Method for evaluating dispatching operation of power spot market based on multiple source factors

Abstract

This application relates to the field of power grid dispatching and operation technology, specifically to a method for evaluating the dispatching and operation of the electricity spot market based on multiple factors. This method includes: labeling all samples based on numerical weather data; combining the output power at each time moment to obtain the predicted power generation value at different time scales; analyzing the error distribution characteristics of the predicted power generation at different time scales to determine the prediction error ratio coefficient; analyzing the numerical distribution of transmission power at each time moment for each line; combining the changes in SOC data over a preset time interval to determine the dispatching fluctuation coefficient at each time moment; and using a neural network based on historical power plant data to obtain the prediction function value at each time moment, comparing it with the dispatching fluctuation coefficient at each time moment to determine the dispatching and operation evaluation value at each time moment. This application aims to quantify the quality of dispatching schemes and use this to evaluate the quality of electricity spot market dispatching and operation schemes.

Description

Technical Field

[0001] This application relates to the field of power grid dispatching and operation technology, specifically to an evaluation method for power spot market dispatching and operation based on multiple factors. Background Technology

[0002] Driven by the global goal of energy transition, building a new power system dominated by new energy sources has become the core direction of my country's energy development. The large-scale grid connection of renewable energy sources such as wind and solar power has profoundly changed the physical characteristics and operating modes of the power system. At the same time, the continuous deepening of power market reforms, with the electricity spot market as a key mechanism for optimizing resource allocation, is crucial for achieving real-time balance between power supply and demand. Its dispatching decisions directly affect the system's safe, economical, and low-carbon operation. However, the strong randomness, volatility, and weak controllability of high-proportion renewable energy sources, coupled with the emergence of new market players such as distributed energy, energy storage, and electric vehicles, as well as the frequent occurrence of extreme weather events, have made the operating environment of the spot market increasingly complex and volatile, posing unprecedented challenges to the scientific nature of dispatching operations and the accuracy of assessments.

[0003] The core technical problem facing existing power spot market dispatch operation assessment technologies is that they analyze dispatch processes at different time scales, such as day-ahead, intraday, and real-time, in a fragmented manner. Relying on static and isolated assessment models, they struggle to quantify the cross-time scale decision-making coupling effects and risk transmission, leading to severe distortions in the assessment of overall system operational efficiency and potential risks. Specifically, traditional assessment methods typically evaluate day-ahead planning, intraday adjustments, and real-time control independently, failing to establish a dynamic assessment framework spanning the entire time scale. This prevents the assessment from capturing the impact of day-ahead dispatch decisions (such as unit combination) on intraday adjustments, and from measuring how intraday market adjustments alter the regulation pressure and safety margin during the real-time balance phase. Since errors in renewable energy generation and load forecasting are continuously corrected as the execution time approaches, the static assessment system ignores the cumulative effect of this cross-time-period "decision-deviation-re-decision" chain, resulting in a lack of systematic approach to power spot market regulation and assessment, and failing to provide a reliable basis for subsequent optimization of the entire lifecycle operation. Summary of the Invention

[0004] In view of the above, it is necessary to provide an evaluation method for the dispatch and operation of the electricity spot market based on multiple factors to solve the above problems.

[0005] One embodiment of this application provides a method for evaluating the operation of electricity spot market dispatch based on multiple factors, the method comprising:

[0006] The power output, numerical weather data, and SOC data of the power station at each moment are used as a sample, and the transmission power of each line of the power station at each moment is collected.

[0007] All samples are labeled based on numerical weather data. Combined with the output power at each time, machine learning models are used to obtain the predicted power generation at different time scales at each time. The error distribution characteristics of the power generation prediction at different time scales at each time are analyzed, and the prediction error ratio coefficient at each time scale is determined.

[0008] Analyze the numerical distribution of transmission power of each line at each time, and combine it with the changes in SOC data at preset time intervals to determine the dispatch fluctuation coefficient at each time.

