces a flexible capacity weighting coefficient generated by machine learning in the dispatching decision-making link, adjusts the available margin of energy storage and thermal machine in real time according to the fluctuation intensity, and then subjects the capacity configuration to a power flow sensitivity check to exclude the network bottleneck, and completes economic sorting with market quotation and feeds back the execution to write the model, continuously improves the judgment accuracy of the next round, makes the capacity scheduling closely follow the instantaneous changes without excessive risk, reduces the frequency deviation, reduces the abandoned wind and light to a negligible level, synchronously converges the dispatching cost with the fine calling of flexible resources, and finally realizes the stability, economy and green of high proportion of new energy grid connection. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The flowchart of the power resource deployment method for new energy access of the present application is shown.

[0038] Figure 2 The structure diagram of the power resource deployment system for new energy access of the present application is shown. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0040] Embodiment 1: Figure 1The power resource deployment method for new energy access is given, comprising:

[0041] S1: receiving wind and light real-time measurement sequence and contemporaneous prediction sequence, calculating difference value to form error trajectory.

[0042] S2: calculating the slope and duration of the error trajectory, and constructing the climbing demand prediction matrix.

[0043] S3: collecting unit start-stop constraints, energy storage state of charge and demand response window, mapping the energy storage amplitude modulation potential, unit load margin and the average slope and duration in the climbing demand prediction matrix through the preset adaptive kernel ridge regression model to generate flexible capacity weighting coefficient, and generating candidate capacity configuration vector according to the flexible capacity weighting coefficient and writing into the dispatch candidate table.

[0044] S4: performing power flow sensitivity check on the capacity configuration vector in the dispatch candidate table, and writing into the bottleneck-free buffer area after eliminating the trigger bottleneck scheme.

[0045] S5: extracting the capacity configuration vector from the bottleneck-free buffer area, implementing economic sorting on the bottleneck-free vector based on the real-time market quotation sequence, outputting the optimal dispatch instruction and issuing the execution layer, and simultaneously feeding back to the adaptive kernel ridge regression model for subsequent iteration.

[0046] Under the background of high proportion of new energy grid connection, the wind and light power generation output has significant uncertainty and volatility, which poses a severe challenge to the real-time scheduling of power resources. The traditional scheduling method usually relies on the predicted net load superimposed on the static uncertainty envelope to reserve the climbing capacity, but when the wind speed is unstable or the cloud cluster is blocked, the predicted error accumulates rapidly, and the preset climbing margin is difficult to adapt to the changes of real-time demand. This static reservation method lacks a closed-loop feedback mechanism with the actual error evolution, which may lead to insufficient capacity, frequency out of control, and even high-cost emergency measures such as power rejection or load shedding. The present application proposes a power resource deployment method for new energy access, which monitors the measurement and prediction data of wind and light power generation in real time, calculates the error trajectory, and provides basic data for subsequent dynamic adjustment of climbing demand and capacity configuration, thereby improving the stability, economy and green operation ability of the system. Step S1 as the starting point of the whole scheme focuses on the generation of error trajectory to provide accurate time-varying input for the subsequent steps.

[0047] The goal of step S1 is to receive the real-time measurement sequence and contemporaneous prediction sequence of wind and light power generation, calculate the difference between the two, and form an error trajectory to lay the data foundation for subsequent climbing demand prediction. The following is the detailed technical logic:

[0048] 1.1, data receiving;

[0049] Firstly, real-time measurement data and contemporaneous prediction data of the wind-solar power plant are needed to capture the characteristics of wind-solar power output over time. Real-time measurement data is obtained through the power system monitoring and data acquisition system, which uses sensors and communication equipment to record the actual power generation of wind turbines or photovoltaic arrays at each moment in real time, with a time resolution of seconds or minutes. Contemporaneous prediction data comes from the results of meteorological models or historical data analysis, which estimates the wind speed, irradiance and other parameters at a certain time in the future through numerical calculation methods, and then calculates the corresponding power generation. The time resolution is consistent with the real-time measurement data. To ensure the consistency of the two sets of data in subsequent processing, the time markers of each real-time measurement data and contemporaneous prediction data are compared and adjusted to make their time stamps completely correspond.

[0050] 1.2, error calculation;

[0051] After completing data reception, real-time measurement data and contemporaneous prediction data are used to calculate the prediction error at each moment to quantify the difference between wind-solar power output prediction and actual performance. For each pair of time-aligned data, the calculation process is as follows: take the real-time measurement power value at a certain moment, subtract the contemporaneous prediction power value at the same moment, and the result is the error value at that moment. In order to preserve the instantaneous characteristics of wind-solar power output fluctuations, no smoothing or averaging operations are introduced in the calculation process, and the original difference value is directly used as the error result. This direct difference calculation method can quickly reflect sudden changes in output, such as sudden drops in photovoltaic output caused by cloud cover or fluctuations in wind power caused by sudden changes in wind speed.

[0052] 1.3, error trajectory formation;

[0053] After obtaining the error value sequence, it is organized in chronological order to form a continuous error trajectory to describe the evolution of prediction error over time. The specific operation is to associate each error value calculated in the previous step with its corresponding time marker to form an ordered time-error data pair set. To ensure the integrity of the trajectory, all calculated error values are included in the sequence without missing any moment data or performing interpolation or deletion operations. In this way, the error trajectory not only records the size of the prediction bias at each moment, but also shows the trend of the bias over time, such as the transition of error from positive to negative or the sustained fluctuation amplitude.

