Double-layer rolling precise scheduling method based on new energy output error sampling calibration

Abstract

The invention relates to a double-layer rolling precise scheduling method based on new energy output error sampling calibration, and belongs to the field of power systems. According to the method, a new energy output fluctuation range is determined by analyzing historical data, a prediction error model is established, a sampling probability interval is corrected, and a representative new energy output scene is generated. Meanwhile, scene reduction is carried out in combination with BTM and FCM clustering algorithms, extreme scenes are reserved, and the diversity of a scene library and the reliability of a scheduling scheme are improved. In addition, the method also provides a day-ahead-intra-day double-layer rolling optimization scheduling model, and by dynamically adjusting a day-ahead scheduling plan, real-time fluctuation of new energy output and load is effectively dealt with, and stable operation of the microgrid is ensured. The method can effectively reduce the average error of the new energy output, stabilize the output fluctuation, improve the operation efficiency and safety of the microgrid, and provide a new thought for the scheduling optimization, risk management and extreme event detection of wind and light power generation.

Description

Technical Field

[0001] The invention belongs to the field of electric power systems and relates to a double-layer rolling precise dispatching method based on new energy output error sampling and calibration. Background Art

[0002] As a carrier for the consumption of new energy, microgrids can achieve energy conservation, emission reduction and multi-energy complementarity, but the fluctuation of renewable energy output and the error of day-ahead scheduling will threaten the stability of microgrid operation. The task of the present invention is to study the method of generating wind and solar output scenarios, realize the smoothing of new energy output fluctuations, cooperate with the intraday double-layer rolling multi-time scale optimization scheduling method, reduce the day-ahead scheduling error, and improve the stability of microgrid operation. The technical process includes multiple key steps. First, the fluctuation range of new energy output is analyzed based on historical data, and then a new energy prediction error model is established based on the prediction box principle and Gaussian mixture model, so as to realize the sampling of new energy time series prediction error and correct the sampling probability interval, and superimpose the sampled error on the predicted value to obtain a new energy output time series scenario set, and reduce the generated scenarios through the FCM clustering algorithm improved by BTM, and obtain representative scenarios of wind and photovoltaic power generation and their corresponding probabilities. Finally, starting from solving the impact of new energy prediction error and uncertainty on microgrid, a microgrid multi-time scale optimization scheduling method considering new energy output scenarios is proposed, and a day-ahead cold-heat-electricity double-layer rolling model is established based on the day-ahead scheduling plan. Through this technical method, the impact of fluctuations in renewable energy output on the stability of the microgrid can be effectively reduced, improving its operational reliability and controllability.

[0003] At present, the research on multi-energy microgrids mainly focuses on model establishment, planning and energy management. The traditional optimization scheduling method optimizes the scheduling of micro-sources based on short-term forecast data, considering only the single time scale of the day-ahead. The results obtained are too one-sided and the control effect is unstable. In addition, the random fluctuations of renewable energy and the load forecast error will significantly affect the day-ahead plan, and the single cooling mode cannot fit the actual cooling energy operation conditions well. The core challenge in the operation of microgrids is to deal with the volatility and uncertainty of renewable energy output. Robust optimization and scenario analysis methods are usually used to deal with system uncertainty. The existing time series scenario generation method based on statistical methods still has some shortcomings: the use of random sampling for time series scenario generation does not consider the correlation of renewable energy in time series, and the time series scenario generation based on Markov chain has the problem of mapping discrete state quantities to continuous quantities, which will cause no fluctuation or drastic fluctuations in the sampling of the previous and next moments. In addition, although the scenario reduction technology reduces the amount of optimization calculations, its norm distance definition based on probability measurement lacks theoretical breakthroughs, which makes it difficult to cover extreme ramping or output events such as wind power. These low-probability long-tail events are often ignored in the scenario reduction process, resulting in the final typical scenario failing to fully consider extreme situations, weakening the system's ability to respond to risks.

[0004] The current optimization dispatching method of microgrid has not yet considered the "suboptimal" problem caused by the fluctuation of wind power output within the day, and has not taken the wind power forecast error of the day into consideration. There may be wind abandonment during the intraday dispatching, resulting in loss of benefits. In addition, the method of generating time series scenarios based on the first-order Marko sky chain in the existing scenario analysis method adopts an indirect power sampling method that first samples the power state and then processes the power continuously, which is easy to cause no fluctuation or violent fluctuation of new energy power at adjacent moments. In contrast, the present invention improves the existing method in the generation of time series scenarios, adjusts the probability interval of prediction error sampling based on the historical volatility of new energy, generates wind and solar output scenarios according to the time series sampling error and superimposes it on the predicted value, and uses the BTM improved FCM algorithm to reduce the scenario, so as to obtain a more representative and comprehensive output scenario. In addition, the commonly used fast forward selection method, K-means clustering algorithm, hierarchical clustering algorithm and other methods realize the quantitative analysis of uncertainty in the generation of wind and solar scenarios, but often ignores the retention of extreme scenarios in the scenario reduction process, and cannot guarantee the feasibility of the planning scheme under extreme events. The FCM algorithm can flexibly handle fuzzy boundaries and adapt to dynamic changes, thereby capturing data features more comprehensively. The algorithm divides scene categories according to the degree of membership, and can quickly identify and generate extreme scenes through data classification, but there is still a situation where the differentiated feature information of the scene data itself is ignored. In order to make up for the shortcomings of the FCM algorithm in the difference characteristics of scene data, the present invention identifies the potential patterns in the wind and solar power output data through BTM co-occurrence analysis, and inputs these patterns as features into the FCM clustering, so as to better identify typical scenes, extreme scenes and fluctuation patterns. Finally, based on the generated wind and solar power output time series scenario, the day-ahead-intraday double-layer rolling optimization scheduling method is used to adjust the day-ahead scheduling plan from different time scales, improve the accuracy of scheduling, and ensure the stable operation of the microgrid. The optimization method used in the present invention can provide new ideas for the scheduling optimization, risk management and extreme event detection of wind and solar power generation. In summary, the innovation of the present invention lies in the optimization method proposed to deal with the uncertainty of new energy output and improve the scheduling accuracy of microgrids, which effectively improves the safety and reliability of microgrid operation. Summary of the invention

[0005] In view of this, the object of the present invention is to provide a double-layer rolling precise scheduling method based on new energy output error sampling and calibration.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A double-layer rolling precise scheduling method based on new energy output error sampling and calibration, the method comprises the following steps:

[0008] S1: Generate a typical renewable energy output timing scenario based on error sampling probability interval correction

[0009] Firstly, the fluctuation range of renewable energy output is analyzed based on historical data, and then a renewable energy prediction error model is established based on the prediction box and Gaussian mixture model (GMM). Then, the prediction error of renewable energy time series is sampled and the sampling probability interval is corrected. The sampled error is superimposed on the predicted value to obtain a set of renewable energy output time series scenarios. Finally, the generated scenarios are reduced through the FCM method optimized by BTM, and representative scenarios of wind and photovoltaic power generation and their corresponding probabilities are obtained, providing scenario support for microgrid optimization scheduling.

