Active power control method oriented to Sagomean new energy base

Through multi-level modeling and predictive control methods, the power distribution of wind power and photovoltaic clusters is optimized, and the active power control problem of Shagohuang New Energy Base is solved, the control accuracy and new energy consumption capacity are improved, and the station is not working hard. Both control commands are over-issued.

CN118944204APending Publication Date: 2024-11-12NORTHEAST DIANLI UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410987356.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The active power control of Shagohuang New Energy Base lacks an adaptive control framework, lack of effective grouping strategies between stations, uneven distribution of output in the wind and light stations, and over-issuance problems of control instructions are prone to over-issuance. The weight setting of the optimization model depends on Inadequate experience, operational and explanatory.

Method used

Multi-level modeling and predictive control methods are adopted, including the base wind and light scheduling layer, cluster group allocation layer, in-group station optimization layer and station power tracking layer. Through predictive control and feedback correction models, the power distribution of wind power and photovoltaic clusters is optimized. , dynamically adjust the station weight to achieve refined power control.

Benefits of technology

The Shagohuang New Energy Base has improved the tracking ability of the dispatching instructions, improved the control accuracy, avoided the problems of uneven output of the station and over-issuance of control instructions, and enhanced the consumption and coordination and control capabilities of new energy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118944204A_ABST
    Figure CN118944204A_ABST
Patent Text Reader

Abstract

The invention discloses an active power control method oriented to a Sagomean new energy base, and belongs to the technical field of active power control. The active power control method comprises the following steps: optimizing power values of a future wind power cluster and a photovoltaic cluster; optimizing the power distribution of the wind and light stations, and obtaining the planned power value of each group by combining with a dynamic grouping strategy of the stations; optimizing the power value of the wind and light station in the future field group, and calculating the output value of the station; the wind and light station performs active power control according to the station output value; and feedback correction is carried out on multi-scale prediction of the prediction model, the corrected prediction value is used for guiding optimization scheduling of each layer, and active power control of the Saggohm new energy base at the next moment is carried out. According to the invention, through an active hierarchical predictive control method, optimal distribution of the active power of the Sagomean new energy base station is realized in a layer-by-layer refinement and feedback correction mode, and the control precision is higher; dynamic grouping is carried out by considering the power output characteristics of energy, and the strategy control effect is effectively improved in combination with the weight; and non-uniform output of the station can be effectively avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of active power control, and in particular relates to an active power control method for a Shagohuang new energy base. Background Art

[0002] At present, my country is stepping up the construction of large-scale wind power and photovoltaic bases in deserts, Gobi and wasteland (abbreviated as Shagohuang). Although the integration of wind and light into the power grid in a clustered manner can effectively smooth the power fluctuations of individual wind and light stations, making wind and light more like "conventional power sources". However, there are many wind and light stations in the Shagohuang new energy base, and the coordination and control problems are prominent, which can easily lead to uneven output of the stations, causing the base power control results to deviate from the dispatching instructions. Therefore, it is of great significance to carry out active power control research based on the Shagohuang new energy base to improve the tracking accuracy of the base output power to the dispatching instructions.

[0003] At present, the active power control of the Shagohuang New Energy Base has the following problems: ① There is a lack of a control framework that is compatible with the active power control of the "Shagohuang" energy base; ② In the Shagohuang new energy scenario, there is a lack of effective grouping strategy between stations; ③ The output distribution between wind and solar stations is uneven, and the abandonment rates of new energy between stations, between stations, and between stations vary greatly, and the problem of "over-issuance" of control instructions is prone to occur, that is, the control instructions are greater than the natural power value of the station; ④ The optimization model proposed by the previous control method relies on experience to set the weight value, and lacks operability and interpretability.

[0004] Therefore, a new technical solution is urgently needed in the existing technology to solve these problems. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an active power control method for the Shagohuang new energy base, which is used to improve the tracking ability of the Shagohuang new energy base for scheduling instructions and solve the above-mentioned technical problems existing in the active power control of the Shagohuang new energy base.