[0009] Using a neural network based on historical power plant data, combined with the error proportionality coefficient and the scheduling fluctuation coefficient, the predicted function value at each time point is obtained and compared with the scheduling fluctuation coefficient at each time point to determine the scheduling operation evaluation value at each time point.

[0010] Specifically, determining the prediction error scaling factor at each time scale for each moment is as follows:

[0011] When making a prediction at time t, the time weight of each moment in each time scale for the prediction at time t is obtained based on the time interval between different moments in each time scale and time t.

[0012] The difference between the actual output power and the corresponding predicted power generation rate at each time scale is obtained, and the prediction weighted mean square error at each time scale at time t is determined by combining the time weight.

[0013] Arrange all time scales in the order of prediction, compare the weighted mean square error of prediction at time t between two adjacent time scales, and obtain the prediction error ratio between two adjacent time scales.

[0014] Time scales include day-ahead forecasts, intraday forecasts, and ultra-short-term forecasts;

[0015] The prediction error ratios obtained at time t across all adjacent time scales are positively fused to obtain the prediction error ratio coefficient at time t in the ultra-short-term prediction.

[0016] The prediction error ratio of the intraday prediction at time t is determined by subtracting the prediction error ratio of the ultra-short-term prediction at time t from the prediction error ratio of the day-ahead prediction and the intraday prediction at time t.

[0017] Calculate the difference between the value 1 and the prediction error ratios obtained between all adjacent time scales, and perform positive fusion of the differences to obtain the prediction error ratio coefficient for the day-ahead prediction at time t.

[0018] The time weight is determined by the normalized value of the negative correlation mapping between the time interval at different times of each time scale and time t.

[0019] Specifically, obtaining the prediction error ratio between two adjacent time scales is as follows:

[0020] Calculate the difference between the prediction weighted mean square error of two adjacent time scales, divide it by the prediction weighted mean square error of the time scale that is ranked first, and obtain the prediction error ratio of the two adjacent time scales.

[0021] Specifically, determining the scheduling fluctuation coefficient at each moment involves:

[0022] Obtain the percentage of power exceeding the limit for each line in the first n time moments. Use the ratio of the maximum safe power of each line to the maximum safe power of all lines as the important weight of each line. Use the important weight as the weight of the percentage of power exceeding the limit for the corresponding line. Sum all lines by weight to obtain the frequency of exceeding the limit at the nth time moment.

[0023] Obtain the rate of change of SOC at the nth time point over a preset time interval, and use the positive correlation mapping result of the rate of change as the energy storage fluctuation component at the nth time point;

[0024] By combining the numerical distribution between the transmission power of all lines at time n and the maximum safe power of the lines, the transmission components of the lines at time n are determined.

[0025] The specific formula for the scheduling fluctuation coefficient is as follows:

[0026]

[0027] In the formula, It is the first The scheduling fluctuation coefficient at each moment, , They are the first Energy storage fluctuation components and transmission line components at any given moment It was before The frequency of exceeding the limit at any given time, where N represents the total number of times.

[0028] Specifically, the transmission power exceeding the limit refers to the transmission power exceeding the maximum safe power of the line.

[0029] The positive correlation mapping result of the rate of change is obtained through the formula. Obtain; where, Represents an exponential function with the natural constant as the base; , They are the first The moment, the first SOC data of the energy storage system at a given moment. This is the preset time interval.

[0030] Specifically, the transmission components of the lines are all the lines in the [number]th [phase]. The average ratio of the actual transmission power to the maximum safe power of the line at each time point.

[0031] In the process of obtaining the prediction function value at each time point by using a neural network based on historical power plant data and combining the error ratio coefficient and the scheduling fluctuation coefficient, the features corresponding to each sample include the prediction error ratio coefficient, sample label, actual output power, and scheduling fluctuation coefficient at all time scales.

[0032] Specifically, determining the scheduling operation evaluation value at each time point involves:

[0033] The ratio of the predicted function value at each time point to the scheduling fluctuation coefficient is used as the scheduling operation evaluation value at each time point.

