Power grid power supply scheduling method and system based on load prediction

Abstract

The invention provides a power grid power supply scheduling method and system based on load prediction, and the method comprises the steps: carrying out the preprocessing of original time series data of the operation of a power grid, and generating net load time series data; predicting a net load mean value and a load fluctuation entropy spectrum in a future scheduling period in parallel to obtain a net load mean value prediction result and a load fluctuation entropy spectrum prediction result; analyzing a fluctuation risk type of a future scheduling period, and obtaining performance parameters of available adjustment resources from the heterogeneous resource feature library; adjusting resources are matched based on the obtained performance parameters, and a prepared adjusting resource pool is determined; and generating a comprehensive scheduling instruction including a basic output plan and an auxiliary service plan in combination with the net load mean value prediction result and the preparatory adjustment resource pool, and issuing the comprehensive scheduling instruction to a target control unit in the power grid. According to the method, the problems that the utilization of heterogeneous adjustment resources is not fine enough and does not have perspectiveness due to the fact that a scheduling strategy mainly pays attention to load mean value prediction but neglects fluctuation characteristic quantization in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of power grid technology, specifically relating to a power grid power dispatching method and system based on load forecasting. Background Technology

[0002] As the penetration rate of renewable energy sources such as wind power and photovoltaic power in the power grid continues to increase, the operating characteristics of the power system are undergoing profound changes. The output of these renewable energy sources is intermittent and fluctuating, resulting in increasingly complex dynamic characteristics of the net load of the power grid (i.e., the total load minus the output of uncontrollable renewable energy sources).

[0003] Existing power grid dispatching methods typically rely on forecasts of future average net load to plan generator output and maintain basic power balance. For random fluctuations in net load, dispatching systems usually address them by allocating a certain amount of reserve capacity, such as spinning reserve and frequency regulation reserve. However, this dispatching model primarily focuses on the average load trend, lacking quantitative analysis and prediction of the specific characteristics of the fluctuations themselves, such as their frequency, amplitude, and pattern.

[0004] Therefore, existing methods of allocating backup resources are insufficient for precision, and the system cannot predict in advance whether future fluctuations will be characterized by high-frequency, small-amplitude disturbances or by large-scale, sustained ramp-ups. This results in a lack of proactive and targeted matching and preparation for heterogeneous regulation resources with different response characteristics and regulation costs, such as conventional generators, hydropower units, and battery energy storage, during scheduling decisions. Consequently, scheduling strategies are often reactive, potentially leading to inefficient utilization of regulation resources or failing to guarantee system stability in the face of extreme fluctuations.

[0005] A search revealed that Chinese invention patent CN115169747A discloses a method, apparatus, and related equipment for predicting non-stationary time series of power load. The method includes: acquiring time series data of power load and preprocessing the time series data to obtain training data; decomposing the power load data of the training data using discrete wavelet transform to obtain low-frequency and high-frequency subsequences; training an initial ARIMAX model and an initial LSTM model based on the low-frequency and high-frequency subsequences respectively to obtain a target ARIMAX model and a target LSTM model; acquiring target feature indicators of the power load to be predicted and inputting these indicators into the target ARIMAX model and the target LSTM model respectively for power load prediction, obtaining low-frequency and high-frequency prediction data of the power load to be predicted; performing wavelet reconstruction and inverse standardization on both the low-frequency and high-frequency prediction data, and superimposing the results to obtain the final prediction result of the power load to be predicted.

[0006] The aforementioned prior art differs from this application in the following ways:

[0007] 1. The core idea of ​​the technical solution in the aforementioned comparative documents is to decompose the power load sequence into low-frequency and high-frequency subsequences using discrete wavelet transform, and then predict them using both the ARIMAX model and the attention-based LSTM model. Finally, the final prediction result is obtained through wavelet reconstruction, denormalization, and superposition. In contrast, the core idea of ​​this application is to predict the net load average and load fluctuation entropy spectrum in parallel, analyze the fluctuation risk type based on the entropy spectrum, match regulation resources from a heterogeneous resource library, and generate comprehensive dispatch instructions. The application objectives of the two are significantly different.

[0008] 2. The core objective of the prior art is to solve a purely predictive accuracy problem. Its focus is on how to more accurately predict future load values ​​from nonlinear, non-stationary load time series; while this application addresses the safety, economy, and refined scheduling of power grid operation. Its focus is on how to utilize predictive information to prepare resources in advance to cope with the risks brought about by load fluctuations. The core objectives of the two are clearly different.

[0009] 3. The prior art treats the "fluctuation characteristics" by considering the load sequence as composed of "components" of different frequencies. It uses wavelet decomposition to split the original sequence into low-frequency (trend component) and high-frequency (detail component), but it neither preserves nor outputs the independent characteristics of the fluctuations. While this application also performs wavelet decomposition, its purpose is not to predict the components themselves, but to calculate the permutation entropy of each scale component, thus forming a vector called the "load fluctuation entropy spectrum." The entropy value quantifies the disorder and complexity of the fluctuations at that scale, and the entropy spectrum is used to identify and classify fluctuation risks (such as high-frequency fluctuation risk, mid-frequency ramp risk). The two applications clearly differ in their treatment of "fluctuation characteristics."