[0054] The above data receiving process provides synchronized input data for error calculation through time alignment, ensuring the accuracy of error values; the error calculation process quantifies the deviation between prediction and actual output through per-time difference operation and passes the result to the error trajectory formation process; the error trajectory formation process finally organizes discrete error values into continuous time series, fully presenting the dynamic characteristics of wind and light power generation output uncertainty.

[0055] Step S1 forms an error trajectory by calculating the difference between the real-time measurement sequence and the contemporaneous prediction sequence, providing basic data for subsequent analysis. However, the error trajectory itself only reflects the deviation between prediction and actual output, and cannot directly guide scheduling decisions, so further extraction of its dynamic characteristics is needed to quantify the ramping demand. Step S2 focuses on analyzing the change characteristics of the error trajectory and constructing a ramping demand prediction matrix to provide time-varying demand input for step S3, thus realizing the transition of scheduling from static margin to dynamic response.

[0056] The goal of step S2 is to extract features from the error trajectory generated in step S1, calculate its slope and duration, and construct a ramping demand prediction matrix to provide quantitative time-varying ramping demand for subsequent capacity allocation. The following is the detailed technical logic:

[0057] 2.1, error trajectory slope calculation;

[0058] First, the slope of the error trajectory is calculated to quantify the rate of change of the prediction error over time. The error trajectory is composed of a series of discrete time points and corresponding error values, which represent the deviation between actual and predicted wind and light power generation output.

[0059] The slope calculation aims to capture the instantaneous rate of error change, providing dynamic characteristics for subsequent analysis. The specific operation process is as follows: for each time point, subtract the error value of the previous time point from the error value of the current time point to obtain the error difference between the two time points, then divide the error difference by the time interval between the two time points to calculate the slope value of the time point, with the unit of megawatt per hour. This method reflects the change trend of the error trajectory at each time point through the difference ratio of adjacent time points.

[0060] Since the wind and light power output fluctuation may contain short-term random interference, directly using the instantaneous slope may lead to unstable results, therefore a three-point sliding window method is further used for smoothing processing. Specifically, for the slope value of each time point, the slope value of the time point is added to the slope value of the previous time point and the slope value of the next time point, and then the sum is divided by three to obtain the smoothed slope value of the time point. For the first time point and the last time point of the sequence, since the previous or next slope value is missing, the slope value of the first time point is added to the slope value of the second time point and then divided by two, or the slope value of the last time point is added to the slope value of the second last time point and then divided by two, as the smoothed value. Through this smoothing processing, short-term fluctuations are effectively filtered out, the medium and long-term change trend of the error trajectory is retained, and the generated smoothed slope sequence provides a more stable dynamic change index for subsequent prediction.

[0061] 2.2, duration calculation;

[0062] After the slope calculation is completed, the duration of the error trajectory is calculated to characterize the duration of the prediction error in a certain state. The error value is divided into positive error and negative error, the positive error represents that the actual output is higher than the predicted value, the negative error represents that the actual output is lower than the predicted value, and the duration calculation aims to reflect the persistence of the error state. The specific operation process is that, for each time point, according to the positive and negative state of the error value of the current time point, the time sequence is traced back, the earliest time point from which the same positive and negative state is continuously maintained since a certain earlier time point is found, and then the time length from the earliest time point to the current time point is calculated as the duration of the current time point.

[0063] For example, if the error value of the current time point is positive, the time sequence is checked forward, and the error value of the previous time point is sequentially checked from the current time point, if they are all positive, the tracing back is continued until the first negative error value or the sequence starting point is encountered, and the time difference from the earliest continuous positive error time point to the current time point is recorded as the duration. If the error value of the current time point is negative, the earliest continuous negative error time point is traced back in the same way, and the time difference is calculated. If the current time point is the sequence starting point, the duration is zero. This backtracking calculation method generates a duration sequence by tracking the time period of continuous same state, and completely reflects the time extension characteristics of the error state, providing quantitative support in the time dimension for subsequent analysis.

[0064] 2.3, construction of ramping demand prediction matrix;

[0065] After the completion of the slope calculation and the duration calculation, the smooth slope sequence and the duration sequence are integrated into a ramping demand prediction matrix to present the time-varying demand characteristics in a structured form. The ramping demand prediction matrix is a two-dimensional matrix, each row corresponding to a time point, and each column representing the smooth slope value and the duration value of the time point respectively, aiming to provide direct input for subsequent capacity configuration.

[0066] The specific construction process is to arrange the smooth slope sequence and the duration sequence in chronological order. For each time point, the smooth slope value of the time point is taken as the value of the first column of the matrix, and the duration value of the time point is taken as the value of the second column of the matrix. All time points are processed in turn to complete the filling of the matrix. For example, for the first time point, the first row and the first column of the matrix are filled with the smooth slope value, and the first row and the second column are filled with the duration value; for the second time point, the second row and the first column of the matrix are filled with the smooth slope value, and the second row and the second column are filled with the duration value, and so on, until all time points are processed.