[0006] An active power control method for a Shagohuang new energy base includes the following steps, and the following steps are performed in sequence:

[0007] Step S1: Modeling of the base wind-solar dispatching layer, cluster sub-clustering allocation layer, intra-cluster station optimization layer, station power tracking layer, feedback correction model and prediction model are performed respectively. The base wind-solar dispatching layer receives the dispatching plan power value issued by the dispatching center, and adopts the prediction control method based on the 30-minute ultra-short-term power prediction information of the wind-solar cluster, with the goal of tracking the planned value of the dispatching center, to optimize the power value of the wind power cluster and photovoltaic cluster in the next 1 hour, and sends the optimization result as the planned power value of the wind power cluster and photovoltaic cluster to the cluster sub-clustering allocation layer;

[0008] Step S2: The cluster grouping allocation layer receives the planned power values ​​of the wind power cluster and photovoltaic cluster issued by the wind and solar dispatching layer of the base, adopts a distributed prediction control method, uses the 15-minute ultra-short-term power prediction information of the cluster and wind and solar stations to optimize the power allocation of wind and solar stations, and then combines the station dynamic grouping strategy to obtain the planned power value of each group, and sends it to the station optimization layer within the group;

[0009] Step S3: The station optimization layer within the cluster receives the planned power value of the cluster group issued by the cluster allocation layer, and optimizes the allocation based on the ultra-short-term power prediction result with a resolution of 5 minutes, with the goal of tracking the planned power of the cluster group. The rolling optimization cycle is 5 minutes, and the power value of the wind and solar stations within the cluster in the next 15 minutes is optimized, that is, the power of the stations within the cluster in the next three time sections is optimized each time, but only the optimization result of the first time section is issued as the planned power value of the station to the station power tracking layer;

[0010] Step S4: The station power tracking layer receives the station planned power value sent by the station optimization layer within the group, and calculates and obtains the station output value based on the ultra-short-term power prediction result with a resolution of 1 minute;

[0011] Step S5: The monitoring system of the wind and solar power station performs active power control according to the station output value obtained by the station power tracking layer, and records the actual control power and natural power, wherein the actual control power is used to construct the climbing constraint of each layer, and the natural power is sent to the feedback correction model;

[0012] Step S6: The prediction model is used for multi-scale prediction of natural power, and the input of the prediction model is multi-source data, including numerical weather forecast data and historical output data of the target station, and numerical weather forecast data and historical output data of other stations;

[0013] The feedback correction model calculates the prediction error based on the natural power sent by the monitoring system of the wind and solar power station and the predicted natural power sent by the prediction model, and predicts the power prediction error at the next moment, performs feedback correction on the multi-scale prediction of the prediction model, obtains the corrected prediction value and uses it to guide the optimization scheduling of each layer, and carry out active power control of the Shagohuang New Energy Base at the next moment.

[0014] The method for modeling the base wind-solar scheduling layer in step S1 is as follows:

[0015] Establish optimization model objective function minJ Base :

[0016]

[0017] Where: c is the number of cluster types, in this invention, c=2, representing wind power cluster and photovoltaic cluster respectively; α h,1 , αh,2 To optimize the adaptive weight; h Adaptive weights for clusters; The allocated power value for cluster h at the optimization moment; 30-minute ultra-short-term power prediction value for cluster optimization; is the actual power value of the cluster at the previous moment; T is the optimization time domain. The base layer optimizes the cluster power in the base for the next 1 hour each time, with an optimization resolution of 30 minutes, that is, T = 2, and as the k value changes, the time t is optimized forward with a step size of Δt until the optimization control in the time domain is completed, and finally the optimization result when k = 1 is issued;

[0018] The above model also considers the scheduling plan constraints, cluster output constraints and cluster ramp rate constraints during the rolling optimization process. The specific formula is as follows:

[0019]

[0020] Where: To optimize the dispatch plan power value issued by the dispatch center at the moment; is the minimum output of cluster h; Install capacity for the cluster; It is the maximum climbing rate of the cluster on a 30-min time scale.

[0021] After the optimization is completed, the optimization results of the base layer will be sent to the cluster distribution layer.

[0022]

[0023] Where: The cluster scheduling instruction sent by cluster h to the cluster distribution layer.