[0010] A search revealed that Chinese invention patent CN118445600A discloses a method, apparatus, medium, and device for power load forecasting based on feature extraction. The method includes: acquiring time-series data; extracting features from the time-series data to obtain a first dataset; the feature extraction methods include discrete Fourier transform, discrete wavelet transform, and statistical methods; compressing the features of the first dataset to obtain a second dataset; selecting multiple data subsets of different output lengths from the second dataset and sequentially inputting them into an Autoformer model to obtain multiple single prediction models corresponding to the data subsets of different output lengths; and integrating the multiple single prediction models to output an integrated prediction model for power load forecasting.

[0011] The aforementioned prior art differs from this application in the following ways:

[0012] 1. The purpose of the aforementioned comparative documents is to improve the accuracy of the final load forecast values. They focus on extracting more useful features from the raw data to improve model learning, using Discrete Wavelet Transform (DWT), Discrete Wavelet Transform (DFT), and statistical methods as feature engineering techniques to expand the dimensionality and information content of the input features. In contrast, the purpose of this application is not to better predict load values, but rather to determine the types of risks future load fluctuations will pose to grid stability. After using Discrete Wavelet Transform (DWT) decomposition, it does not predict the specific values ​​of each component, but rather calculates the permutation entropy of each component. The application objectives of the two are significantly different.

[0013] 2. The output of the comparative document's predictive model is a single (or set of) load forecast values. These forecast values ​​can be provided to dispatchers as reports or used as input data for another independent dispatching system. They do not involve any dispatching decisions themselves. In contrast, the output of the predictive model in this application is a set of structured, decision-oriented information packets. These outputs are directly fed into the decision and command generation module to generate dispatching commands for the power grid, serving as the starting point for control behavior. The output properties of the two are completely different from their subsequent applications.

[0014] 3. The prior art document does not involve resource pools, resource matching, or any form of scheduling decision-making. All its innovations are concentrated on data preprocessing (feature extraction and compression) and predictive models (Autoformer integration), belonging to the pure data analysis level. In contrast, this application explicitly includes a "scheduling decision-making and resource pre-matching module," which has a built-in heterogeneous resource feature library that stores performance parameters (such as response time and regulation rate) of various resources (such as thermal power units and energy storage batteries). Based on the diagnosed risk type, the system automatically selects and matches the most suitable resources from the library to form a "prepared regulation resource pool." The two are fundamentally different in whether they include "resource matching" and "scheduling decision-making" stages. Summary of the Invention

[0015] To address the shortcomings of existing technologies, this invention provides a power grid dispatching method and system based on load forecasting. This solves the problem that existing dispatching strategies mainly focus on load average forecasting while ignoring the quantification of fluctuation characteristics, resulting in insufficient precision and lack of foresight in the utilization of heterogeneous regulation resources.

[0016] The specific plan is as follows:

[0017] A power grid dispatching method based on load forecasting includes the following steps: S100: Collect raw time series data of power grid operation, and preprocess the raw time series data to generate net load time series data; S200. Based on net load time series data, the average net load and load fluctuation entropy spectrum for future scheduling periods are predicted in parallel, and the prediction results of the average net load and the load fluctuation entropy spectrum are obtained respectively. S300: Based on the load fluctuation entropy spectrum prediction results, analyze the fluctuation risk type of the future scheduling period, and obtain the performance parameters of available adjustable resources from the heterogeneous resource feature library according to the fluctuation risk type. S400: Based on the obtained performance parameters, adjust the resources accordingly and determine the reserve adjustment resource pool; S500, combining the net load average forecast results and the reserve regulation resource pool, generates a comprehensive dispatch instruction that includes the basic output plan and the ancillary service plan, and issues it to the target control unit in the power grid.

[0018] Furthermore, the raw time series data in step S100 includes core power data of the power grid and auxiliary data; the core power data of the power grid includes total load time series data of the power grid and output time series data of uncontrollable renewable energy; the auxiliary data includes meteorological time series data and electricity market time series data; the preprocessing includes data cleaning and data alignment. The net load time series data is calculated according to the following formula: ; in, Represents the moment Net load power value; Represents the moment The total power load value of the power grid after collection and preprocessing; Represents the moment The actual aggregate power output of all uncontrollable renewable energy sources.

[0019] Furthermore, the method for predicting the average net load in step S200 is as follows: S211. Input the net load time series data within a preset historical time window, as well as the auxiliary data corresponding to the future forecast period, into the time series forecast model to capture and memorize the long-term dependency features. S212. Output the predicted sequence of net load mean for one or more future time steps, denoted as... ,in For the current moment, The prediction step size is greater than zero; The method for predicting the load fluctuation entropy spectrum is as follows: S221. Perform multi-scale decomposition of the net load sequence; Discrete wavelet transform technology is used to process net load time series data within a preset historical time window. conduct After layer decomposition, the original net load sequence is represented as: ; in, This is the preset maximum number of decomposition levels. It is the first Approximate components of the layer, It is the first Layer detail components; S222. Calculate the permutation entropy of the wave components at each scale; For any detail component sequence Its permutation entropy The calculation process is as follows: First, define the embedding dimension. and time delay Two parameters, through phase space reconstruction technology, transform a one-dimensional detail component sequence. Mapped to a series dimensional vector : ; Secondly, N is a positive integer, for each reconstructed vector In Sort the elements in ascending order and record the arrangement pattern formed by their original indices. There are a total of Possible arrangement patterns; Then, count every permutation pattern in all reconstructed vectors. Number of times And calculate its probability of occurrence. : ; Finally, the detail component is calculated. The corresponding permutation entropy : ; Among them, when When, define ; Construct and predict the load fluctuation entropy spectrum, and include all The permutation entropy calculation results corresponding to each detailed component are combined into a vector in order from high frequency to low frequency. This vector is the load fluctuation entropy spectrum for that historical time window. ; S223. Predict the load fluctuation entropy spectrum for future scheduling periods; The load fluctuation entropy spectrum sequence of historical time windows, along with auxiliary data related to volatility, is input into a multi-output prediction model to output the predicted load fluctuation entropy spectrum for one or more future scheduling periods. .