[0034] This application has at least the following beneficial effects:

[0035] This application uses the power output, numerical weather data, and SOC data of a power plant at each moment as a sample, and collects the transmission power of each line of the power plant at each moment to construct a complete multi-source sample, providing multi-dimensional input for subsequent prediction. SOC data can reflect the energy storage regulation capacity and assist in optimizing dispatch strategies. The transmission power of each line provides the basis for subsequent load dispatch and constraint analysis. Based on numerical weather data, the samples are labeled, and combined with the power output at each moment, a machine learning model is used to predict the power generation at different time scales at each moment. Weather information is used to improve the accuracy of power generation prediction, especially for renewable energy (photovoltaic and wind power). Predictions at different time scales can reflect short-term fluctuations and intraday / interday trends, providing multi-layered decision-making information for dispatch, helping to identify the patterns of prediction errors, and providing a basis for error compensation strategies. The distribution of power generation prediction errors at different time scales at each moment is analyzed, and the prediction error proportionality coefficient for each time scale at each moment is determined for quantification. To predict uncertainties, a safety margin is provided for dispatching, and energy storage or reserve capacity is adjusted during periods of high error to improve dispatching reliability and add an error compensation dimension to the subsequent neural network input. The power distribution of each line at different times is analyzed, and the dispatching fluctuation coefficient is determined by combining the changes in SOC over a preset time interval. By identifying the load fluctuation patterns of the lines, potential overload risks are assessed. The dispatching fluctuation coefficient can quantify dispatching flexibility and mobilization capability, providing a reference for actual operation and assisting in optimizing energy storage participation in regulation strategies to smooth the load curve. Finally, based on historical power plant data, a neural network is trained to predict power function values. The predicted function values ​​are combined with the error proportionality coefficient and the dispatching fluctuation coefficient to compare the actual dispatching with the fluctuation situation, obtaining a dispatching operation evaluation value. The neural network can capture nonlinear relationships, improving prediction accuracy. The comprehensive evaluation of the dispatching scheme based on error and fluctuation coefficients can identify potential risk points and optimization space, providing quantitative indicators for dispatching optimization and reserve capacity planning. Attached Figure Description

[0036] Figure 1 A flowchart of the power spot market dispatch operation evaluation method based on multi-source factors provided for this application;

[0037] Figure 2 This is a schematic diagram illustrating the acquisition of the prediction error proportionality coefficient at each time scale for time t, as provided in this application. Detailed Implementation

[0038] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0040] It should also be noted that the terms "first" and "second" in this application and its accompanying drawings are used to distinguish similar objects, rather than to describe a specific order or sequence. The methods disclosed in the embodiments of this application or the methods shown in the flowcharts include one or more steps for implementing the method. Without departing from the scope of protection of this application, the execution order of multiple steps can be interchanged, and some steps can also be deleted.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0042] This application proposes a multi-source factor-based evaluation method for electricity spot market dispatch operation, which is applied to the field of power grid dispatch operation technology. (See attached document.) Figure 1 The method includes the following steps:

[0043] Step 1: Take the power output, numerical weather data, and SOC data of the power station at each time as a sample, and collect the transmission power of each line of the power station at each time.

[0044] On the power generation side, the monitoring systems of the booster stations at each wind farm / photovoltaic power station collect the output power of different new energy power supply stations. Then, through meteorological stations within the wind farm / photovoltaic power station or the China Meteorological Data Network, numerical weather prediction (NWP) data for each wind farm / photovoltaic power station is obtained, including but not limited to wind speed, wind direction, temperature and humidity, solar radiation intensity, and cloud opacity. In the energy storage system, the battery management system monitors the state of charge (SOC) of each energy storage battery module during charging and discharging; changes in the SOC of the energy storage batteries reflect the scheduling status and speed of the energy storage system. In the power grid, the transmission power of each substation and key transmission line is collected through the SCADA system; changes in the transmission power reflect system load fluctuations and line safety margins. Except for wind direction data, all other collected data are processed using the maximum-minimum normalization method.