[0024] The optimized adaptive weight α h,1 , α h,2 The parameter calculation formula is as follows:

[0025]

[0026] Where: are the power prediction values ​​of the wind power cluster and the photovoltaic cluster at the optimization time respectively;

[0027] The cluster adaptive weight λ h Take the wind power cluster weight λ1 and the photovoltaic cluster weight λ2 respectively. The larger the weight value, the greater the adjustment of the cluster output relative to the optimization target. The specific formula is as follows:

[0028]

[0029] The method for clustering and allocation layer modeling in step S1 is as follows:

[0030] Establish optimization model objective function minJ Clu :

[0031]

[0032] Where: is the optimal allocation result of the optimal time of field station j in cluster i; θ i is the adaptive weight of the field group; m is the number of field groups in the cluster; n is the number of wind farms or photovoltaic power stations in field group i; Optimize the 15-minute ultra-short-term power forecast value for cluster-level stations; is the actual power value of the cluster layer station at the previous moment;

[0033] The above optimization constraints include cluster planning constraints, station output constraints and station ramp rate constraints. The specific formulas are as follows:

[0034]

[0035] Where: The minimum output of the station; Installed capacity for the station; It is the climbing rate limit of the station on a time scale of 15 minutes.

[0036] The dynamic clustering strategy of the midfield stations in step S2 is specifically as follows:

[0037] ① Dynamic grouping strategy of wind power clusters:

[0038] The dynamic clustering strategy of wind farms is based on the ultra-short-term power forecast value of the wind farm from the current moment to the next 15 minutes, with a resolution of 5 minutes. The wind farm group is divided into an up-climbing group, a down-climbing group, a stable group and a fluctuating group by judging the trend factor of the wind farm output power change. The trend factor k j The calculation formula is as follows:

[0039]

[0040] in:

[0041]

[0042] Where: is the power forecast value of wind farm j in wind farm cluster h1 at time t+nΔt. When n-1=0, it represents the wind farm power value at this moment. N wp is the number of future time sections. The ultra-short-term power forecast of 5 minutes within 15 minutes has 3 time sections, namely N wp =3; C(x) is the section trend judgment function; η is the wind farm power fluctuation threshold; Installed capacity for wind farms;

[0043] ② Photovoltaic cluster dynamic grouping strategy

[0044] The dynamic clustering of photovoltaic clusters is based on the predictable factor. The clustering is performed according to the ultra-short-term power prediction error 15 minutes before the optimization time. The resolution is 5 minutes and the clusters are divided into strong prediction group, medium prediction group and weak prediction group. The calculation formula of the predictable factor μ is as follows:

[0045]

[0046] Where: are the power prediction value and natural power value of PV power station j in PV cluster h2 at time t+(nN)Δt respectively; is the installed capacity of photovoltaic power station; N pv is the number of historical time sections;

[0047] After rolling optimization, the optimization allocation results of each cluster in the cluster are calculated according to the dynamic grouping results, and the results are sent to the cluster station optimization layer. The optimization results of each cluster in the cluster are as follows:

[0048]

[0049] In the formula, Optimization assignment results for field group i.

[0050] The method for modeling the optimization layer of the stations within the cluster in step S1 is as follows:

[0051] Establish optimization model objective function minJ SubClu :

[0052]

[0053] Where: The 5-minute ultra-short-term power prediction value of the inner-layer station j at the optimized time; is the actual power value of the inner-layer station in the group at the previous moment; Optimizing the allocation results of the optimized time for the stations within the group;

[0054] The optimization constraints of the above model include the station group planning constraints, station output constraints and station ramp rate constraints. The specific formula is as follows:

[0055]

[0056] Where: is the lowest output of station j in the group; The installed capacity of the stations within the cluster; It is the climbing rate limit of the station on a 5-minute time scale;

[0057] Send the optimization allocation results of the inner layer of the group to the power tracking layer of the station:

[0058]

[0059] Where: The station dispatching instruction sent from station j within the group to the station power tracking layer.

[0060] The method for modeling the field station power tracking layer in step S1 is as follows:

[0061] Establish optimization model objective function minJ Sta :

[0062]

[0063] Where: is the actual power value of station j at the previous moment in the station layer; The optimization allocation result of the station optimization time at the station level;

[0064] The optimization constraints of the above model include station planning constraints, station ramp rate constraints and station output constraints. The specific formula is as follows:

[0065]

[0066] Where: is the 1-minute ultra-short-term wind power forecast value at the optimization time of station j; It is the climbing rate limit of the station on a 1-minute time scale.