[0020] Furthermore, the volatility risk type in step S300 is obtained through qualitative analysis of volatility risk, as follows: S311. Determine the baseline threshold for multi-scale entropy values; Based on the load fluctuation entropy spectrum dataset calculated over a historical period, a baseline threshold vector with the same dimension as the load fluctuation entropy spectrum is set or calculated. ; S312, Identification of volatility risk types; The load fluctuation entropy spectrum prediction results for future scheduling periods , with baseline threshold vector By comparing and determining the type of volatility risk, as follows: a. If the load fluctuation entropy spectrum prediction results contain one or more predicted entropy values ​​corresponding to high-frequency components... Greater than its corresponding baseline threshold ,Right now Then, the volatility risk type of the future scheduling period will be analyzed as high-frequency volatility risk; b. If in the load fluctuation entropy spectrum prediction results, one or more predicted entropy values ​​corresponding to the mid-frequency component... Greater than its corresponding baseline threshold Then the volatility risk type will be analyzed as mid-frequency ramp risk; c. If the above judgment conditions for high-frequency volatility risk and mid-frequency ramp risk are met at the same time, that is, the predicted entropy values ​​of both high-frequency component and mid-frequency component exceed the limit, then the volatility risk type is analyzed as a composite volatility risk. d. If all predicted entropy values ​​in the load fluctuation entropy spectrum prediction results All are less than or equal to their corresponding baseline thresholds Then the volatility risk type will be analyzed as a low-risk stable state; The specific steps for obtaining the performance parameters of available adjustable resources from the heterogeneous resource feature library based on the volatility risk type are as follows: When the volatility risk type is high-frequency volatility risk, adjustable resources with response times less than the first preset time threshold are selected from the heterogeneous resource feature library; When the volatility risk type is mid-frequency ramp risk, adjustable resources with adjustment rates greater than the first preset rate threshold and adjustment duration greater than the first preset duration threshold are selected from the heterogeneous resource feature library. When the volatility risk type is complex volatility risk, adjustable resources that meet both of the above conditions are selected from the heterogeneous resource feature library; When the volatility risk type is low-risk and stable, no screening operation is performed; The performance parameters include response time, adjustment rate, adjustment capacity, duration of continuous adjustment, and adjustment cost.

[0021] Furthermore, the pre-adjustment resource pool in step S400 is a set containing one or more unique identifiers of adjustment resources, and is sorted based on adjustment cost.

[0022] Further, step S500 specifically includes: Step S500 specifically includes: S501. Generate a basic output plan based on the net load average forecast results; S502. Generate ancillary service plans based on the pre-adjusted resource pool; S503. Using the capacity reservation requirements for adjustment resources in the auxiliary service plan as constraints, we incorporate them into the optimization scheduling model used to generate the basic output plan for unified solution.

[0023] Furthermore, the heterogeneous resource feature library in step S300 is a collection of heterogeneous regulating resources with performance parameters; the heterogeneous regulating resources include thermal power generating units, hydropower generating units, battery energy storage systems, pumped storage power stations, and interruptible loads; the performance parameters include response time. Adjustment rate Adjustable capacity Adjusting costs And the opportunity costs incurred by various heterogeneous regulatory resources in providing regulatory services.

[0024] The present invention also provides a power grid power supply dispatching system based on load forecasting, for implementing the above-mentioned power grid power supply dispatching method based on load forecasting. The system includes: a data acquisition and preprocessing module, for acquiring raw time series data of power grid operation and preprocessing the raw time series data to generate net load time series data. The multidimensional characteristic prediction module is used to predict the average net load and the load fluctuation entropy spectrum for future scheduling periods in parallel based on net load time series data, and obtain the prediction results of the average net load and the load fluctuation entropy spectrum, respectively. The scheduling decision and resource pre-matching module analyzes the fluctuation risk type of the future scheduling period based on the load fluctuation entropy spectrum prediction results, and matches adjustment resources and determines the reserve adjustment resource pool according to the fluctuation risk type and the performance parameters of available adjustment resources obtained from the heterogeneous resource feature library module. The integrated dispatch instruction generation module is used to combine the net load average forecast results and the reserve adjustment resource pool to generate integrated dispatch instructions that include basic output plans and ancillary service plans, and issue them to the target control unit in the power grid.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. By introducing and predicting the load fluctuation entropy spectrum, this invention can quantify the complexity of future net load fluctuations at multiple time scales, rather than simply predicting the mean. Based on the different risk types such as high-frequency fluctuations and mid-frequency ramps identified, it can match regulation resources with corresponding performance advantages, thereby improving the effectiveness of power balance control.

[0027] 2. This invention transforms the traditional reactive backup scheduling into a proactive resource allocation by proactively analyzing future fluctuation risk types and adjusting resource pre-matching, which can effectively suppress the impact of events such as sudden changes in new energy output on the power grid frequency.