[0010] S2: Establishing a dual-layer rolling optimization model for day-ahead scheduling and intraday scheduling

[0011] S21: Establishing the day-ahead scheduling model

[0012] S211: Establishing the objective function

[0013] The day-ahead scheduling model is a typical mixed integer linear programming problem, and its goal is to minimize the operating cost:

[0014]

[0015] Where: F day-ahead Plan operating costs for day-ahead dispatch; is the probability of the i-th wind and solar power output scenario; For scene S i The interaction power cost between the lower and upper grids; For scene S i Lower fuel costs; For scene S i The aging cost of battery charging and discharging;

[0016] 1) Grid interaction power cost

[0017]

[0018] Where: T represents the day-ahead dispatch period; is the unit price of electricity purchased during period T; For scene S i Purchase power from the grid during the next T period;

[0019] 2) Fuel costs

[0020]

[0021] Where: The unit price of natural gas purchased during period T; and The scene S i The natural gas consumption of micro gas turbines, gas boilers and fuel cells during the next T period; H FUEL It is the low calorific value of natural gas;

[0022] 3) Battery charging and discharging aging cost

[0023]

[0024] Where: is the battery unit aging cost during period T; and For scene S i Battery charge and discharge state variables during the next T period;

[0025] S212: Establishing Constraints

[0026] The system considers equipment operation constraints, power balance constraints, and power interaction constraints;

[0027] 1) Equipment operation constraints

[0028]

[0029] Where: P e T is the output power of device e during period T; is the start / stop state variable of device e during period T; P e max and P e min is the maximum and minimum output power of device e; P e up and P e dn The upper and lower limits of the climbing power of device e;

[0030] 2) Power balance constraints

[0031]

[0032] In the formula: The following are all variables in the T period; and The scene S i Wind power and photovoltaic output power under and are the output electrical power of the micro gas turbine and the fuel cell respectively; and They are the charging and discharging power of BT respectively; and are the electric power consumed by ice storage air conditioner, electric boiler, CCS and P2G respectively; Output cooling power for absorption chiller; and They are electricity, cooling and heating loads respectively; and are the output thermal power of waste heat boiler, electric boiler and gas boiler respectively; and are the heat storage and release power of the heat storage tank respectively; is the heat power absorbed by the absorption refrigerator;

[0033] 3) Power interaction constraints

[0034]

[0035] Where: and are the maximum and minimum values ​​of the system's power purchase respectively;

[0036] S213: Establishing a two-layer rolling optimization model within a day

[0037] The intraday rolling optimization is divided into a double-layer structure of hot-cold rolling and electric rolling. Hot-cold rolling controls hot-cold power, and electric rolling controls electric power, so as to smooth out the actual power fluctuation. The hot-cold rolling control time domain X is 1 hour, and the scheduling time domain is 2 hours. The electric rolling control time domain Y is 5 minutes, and the scheduling time domain is 1 hour. In the hot-cold rolling scheduling, at t 0 time, predict t 0 +X to t 0 +X+2 period of cooling and heating load, so as to adjust t 0 +X to t 0 +2X period unit equipment output plan; at the same time, in the electric rolling dispatch, at t 0 Time prediction t 0 +Y to t 0 +Y+1 period of electrical load, adjust t 0 +Y to t 0 +2Y unit power generation plan;

[0038] (1) Hot and cold rolling optimization scheduling model

[0039] Based on the equipment operation status and the cooling and heating power dispatch plan on the previous day, the output plan of each unit is adjusted according to the change of cooling and heating load in the t period; the objective function is:

[0040]

[0041] Where: t represents the scheduling period within the day; F day,up Optimize dispatch costs for intraday hot and cold rolling; is the hot and cold rolling fuel cost for period t; is the power change cost of the electric boiler during period t; is the power change cost of ice storage air conditioning during period t; is the cost of heat storage and release power change of the heat storage tank during period t;

[0042]

[0043] Where: and are the gas consumption adjustment of the micro gas turbine and gas boiler during period t; μ MT , μ GB , μ EB , μ ISAC and μ h The unit penalty costs for micro gas turbine, gas boiler, electric boiler, ice storage air conditioner and thermal storage tank adjustment are respectively; and is the cooling and heating adjustment power of the micro gas turbine, electric boiler, gas boiler and ice storage air conditioner during period t; and They are respectively the heat storage and release power adjustment values ​​of the heat storage tank in period t;

[0044] Constraints:

[0045] 1) Power balance constraints

[0046]

[0047] 2) Power adjustment constraints

[0048]

[0049] In the cold and hot rolling dispatch, during the off-peak electricity price period, when the energy supply is greater than the demand, the output of the gas boiler, micro gas turbine, electric boiler, and ice storage air conditioner is adjusted in sequence; when the energy supply is less than the demand, the output of the ice storage air conditioner, gas boiler, electric boiler, and micro gas turbine is adjusted in sequence; during the non-off-peak electricity price period, when the energy supply is greater than the demand, the output of the gas boiler, ice storage air conditioner, electric boiler, and micro gas turbine is adjusted in sequence; when the energy supply is less than the demand, the output of the gas boiler, micro gas turbine, electric boiler, and ice storage air conditioner is adjusted in sequence;

[0050] (2) Establishing an electric rolling optimization scheduling model

[0051] Based on the day-ahead battery charge and discharge status, and with the fluctuation of wind and solar power output and the changes in the power and electric load of the cold and hot rolling dispatching equipment, the day-ahead dispatching plan is adjusted; the objective function is:

[0052]

[0053] Where: F day,lowOptimize dispatching costs for intraday electricity rolling; is the cost of purchasing electricity from the power grid during period t; is the rolling fuel cost of electricity during period t; is the cost of battery charging and discharging power change; β is the SOC penalty factor; is the SOC value of the battery during the T period in the day-ahead dispatch; is the battery SOC value during period t in intraday scheduling;

[0054]

[0055] Where: is the adjustment value of purchased power in period t; μ GRID is the unit penalty cost of purchasing power adjustment; μ FC unit penalty costs adjusted for fuel cells; is the gas consumption adjustment of the fuel cell during period t; and are the adjusted electric power of the micro gas turbine and the fuel cell during period t; μ p Unit penalty costs adjusted for battery charging and discharging; and They are the battery charging and discharging power adjustment values ​​for period t respectively;

[0056] In electric rolling dispatch, power balance constraints and power adjustment constraints are met, which are similar to hot and cold rolling constraints; in addition, energy storage constraints are also met:

[0057]

[0058] Where: η BT,ch and η BT,dis are the charge and discharge efficiency of the battery respectively; and They are the upper and lower limits of the battery SOC value respectively; is the change of SOC in period t.