[0067] The generated matrix combines the dynamic change rate represented by the smooth slope and the persistent length represented by the duration through the structured organization of columns, completely describes the time-varying characteristics of the error trajectory, and provides quantified data that can be directly called by the scheduling system.

[0068] The error trajectory slope calculation generates a stable dynamic change index through smoothing processing, and the duration calculation extracts the persistent characteristics of the error state through the backtracking method. The results of the two are integrated into a structured output in the construction of the ramping demand prediction matrix.

[0069] Step S2 extracts the slope and duration characteristics based on the error trajectory, constructs the ramping demand prediction matrix, and clearly defines the time-varying characteristics of the demand side. However, relying solely on demand-side characteristics is difficult to fully address the complex constraints in the new energy grid-connected scenario, and further integration of supply-side resource characteristics and operation restrictions is needed to achieve dynamic matching of capacity configuration. Step S3 takes the output of the ramping demand prediction matrix and focuses on integrating thermal power unit start-stop constraints, energy storage state of charge, and demand response windows. By quantifying the flexibility potential of the supply side and introducing an adaptive method, a candidate capacity configuration vector is generated to provide alternative solutions for subsequent power flow checking and economic optimization.

[0070] The goal of step S3 is to calculate the flexibility capability index of energy storage and thermal power units based on the ramping demand prediction matrix, thermal power unit start-stop constraints, energy storage state of charge, and demand response windows, and to generate flexible capacity weighting coefficients through adaptive kernel ridge regression to finally form a candidate capacity configuration vector. The following is the detailed technical logic:

[0071] 3.1, gather resource constraints and states;

[0072] Firstly, real-time information of both supply and demand sides needs to be integrated to fully grasp the operating conditions of the power system. The specific process is to extract the start-stop time limit and ramp rate limit of thermal power units from the power market management database through the real-time data interface, where the start-stop time limit refers to the minimum time required for a thermal power unit to start or stop, and the ramp rate limit refers to the power change rate that a thermal power unit can adjust in a unit of time. At the same time, the state of charge of the energy storage unit is obtained from the SCADA database, which represents the current power level of the energy storage unit, ranging from 0% to 100%, reflecting the available charging or discharging capacity of the energy storage unit. In addition, the available time window and maximum response power of demand response resources are collected from the demand side management database, where the available time window refers to the time period within which the demand side resources can be adjusted within a specific time range, and the maximum response power refers to the maximum power that the demand side resources can provide within that time period. After extraction, the start-stop time limit and ramp rate limit of thermal power units, the state of charge of energy storage units, and the available time window and maximum response power of demand response resources are classified by resource type and stored in the resource state database. Through this process, a unified resource state view is formed, providing comprehensive data support for subsequent flexible capacity assessment, ensuring the visualization and controllability of system operating conditions.

[0073] 3.2, Calculate the energy storage amplitude release index;

[0074] After completing the collection of resource constraints and states, the adjustment capacity of energy storage units within the dispatch window needs to be evaluated, so the calculation of the energy storage amplitude release index is introduced. The specific process is as follows: first, for each energy storage unit, extract its maximum charging power and maximum discharging power from the resource state database, where the maximum charging power refers to the maximum power that the energy storage unit can absorb in a unit of time, and the maximum discharging power refers to the maximum power that the energy storage unit can release in a unit of time. Next, calculate the remaining charge-discharge power difference for each energy storage unit, which is calculated by subtracting the maximum discharging power from the maximum charging power. Then, add up the remaining charge-discharge power differences of all energy storage units to get the total remaining charge-discharge power difference. After that, determine the dispatch window duration, which refers to the time range of dispatch decisions, and obtain the base load power, which refers to the base load level within the dispatch range. Based on the data in the resource state database, calculate the remaining charge-discharge power difference for each energy storage unit, and divide the power difference by the base load power to get the normalized power difference; sum up the normalized power differences of all energy storage units and multiply by the dispatch window duration to get the energy storage amplitude release index. The energy storage amplitude release index is a dimensionless quantity that reflects the adjustment capacity of energy storage units relative to the base load within the dispatch window. Through this calculation, the dynamic adjustment potential of energy storage units can be clearly grasped, providing a key quantitative basis for the optimization of capacity configuration.

[0075] 3.3, computer group load response margin index;

[0076] After the calculation of the energy storage amplitude modulation release index, the regulation capacity of the thermal power unit needs to be further evaluated, so the calculation of the unit load response margin index is introduced. The specific process is as follows: first, for each thermal power unit, extract its maximum allowed ramp rate, current ramp rate and remaining online time proportion from the resource state database, wherein the maximum allowed ramp rate refers to the maximum regulation rate that the thermal power unit can theoretically achieve, the current ramp rate refers to the regulation rate in the actual operation of the thermal power unit, and the remaining online time proportion refers to the proportion of the running time of the thermal power unit in the total scheduling window. Then, calculate the ramp rate difference of each thermal power unit, which is obtained by subtracting the current ramp rate from the maximum allowed ramp rate. Then, multiply the ramp rate difference of each thermal power unit by its corresponding remaining online time proportion to obtain the weighted ramp rate difference. After that, add the weighted ramp rate difference of all thermal power units one by one to obtain the total weighted ramp rate difference, and add the maximum allowed ramp rate of all thermal power units one by one to obtain the total maximum allowed ramp rate. Divide the total weighted ramp rate difference by the total maximum allowed ramp rate to obtain the unit load response margin index. The unit load response margin index is a dimensionless quantity, which reflects the overall remaining response capacity of the thermal power unit. Through this calculation, the flexible regulation space of the thermal power unit can be accurately quantified, providing a complementary evaluation perspective for the overall capacity allocation of the system.