[0028] 3. By matching fluctuation risk with resource performance parameters, this invention can ensure that resources with fast response but high cost are used to smooth high-frequency disturbances, while resources with large adjustment capacity are used to deal with ramp-up events. This avoids the waste and improper use of adjustment capacity, and reduces the overall scheduling cost while ensuring system safety. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention;

[0030] Figure 2 This is the system architecture diagram of the present invention.

[0031] The module includes: 100, data acquisition and preprocessing module; 200, multi-dimensional characteristic prediction module; 300, heterogeneous resource feature library module; 400, scheduling decision and resource pre-matching module; and 500, comprehensive scheduling instruction generation module. Detailed Implementation

[0032] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. The system architecture diagram is shown below. Figure 2 As shown in the accompanying drawings and specific embodiments, the specific implementation of the present invention will be described in detail below.

[0033] Example 1:

[0034] This embodiment provides a power grid power dispatching method based on load forecasting, wherein the flowchart is as follows: Figure 1 As shown, it includes the following steps:

[0035] S100: Collect raw time series data of power grid operation, and preprocess the raw time series data to generate net load time series data;

[0036] The time resolution of the raw time series data in step S100 is no less than that at the minute level, including core power data of the power grid and auxiliary data.

[0037] Core power data of the power grid is used to characterize the actual supply and demand status of the power grid. It includes: total load time series data of the power grid, which records the total power of all power terminals in the power grid at each time; and output time series data of uncontrollable renewable energy, such as the actual aggregated grid-connected power of wind power plants and photovoltaic power plants.

[0038] The auxiliary data used for prediction, as additional feature inputs to improve the performance of the prediction model, includes: meteorological time series data, such as wind speed and wind direction data related to wind power output, and data such as light intensity, temperature, cloud thickness, and cloud movement speed related to photovoltaic power output; electricity market time series data, such as time-of-use electricity prices and ancillary service market prices; and calendar information, including holiday types and weekday or rest day identifiers.

[0039] Preprocessing aims to eliminate noise, errors, and inconsistencies in the raw data, providing high-quality, standardized data input for subsequent prediction modules.

[0040] Preprocessing includes data cleaning and data alignment;

[0041] Data cleaning: For missing or null values ​​that may exist in the data sequence, one or more interpolation methods can be used to fill them. For example, linear interpolation can be used to estimate the value of the missing point using the values ​​of the two valid data points before and after it. In scenarios with higher requirements for data quality, methods such as cubic spline interpolation can also be used to ensure the smoothness of the filled sequence. For outliers that may exist in the data due to communication or measurement errors, statistical methods can be used for identification and processing. For example, the 3σ criterion can be used for identification, that is, when the value of a data point deviates from the mean of the data in its window by more than three times the standard deviation, the point is identified as an outlier. Identified outliers can be removed and filled using the same method as for missing values.

[0042] Data alignment: Since the original time series data from different sources may have slight deviations in their timestamps or inconsistent sampling frequencies, in order to ensure the consistency of each time series in the time dimension, it is necessary to align all data series to a unified time grid. For example, all series can be resampled to a fixed time interval, such as 1 minute. For the specific implementation of this data resampling, those skilled in the art can use methods such as average sampling, forward padding, or backward padding, which are well-known technologies in the field and will not be described in detail here.

[0043] Net load time series data are calculated using the following formula:

[0044] ;

[0045] in, Represents the moment Net load power value; Represents the moment The total power load value of the power grid after collection and preprocessing; Represents the moment The actual aggregate power output of all uncontrollable renewable energy sources.

[0046] To improve the training efficiency and convergence of subsequent prediction models, the calculated net load sequence was also... Normalize other auxiliary data features, for example, by linearly mapping them to or Within the range.

[0047] S200. Based on net load time series data, the average net load and load fluctuation entropy spectrum for future scheduling periods are predicted in parallel, and the prediction results of the average net load and the load fluctuation entropy spectrum are obtained respectively.

[0048] The method for predicting the average net load in step S200 is as follows; S211. Input the net load time series data within a preset historical time window, as well as the auxiliary data corresponding to the future prediction period, into the time series prediction model to capture and memorize the long-term dependency features. The time series prediction model adopts a long short-term memory network model. After training, this LSTM model can learn the time dependency relationship within the net load time series data, as well as the nonlinear mapping relationship between the net load and various auxiliary data. S212. Output the predicted sequence of net load mean for one or more future time steps, denoted as... ,in For the current moment, The prediction step size is greater than zero;

[0049] The LSTM model processes the input feature sequence, which contains historical and future auxiliary information, through its internal input, forget, and output gates. It captures and remembers long-term dependency features, and the model's output is the predicted sequence of net load mean for one or more future time steps, which can be denoted as... ,in, For the current moment, With a prediction step size greater than zero, the construction and training of this LSTM model can employ backpropagation algorithms and optimizers such as Adam, aiming to minimize the mean squared error between the model's predicted values ​​and historical true values, and train the network weight parameters of the model.

[0050] Besides the LSTM model, other models capable of handling time series forecasting tasks, such as gated recurrent unit networks, attention-based Transformer networks, or traditional autoregressive integral moving average models, can also be used to implement the net load average forecasting function in this embodiment. These models can all receive historical series data and auxiliary data, and output a forecast of the future load average.