[0059] Further, the S1 comprises the following steps:

[0060] S11: Determine the fluctuation range of wind and solar power output

[0061] The actual wind and solar output data of similar weather days in history are obtained by collecting historical weather information, and the fluctuation range of wind and solar output data in each adjacent period of similar weather days is obtained by performing a volatility analysis on the wind and solar output curves on similar weather days; the fluctuation range calculation formula is as follows:

[0062] ΔP T =P t+T -P t (1)

[0063] Where: ΔP T is the output fluctuation value before and after the interval T; T is the interval period; P t The wind and solar power output during period t;

[0064] The output fluctuation data of wind and solar power in adjacent time periods are collected at intervals of T = 1h and saved in the wind and solar power output fluctuation set θ:

[0065]

[0066] Where: k t→t+1,up and k t→t+1,dn They are the upper and lower limits of wind and solar power fluctuation in adjacent periods; ΔP t,q is the fluctuation of wind and solar power output from t to t+1 on similar days q; S similar is a collection of similar days;

[0067] S12: Establishing a wind and solar power output prediction error model

[0068] Based on the historical forecast data and historical actual operation data of wind and solar power output, a wind and solar power output forecast error model is established as follows:

[0069] P t =P t,pre +ξ t (3)

[0070] Where: P t,pre is the predicted value of wind and solar power output in period t; t is the wind and solar power output prediction error value in period t;

[0071] Fitting modeling of wind and solar power output prediction error is performed based on prediction box and Gaussian mixture model;

[0072] S13: Time series scene sampling based on sampling probability interval correction

[0073] First, the Latin hypercube sampling (LHS) method is used to extract the initial prediction error values ​​of all time series scenarios, and the prediction values ​​are superimposed to obtain the wind and solar output power in the first period of all time series scenarios; then, for each time series scenario, the sampling interval of the prediction error is corrected by considering the fluctuation amplitude of wind and solar output in adjacent periods one by one, and the error is randomly sampled within the corrected sampling interval; the "LHS to interval random sampling" method is used to effectively simulate the fluctuation characteristics of wind and solar output at adjacent time points to generate time series scenarios;

[0074] (1) Step 1:

[0075] Suppose that N time-series wind and solar power output scenarios are generated. Based on the cumulative distribution of wind and solar power output power prediction errors in the first period, N prediction error values ​​are randomly extracted through LHS, and the prediction values ​​are superimposed to obtain the power sampling values ​​of the first period of N wind and solar power output scenarios.

[0076] (2) Step 2:

[0077] For each time series scenario S i , i = 1, 2, ..., N for prediction error sampling; the output power values ​​in adjacent time periods must satisfy:

[0078]

[0079] Where: and They are respectively the time series scenes S i The wind and solar power output sampling values ​​at time t and time t+1;

[0080] Before sampling the prediction error, the sampling interval needs to be corrected. When the wind and solar power output sampling value of period t is known, the prediction error sampling range after correction in period t+1 is obtained according to equations (3) and (4) and the wind and solar power output prediction value in period t+1:

[0081]

[0082] Where: The time series scenario S is in the t+1 period i The sample value of wind and solar power output prediction error; For the time series scene S i The difference between the wind and solar power output sampling value obtained in period t and the wind and solar power output prediction value in period t+1;

[0083] Will The sampling range is brought into the prediction error cumulative distribution function F of the t+1 period, and the corresponding probability interval [b 1 , b 2 ],in:

[0084]

[0085] Where: b 1 and b 2 are the lower and upper limits of the probability interval respectively;

[0086] In the probability interval [b 1 , b 2 ], and obtain the wind and solar power output prediction error in the t+1 period, which is superimposed with the predicted value to obtain the time series scene S i The wind and solar power output sampling value considering the historical wind and solar power output fluctuation in the t+1 period;

[0087] (3) Step 3:

[0088] Repeat step 2 until the wind and solar output sampling is completed for any period of all time series scenes;

[0089] S14: Wind and solar output timing reduction scenario

[0090] S141: BTM Model

[0091] The FCM clustering algorithm is improved by using the Biterm Topic Model (BTM) to enhance the pattern extraction, dynamic scene recognition and extreme event detection capabilities of time series data. The co-occurrence relationship of feature pairs of wind and solar power time series scenes is analyzed by BTM, and the topic probability distribution of each time window is extracted to reveal the typical output mode and generate different scene themes of wind and solar power output.

[0092] Let γ and δ be Dirichlet prior parameters, establish the BTM model, T is the number of manually set topic words, α is the topic word distribution on the text corpus, ψ is the word item distribution under the topic word, C is the Dirichlet distribution of the topic word α, E is the word pair set; Q α , Q is the subject word of the sample;

[0093] The specific steps are as follows:

[0094] (1) For each topic C: Calculate the word distribution φ for a specific topic C ~Dir(δ); Calculate the topic distribution of the entire corpus α~Dir(γ);

[0095] (2) For each word pair E in the word pair set e: calculate the topic distribution C~Multi(α); describe the two words Q α , Q~Multi(φ C ); the joint probability of a word corresponding to e is expressed as:

[0096]

[0097] The probability calculation formula for the entire corpus is expressed as:

[0098]

[0099] That is, the scenario generation probability of the entire time series data set is generated; BTM uses vocabulary patterns to express the semantics of the topic, and vocabulary pairs can better reflect the content of the topic and improve the understanding of the topic;

[0100] S142: FCM clustering algorithm

[0101] Expand the connection between the target and the category to any value in the closed interval [0, 1]; determine the degree to which the target is more inclined to belong to a certain category based on its membership; let U be the domain:

[0102]

[0103] A specific membership value is used to describe the degree to which each data object belongs to a certain category, but it is not strictly divided into a certain category;

[0104] The i-th data sample u in the data set U i The membership degree γ belonging to class j ij It is expressed by the following numerical relationship:

[0105]

[0106] Membership value γ ij The calculation is as follows:

[0107]

[0108] S ij =||u i -w j || represents the Euclidean distance from the i-th data sample to the j-th cluster center; the process of the FCM algorithm is as follows:

[0109] 1) Set the number of clusters z, the number of iterations t = 0, the maximum number of iterations K, the termination error σ, and initialize the cluster center W 0 ;

[0110] 2) Update the membership matrix O using formula (10) t ;

[0111] 3) Use equations (9) and (10) to update the next cluster center W t+1 ;

[0112] 4) If ||O is satisfied t+1 -O t ||<σ, then the algorithm ends; otherwise, let t=t+1 and return to step 2);

[0113] S143: BTM-improved FCM clustering algorithm

[0114] Use BTM to improve FCM clustering to analyze the wind and solar power processing timing scenarios and reduce the wind and solar power output timing scenarios. The specific steps are as follows:

[0115] Extract potential scenarios through BTM: First, the scenario themes generated by BTM are input into FCM as features to improve FCM's ability to predict wind and solar power output in microgrids. At this time, the wind and solar power output features of each time window are mapped to multiple theme categories, and the main output mode in the time period is identified.