[0077] 3.4, adaptive kernel ridge regression to generate flexible capacity weighting coefficients;

[0078] After the calculation of the energy storage amplitude modulation release index and the unit load response margin index, the flexibility of the supply side needs to be combined with the time-varying characteristics of the demand side, so the adaptive kernel ridge regression method is introduced to generate the flexible capacity weighting coefficient. The specific processing process is as follows: first, the energy storage amplitude modulation release index and the unit load response margin index are extracted from the dispatching database, and the average slope and the duration are extracted from the climbing demand prediction matrix, wherein the average slope refers to the smooth slope value of each time point in the climbing demand prediction matrix, and the duration refers to the duration of each time point in the climbing demand prediction matrix, which is in hours. Then, the energy storage amplitude modulation release index, the unit load response margin index, the average slope and the duration are input into the pre-trained adaptive kernel ridge regression model. The adaptive kernel ridge regression model outputs the flexible capacity weighting coefficient by weighting and nonlinear mapping the input data. The flexible capacity weighting coefficient is a dimensionless quantity, which is used to adjust the amplification factor of the candidate capacity configuration vector. Then, the initial candidate capacity configuration vector is obtained, which represents the preliminary capacity allocation scheme, and is multiplied by the flexible capacity weighting coefficient to obtain the adjusted candidate capacity configuration vector. The adjusted candidate capacity configuration vector is written into the dispatching candidate table. Through this process, the effective coupling of the supply side flexibility index and the dynamic demand of the demand side is realized, and flexible and adaptive alternative schemes are provided for subsequent power flow checking and economic optimization.

[0079] The adaptive kernel ridge regression model is an online updating implementation form of kernel ridge regression, belongs to kernel ridge regression, and only performs adaptive re-estimation on hyperparameters and sample dictionaries during the running period. The kernel ridge regression solves the target function of "empirical error term + regularization term" as a criterion, and the target function is composed of mean square error and ridge term. Linear regression is completed in the feature space after mapping by using a kernel function; for example, when the Gaussian kernel is taken as the kernel function, the hyperparameters include the regularization coefficient and the kernel width, and the model parameters are the regression coefficients or their dual coefficients. The adaptive meaning is limited to: when the preset updating period comes, the latest feedback sample is introduced, and the historical training sample is combined into the current training set, and the above minimum target function is taken as the only criterion to jointly update the regularization coefficient, the kernel width and the coefficient solution, and the size of the training sample set is maintained according to the sparse dictionary strategy.

[0080] The retraining and updating process is defined as follows: after merging the training set, the hyperparameters are fixed first, the kernel matrix is constructed according to the merged set, and the coefficient vector of the closed solution is obtained; then, the value of the objective function on the merged set is used as the evaluation quantity to perform one-dimensional search or Newton iteration on the regularization coefficient, and the value that minimizes the objective function is obtained; after the regularization coefficient is updated, the residual variance ratio of the merged set and the previous period validation set is used as the drift index, if the index exceeds the threshold, the kernel width is scaled by a certain proportion and the kernel matrix is reconstructed, and the kernel width that makes the objective function decrease is selected; in order to control the size and suppress redundant samples, the leverage score or information gain of each sample is calculated, the historical samples with a score lower than the threshold are removed, and the samples with the highest score are supplemented from the feedback samples to the upper limit capacity to form the updated sample dictionary; the Woodbury identity or incremental Cholesky decomposition is used for recursive updating instead of full calculation to obtain the new coefficient solution and the new objective function value. The difference between before and after updating is that the hyperparameter value, the sample dictionary composition and the kernel matrix and coefficient solution generated thereby are all changed, the mapping relationship is adjusted, and different flexible capacity weighting coefficients are output.

[0081] The adaptive kernel ridge regression model of the application takes Gaussian kernel as the mapping base and maintains the kernel matrix inverse array in a sliding window recursive manner. The input end receives the energy storage amplitude modulation release index and the unit load response margin index, and the two indices are stacked in time sequence as a feature vector stream; at each new time, the model first calculates the residual sum of squares change rate, dynamically adjusts the kernel width coefficient according to the preset threshold, then calls the sparse dictionary screening strategy to remove information redundant samples, and finally updates the double regularization term weight in the form of incremental Cholesky decomposition, and outputs the flexible capacity weighting coefficient. Dynamic kernel width adjustment ensures that the model is sensitive to error variance drift, sparse dictionary compression controls the computational complexity, and incremental decomposition ensures the stability of the weight solution; the above three work together to make the model stable in the fast fluctuation of the new energy scene, and to map the flexible measurement and capacity demand.

[0082] The processing process of gathering resource constraints and state integrates multi-source data to form a comprehensive view of the system operation state; the energy storage amplitude modulation release index and the unit load response margin index respectively quantify the regulation ability of the energy storage unit and the thermal power unit, and provide specific evaluation of the flexible potential of the supply side; the adaptive kernel ridge regression generates flexible capacity weighting coefficients, which combines the flexible capacity of the supply side with the time-varying characteristics of the demand side, and completes the dynamic adjustment of capacity configuration.