[0051] The method for predicting the load fluctuation entropy spectrum using a multi-output support vector regression model is as follows:

[0052] S221. Perform multi-scale decomposition of the net load sequence;

[0053] To analyze the different characteristics of load fluctuations at different time scales, this step employs discrete wavelet transform technology to process net load time series data within a preset historical time window. conduct Layer decomposition, or Discrete Wavelet Transform, is a technique that enables multi-scale decomposition of signals. After decomposition, the original net load sequence is represented as:

[0054] ;

[0055] in, The maximum preset number of decomposition layers can be determined based on the data sampling frequency and the range of fluctuation frequencies that need to be monitored.

[0056] It is the first The approximate components of the layer represent the low-frequency trend of the net load sequence;

[0057] It is the first The detailed components of the layer, which represent the net load sequence in the 1st layer, are... Fluctuation details within a specific frequency band (scale). By convention, Corresponding to the highest frequency fluctuation component, as As the frequency of the component increases, the frequency of the component gradually decreases. For example, Daubechies wavelet or Symlet wavelet can be selected as the basis function of wavelet transform.

[0058] S222. Calculate the permutation entropy of the wave components at each scale;

[0059] To quantify each detailed component The inherent dynamic complexity arises from this step, which independently calculates the permutation entropy for each detail component time series. Permutation entropy is a measure of the degree of disorder in a time series. For any detail component sequence... Its permutation entropy The calculation process is as follows:

[0060] First, define the embedding dimension. and time delay Two parameters, through phase space reconstruction technology, transform a one-dimensional detail component sequence. Mapped to a series dimensional vector :

[0061] ;

[0062] Secondly, for each reconstructed vector In Sort the elements in ascending order and record the arrangement pattern formed by their original indices. There are a total of Possible arrangement patterns;

[0063] Then, count every permutation pattern in all reconstructed vectors. Number of times And calculate its probability of occurrence. :

[0064] ;

[0065] Finally, the detail component is calculated. The corresponding permutation entropy :

[0066] ;

[0067] Among them, when When, define ; The larger the value, the more disordered the net load fluctuations at that scale, and the closer its dynamic behavior is to random noise;

[0068] Construct and predict the load fluctuation entropy spectrum, and include all The permutation entropy calculation results corresponding to each detailed component are combined into a vector in order from high frequency to low frequency. This vector is the load fluctuation entropy spectrum for that historical time window.

[0069] ;

[0070] This entropy spectral vector, across multiple time scales, collectively characterizes the intrinsic dynamic structure of net load fluctuations.

[0071] S223. Predict the load fluctuation entropy spectrum for future scheduling periods;

[0072] The load fluctuation entropy spectrum sequence of historical time windows, along with auxiliary data related to volatility, is input into a multi-output prediction model to output the predicted load fluctuation entropy spectrum for one or more future scheduling periods. .

[0073] S300: Based on the load fluctuation entropy spectrum prediction results, analyze the fluctuation risk type of the future scheduling period, and obtain the performance parameters of available adjustable resources from the heterogeneous resource feature library according to the fluctuation risk type.

[0074] S311. Determine the baseline threshold for multi-scale entropy values;

[0075] Based on the load fluctuation entropy spectrum dataset calculated over a historical period, a baseline threshold vector with the same dimension as the load fluctuation entropy spectrum is set or calculated. It can display entropy values ​​at various scales in historical datasets. The 95th percentile or a higher percentile is used as the corresponding baseline threshold. ;

[0076] S312, Identification of volatility risk types;

[0077] The load fluctuation entropy spectrum prediction results for future scheduling periods , with baseline threshold vector By comparing and determining the type of volatility risk, as follows:

[0078] a. If the load fluctuation entropy spectrum prediction results contain one or more predicted entropy values ​​corresponding to high-frequency components... (For example, when the number of decomposition layers) When it is large, (1 or 2) is greater than its corresponding baseline threshold ,Right now This classifies the volatility risk during future scheduling periods as high-frequency volatility risk. This risk type characterizes the possibility that the power grid may face dense, small-amplitude random disturbances in the future.

[0079] b. If in the load fluctuation entropy spectrum prediction results, one or more predicted entropy values ​​corresponding to the mid-frequency component... (For example, when) When within the middle range, it is greater than its corresponding baseline threshold. The volatility risk type is then analyzed as mid-frequency ramp risk, which represents the possibility of a significant and sustained increase or decrease in power in the power grid in the future.

[0080] c. If the above-mentioned judgment conditions for high-frequency volatility risk and mid-frequency ramp risk are met simultaneously, that is, the predicted entropy values ​​of both the high-frequency component and the mid-frequency component exceed the limit, then the volatility risk type is analyzed as a composite volatility risk.

[0081] d. If all predicted entropy values ​​in the load fluctuation entropy spectrum prediction results All are less than or equal to their corresponding baseline thresholds If the volatility risk type is analyzed as a low-risk stable state, this state indicates that the net load fluctuation of the power grid in the future period will be relatively smooth, with no significant disturbance risk.

[0082] Specifically, the process of obtaining performance parameters of available adjustable resources from a heterogeneous resource feature library based on volatility risk type is as follows:

[0083] When the volatility risk type is high-frequency volatility risk, the adjustment resources with response time less than the first preset time threshold are selected from the heterogeneous resource feature library, mainly including battery energy storage systems, supercapacitors, etc.

[0084] When the volatility risk type is mid-frequency ramp risk, the regulating resources with a regulation rate greater than the first preset rate threshold and a regulation duration greater than the first preset duration threshold are selected from the heterogeneous resource feature library. These mainly include hydroelectric generator sets, gas turbine sets and pumped storage power stations.