[0067] The actual control power calculation method in step S5 is:

[0068]

[0069] Where: are the actual control results and natural power values ​​of station j at time t+Δt respectively; if the optimization result is less than the natural power of the station, the wind-solar station will operate with limited power, and the actual control result will be the optimization result; conversely, if the optimization result is greater than the natural power of the station, the wind-solar station will track the maximum power point, and the actual control result will be the natural power value.

[0070] The prediction model in steps S1 and S6 is composed of a convolutional neural network, a Transformer model and a feedforward neural network.

[0071] The prediction error calculation formula in step S6 is:

[0072] e t =P t pre-P t nat ;

[0073] Where: P t pre is the output value of the prediction model before correction; e t is the prediction error at time t; P t nat is the natural power value at time t;

[0074] The power prediction error at the next moment is obtained by the following formula

[0075]

[0076] Where: f(x) is the exponential smoothing model; e t-sΔt is the prediction error at time t-sΔt;

[0077] The calculation formula for feedback correction of the multi-scale prediction of the prediction model in step S6 is:

[0078]

[0079] Where: is the predicted power correction value at the optimization time; is the power prediction value at the optimization time.

[0080] Through the above design scheme, the present invention can bring the following beneficial effects:

[0081] 1) The present invention realizes the optimal allocation of active power of the Shagohuang New Energy Base Station through the active layered predictive control method in a layer-by-layer refinement and feedback correction manner. Compared with other methods, the control accuracy of the method of the present invention is higher.

[0082] 2) In the active power control tasks with various types of energy such as Shagohuang, the present invention considers the power output characteristics of the energy to perform dynamic grouping, and combines the field group weights to effectively improve the strategy control effect.

[0083] 3) The present invention can dynamically plan wind and solar power output and optimization targets according to existing information of the control system through adaptive weights. The dynamic optimization strategy can effectively coordinate the output of stations and avoid uneven output of stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The present invention is further described below with reference to the accompanying drawings and specific embodiments:

[0085] Figure 1 A framework diagram of an active power control method for the Shagohuang new energy base according to the present invention;

[0086] Figure 2 It is a schematic diagram of a feedback correction model in an active power control method for Shagohuang new energy base of the present invention. DETAILED DESCRIPTION

[0087] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical scheme in the present invention is clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the present invention is not limited to the following embodiments, and the specific implementation method can be determined according to the technical scheme of the present invention and the actual situation. In order to avoid confusing the essence of the present invention, the known methods, processes and flows are not described in detail.

[0088] The present invention proposes an active power control method for the Shagohuang new energy base, see Figure 1 , the method comprises the following steps:

[0089] Step S1: The base wind and solar dispatching layer receives the dispatching plan power value issued by the dispatching center, and based on the 30-minute ultra-short-term power forecast information of the wind and solar cluster, adopts the predictive control method to track the planned value of the dispatching center, optimizes the power value of the wind power cluster and photovoltaic cluster in the next 1 hour, and sends the optimization result to the cluster grouping allocation layer;

[0090] The modeling method of the base wind and solar scheduling layer is as follows:

[0091] Establish the optimization model objective function:

[0092]

[0093] Where: c is the number of cluster types, in this invention, c=2, representing wind power cluster and photovoltaic cluster respectively; α h,1 , α h,2 To optimize the adaptive weight; h Adaptive weights for clusters; The allocated power value for cluster h at the optimization moment; 30-minute ultra-short-term power prediction value for cluster optimization; is the actual power value of the cluster at the previous moment; T is the optimization time domain. The base layer optimizes the cluster power in the base for the next 1 hour each time, with an optimization resolution of 30 minutes, that is, T=2. As the k value changes, the time t is optimized forward with a step size of Δt until the optimization control in the time domain is completed, and finally the optimization result when k=1 is issued.

[0094] The above model also needs to consider the scheduling plan constraints, cluster output constraints and cluster ramp rate constraints during the rolling optimization process. The specific formula is as follows:

[0095]

[0096] Where: To optimize the dispatch plan power value issued by the dispatch center at the moment; is the minimum output of cluster h; Install capacity for the cluster; It is the maximum climbing rate of the cluster on a 30-min time scale.

[0097] After the optimization is completed, the optimization results of the base layer will be sent to the cluster distribution layer.