[0116] Feature conversion and clustering preparation: The topic distribution and other key features extracted by BTM are combined into a feature vector and input into FCM. At this time, the wind and solar power output characteristics of each time window are mapped to multiple topic categories, and the main output mode in this time period is identified.

[0117] FCM clustering: FCM is used to perform fuzzy clustering on time windows. Each window is divided into different cluster centers according to the membership value, representing different output scenarios. This process automatically groups the data into several representative scenarios.

[0118] Scenario reduction: After FCM clustering is completed, each cluster center represents a typical output scenario;

[0119] Identification and retention of extreme scenarios and typical scenarios: For the identified representative scenarios, retain the extreme scenarios that have a great impact on the system and common typical scenarios.

[0120] The beneficial effects of the present invention are:

[0121] (1) The present invention can effectively reduce the average error of renewable energy output, smooth out output fluctuations, and make output prediction closer to the actual situation by performing probability interval correction on the sampling of renewable energy output scenes;

[0122] (2) The present invention innovatively combines BTM (Biterm Topic Model) and FCM (Fuzzy Clustering Algorithm) to generate and optimize new energy output scenarios, while reducing redundant scenarios and retaining extreme scenarios, which can effectively improve the diversity of the scenario library and the reliability of the scheduling scheme, enable the system to cope with extreme weather conditions, and improve the adaptability of the scheduling system;

[0123] (3) The day-ahead and intra-day dual-layer rolling optimization scheduling method proposed in the present invention can effectively cope with the real-time fluctuations of renewable energy output and load by dynamically adjusting the day-ahead scheduling plan on different time scales, thus ensuring the stable operation of the microgrid in a complex environment.

[0124] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0125] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0126] Figure 1 Modeling process for wind and solar output forecast error fitting;

[0127] Figure 2 Generate processes for wind and solar output scenarios;

[0128] Figure 3 It is the BTM model;

[0129] Figure 4 Flowchart of the scenario reduction method for optimizing the FCM clustering algorithm for BTM;

[0130] Figure 5 Schematic diagram for multi-time scale optimization scheduling;

[0131] Figure 6 This is the workflow diagram of the multi-time scale optimization scheduling method for microgrids based on the corrected error sampling interval. DETAILED DESCRIPTION

[0132] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0133] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0134] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "back" and the like indicate directions or positional relationships, they are based on the directions or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0135] 1 Generation of typical renewable energy output timing scenarios based on error sampling probability interval correction

[0136] The uncertainty of renewable energy output is smoothed by using the method of generating typical wind and solar time series scenarios. First, the fluctuation range of renewable energy output is analyzed based on historical data, and then a new energy prediction error model is established based on the prediction box and Gaussian mixture model (GMM). Then, the new energy time series prediction error is sampled and the sampling probability interval is corrected. The sampled error is superimposed on the predicted value to obtain a set of renewable energy output time series scenarios. Finally, the FCM method optimized by BTM is used to reduce the generated scenarios, and representative scenarios of wind and photovoltaic power generation and their corresponding probabilities are obtained, providing scenario support for microgrid optimization scheduling.

[0137] 1.1 Determination of the fluctuation range of wind and solar power output

[0138] Based on the historical weather information, we collected the actual wind and solar output data on similar weather days in history, conducted a volatility analysis on the wind and solar output curves on similar weather days, and obtained the fluctuation range of wind and solar output data in each adjacent period on similar weather days. The fluctuation range calculation formula is as follows:

[0139] ΔP T =P t+T -P t (1)

[0140] Where: ΔP T is the output fluctuation value before and after the interval T; T is the interval period; P t Contribute to the wind and solar power during period t.

[0141] The output fluctuation data of wind and solar power in adjacent time periods are collected at intervals of T = 1h and saved in the wind and solar power output fluctuation set θ:

[0142]

[0143] Where: k t→t+1,up and kt→t+1,dn They are the upper and lower limits of wind and solar power fluctuation in adjacent periods; ΔP t,q is the fluctuation of wind and solar power output from t to t+1 on similar days q; S similar A collection of similar days.

[0144] 1.2 Establishment of wind and solar power output prediction error model

[0145] Based on the historical forecast data and historical actual operation data of wind and solar power output, a wind and solar power output forecast error model is established as follows:

[0146] P t =P t,pre +ξ t (3)

[0147] Where: P t,pre is the predicted value of wind and solar power output in period t; t is the wind and solar power output prediction error value in period t.

[0148] Based on the prediction box and Gaussian mixture model, the wind and solar power output prediction error is fitted and modeled. The specific process is as follows: Figure 1 shown.

[0149] 1.3 Time series scene sampling based on sampling probability interval correction

[0150] First, the Latin hypercube sampling (LHS) method is used to extract the initial prediction error values ​​of all time series scenarios, and the prediction values ​​are superimposed to obtain the wind and solar output power in the first period of all time series scenarios. Then, for each time series scenario, the sampling interval of the prediction error is corrected by considering the fluctuation amplitude of wind and solar output in adjacent periods one by one, and the error is randomly sampled within the corrected sampling interval. The "LHS to interval random sampling" method is used to effectively simulate the fluctuation characteristics of wind and solar output at adjacent time points and generate time series scenarios.

[0151] (1) Step 1:

[0152] Suppose that N time-series wind and solar power output scenarios are generated. Based on the cumulative distribution of wind and solar power output power prediction error in the first period, N prediction error values ​​are randomly extracted through LHS, and the prediction values ​​are superimposed to obtain the power sampling values ​​of the first period of N wind and solar power output scenarios.

[0153] (2) Step 2:

[0154] For each time series scenario S i (i=1,2,…,N) to sample the prediction error. The output power values ​​of adjacent time periods must meet the following requirements:

[0155]

[0156] Where: and They are respectively the time series scenes S i The wind and solar power output sampling values ​​in period t and period t+1.

[0157] Before sampling the prediction error, the sampling interval needs to be corrected. When the wind and solar power output sampling value of period t is known, according to equations (3) and (4) and the wind and solar power output prediction value of period t+1, the prediction error sampling range after correction in period t+1 is obtained:

[0158]

[0159] Where: The time series scenario S is in the t+1 period i The sample value of wind and solar power output prediction error; For the time series scene S i The difference between the wind and solar power output sampling value obtained in period t and the wind and solar power output prediction value in period t+1.

[0160] Will The sampling range is brought into the prediction error cumulative distribution function F of the t+1 period, and the corresponding probability interval [b 1 , b 2 ],in:

[0161]

[0162] Where: b 1 and b 2 are the lower and upper bounds of the probability interval respectively.