[0045] Step 2: Classify all samples based on numerical weather data, combine the output power at each time point, use machine learning models to obtain the predicted power generation at different time scales at each time point, analyze the error distribution characteristics of the predicted power generation at different time scales at each time point, and determine the prediction error ratio coefficient at each time point and each time scale.

[0046] The electricity spot market optimizes resource allocation and ensures a balance between power supply and demand in the power grid. The quality of its dispatching schemes directly impacts the stable and secure operation of the power grid. With a high proportion of renewable energy sources connecting to the grid, the electricity spot market environment is becoming increasingly complex, posing a more severe challenge to the accuracy of dispatching operation assessments. Due to the uncertainties brought about by renewable energy integration, existing technologies often require evaluating the dispatching process at multiple different time scales. However, current assessments often rely on static, isolated models, ignoring the coupling between multi-scale decisions and simplifying the assessment process without establishing a dynamic assessment model spanning all time scales. This current dispatching assessment method cannot effectively handle the errors caused by the uncertainties of renewable energy integration, resulting in the inability to fully guarantee the reliability of electricity spot market dispatching operations throughout their entire lifecycle.

[0047] In the electricity spot market, fluctuations in renewable energy generation are one of the sources of uncertainty in the system. Therefore, it is usually necessary to forecast renewable energy generation to mitigate the impact of renewable energy fluctuations on the electricity spot market to some extent. Existing renewable energy generation forecasts mainly focus on three time scales: day-ahead forecasts, intraday forecasts, and ultra-short-term forecasts. Day-ahead forecasts are full-day forecasts based on the next day's weather forecast combined with historical generation data and load demand. Intraday forecasts are rolling corrections to day-ahead forecasts within a short period (e.g., 4 hours) based on the latest meteorological data of the day. Ultra-short-term forecasts are generation forecasts for a shorter future period (e.g., 15 minutes) based on real-time meteorological parameters and power plant operating status.

[0048] In the process of evaluating the dispatch of the electricity spot market, it is necessary to determine the degree of impact of the uncertainty of renewable energy generation on the quality of the dispatch plan. Therefore, it is necessary to evaluate the prediction error of renewable energy generation. In the process of renewable energy generation prediction, the final renewable energy generation prediction is obtained by rolling correction between the prediction results at three time scales: day-ahead, intraday, and ultra-short-term. During the rolling correction process, the contribution of the prediction at different time scales to the final renewable energy generation prediction is also different. However, it should be noted that this difference in contribution is not due to the prediction model. The closer to the time to be predicted, the more accurate the NWP data is, which in turn makes the prediction accuracy at the corresponding scale higher.

[0049] In the dispatching and operation of the electricity spot market, reasonable dispatching planning is usually required in advance to avoid abrupt equipment shutdowns and restarts caused by temporary dispatching, thereby reducing the impact on compensation equipment and power transmission processes, and preventing equipment damage and system instability. Therefore, in the subsequent electricity spot market dispatching assessment, it is necessary to decouple dispatching at different time scales to determine the error situation of dispatching at different time scales. Similarly, in the renewable energy generation forecasting stage, it is necessary to decouple renewable energy generation forecasts at different time scales to determine the contribution of forecasts at different time scales to the final renewable energy generation forecast.