[0085] When the volatility risk type is complex volatility risk, adjustable resources that meet both of the above conditions are selected from the heterogeneous resource feature library;

[0086] When the volatility risk type is low-risk and stable, no screening operation is performed;

[0087] The above performance parameters include response time, regulation rate, regulation capacity, duration of continuous regulation, and regulation cost.

[0088] The heterogeneous resource feature library is a collection of heterogeneous regulating resources with performance parameters; heterogeneous regulating resources include thermal power generating units, hydropower generating units, battery energy storage systems, pumped storage power stations, and interruptible loads; performance parameters include response time. Adjustment rate ,

[0089] Adjustable capacity Adjusting costs And the opportunity costs incurred by various heterogeneous regulatory resources in providing regulatory services.

[0090] Response time This parameter quantifies the time delay between receiving a scheduling command and the actual change in a resource's power output, measured in seconds or minutes. It is a core indicator for measuring resource response speed. The value indicates that the resource can respond to scheduling requests more quickly.

[0091] Adjustment rate This parameter quantifies the maximum power adjustment a resource can achieve per unit time, expressed in megawatts per minute. It characterizes the resource's power ramp-up or ramp-down capability, and a larger value indicates this capability. The value indicates that the resource can provide or absorb high power more quickly.

[0092] Adjustable capacity This parameter quantifies the total power range that resources can provide for regulation services, measured in megawatts. For generator sets, it is typically the range between their technical output upper and lower limits; for energy storage systems, it is the upper limit of their charging and discharging power and the duration of continuous regulation. This parameter is used to quantify the maximum time that an energy-constrained resource can continuously provide regulation services at a specific output power. The unit can be minutes or hours. This parameter is determined by the resource's available energy and its current regulation power. For unconstrained resources, this value can be regarded as a maximum value.

[0093] Adjusting costs This parameter is used to quantify the economic cost of providing regulation services for resources, and the unit can be yuan / megawatt-hour or yuan / time.

[0094] S400. Based on the obtained performance parameters, match and adjust resources to determine a preliminary adjustment resource pool. The preliminary adjustment resource pool is a set containing one or more unique identifiers of adjustment resources, and is sorted according to adjustment cost. When multiple resources can meet the performance requirements, the resource with lower adjustment cost is selected first.

[0095] S500. Combining the net load average forecast results and the reserve regulation resource pool, generate a comprehensive dispatch instruction including a basic output plan and ancillary service plans, and issue it to the target control unit in the power grid; S501. Based on the net load average forecast results, generate a basic output plan. The goal of this basic output plan is to plan the basic output curves of each conventional generator unit during future dispatch periods, so as to macroscopically track and meet the predicted net load average, while minimizing system operating costs; S502. Based on the reserve regulation resource pool, generate an ancillary service plan. This ancillary service plan is a configuration and status preparation scheme for various resources in the reserve regulation resource pool. It pre-specifies the type of regulation task that each resource in the pool needs to undertake during future dispatch periods and reserves the corresponding regulation capacity for it; S503. Using the capacity reservation requirements for regulation resources in the ancillary service plan as constraints, incorporate them into the optimization dispatch model used to generate the basic output plan for unified solution, as follows:

[0096] If resources Designated to provide in the ancillary services plan The upward reserve capacity, then when solving the basic output plan, the output of the generator unit. The following constraints must be met:

[0097] ;

[0098] in, This is the technical maximum output of the generator set. This constraint ensures that while the unit is executing its basic output plan, it always has sufficient capacity at the top to cope with possible sudden power demands.

[0099] If resources If a service is designated to provide bidirectional high-frequency regulation services in the ancillary services plan, its own charge and discharge plans will be constrained when solving the base output plan to ensure its state of charge. Maintain within a preset central range This ensures that the energy storage system has sufficient energy for discharging and sufficient space for charging when needed. By incorporating these constraints derived from the ancillary service plan into the optimization model of the basic output plan, this embodiment achieves a comprehensive dispatch plan that integrates conventional energy dispatch with forward-looking risk reserves. This plan is not only economically optimized but also structurally prepared to address predicted volatility risks.

[0100] Example 2:

[0101] This embodiment provides a power grid dispatching system based on load forecasting. The system architecture diagram is as follows: Figure 2As shown, the system used to implement the scheduling method in Embodiment 1 includes a data acquisition and preprocessing module 100, a multi-dimensional characteristic prediction module 200, a heterogeneous resource feature library module 300, a scheduling decision and resource pre-matching module 400, and a comprehensive scheduling instruction generation module 500.

[0102] The data acquisition and preprocessing module 100 is used to acquire raw data of power grid operation from multiple data sources such as power system data centers and meteorological information systems, and to standardize the raw data to generate net load time series data for subsequent analysis.

[0103] The data acquisition and preprocessing module 100 can connect to the data interfaces of the power grid monitoring and control and data acquisition system, wide-area measurement system, energy management system, meteorological information service system and power trading center through standard communication protocols to periodically and automatically acquire the above-mentioned types of data.

[0104] The multidimensional characteristic prediction module 200, connected to the data acquisition and preprocessing module 100, is used to receive net load time series data and perform parallel predictions of its characteristics in two different dimensions over a future period, outputting the mean net load prediction result and the load fluctuation entropy spectrum prediction result, respectively.

[0105] The heterogeneous resource feature library module 300 exists as a structured database used to store performance parameters of various dispatchable and adjustable resources in the power grid. These performance parameters include response time, regulation rate, regulation capacity, duration of regulation, and regulation cost.