[0163] In the probability interval [b 1 , b 2 ], and obtain the wind and solar power output prediction error in the t+1 period, which is superimposed with the predicted value to obtain the time series scene S i The wind and solar power output sampling value considering the historical wind and solar power output volatility in the t+1 period.

[0164] (3) Step 3:

[0165] Repeat step 2 until the wind and solar output sampling is completed for any period of all time series scenarios.

[0166] The process of generating wind and solar power output timing scenarios based on sampling probability interval correction is as follows: Figure 2 shown.

[0167] 1.4 Reduction of wind and solar power output timing scenarios

[0168] 1.4.1BTM Model

[0169] This paper uses the Biterm Topic Model (BTM) to improve the FCM clustering algorithm to enhance the pattern extraction, dynamic scene recognition and extreme event detection capabilities of time series data. The BTM is used to analyze the co-occurrence relationship of feature pairs of wind and solar power time series scenes, extract the topic probability distribution of each time window, reveal the typical output mode, and generate different scene themes of wind and solar power output.

[0170] BTM was originally designed for short texts, but over time, this method has been gradually used for long texts and has achieved the same results as classic topic models such as Latent Dirichlet Allocation (LDA). This method is a new method for extracting text topics using the entire corpus, treating the text set as a topic set, and extracting each vocabulary pair from different topics. The present invention regards the set of wind and solar power output time series scenes as a topic set, generates vocabulary pairs for processing feature pairs at adjacent moments under the topic distribution, and helps to better identify the output scenes.

[0171] The possibility of pairing depends on whether the two words come from the same topic. Assuming that γ and δ are Dirichlet prior parameters, the established BTM model is as follows Figure 3 As shown in the figure: T is the number of manually set topic words, α is the distribution of topic words on the text corpus, ψ is the distribution of terms under the topic word, C is the Dirichlet distribution of the topic word α, and E is the word pair set. α , Q is the keyword of the sample.

[0172] The specific steps are as follows:

[0173] (1) For each topic C: 1) Calculate the word distribution φ for a specific topic C ~Dir(δ). 2) Calculate the topic distribution of the entire corpus α~Dir(γ).

[0174] (2) For each word pair E in the word pair set e: 1) Calculate the topic distribution C~Multi(α). 2) Describe the two words Q α , Q~Multi(φ C ). The joint probability of a word corresponding to e is expressed as:

[0175]

[0176] Therefore, the probability calculation formula for the entire corpus is expressed as:

[0177]

[0178] That is, the scenario generation probability of the entire time series data set is generated. BTM uses vocabulary patterns to express the semantics of the topic. Vocabulary pairs can better reflect the content of the topic and improve the understanding of the topic. Compared with the traditional LDA method, the BTM method uses the alpha value of the topic to assign topics, thereby avoiding the excessive dependence of the LDA algorithm on the topic. The relationship between words is retained, thereby achieving the synchronization of multiple topics in the text. The advantage of BTM is that it can effectively identify potential patterns in short texts or sparse data, and is particularly suitable for processing data scenarios with less contextual information. By analyzing the co-occurrence relationship between word pairs, BTM can mine the topic structure without relying on the context of long texts, so that it can also extract dynamic patterns, identify extreme events and enhance data feature dimensions when processing complex time series data (such as wind and solar output).

[0179] 1.4.2 FCM Clustering Algorithm

[0180] FCM is a clustering method based on segmentation. Different from the previous strict "either-or" "hard segmentation", it expands the connection between the target and the category to any value in the closed interval [0, 1]. Therefore, according to its membership, the degree to which the target is more inclined to belong to a certain category can be determined. Let U be the domain:

[0181]

[0182] A specific membership value is used to describe the degree to which each data object belongs to a certain category, but it is not strictly divided into a certain category.

[0183] The i-th data sample u in the data set U i The membership degree γ belonging to class j ij It is expressed by the following numerical relationship:

[0184]

[0185] Membership value γ ij The calculation is as follows:

[0186]

[0187] S ij =||u i -w j || represents the Euclidean distance from the i-th data sample to the j-th cluster center. The process of the FCM algorithm can be summarized as follows:

[0188] 1) Set the number of clusters z, the number of iterations t = 0, the maximum number of iterations K, the termination error σ, and initialize the cluster center W 0 .

[0189] 2) Update the membership matrix O using formula (10)t .

[0190] 3) Use equations (9) and (10) to update the next cluster center W t+1 .

[0191] 4) If ||O is satisfied t+1 -O t ||<σ, then end the algorithm; otherwise, let t=t+1 and return to step 2).

[0192] 1.4.3BTM Improved FCM Clustering Algorithm

[0193] Use BTM to improve FCM clustering to analyze the wind and solar power processing timing scenarios and achieve wind and solar power output timing scenario reduction. The flow chart is as follows: Figure 4 shown.

[0194] The specific steps are as follows:

[0195] (1) Extracting potential scenarios through BTM:

[0196] First, the scenario themes generated by BTM are input into FCM as features to improve FCM's ability to predict wind and solar power output in microgrids. At this point, the wind and solar power output features of each time window are mapped to multiple theme categories, and the main output mode in that time period is identified.

[0197] (2) Feature conversion and clustering preparation:

[0198] The topic distribution and other key features extracted by BTM are combined into a feature vector and input into FCM. At this time, the wind and solar power output characteristics of each time window are mapped to multiple topic categories, and the main output mode in this time period is identified.

[0199] (3) Fuzzy clustering (FCM clustering):

[0200] Then, FCM is used to perform fuzzy clustering on the time windows, and each window is divided into different cluster centers (representing different output scenarios) according to the membership value. This process automatically groups the data into several representative scenarios.

[0201] (4) Scene reduction:

[0202] After FCM clustering is completed, each cluster center represents a typical output scenario. In this step, by screening these cluster centers, a small number of representative typical high output, low output, fluctuation mode and other scene themes are selected, and redundant or highly similar scenes are ignored. The distance and membership threshold between adjacent cluster centers are used to judge the similarity of the scenes and reduce repeated or redundant scenes.

[0203] (5) Identification and retention of extreme and typical scenarios:

[0204] For the identified representative scenarios, extreme scenarios and common typical scenarios with greater impact on the system are retained to ensure that the diversity and integrity of the scenario library are retained even if the number of scenarios is reduced.

[0205] 2-day-ahead scheduling-intraday double-layer rolling optimization model

[0206] The present invention proposes a multi-time scale optimization scheduling model that considers the day-ahead scheduling plan and the intra-day double-layer rolling optimization. The cooling and heating loads and the electric loads in the intra-day rolling optimization have a coupling relationship and a time sequence. The schematic diagram of multi-time scale optimization scheduling is shown in Figure 5 shown.