[0050] Using historical NWP data at various time points as input, the DBSCAN clustering algorithm is employed for clustering. Data at each time point is labeled based on the clustering results, with data in the same cluster sharing the same label. Then, 70% of the labeled historical data is randomly selected as the training set, and the remaining samples are used as the test set. The sample data and corresponding output power of the training set are used as input. The number of trees is set to 1000, the tree depth to 5, the L2 regularization term to 3, the split threshold to 0.2, and the learning rate to 0.05. The XGBoost model is used to train the new energy power generation prediction model. The DBSCAN clustering algorithm and XGBoost model training are well-known techniques and will not be elaborated further. After obtaining the new energy power generation prediction model, the NWP data at each time scale is used as input, and the predicted new energy power generation value for each time scale is output. Specifically, the predicted value is calculated based on the NWP data at each time scale. day Taking time-based predictions as an example, the current-day prediction is based on the [number]th [time]. The first day obtained The NWP data at various times of the day predicts the power generation at each time point. The intraday forecast is based on the data from the [missing information - likely a specific data point or data point]. day Obtained 4 hours before the time The power generation predicted from NWP data at a given time point is based on the ultra-short-term forecast. day Obtained 15 minutes before the time The power generation predicted from the NWP data at a given time.

[0051] Based on the above analysis, and according to the prediction errors of new energy power generation at different time scales, a prediction error ratio coefficient is calculated to measure the contribution of predictions at different time scales to the final new energy power generation prediction. Specifically, when predicting at time t, based on the time interval between different times at each time scale and time t, the time weight of each time point in each time scale for the prediction at time t is obtained. In this embodiment, the time weight of each time point n for the prediction at time t is... The specific formula is: In the formula, t represents time t; n and i represent time n and time i at each time scale; e represents the natural constant. This indicates the preset attenuation rate parameter. The larger the value, the faster the weight decays. It is recommended that the value range be [0.4, 0.7]. In this embodiment, the value is 0.5; N represents the total number of moments at each time scale. This indicates a negative correlation between the time interval at different times on each time scale and time t. This represents the normalized value of the negative correlation mapping between the time interval at different times on each time scale and time t.

[0052] The difference between the actual output power and the corresponding predicted power generation rate at each moment of each time scale is obtained. Combined with the time weight, the prediction weighted mean square error at time t at each time scale is determined. In this embodiment, the formula for the prediction weighted mean square error is specifically as follows: In the formula, , They are time points Actual output power and predicted power generation at each time scale; N represents the total number of moments at each time scale; This represents the weighted mean square error of the prediction at each time scale at time t; This represents the time weight of the prediction at time t for each time scale n.

[0053] All time scales are arranged in the order of prediction. The weighted mean square error of prediction at time t is compared between two adjacent time scales to obtain the prediction error ratio between the two adjacent time scales. In this embodiment, the prediction error ratio between time scale u and time scale v is denoted as... Its formula is as follows: In the formula, , yes Each moment is on a different time scale Time scale The prediction weighted mean square error is given below, where the prediction order of time scale u precedes the prediction order of time scale v.

[0054] Finally, since this application considers three time scales—day-ahead forecast, intraday forecast, and ultra-short-term forecast—the prediction error ratios obtained at time t across all adjacent time scales are forward-fused to obtain the prediction error proportion coefficient for ultra-short-term forecast at time t. The prediction error ratio between day-ahead and intraday forecasts at time t is subtracted from the prediction error proportion coefficient for ultra-short-term forecast at time t to determine the prediction error proportion coefficient for intraday forecast at time t. The differences between the value 1 and the prediction error ratios obtained across all adjacent time scales are calculated, and these differences are forward-fused to obtain the prediction error proportion coefficient for day-ahead forecast at time t. In this embodiment, the forward fusion uses a multiplication calculation method; specifically, the formula for the prediction error proportion coefficient at each time scale at time t is as follows:

[0055]

[0056] In the formula, , , They are the first day The prediction error ratio coefficients for day-ahead, intraday, and ultra-short-term forecasts at specific times. It is the ratio of the forecast error between the previous day's forecast and the intraday forecast. This is the ratio of the forecast error between intraday forecasts and ultra-short-term forecasts, where the subscript is... , , These correspond to daily forecasts, intraday forecasts, and ultra-short-term forecasts, respectively.

[0057] The diagram illustrating the acquisition of the prediction error proportion coefficient at each time scale at time t is shown below. Figure 2 As shown.