[0106] To obtain real-time values ​​of these performance parameters, the heterogeneous resource feature library module 300 needs to establish communication connections with relevant external data systems. For example, it can connect with the battery management system of a battery storage station to obtain real-time SOC data; or connect with the monitoring system of a power plant or the energy management system of the power grid to obtain the unit's operating status and availability information.

[0107] Data updates within the heterogeneous resource feature library module 300 can be triggered in various ways. For parameters that change slowly, periodic updates can be used, such as daily or hourly updates. For parameters that change dynamically, near real-time updates or event-triggered updates can be used. For example, when the BMS detects a significant change in SOC or a change in unit operating status, the relevant system will proactively push the updated data to the heterogeneous resource feature library module 300. Alternatively, the feature library module 300 can proactively query the database at a relatively short interval. After receiving the updated data, the module will then update the corresponding field value of the resource ID in the database.

[0108] Through the aforementioned construction and maintenance mechanisms, the heterogeneous resource feature library module 300 can ensure the accuracy and timeliness of the performance parameters it stores, providing a reliable data foundation for the precise formulation of scheduling decisions.

[0109] The scheduling decision and resource pre-matching module 400 is connected to the multi-dimensional characteristic prediction module 200 and the heterogeneous resource feature library module 300, respectively. This module receives the net load average prediction results and the load fluctuation entropy spectrum prediction results, and analyzes the fluctuation risk type for future scheduling periods based on the load fluctuation entropy spectrum prediction results. Subsequently, according to the analyzed fluctuation risk type, this module retrieves and filters adjustment resources with matching performance indicators from the heterogeneous resource feature library module 300 to form a forward-looking reserve adjustment resource pool.

[0110] The integrated dispatching instruction generation module 500 is connected to the dispatching decision and resource pre-matching module 400. This module is used to integrate basic load output demand with forward-looking risk response demand. Specifically, it combines the basic output plan corresponding to the net load average forecast with the ancillary service plan corresponding to the reserve adjustment resource pool to generate a comprehensive power supply dispatching instruction, which is then sent to the relevant control units in the power grid through the communication interface.

[0111] The above embodiments are for illustrative purposes only and are not intended to limit the scope of this application. Although this application has been described in detail with reference to the embodiments, those skilled in the art should understand that various combinations, modifications, or equivalent substitutions of the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application and should be covered within the scope of the claims of this application.

Claims

1. A power grid supply dispatching method based on load forecasting, characterized in that, Includes the following steps: S100: Collect raw time series data of power grid operation, and preprocess the raw time series data to generate net load time series data; S200. Based on net load time series data, the average net load and load fluctuation entropy spectrum for future scheduling periods are predicted in parallel, and the prediction results of the average net load and the load fluctuation entropy spectrum are obtained respectively. S300: Based on the load fluctuation entropy spectrum prediction results, analyze the fluctuation risk type of the future scheduling period, and obtain the performance parameters of available adjustable resources from the heterogeneous resource feature library according to the fluctuation risk type. S400: Based on the obtained performance parameters, adjust the resources accordingly and determine the reserve adjustment resource pool; S500, combining the net load average forecast results and the reserve regulation resource pool, generates a comprehensive dispatch instruction that includes the basic output plan and the ancillary service plan, and issues it to the target control unit in the power grid.

2. The power grid supply dispatching method based on load forecasting according to claim 1, characterized in that, The raw time-series data in step S100 includes core power grid data and auxiliary data; the core power grid data includes total power grid load time-series data and uncontrollable renewable energy output time-series data; the auxiliary data includes meteorological time-series data and electricity market time-series data; the preprocessing includes data cleaning and data alignment. The net load time series data is calculated according to the following formula: ； in, Represents the moment Net load power value; Represents the moment The total power load value of the power grid after collection and preprocessing; Represents the moment The actual aggregate power output of all uncontrollable renewable energy sources.

3. The power grid supply dispatching method based on load forecasting according to claim 2, characterized in that, The method for predicting the average net load in step S200 is as follows: S211. Input the net load time series data within a preset historical time window, as well as the auxiliary data corresponding to the future forecast period, into the time series forecast model to capture and memorize the long-term dependency features. S212. Output the predicted sequence of net load mean for one or more future time steps, denoted as... ,in For the current moment, The prediction step size is greater than zero; The method for predicting the load fluctuation entropy spectrum is as follows: S221. Perform multi-scale decomposition of the net load sequence; Discrete wavelet transform technology is used to process net load time series data within a preset historical time window. conduct After layer decomposition, the original net load sequence is represented as: ； in, This is the preset maximum number of decomposition levels. It is the first Approximate components of the layer, It is the first Layer detail components; S222. Calculate the permutation entropy of the wave components at each scale; For any detail component sequence Its permutation entropy The calculation process is as follows: First, define the embedding dimension. and time delay Two parameters, through phase space reconstruction technology, transform a one-dimensional detail component sequence. Mapped to a series dimensional vector : ； Secondly, N is a positive integer, for each reconstructed vector In Sort the elements in ascending order and record the arrangement pattern formed by their original indices. There are a total of Possible arrangement patterns; Then, count every permutation pattern in all reconstructed vectors. Number of times And calculate its probability of occurrence. : ； Finally, the detail component is calculated. The corresponding permutation entropy : ； Among them, when When, define ; Construct and predict the load fluctuation entropy spectrum, and include all The permutation entropy calculation results corresponding to each detailed component are combined into a vector in order from high frequency to low frequency. This vector is the load fluctuation entropy spectrum for that historical time window. ； S223. Predict the load fluctuation entropy spectrum for future scheduling periods; The load fluctuation entropy spectrum sequence of historical time windows, along with auxiliary data related to volatility, is input into a multi-output prediction model to output the predicted load fluctuation entropy spectrum for one or more future scheduling periods. .