[0207] 2.1 Day-ahead scheduling model

[0208] 2.1.1 Objective Function

[0209] The day-ahead scheduling model is a typical mixed integer linear programming problem, and its goal is to minimize the operating cost:

[0210]

[0211] Where: F day-ahead Plan operating costs for day-ahead dispatch; is the probability of the i-th wind and solar power output scenario; For scene S i The interaction power cost between the lower and upper grids; For scene S i Lower fuel costs; For scene S i The aging cost of battery charging and discharging.

[0212] 1) Grid interaction power cost

[0213]

[0214] Where: T represents the day-ahead scheduling period. is the unit price of electricity purchased during period T; For scene S i The next T period is to purchase electricity from the grid.

[0215] 2) Fuel costs

[0216]

[0217] Where: The unit price of natural gas purchased during period T; and The scene S iThe natural gas consumption of micro gas turbines, gas boilers and fuel cells during the next T period; H FUEL It is the low calorific value of natural gas.

[0218] 3) Battery charging and discharging aging cost

[0219]

[0220] Where: is the battery unit aging cost during period T; and For scene S i The battery charging and discharging state variables during the next T period.

[0221] 2.1.2 Constraints

[0222] The system considers equipment operation constraints, power balance constraints, and power interaction constraints.

[0223] 1) Equipment operation constraints

[0224]

[0225] Where: P e T is the output power of device e during period T; is the start / stop state variable of device e during period T; P e max and P e min is the maximum and minimum output power of device e; P e up and P e dn The upper and lower limits of the climbing power of device e.

[0226] 2) Power balance constraints

[0227]

[0228] Where: The following are variables in the T period. and The scene S i Wind power and photovoltaic output power under and are the output electrical power of the micro gas turbine and the fuel cell respectively; and They are the charging and discharging power of BT respectively; and are the electric power consumed by ice storage air conditioner, electric boiler, CCS and P2G respectively; Output cooling power for absorption chiller; and They are electricity, cooling and heating loads respectively; and are the output thermal power of waste heat boiler, electric boiler and gas boiler respectively; and are the heat storage and release power of the heat storage tank respectively; It is the heat power absorbed by the absorption refrigerator.

[0229] 3) Power interaction constraints

[0230]

[0231] Where: and are the maximum and minimum values ​​of the system's power purchase respectively.

[0232] 2.1.3 Intraday Double-Layer Rolling Optimization Model

[0233] Intraday rolling optimization is divided into a double-layer structure of hot and cold rolling and electric rolling. Hot and cold rolling controls hot and cold power, and electric rolling controls electric power, so as to smooth out the actual power fluctuation. The hot and cold rolling control time domain X is 1 hour, and the scheduling time domain is 2 hours; the electric rolling control time domain Y is 5 minutes, and the scheduling time domain is 1 hour. In hot and cold rolling scheduling, at t 0 time, predict t 0 +X to t 0 +X+2 period of cooling and heating load, so as to adjust t 0 +X to t 0 +2X period unit equipment output plan. At the same time, in the power rolling dispatch, at t 0 Time prediction t 0 +Y to t 0 +Y+1 period of electrical load, adjust t 0 +Y to t 0 +2Y unit power generation plan.

[0234] (1) Hot and cold rolling optimization scheduling model

[0235] Based on the equipment operation status and the cooling and heating power dispatch plan on the previous day, the output plan of each unit is adjusted according to the change of cooling and heating load in the t period. The objective function is:

[0236]

[0237] Where: t represents the scheduling period within the day. day,up Optimize dispatch costs for intraday hot and cold rolling; is the hot and cold rolling fuel cost for period t; is the power change cost of the electric boiler during period t; is the power change cost of ice storage air conditioning during period t; is the cost of heat storage and release power change of the heat storage tank during period t.

[0238]

[0239] Where: and are the gas consumption adjustment of the micro gas turbine and gas boiler during period t; μ MT , μ GB , μ EB , μ ISAC and μ h The unit penalty costs for micro gas turbine, gas boiler, electric boiler, ice storage air conditioner and thermal storage tank adjustment are respectively; and is the cooling and heating adjustment power of the micro gas turbine, electric boiler, gas boiler and ice storage air conditioner during period t; and They are respectively the heat storage and release power adjustment values ​​of the heat storage tank in period t.

[0240] Constraints:

[0241] 1) Power balance constraints

[0242]

[0243] 2) Power adjustment constraints

[0244]

[0245] In the cold and hot rolling dispatch, during the off-peak electricity price period, when the energy supply is greater than the demand, the output of the gas boiler, micro gas turbine, electric boiler, and ice storage air conditioner is adjusted in sequence; when the energy supply is less than the demand, the output of the ice storage air conditioner, gas boiler, electric boiler, and micro gas turbine is adjusted in sequence. During the non-off-peak electricity price period, when the energy supply is greater than the demand, the output of the gas boiler, ice storage air conditioner, electric boiler, and micro gas turbine is adjusted in sequence; when the energy supply is less than the demand, the output of the gas boiler, micro gas turbine, electric boiler, and ice storage air conditioner is adjusted in sequence.

[0246] (2) Electric rolling optimization scheduling model

[0247] Based on the day-ahead battery charge and discharge status, and with the fluctuation of wind and solar power output and the changes in the power and electrical load of the cold and hot rolling dispatching equipment, the day-ahead dispatching plan is adjusted. The objective function is:

[0248]

[0249] Where: F day,low Optimize dispatching costs for intraday electricity rolling; is the cost of purchasing electricity from the power grid during period t; is the rolling fuel cost of electricity during period t; is the cost of battery charging and discharging power change; β is the SOC penalty factor; is the SOC value of the battery during the T period in the day-ahead dispatch; It is the battery SOC value during period t in intraday scheduling.

[0250]

[0251] Where: is the adjustment value of purchased power in period t; μ GRID is the unit penalty cost of purchasing power adjustment; μ FC unit penalty costs adjusted for fuel cells; is the gas consumption adjustment of the fuel cell during period t; and are the adjusted electric power of the micro gas turbine and the fuel cell during period t; μ p Unit penalty costs adjusted for battery charging and discharging; and They are the battery charging and discharging power adjustment values ​​during period t respectively.

[0252] In the electric rolling dispatch, the power balance constraint and power adjustment constraint should also be met, which is similar to the cold and hot rolling constraints and will not be repeated here. In addition, the energy storage constraint should also be met:

[0253]

[0254] Where: η BT,ch and η BT,dis are the charge and discharge efficiency of the battery respectively; and They are the upper and lower limits of the battery SOC value respectively; is the change of SOC in period t.

[0255] Figure 6 This is the workflow diagram of the multi-time scale optimization scheduling method for microgrids based on the corrected error sampling interval.