[0058] It is understandable that the weighted mean square error measures the prediction error between different time scales. Since changes in meteorological conditions have a certain degree of continuity, the closer the prediction is to the time to be predicted, the higher the weight of the prediction accuracy should have in measuring the error at the time to be predicted. Thus, the larger the prediction error ratio of the ultra-short-term prediction, the higher the contribution of the ultra-short-term prediction. It should be further noted that the same applies to intraday and day-ahead predictions. However, since day-ahead predictions do not have a comparable prediction at a larger time scale, the contribution of day-ahead predictions is measured by the error situation of intraday and ultra-short-term predictions.

[0059] Step 3: Analyze the numerical distribution of transmission power of each line at each time, and combine it with the changes in SOC data at preset time intervals to determine the dispatch fluctuation coefficient at each time.

[0060] The prediction error proportionality coefficient is a multi-scale analysis of the uncertainty in power supply at the generation end of the system. In the electricity spot market, the power supply and transmission during dispatch operations are also key factors in evaluating the quality of electricity spot market dispatch schemes. On the one hand, the dynamic dispatch of energy storage devices in the power grid, which play a role in peak shaving and valley filling, reflects the power supply and renewable energy consumption within the system. On the other hand, the power transmission in the transmission lines is affected by load changes, unit start-up and shutdown, and fluctuations in renewable energy generation, reflecting the dispatch quality of the electricity spot market. Simultaneously, line safety is crucial for ensuring the safety of the entire power grid; problems such as line overload pose a serious threat to grid safety. Therefore, it is necessary to analyze the dynamic fluctuations of energy storage devices and critical lines in the power grid to measure the quality of electricity spot market dispatch operation schemes from the transmission and load sides. Based on the above analysis, a dispatch fluctuation coefficient is calculated to measure the quality of dispatch schemes and ensure the stability and security of the power grid.

[0061] Specifically, the percentage of transmission power exceeding the limit for each line in the first n time points is obtained. The ratio of the maximum safe power of each line to the maximum safe power of all lines is used as the weight of each line. This weight is then used as the weight of the percentage of transmission power exceeding the limit for the corresponding line. The weighted summation of all lines yields the frequency metric for exceeding the limit at the nth time point. It should be noted that exceeding the transmission power limit refers to transmission power exceeding the maximum safe power of the line; the maximum safe power of the line can be obtained from the power grid company or the line's factory parameters.

[0062] Furthermore, the rate of change of SOC at the nth time point over a preset time interval is obtained, and the positive correlation mapping result of the rate of change is used as the energy storage fluctuation component at the nth time point. In this embodiment, the rate of change is expressed by the formula... Obtain, among which, , They are the first The moment, the first SOC data of the energy storage system at a given moment. This is a preset time interval used to control the measurement of energy storage system scheduling based on the rate of change of SOC data within the preset time period. In this embodiment, the value is 15 minutes. Implementers can adjust it according to actual conditions. It must be at least greater than the time interval for ultra-short-term forecasting (15 minutes). The positive correlation mapping result of the rate of change is obtained through the formula... Obtain; where, This represents an exponential function with the natural constant as its base.

[0063] Finally, the scheduling fluctuation coefficient is obtained, and the specific formula is as follows:

[0064]

[0065] In the formula, It is the first The scheduling fluctuation coefficient at each moment, , They are the first Energy storage fluctuation components and transmission line components at any given moment It was before The frequency of exceeding the limit at each time point is quantified, where N represents the total number of time points. In this embodiment, All lines are at the The average ratio of the actual transmission power to the maximum safe power of the line at a given moment. Similarly, the maximum safe power of the line can be obtained from the power grid company or the line's factory parameters.

[0066] Understandably, in the dispatching process of the electricity spot market, fluctuations in energy storage systems and transmission power need to be controlled. For a high-quality dispatching scheme, firstly, it is necessary to avoid drastic and rapid charging and discharging of energy storage batteries, meaning a small average rate of change in the corresponding State of Charge (SOC). Secondly, it is necessary to avoid continuous high-power transmission during transmission. While occasional limit-crossing issues are unavoidable due to the complexity of the transmission process, the probability of such issues should be kept low. Furthermore, lines with a greater impact on the grid should have their safety guaranteed. Typically, transmission lines with a greater impact on the grid bear a larger load, thus requiring larger transmission line capacities. This can be achieved by weighting the importance of different lines according to their maximum safe power ratio. In summary, the better the quality of the dispatching scheme, the smaller the dispatching fluctuation coefficient.