[0098]

[0099] Where: The cluster scheduling instruction sent by cluster h to the cluster distribution layer.

[0100] In the above modeling, the adaptive weight α is optimized h,1 , α h,2 The parameter setting method is:

[0101] When the dispatch plan value is closer to the output capacity of the wind and solar cluster, the model has insufficient optimization space, and the model should focus on maximizing the output; conversely, when the dispatch plan value is far away from the output capacity of the wind and solar cluster, the model has sufficient optimization space, and the model should reduce the maximum output bias and increase the wave suppression bias. The mathematical calculation formula is as follows:

[0102]

[0103] In the above modeling, the cluster adaptive weight λ h Take the wind power cluster weight λ1 and the photovoltaic cluster weight λ2 respectively, and the setting method is:

[0104]

[0105] In the formula: λ1 and λ2 represent the weight of wind power cluster and photovoltaic cluster respectively. The larger the weight value, the greater the adjustment of cluster output relative to the optimization target; They are the power prediction values ​​of the wind power cluster and photovoltaic cluster at the optimization time respectively.

[0106] If wind power is the main output object in the Shagohuang New Energy Base, the wind power cluster weight λ1 is smaller than the photovoltaic cluster weight λ2. At this time, the adjustment of the wind power cluster relative to the optimization target is smaller, and the adjustment of the photovoltaic cluster relative to the optimization target is larger. During the optimization process, the model increases the photovoltaic output.

[0107] Step S2: The cluster grouping allocation layer receives the planned power values ​​of the wind power cluster and photovoltaic cluster issued by the wind and solar dispatching layer of the base, adopts a distributed prediction control method, uses the 15-minute ultra-short-term power prediction information of the cluster and wind and solar stations to optimize the power allocation of wind and solar stations, and then combines the station dynamic grouping strategy to obtain the planned power value of each group, and sends it to the station optimization layer within the group;

[0108] The modeling method of the cluster allocation layer is as follows:

[0109] Establish the optimization model objective function:

[0110]

[0111] Where: is the optimal allocation result of the optimal time of field station j in cluster i; θ i is the adaptive weight of the field group; m is the number of field groups in the cluster; n is the number of wind farms or photovoltaic power stations in field group i; Optimize the 15-minute ultra-short-term power forecast value for cluster-level stations; It is the actual power value of the cluster layer station at the previous moment.

[0112] The optimization constraints include cluster planning constraints, station output constraints, and station ramp rate constraints. The specific formulas are as follows:

[0113]

[0114] Where: Contribute to the Cluster H program; The minimum output of the station; Installed capacity for the station; It is the climbing rate limit of the station on a time scale of 15 minutes.

[0115] After rolling optimization, the optimization allocation results of each cluster in the cluster are calculated based on the dynamic grouping results, and the results are sent to the cluster station optimization layer. The optimization results of each cluster in the cluster are as follows:

[0116]

[0117] In the formula, Optimization assignment results for field group i.

[0118] The specific implementation method of the dynamic clustering strategy of midfield stations in the above modeling is:

[0119] ① Dynamic grouping strategy of wind power clusters:

[0120] The dynamic clustering strategy of wind farms is based on the ultra-short-term power forecast value of the wind farm from the current moment to the next 15 minutes (resolution 5 minutes), and considers the change trend of wind farm output power. The wind farm group is divided into an up-climbing group, a down-climbing group, a stable group and a fluctuating group. The trend factor k is used to judge the wind farm power output. j The calculation formula is as follows:

[0121]

[0122] in:

[0123]

[0124] Where: is the power forecast value of wind farm j in wind farm cluster h1 at time t+nΔt. When n-1=0, it indicates the power value of the wind farm at this moment. N wp is the number of future time sections. The ultra-short-term power forecast of 5 minutes within 15 minutes has 3 time sections, namely N wp =3; C(x) is the section trend judgment function, when x>η, C(x)=4, when -η≤x≤η, C(x)=1, when x<-η, C(x)=0; η is the wind farm power fluctuation threshold; is the installed capacity of the wind farm. The dynamic grouping standard of wind farms based on trend factors is shown in Table 1:

[0125] Table 1. Criteria for dynamic grouping of wind farms

[0126]