[0083] Step S3 has formed candidate capacity configuration vectors to meet the ramping demand through real-time data analysis and feature extraction. However, these candidate capacity configuration vectors need to be further verified for their potential impact on power network flow before actual scheduling to avoid network bottleneck problems caused by branch overload. Step S4 focuses on checking the flow sensitivity of the capacity configuration vectors in the scheduling candidate table to ensure the safety and feasibility of the scheduling scheme, providing a reliable foundation for subsequent economic optimization.

[0084] The goal of step S4 is to perform flow sensitivity checking on the capacity configuration vectors in the scheduling candidate table generated in step S3, eliminate schemes that may trigger network bottlenecks, and write the passing schemes to the bottleneck-free buffer area. The following is the detailed technical logic:

[0085] 4.1, flow sensitivity calculation;

[0086] When processing the capacity configuration vectors in the scheduling candidate table, it is necessary to quantify the impact of these vectors on the key branch flow in the power network to assess whether there is an overload risk. The specific processing process is as follows: First, extract the flow sensitivity matrix and the maximum allowed flow of each key branch from the power network database. The flow sensitivity matrix is a pre-computed matrix, where the elements represent the degree of influence of the change in power injection at any node in the power network on the flow of any branch, which is a dimensionless quantity.

[0087] Among them, the key branch refers to the transmission line or transformer branch that is highly sensitive to power grid safety margin and flow distribution within the scheduling window, and its determination is based on the following aspects:

[0088] First, in the recent operation record, if the load rate of a branch is close to its long-term thermal stability limit in most periods, or close to or even reaches the allowed upper limit at peak periods, it is determined to be a key branch;

[0089] Second, when any unit fails, use the line flow transfer coefficient to calculate whether the flow of the branch is close to or exceeds its short-term emergency limit, and determine it as a key branch;

[0090] Third, based on flow distribution factor analysis, if the unit power injection of any bus causes significant flow change in the branch, it is determined to be a key branch;

[0091] Fourth, the branch is located in the minimum cut set of the network, and once removed, it will cause some load nodes to lose the only communication path with the main power supply area;

[0092] Fifth, when the number of branches meeting any of the above conditions is too large, the load rate peak can be sorted from high to low according to the recent operation record, and a certain number is selected to form the final key branch set.

[0093] The operation limits, sensitivity coefficients and statistical results required for the above determination are derived from the real-time power grid model and operation database, and are dynamically updated through a sliding time window to ensure the uniformity, reviewability and direct availability in the process of checking the power flow sensitivity.

[0094] For each capacity configuration vector in the dispatch candidate table, the power flow increment of each key branch is calculated one by one. When calculating the power flow increment, the capacity adjustment amount of each node in the capacity configuration vector is multiplied by the sensitivity value of the corresponding node to a certain branch in the power flow sensitivity matrix to obtain the contribution value of each node, and then the contribution values of all nodes are added to obtain the power flow increment of the capacity configuration vector to the certain branch.

[0095] Then, the calculated power flow increment is divided by the maximum allowed power flow of the certain branch to obtain the power flow sensitivity, which is a dimensionless quantity reflecting the influence degree of the capacity configuration vector on the load pressure of the branch. The above calculation is repeated for each capacity configuration vector and all key branches to finally generate a power flow sensitivity matrix, in which the rows correspond to the key branches and the columns correspond to the capacity configuration vectors. This process quickly evaluates the impact of a large number of candidate solutions on the network through simplified matrix operations without complex power flow simulation, thereby greatly reducing the consumption of computing resources while accurately monitoring the load capacity of the branch.

[0096] 4.2, network bottleneck identification;

[0097] After the power flow sensitivity calculation is completed, the capacity configuration vectors that may cause the branch to be overloaded need to be selected from the dispatch candidate table to ensure that the dispatch scheme meets the network security requirements. The specific processing process is as follows: first, a power flow sensitivity threshold is determined in advance according to the historical operation data and the network security margin, which is a dimensionless quantity used to judge whether the branch is close to an overloaded state. From the power flow sensitivity matrix, the power flow sensitivity values of all key branches corresponding to each capacity configuration vector are extracted one by one. Then, for each capacity configuration vector, the power flow sensitivity values of all key branches are checked, and the specific method is to compare the size of the power flow sensitivity value of each key branch with the power flow sensitivity threshold. If the power flow sensitivity value of a certain key branch is greater than the power flow sensitivity threshold, it is considered that the capacity configuration vector may trigger a network bottleneck, and the capacity configuration vector is marked as "bottleneck exists"; if the power flow sensitivity values of all key branches are less than or equal to the power flow sensitivity threshold, the capacity configuration vector is marked as "no bottleneck". The above check is performed on all capacity configuration vectors in the dispatch candidate table, and the marking results are attached to the dispatch candidate table.

[0098] Through this process of comparison, the capacity configuration vector that may cause branch overload can be quickly identified, ensuring the safety of the scheduling scheme under network constraints, avoiding emergency adjustment due to overload, and enhancing the reliability of the scheme.