[0067] Step 4: Using a neural network based on historical power plant data, obtain the prediction function value at each time point, and compare it with the scheduling fluctuation coefficient at each time point to determine the scheduling operation evaluation value at each time point.

[0068] Historical data from the electricity spot market was obtained. This application acquired six months of historical data with a time granularity of 15 minutes, resulting in all historical sample data. Each sample corresponds to various monitoring data at a specific time point, with a 15-minute time interval between adjacent samples. It should be noted that the monitoring data at each time point includes at least the monitoring data collected in step one, as well as the predicted power generation at the day-ahead, intraday, and ultra-short-term time scales. Then, each sample was labeled according to steps two and three, and the prediction error proportionality coefficient and dispatch fluctuation coefficient at each time point were calculated. 70% of the historical samples were randomly selected as the training set, and the remaining samples were used as the test set. All sample data in the training set constituted a sample matrix, with each row corresponding to a sample data point and each column corresponding to a feature. The features included the prediction error proportionality coefficient at the three scales, the sample label, the actual output power of the corresponding renewable energy generation, and the dispatch fluctuation coefficient. A quantile regression algorithm was used for model training, with a target quantile of 0.1 and a maximum number of iterations set to 1000. The trained quantile regression model was output, in which the dispatch fluctuation coefficient was the dependent variable, and the other features were independent variables. Quantile regression algorithm is a well-known technique, so I will not go into details here.

[0069] When evaluating the operation of the electricity spot market dispatch at a subsequent time T, the prediction error ratio coefficient, sample label, and actual output power of the corresponding renewable energy generation at three scales are obtained according to the above steps at time T. The data corresponding to these characteristics are then input into the quantile regression model to output the corresponding function value. That is, the theoretical value of electricity spot market dispatch is determined by the quantile regression model based on historical operation, and it is compared with the dispatch fluctuation coefficient obtained from the actual data.

[0070]

[0071] This is the dispatch operation assessment value of the electricity spot market at time T. The actual scheduling fluctuation coefficient is calculated based on the actual data at time T. It is the output function value of the quantile regression model.

[0072] Understandably, quantile regression models can be used to determine the theoretical level that the dispatch system can achieve under the uncertainty of new energy power generation forecasts at different given time scales. The percentage difference between the actual dispatch fluctuation and the optimal dispatch fluctuation can then be used to assess the dispatch quality of the electricity spot market. Since a smaller dispatch fluctuation coefficient corresponds to a better quality dispatch scheme, a larger assessment value that is closer to the theoretical level indicates better dispatch in the electricity spot market.