4. The power grid power dispatching method based on load forecasting according to claim 3, characterized in that, The volatility risk type in step S300 is obtained through qualitative analysis of volatility risk, as follows: S311. Determine the baseline threshold for multi-scale entropy values; Based on the load fluctuation entropy spectrum dataset calculated over a historical period, a baseline threshold vector with the same dimension as the load fluctuation entropy spectrum is set or calculated. ; S312, Identification of volatility risk types; The load fluctuation entropy spectrum prediction results for future scheduling periods , with baseline threshold vector By comparing and determining the type of volatility risk, as follows: a. If the load fluctuation entropy spectrum prediction results contain one or more predicted entropy values ​​corresponding to high-frequency components... Greater than its corresponding baseline threshold ,Right now Then, the volatility risk type of the future scheduling period will be analyzed as high-frequency volatility risk; b. If in the load fluctuation entropy spectrum prediction results, one or more predicted entropy values ​​corresponding to the mid-frequency component... Greater than its corresponding baseline threshold Then the volatility risk type will be analyzed as mid-frequency ramp risk; c. If the above judgment conditions for high-frequency volatility risk and mid-frequency ramp risk are met at the same time, that is, the predicted entropy values ​​of both high-frequency component and mid-frequency component exceed the limit, then the volatility risk type is analyzed as a composite volatility risk. d. If all predicted entropy values ​​in the load fluctuation entropy spectrum prediction results All are less than or equal to their corresponding baseline thresholds Then the volatility risk type will be analyzed as a low-risk stable state; The specific steps for obtaining the performance parameters of available adjustable resources from the heterogeneous resource feature library based on the volatility risk type are as follows: When the volatility risk type is high-frequency volatility risk, adjustable resources with response times less than the first preset time threshold are selected from the heterogeneous resource feature library; When the volatility risk type is mid-frequency ramp risk, adjustable resources with adjustment rates greater than the first preset rate threshold and adjustment duration greater than the first preset duration threshold are selected from the heterogeneous resource feature library. When the volatility risk type is complex volatility risk, adjustable resources that meet both of the above conditions are selected from the heterogeneous resource feature library; When the volatility risk type is low-risk and stable, no screening operation is performed; The performance parameters include response time, adjustment rate, adjustment capacity, duration of continuous adjustment, and adjustment cost.

5. The power grid supply dispatching method based on load forecasting according to claim 4, characterized in that, The pre-adjustment resource pool in step S400 is a set containing one or more unique identifiers of adjustment resources, and is sorted according to adjustment cost.

6. The power grid supply dispatching method based on load forecasting according to claim 5, characterized in that, The specific steps of step S500 include: S501. Generate a basic output plan based on the net load average forecast results; S502. Generate ancillary service plans based on the pre-adjusted resource pool; S503. Using the capacity reservation requirements for adjustment resources in the auxiliary service plan as constraints, we incorporate them into the optimization scheduling model used to generate the basic output plan for unified solution.

7. A power grid power dispatching method based on load forecasting according to claim 2, characterized in that, In step S100, the time resolution of the original time series data is no less than that of minutes, and the net load sequence obtained in step S100... The auxiliary data features also need to be normalized.

8. The power grid supply dispatching method based on load forecasting according to claim 3, characterized in that, In step S200, a long short-term memory network is used to predict the net load average, and a multi-output support vector regression model is used to predict the load fluctuation entropy spectrum for future scheduling periods.

9. A power grid power dispatching method based on load forecasting according to claim 4, characterized in that, The heterogeneous resource feature library in step S300 is a collection of heterogeneous regulating resources with performance parameters; the heterogeneous regulating resources include thermal power generating units, hydropower generating units, battery energy storage systems, pumped storage power stations, and interruptible loads; the performance parameters include response time. Adjustment rate , Adjustable capacity Adjusting costs And the opportunity costs incurred by various heterogeneous regulatory resources in providing regulatory services.

10. A power grid power supply dispatching system based on load forecasting, used to implement the power grid power supply dispatching method based on load forecasting as described in any one of claims 1-9, characterized in that, The system includes: a data acquisition and preprocessing module, used to acquire raw time series data of power grid operation and preprocess the raw time series data to generate net load time series data; The multidimensional characteristic prediction module is used to predict the average net load and the load fluctuation entropy spectrum for future scheduling periods in parallel based on net load time series data, and obtain the prediction results of the average net load and the load fluctuation entropy spectrum, respectively. The scheduling decision and resource pre-matching module analyzes the fluctuation risk type of the future scheduling period based on the load fluctuation entropy spectrum prediction results, and matches the adjustment resources according to the fluctuation risk type and the performance parameters of the available adjustment resources obtained from the heterogeneous resource feature library to determine the reserve adjustment resource pool. The integrated dispatch instruction generation module is used to combine the net load average forecast results and the reserve adjustment resource pool to generate integrated dispatch instructions that include basic output plans and ancillary service plans, and issue them to the target control unit in the power grid.

Images