[0099] 4.3, write the bottleneck-free buffer zone;

[0100] After completing the network bottleneck identification, the capacity configuration vectors that meet the safety requirements need to be sorted and stored for subsequent economic optimization steps. The specific processing process is as follows: all capacity configuration vectors marked as "no bottleneck" are selected from the scheduling candidate table, and these vectors are referred to as passed capacity configuration vectors. These passed capacity configuration vectors are written into the bottleneck-free buffer zone in chronological order or priority order, and metadata such as generation time is attached to each passed capacity configuration vector to facilitate subsequent tracing and calling. After writing is completed, the subsequent steps are notified to prepare to call the data in the bottleneck-free buffer zone. This process clearly transfers the set of safe and checked schemes to the optimization link through ordered storage and metadata management, ensuring smooth connection from safety evaluation to economic optimization, and improving the execution efficiency of the overall scheduling.

[0101] The power flow sensitivity calculation provides accurate data support required for network bottleneck identification by quantifying the impact of capacity configuration vectors on critical branches; the network bottleneck identification ensures that the scheduling scheme meets the network safety constraints by screening safe capacity configuration vectors through threshold comparison; and the writing of the bottleneck-free buffer zone organizes and stores the passed schemes, providing a reliable set of alternative schemes for subsequent economic optimization.

[0102] The preceding steps S1 to S4 successfully generate a set of safe and feasible capacity configuration vectors in the bottleneck-free buffer zone through real-time measurement and prediction error trajectory analysis, construction of climbing demand prediction matrix, flexible measurement and adaptive kernel ridge regression to generate candidate capacity configuration vectors, and power flow sensitivity check. These steps effectively solve the problem of cumulative prediction error accumulation and dynamic matching of real-time climbing demand in the scenario of high proportion of new energy grid connection, ensuring that the capacity configuration scheme meets the network safety constraints. However, in actual scheduling, only meeting technical feasibility is not enough to cope with the economic challenges brought by wind and light output fluctuations, and economic evaluation needs to be introduced into the safe scheme to minimize scheduling cost and improve resource utilization efficiency. Step S5 takes the output of step S4, performs economic optimization on the capacity configuration vectors in the bottleneck-free buffer zone based on real-time market quotes, generates optimal scheduling instructions, and continuously improves the model through a feedback mechanism, providing closed-loop support for stable and economic consideration in the scenario of new energy grid connection.

[0103] The objective of step S5 is to perform economic evaluation and optimization on the capacity configuration vectors in the bottleneck-free buffer, select the optimal dispatching instruction based on the real-time market quotation sequence ranking, and issue it for execution, while feeding back the execution feedback to the adaptive kernel ridge regression model to realize parameter updating. The following is the detailed technical logic:

[0104] 5.1 Economic evaluation

[0105] When processing the capacity configuration vectors in the bottleneck-free buffer, it is necessary to compare the costs of different schemes in a quantitative manner to select a more economical resource allocation method under the premise of meeting safety constraints. The specific processing process is as follows: First, read the real-time market quotation sequence from the power market database. The real-time market quotation sequence includes the current marginal cost or market quotation of thermal power units, energy storage units and demand response resources. Next, extract all capacity configuration vectors from the bottleneck-free buffer. Each capacity configuration vector represents a feasible capacity allocation scheme that has been safety verified. Then, for each capacity configuration vector, identify the types of resources involved, such as thermal power units, energy storage units or demand response resources, and extract the corresponding quotations from the real-time market quotation sequence. After that, obtain the preset dispatching duration. The dispatching duration refers to the execution time range of the dispatching instruction. Multiply the capacity adjustment amount of each resource in the capacity configuration vector by the real-time market quotation of the corresponding resource, and then multiply it by the dispatching duration to calculate the cost of the resource in this dispatching. Then, add the costs of all resources in the capacity configuration vector one by one to obtain the total dispatching cost of the capacity configuration vector. Repeat the above calculation steps for each capacity configuration vector in the bottleneck-free buffer, and add the calculated total dispatching cost to the record of the corresponding capacity configuration vector to form a vector set with cost annotations. In this way, the cost differences of each safe and feasible scheme under the current market conditions can be accurately grasped, providing clear basis for subsequent screening, while adapting to market price fluctuations, ensuring the real-time and accuracy of cost calculation.

[0106] 5.2 Economic ranking and optimal dispatching instruction generation

[0107] After the cost calculation is completed, the scheme with the lowest cost needs to be quickly selected from a large number of schemes to optimize the economic benefit of resource use while meeting the technical requirements. The specific processing process is as follows: all capacity configuration vectors and their corresponding total scheduling costs are extracted from the bottleneck-free buffer zone to construct a paired data set. The paired data set contains the combination of each capacity configuration vector and the corresponding total scheduling cost. All total scheduling costs in the paired data set are arranged in ascending order to form an ascending scheduling cost sorting sequence. Then, the pair with the minimum cost value is selected from the scheduling cost sorting sequence, and the corresponding capacity configuration vector is extracted to determine it as the optimal scheduling instruction. Then, the optimal scheduling instruction is converted into an instruction structure that can be recognized by the execution layer, and is prepared to be issued to the execution layer for implementation. Through direct comparison and sorting of the cost, the best economic instruction can be quickly determined among a plurality of safety schemes, ensuring the efficiency of resource allocation and reducing the additional expenditure caused by selecting a suboptimal scheme.