[0256] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A double-layer rolling precise dispatching method based on new energy output error sampling and calibration, characterized by: The method comprises the following steps: S1: Generate a typical renewable energy output timing scenario based on error sampling probability interval correction Firstly, the fluctuation range of renewable energy output is analyzed based on historical data, and then a renewable energy prediction error model is established based on the prediction box and Gaussian mixture model (GMM). Then, the prediction error of renewable energy time series is sampled and the sampling probability interval is corrected. The sampled error is superimposed on the predicted value to obtain a set of renewable energy output time series scenarios. Finally, the generated scenarios are reduced through the FCM method optimized by BTM, and representative scenarios of wind and photovoltaic power generation and their corresponding probabilities are obtained, providing scenario support for microgrid optimization scheduling. S2: Establishing a dual-layer rolling optimization model for day-ahead scheduling and intraday scheduling S21: Establishing the day-ahead scheduling model S211: Establishing the objective function The day-ahead scheduling model is a typical mixed integer linear programming problem, and its goal is to minimize the operating cost: Where: F day-ahead Plan operating costs for day-ahead dispatch; is the probability of the i-th wind and solar power output scenario; For scene S i The interaction power cost between the lower and upper grids; For scene S i Lower fuel costs; For scene S i The aging cost of battery charging and discharging; 1) Grid interaction power cost Where: T represents the day-ahead dispatch period; is the unit price of electricity purchased during period T; For scene S i Purchase power from the power grid during the next T period; 2) Fuel costs Where: The unit price of natural gas purchased during period T; and The scene S i The natural gas consumption of micro gas turbines, gas boilers and fuel cells during the next T period; H FUEL It is the low calorific value of natural gas; 3) Battery charging and discharging aging cost Where: is the battery unit aging cost during period T; and For scene S i Battery charge and discharge state variables during the next T period; S212: Establishing Constraints The system considers equipment operation constraints, power balance constraints, and power interaction constraints; 1) Equipment operation constraints Where: is the output power of device e during period T; is the start / stop state variable of device e during period T; and is the maximum and minimum output power of device e; and The upper and lower limits of the climbing power of device e; 2) Power balance constraints In the formula: The following are all variables in the T period; and The scene S i Wind power and photovoltaic output power under and are the output electrical power of the micro gas turbine and the fuel cell respectively; and They are the charging and discharging power of BT respectively; and are the electric power consumed by ice storage air conditioner, electric boiler, CCS and P2G respectively; Output cooling power for absorption chiller; and They are electricity, cooling and heating loads respectively; and are the output thermal power of waste heat boiler, electric boiler and gas boiler respectively; and are the heat storage and release power of the heat storage tank respectively; is the heat power absorbed by the absorption refrigerator; 3) Power interaction constraints Where: and are the maximum and minimum values ​​of the system's power purchase respectively; S213: Establishing a two-layer rolling optimization model within a day The intraday rolling optimization is divided into a double-layer structure of cold and hot rolling and electric rolling. The cold and hot rolling controls the cold and hot power, and the electric rolling controls the electric power, so as to smooth the actual power fluctuation; the cold and hot rolling control time domain X is 1h, and the scheduling time domain is 2h; the electric rolling control time domain Y is 5min, and the scheduling time domain is 1h; in the cold and hot rolling scheduling, at time t0, the cold and hot loads from t0+X to t0+X+2 are predicted, so as to adjust the unit equipment output plan from t0+X to t0+2X; at the same time, in the electric rolling scheduling, the electric load from t0+Y to t0+Y+1 is predicted at time t0, and the unit power generation plan from t0+Y to t0+2Y is adjusted; (1) Hot and cold rolling optimization scheduling model Based on the equipment operation status and the cooling and heating power dispatch plan on the previous day, the output plan of each unit is adjusted according to the change of cooling and heating load in the t period; the objective function is: Where: t represents the scheduling period within the day; F day,up Optimize dispatch costs for intraday hot and cold rolling; is the hot and cold rolling fuel cost for period t; is the power change cost of the electric boiler during period t; is the power change cost of ice storage air conditioner during period t; is the cost of heat storage and release power change of the heat storage tank during period t; Where: and are the gas consumption adjustment of the micro gas turbine and gas boiler during period t; μ MT , μ GB , μ EB , μ ISAC and μ h The unit penalty costs for micro gas turbine, gas boiler, electric boiler, ice storage air conditioner and thermal storage tank adjustment are respectively; and is the cooling and heating adjustment power of the micro gas turbine, electric boiler, gas boiler and ice storage air conditioner during period t; and They are respectively the heat storage and release power adjustment values ​​of the heat storage tank in period t; Constraints: 1) Power balance constraints 2) Power adjustment constraints In the cold and hot rolling dispatch, during the off-peak electricity price period, when the energy supply is greater than the demand, the output of the gas boiler, micro gas turbine, electric boiler, and ice storage air conditioner is adjusted in sequence; when the energy supply is less than the demand, the output of the ice storage air conditioner, gas boiler, electric boiler, and micro gas turbine is adjusted in sequence; during the non-off-peak electricity price period, when the energy supply is greater than the demand, the output of the gas boiler, ice storage air conditioner, electric boiler, and micro gas turbine is adjusted in sequence; when the energy supply is less than the demand, the output of the gas boiler, micro gas turbine, electric boiler, and ice storage air conditioner is adjusted in sequence; (2) Establishing an electric rolling optimization scheduling model Based on the day-ahead battery charge and discharge status, and with the fluctuation of wind and solar power output and the changes in the power and electric load of the cold and hot rolling dispatching equipment, the day-ahead dispatching plan is adjusted; the objective function is: Where: F day,low Optimize dispatching costs for intraday electricity rolling; is the cost of purchasing electricity from the power grid during period t; is the rolling fuel cost of electricity during period t; is the cost of battery charging and discharging power change; β is the SOC penalty factor; is the SOC value of the battery during the T period in the day-ahead dispatch; is the battery SOC value during period t in intraday scheduling; Where: μ is the power adjustment value of electricity purchased during period t; GRID is the unit penalty cost of purchasing power adjustment; μ FC unit penalty costs adjusted for fuel cells; is the gas consumption adjustment of the fuel cell during period t; and are the adjusted electric power of the micro gas turbine and the fuel cell during period t; μ p Unit penalty costs adjusted for battery charging and discharging; and They are the battery charging and discharging power adjustment values ​​for period t respectively; In electric rolling dispatch, power balance constraints and power adjustment constraints are met, which are similar to hot and cold rolling constraints; in addition, energy storage constraints are also met: Where: η BT,ch and η BT,dis are the charge and discharge efficiency of the battery respectively; and They are the upper and lower limits of the battery SOC value respectively; is the change of SOC in period t.