[0073] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0074] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for evaluating the dispatching operation of a power spot market based on multiple source factors, characterized in that, The method comprises the following steps: The output power of the power station at each time, numerical weather data and SOC data are taken as a sample, and the power transmission of each line of the power station at each time is collected; All samples are labeled based on numerical weather data, combined with the output power at each time, and a machine learning model is used to obtain the power generation prediction value at each time at different time scales, analyze the error distribution characteristics of the power generation prediction at each time at different time scales, and determine the prediction error proportion coefficient of each time at each time scale; The numerical distribution of the power transmission of each line at each time is analyzed, combined with the change of the SOC data in the preset time interval, and the scheduling fluctuation coefficient at each time is determined; Based on the historical power station data, a neural network is used, combined with the error proportion coefficient and the scheduling fluctuation coefficient, to obtain the predicted function value at each time, and compared with the scheduling fluctuation coefficient at each time, to determine the scheduling operation evaluation value at each time; The determination of the prediction error proportion coefficient of each time at each time scale is specifically: When predicting t time, based on the time interval between different time scales and t time, the time weight of each time for predicting t time in each time scale is obtained; The difference between the actual output power and the corresponding predicted power generation of each time in each time scale is obtained, combined with the time weight, to determine the prediction weighted mean square error of t time in each time scale; All time scales are arranged in the prediction order, the prediction weighted mean square errors of adjacent two time scales at t time are compared, and the prediction error ratio of adjacent two time scales is obtained; The time scale includes day-ahead prediction, intra-day prediction and ultra-short-term prediction; The prediction error ratio obtained between all adjacent time scales at t time is forward fused to obtain the prediction error proportion coefficient of t time in ultra-short-term prediction; The prediction error ratio between t time in day-ahead prediction and intra-day prediction is subtracted by the prediction error proportion coefficient of t time in ultra-short-term prediction to determine the prediction error proportion coefficient of t time in intra-day prediction; The difference between numerical 1 and the prediction error ratio obtained between all adjacent time scales is calculated, and the difference is forward fused to obtain the prediction error proportion coefficient of t time in day-ahead prediction; The determination of the scheduling fluctuation coefficient at each time is specifically: The power transmission over-limit proportion of each line at the previous n times is obtained, the ratio of the maximum safe power of each line to the maximum safe power of all lines is taken as the importance weight of each line, the importance weight is taken as the weight of the power transmission over-limit proportion of the corresponding line, and the over-limit frequency measurement of the previous n times is obtained by weighted summation of all lines; The change rate of the SOC of the nth time in the preset time interval is obtained, and the positive correlation mapping result of the change rate is taken as the energy storage fluctuation component of the nth time; The numerical distribution between the power transmission of each line and the maximum safe power of the line at the nth time is integrated to determine the line power transmission component of the nth time; The specific formula of the scheduling fluctuation coefficient is: ； In the formula, is the scheduling fluctuation coefficient at the th time point, , are the energy storage fluctuation component and the line transmission component at the th time point, respectively, is the out-of-limit frequency metric of the previous time points, and N represents the total number of time points.

2. The multi-source factor based power spot market dispatch operation evaluation method according to claim 1, wherein, The time weight is determined by the negative correlation mapping of the time interval between different time scales and t time.

3. The multi-source based power spot market dispatch operation evaluation method according to claim 1, wherein, The prediction error ratio of the two adjacent time scales is obtained, specifically: The difference between the prediction weighted mean square errors of the two adjacent time scales is calculated, and the prediction weighted mean square error corresponding to the time scale with the previous prediction order is divided to obtain the prediction error ratio of the two adjacent time scales.

4. The multi-source based power spot market dispatch operation evaluation method according to claim 1, wherein, The power transmission power limit is specifically that the power transmission power is greater than the maximum safe power of the line.

5. The multi-source based power spot market dispatch operation evaluation method according to claim 1, wherein, The positive correlation mapping result of the change rate is obtained by a formula is obtained; wherein, represents an exponential function with a natural constant as a base number; , are SOC data of the energy storage system at the first time, and the first time, respectively, is a preset time interval.

6. The multi-source based power spot market dispatch operation evaluation method according to claim 1, wherein, The line power transmission component is specifically the average of the ratio of the actual power transmission of all lines at the first time to the maximum safe power of the lines.

7. The multi-source based power spot market dispatch operation evaluation method according to claim 1, wherein, In the process of obtaining the prediction function value of each time point based on the historical power station data, the neural network, the error proportion coefficient and the scheduling fluctuation coefficient, the features corresponding to each sample include the prediction error proportion coefficient in all time scales, the sample label, the actual output power and the scheduling fluctuation coefficient.

8. The multi-source based power spot market dispatch operation evaluation method according to claim 1, wherein, The scheduling operation evaluation value of each time point is determined, specifically: The ratio of the prediction function value of each time point to the scheduling fluctuation coefficient is taken as the scheduling operation evaluation value of each time point.

Images