[0108] 5.3, execution feedback and model updating;

[0109] Due to the continuous changes of market conditions and network states under the new energy grid-connected scenario, the prediction model needs to be improved by using the actual execution results to improve the accuracy and adaptability of future scheduling. The specific processing process is as follows: after the optimal scheduling instruction is executed, feedback data is collected from the execution layer. The feedback data includes the actual executed capacity configuration vector, the actual occurred scheduling cost, and the network running state and other information. These feedback data are stored in the feedback database. Then, according to the preset update period, the historical feedback data and the newly added feedback data are extracted from the feedback database, and these data are combined with the initial training data of the adaptive kernel ridge regression model. The adaptive kernel ridge regression model is retrained using the combined data set, and the model parameters are adjusted to better reflect the current operating environment. After retraining is completed, the updated adaptive kernel ridge regression model is deployed in the system for the next round of scheduling decision. By continuously collecting feedback data and optimizing model parameters, the adaptability of the scheduling instruction to the dynamic environment can be gradually improved, ensuring that the system remains efficient and stable in the long run, and providing more accurate support for capacity configuration prediction.

[0110] The economic evaluation accurately calculates the cost of each scheme by combining real-time market quotes and resource call quantities; the economic sorting and optimal scheduling instruction generation quickly select the best economic instruction based on cost sorting; the execution feedback and model updating optimize the model using actual operation data to improve the accuracy of subsequent scheduling. From cost quantification to scheme selection to strategy improvement, the economic efficiency and adaptability of resource allocation are comprehensively improved to ensure the stable and efficient operation of the scheduling system under the new energy high proportion grid-connected scenario.

[0111] Embodiment 2:Figure 2 The power resource deployment system for new energy access of the present application is given, comprising:

[0112] Difference track generation module: receiving wind and light real-time measurement sequence and synchronous prediction sequence, calculating difference value to form error track;

[0113] Demand shaping module: calculating slope and duration of error track, constructing ramping demand prediction matrix;

[0114] Flexible mapping module: collecting unit start-stop constraint, energy storage state of charge and demand response window, mapping energy storage amplitude modulation potential, unit load margin and average slope and duration in ramping demand prediction matrix through preset adaptive kernel ridge regression model to generate flexible capacity weighting coefficient, thereby generating candidate capacity configuration vector and writing into dispatch candidate table;

[0115] Bottleneck screening module: performing power flow sensitivity check on capacity configuration vector in dispatch candidate table, removing bottleneck scheme to write into bottleneck-free buffer area;

[0116] Economic dispatch module: extracting capacity configuration vector from bottleneck-free buffer area, implementing economic sorting on bottleneck-free vector based on real-time market quotation sequence, outputting optimal dispatch instruction and issuing execution layer, and simultaneously feeding back to adaptive kernel ridge regression model for subsequent iteration.

[0117] The above formulas are all dimensionless values, and the formulas are obtained by collecting a large amount of data to simulate the latest real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0118] It should be noted that the system of the present application can be deployed in the device itself to realize embedded application, or can be run on PC or other terminal with user interface, so as to meet various hardware environments and use requirements.

[0119] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above figures and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

[0120] It has to be noted that, in the present document, the terms "first", "second", etc. merely serve the purpose of differentiation and do not require or imply any kind of ordering or sequence of the entities or actions associated therewith. Furthermore, the terms "comprising", "containing", etc. are to be interpreted as non- exclusive in the sense that a process, method, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a", "contains... a", etc. does not, without further restriction, exclude the presence of additional identical elements in the process, method, article, or apparatus.

[0121] The above description is only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A power resource dispatching method for new energy access, characterized in that, The method comprises the steps of: S1: receiving wind and light real-time measurement sequence and synchronous prediction sequence, calculating difference value to form error trajectory; S2: calculating slope and duration of error trajectory, constructing climbing demand prediction matrix; S3: collecting unit start-stop constraints, energy storage state of charge and demand response window, mapping the average slope and duration in the climbing demand prediction matrix, the energy storage amplitude modulation potential, the unit load margin and the adaptive kernel ridge regression model to generate a flexible capacity weighting coefficient, and generating a candidate capacity configuration vector according to the flexible capacity weighting coefficient and writing it into a scheduling candidate table; Step S3 includes the following contents: Through the real-time data interface, the start-stop time limit of the thermal power unit, the climbing rate limit, the state of charge of the energy storage unit, the available time window and the maximum response power of the demand response resource are obtained from the power market management database, the SCADA database and the demand side management database respectively, and these data are classified and stored into the resource state database; based on the data in the resource state database, the remaining charge and discharge power difference of each energy storage unit is calculated, and the remaining charge and discharge power difference is divided by the basic load power to obtain the normalized power difference; the normalized power difference of all energy storage units is summed and multiplied by the scheduling window length to obtain the energy storage amplitude modulation release index, which is used to represent the energy storage amplitude modulation potential; at the same time, the climbing rate difference and the weighted climbing rate difference of each thermal power unit are calculated, and then the unit load response margin index is calculated, which is obtained by dividing the weighted climbing rate difference of all thermal power units by the total maximum allowed climbing rate, and is used to represent the unit load margin; S4: performing power flow sensitivity check on the capacity configuration vector in the scheduling candidate table, and writing it into the bottleneck-free buffer area after eliminating the trigger bottleneck scheme; S5: extracting the capacity configuration vector from the bottleneck-free buffer area, implementing economic sorting on the bottleneck-free vector based on the real-time market quotation sequence, outputting the optimal scheduling instruction and issuing it to the execution layer, and at the same time, feeding back the adaptive kernel ridge regression model for subsequent iteration.