2. The double-layer rolling precise scheduling method based on new energy output error sampling and calibration according to claim 1 is characterized in that: The S1 comprises the following steps: S11: Determine the fluctuation range of wind and solar power output The actual wind and solar output data of similar weather days in history are obtained by collecting historical weather information, and the fluctuation range of wind and solar output data in each adjacent period of similar weather days is obtained by performing a volatility analysis on the wind and solar output curves on similar weather days; the fluctuation range calculation formula is as follows: ΔP T =P t+T -P t (1) Where: ΔP T is the output fluctuation value before and after the interval T; T is the interval period; P t The wind and solar power output during period t; The output fluctuation data of wind and solar power in adjacent time periods are collected at intervals of T = 1h and saved in the wind and solar power output fluctuation set θ: Where: k t→t+1,up and k t→t+1,dn They are the upper and lower limits of wind and solar power fluctuation in adjacent periods; ΔP t,q is the fluctuation of wind and solar power output from t to t+1 on similar days q; S similar is a collection of similar days; S12: Establishing a wind and solar power output prediction error model Based on the historical forecast data and historical actual operation data of wind and solar power output, a wind and solar power output forecast error model is established as follows: P t =P t,pre +ξ t (3) Where: P t,pre is the predicted value of wind and solar power output in period t; t is the wind and solar power output prediction error value in period t; Fitting modeling of wind and solar power output prediction error is performed based on prediction box and Gaussian mixture model; S13: Time series scene sampling based on sampling probability interval correction Firstly, the Latin hypercube sampling (LHS) method is used to extract the initial prediction error values ​​of all time series scenarios, and the prediction values ​​are superimposed to obtain the wind and solar power output power in the first period of all time series scenarios; then, for each time series scenario, the sampling interval of the prediction error is corrected by considering the fluctuation amplitude of wind and solar power output in adjacent periods one by one, and the error is randomly sampled within the corrected sampling interval; the "LHS to interval random sampling" method is used to effectively simulate the fluctuation characteristics of wind and solar power output at adjacent time points to generate time series scenarios; (1) Step 1: Suppose that N time-series wind and solar power output scenarios are generated. Based on the cumulative distribution of wind and solar power output power prediction errors in the first period, N prediction error values ​​are randomly extracted through LHS, and the prediction values ​​are superimposed to obtain the power sampling values ​​of the first period of N wind and solar power output scenarios. (2) Step 2: For each time series scenario S i , i = 1, 2, ..., N for prediction error sampling; the output power values ​​in adjacent time periods must satisfy: P Si,t+1,sample -P Si,t,sample ∈[k t→t+1,dn ,k t→t+1,up ] (4) Where: P Si,t,sample and P Si,t+1,sample They are respectively the time series scenes S i The wind and solar power output sampling values ​​at time t and time t+1; Before sampling the prediction error, the sampling interval needs to be corrected. When the wind and solar power output sampling value of period t is known, the prediction error sampling range after correction in period t+1 is obtained according to equations (3) and (4) and the wind and solar power output prediction value in period t+1: Where: P Si,t+1,err The time series scenario S is in the t+1 period i The sample value of the wind and solar power output prediction error; P Si,Δ,t→t+1 For the time series scenario S i The difference between the wind and solar power output sampling value obtained in period t and the wind and solar power output prediction value in period t+1; P Si,t+1,err Substitute the prediction error cumulative distribution function F of the sampling range into the t+1 period, and calculate the corresponding probability interval [b1, b2], where: Where: b1 and b2 are the lower and upper limits of the probability interval respectively; Random sampling is performed within the probability interval [b1, b2] to obtain the wind and solar power output prediction error for the period t+1. When superimposed with the predicted value, the time series scenario S can be obtained. i The wind and solar power output sampling value considering the historical wind and solar power output fluctuation in the t+1 period; (3) Step 3: Repeat step 2 until the wind and solar output sampling is completed for any period of all time series scenes; S14: Wind and solar output timing reduction scenario S141: BTM Model The FCM clustering algorithm is improved by using the Biterm Topic Model (BTM) to enhance the pattern extraction, dynamic scene recognition and extreme event detection capabilities of time series data. The co-occurrence relationship of feature pairs of wind and solar power time series scenes is analyzed by BTM, and the topic probability distribution of each time window is extracted to reveal the typical output mode and generate different scene themes of wind and solar power output. Let γ and δ be Dirichlet prior parameters, establish the BTM model, T is the number of manually set topic words, α is the topic word distribution on the text corpus, ψ is the word item distribution under the topic word, C is the Dirichlet distribution of the topic word α, E is the word pair set; Q α , Q is the subject word of the sample; The specific steps are as follows: (1) For each topic C: Calculate the word distribution φ for a specific topic C ~Dir(δ); Calculate the topic distribution of the entire corpus α~Dir(γ); (2) For each word pair E in the word pair set e: calculate the topic distribution C~Multi(α); describe the two words Q α , Q~Multi(φ C ); the joint probability of a word corresponding to e is expressed as: The probability calculation formula for the entire corpus is expressed as: That is, the scenario generation probability of the entire time series data set is generated; BTM uses vocabulary patterns to express the semantics of the topic, and vocabulary pairs can better reflect the content of the topic and improve the understanding of the topic; S142: FCM clustering algorithm Expand the connection between the target and the category to any value in the closed interval [0, 1]; determine the degree to which the target is more inclined to belong to a certain category based on its membership; let U be the domain: A specific membership value is used to describe the degree to which each data object belongs to a certain category, but it is not strictly divided into a certain category; The i-th data sample u in the data set U i The membership degree γ belonging to class j ij It is expressed by the following numerical relationship: Membership value γ ij The calculation is as follows: S ij =||u i -w j || represents the Euclidean distance from the i-th data sample to the j-th cluster center; the process of the FCM algorithm is as follows: 1) Set the number of clusters z, the number of iterations t = 0, the maximum number of iterations K, the termination error σ, and initialize the cluster center W 0 ; 2) Update the membership matrix O using formula (10) t ; 3) Use equations (9) and (10) to update the next cluster center W t+1 ; 4) If ||O is satisfied t+1 -O t ||<σ, then the algorithm ends; otherwise, let t=t+1 and return to step 2); S143: BTM-improved FCM clustering algorithm Use BTM to improve FCM clustering to analyze the wind and solar power processing timing scenarios and reduce the wind and solar power output timing scenarios. The specific steps are as follows: Extract potential scenarios through BTM: First, the scenario themes generated by BTM are input into FCM as features to improve FCM's ability to predict wind and solar power output in microgrids. At this time, the wind and solar power output features of each time window are mapped to multiple theme categories, and the main output mode in the time period is identified. Feature conversion and clustering preparation: The topic distribution and other key features extracted by BTM are combined into a feature vector and input into FCM. At this time, the wind and solar power output characteristics of each time window are mapped to multiple topic categories, and the main output mode in this time period is identified. FCM clustering: FCM is used to perform fuzzy clustering on time windows. Each window is divided into different cluster centers according to the membership value, representing different output scenarios. This process automatically groups the data into several representative scenarios. Scenario reduction: After FCM clustering is completed, each cluster center represents a typical output scenario; Identification and retention of extreme scenarios and typical scenarios: For the identified representative scenarios, retain the extreme scenarios and typical scenarios that have a great impact on the system.

Images