2. The method of claim 1, wherein, Step S1 includes the following contents: Real-time measurement data and synchronous prediction data of the wind and light power plant are obtained through the monitoring and data acquisition system of the power system, and the time markers of the real-time measurement data and the synchronous prediction data are aligned; for each pair of time-aligned real-time measurement data and synchronous prediction data, the difference between the real-time measurement power value and the synchronous prediction power value is calculated to generate an error value sequence; the error value sequence is arranged in chronological order to form an error trajectory.

3. The method of claim 2, wherein, Step S2 includes the following contents: The slope calculation is performed on the error trajectory, the ratio of the error difference value of adjacent time points to the time interval is used, and the three-point sliding window method is used for smoothing to generate a smooth slope sequence; the duration calculation is performed on the error trajectory, the backtracking method is used to determine the duration of each time point error state to generate a duration sequence; the smooth slope sequence and the duration sequence are combined in chronological order to form a climbing demand prediction matrix, each row of the matrix corresponds to a time point, including the smooth slope value and the duration value of the time point. 4.The method of claim 1, wherein, Step S3 also includes the following contents: ​ The energy storage amplitude modulation release index, the unit load response margin index, the average slope and duration in the climbing demand prediction matrix are input into the pre-trained adaptive kernel ridge regression model to generate a flexible capacity weighting coefficient, and the initial candidate capacity configuration vector is adjusted by using the flexible capacity weighting coefficient to obtain an adjusted candidate capacity configuration vector, which is written into a dispatch candidate table.

5. The method of claim 4, wherein, Step S4 includes the following contents: The power flow sensitivity matrix and the maximum allowable power flow of the key branch are extracted from the power network database; for each capacity configuration vector in the dispatch candidate table, the power flow increment of the capacity configuration vector on all key branches is calculated, and the power flow increment is calculated by multiplying the capacity adjustment amount of each node in the capacity configuration vector by the sensitivity value of the corresponding node pair to a specific branch in the power flow sensitivity matrix and summing up; The power flow sensitivity is calculated by dividing the power flow increment by the maximum allowable power flow of the key branch.

6. The method of claim 5, wherein, Step S4 also includes the following contents: According to a predetermined power flow sensitivity threshold, the power flow sensitivity values of each capacity configuration vector on all key branches are checked, and if the power flow sensitivity value of any key branch is greater than the power flow sensitivity threshold, the corresponding capacity configuration vector is marked as having a bottleneck, and if the power flow sensitivity values of all key branches are less than or equal to the power flow sensitivity threshold, the capacity configuration vector is marked as having no bottleneck; the capacity configuration vectors marked as having no bottleneck are selected from the dispatch candidate table, and these capacity configuration vectors are written into a bottleneck-free buffer area and are additionally provided with generation time metadata.

7. The method of claim 6, wherein, Step S5 includes the following contents: The real-time market quotation sequence is obtained from the power market database, and the capacity configuration vectors are extracted from the bottleneck-free buffer area; for each capacity configuration vector, its resource type is identified, the corresponding real-time market quotation is extracted, the dispatch cost is calculated, and the dispatch cost is attached to the record of the capacity configuration vector; the dispatch costs of the capacity configuration vectors are arranged in ascending order, the capacity configuration vector with the lowest dispatch cost is selected as the optimal dispatch instruction, and it is formatted and then issued to the execution layer; feedback data is collected after execution, stored in a feedback database, and used to update the parameters of the adaptive kernel ridge regression model according to a preset period to optimize subsequent dispatch.

8. The power resource dispatching system for new energy access, used for realizing the power resource dispatching method for new energy access in any of claims 1-7, characterized in that, It includes: Difference track generation module: receives real-time measurement sequence and contemporaneous prediction sequence of wind and light, calculates difference to form error track; Demand shaping module: calculates slope and duration of error track, constructs climbing demand prediction matrix; Flexible mapping module: collects unit start-stop constraints, energy storage state of charge and demand response window, maps energy storage amplitude modulation potential, unit load margin and average slope and duration in the climbing demand prediction matrix through a preset adaptive kernel ridge regression model to generate a flexible capacity weighting coefficient, generates a candidate capacity configuration vector according to the flexible capacity weighting coefficient, and writes it into a dispatch candidate table; Bottleneck screening module: performs power flow sensitivity checking on the capacity configuration vectors in the dispatch candidate table, and writes them into a bottleneck-free buffer area after removing the bottleneck triggering scheme; Economic dispatch module: extract the capacity configuration vector from the bottleneck-free buffer zone, implement economic sorting on the bottleneck-free vector based on the real-time market quotation sequence, output the optimal dispatch instruction and issue it for execution layer, and at the same time, feedback the adaptive kernel ridge regression model for subsequent iteration.

Images