[0127] ② Photovoltaic cluster dynamic grouping strategy

[0128] The dynamic clustering of photovoltaic clusters is based on the predictability factor, taking into account the ultra-short-term power prediction error 15 minutes (resolution 5 minutes) before the optimization time, and is divided into strong prediction group, medium prediction group and weak prediction group. The calculation formula of the predictability factor μ is as follows:

[0129]

[0130] Where: are the power prediction value and natural power value of PV power station j in PV cluster h2 at time t+(nN)Δt respectively; is the installed capacity of photovoltaic power station; N pv is the number of sections in historical time. The photovoltaic dynamic grouping standard based on predictable factors is shown in Table 2:

[0131] Table 2 Criteria for judging dynamic grouping of photovoltaic power stations

[0132]

[0133] The parameter setting method of the adaptive weight of the field group in the above modeling is:

[0134] ① Wind farm group weight

[0135] The present invention divides wind power clusters into climbing groups, downhill groups, stable groups and fluctuating groups. In the process of output optimization and allocation, the weights of various types of field groups are set according to the changing trend of the cluster scheduling instructions received by the cluster sub-group allocation layer. When the scheduling value is climbing up, the output of the climbing group is prioritized; when the scheduling value is climbing down, the output of the downhill group is prioritized; when the scheduling value is stable, the output of the stable group is prioritized. Similar to the cluster weight, the larger the value of the field group weight, the greater the adjustment of the field group output relative to the optimization target. When there are multiple field groups, the weight setting steps are as follows:

[0136] Step S21: Obtain scheduling value trend and field group type;

[0137] Step S22: Determine whether there is a field group C of the same type as the scheduling value trend d .

[0138] Step S23: In step S22, if there is no C d , then the field group weights are equally distributed; if there is C d , then field group C d Set a larger weight;

[0139] ② Photovoltaic field group weight

[0140] The present invention divides photovoltaic clusters into strong prediction groups, medium prediction groups and weak prediction groups. In the process of output optimization allocation, when there are multiple field groups, the field group weight setting rule is: the stronger prediction group is set with a larger weight, and the weaker prediction group is set with a smaller weight.

[0141] Step S3: The station optimization layer within the cluster receives the planned power value of the cluster sent by the cluster allocation layer, and optimizes the allocation based on the ultra-short-term power prediction result with a resolution of 5 minutes, with the goal of tracking the cluster plan. The rolling optimization cycle is 5 minutes, and the power value of the wind and solar stations in the cluster in the next 15 minutes is optimized, that is, the power of the stations in the cluster in the next three time sections is optimized each time, but only the optimization result of the first time section is sent to the station power tracking layer;

[0142] The modeling method of the station optimization layer within the cluster is as follows:

[0143] Establish the optimization model objective function:

[0144]

[0145] Where: The 5-minute ultra-short-term power prediction value of the inner-layer station j at the optimized time; is the actual power value of the inner-layer station in the group at the previous moment; The optimized allocation results of the optimized time for the stations within the group.

[0146] The optimization constraints include the farm group planning constraints, the farm output constraints and the farm ramp rate constraints. The specific formula is as follows:

[0147]

[0148] Where: is the lowest output of station j in the group; The installed capacity of the stations within the cluster; It is the climbing rate limit of the station on a 5-minute time scale.

[0149] Send the optimization allocation results of the inner layer of the group to the power tracking layer of the station:

[0150]

[0151] Where: The station dispatching instruction sent from station j within the group to the station power tracking layer.

[0152] Step S4: The station power tracking layer receives the station planned power value sent by the station optimization layer within the group, and calculates the station output value based on the ultra-short-term power prediction result with a resolution of 1 minute;

[0153] The modeling method of the station power tracking layer is as follows:

[0154] Establish the optimization model objective function:

[0155]

[0156] Where: is the actual power value of station j at the previous moment in the station layer; The optimized allocation result of the station optimization time at the station level.

[0157] The optimization constraints include station planning constraints, station ramp rate constraints, and station output constraints. The specific formulas are as follows:

[0158]

[0159] Where: is the 1-minute ultra-short-term wind power forecast value at the optimization time of station j; It is the climbing rate limit of the station on a 1-minute time scale.

[0160] Step S5: The wind and solar power station performs active power control according to the station output value of the station power tracking layer, and records the actual control power and natural power. The actual control power is used to construct the ramp constraint of each layer, and the natural power is sent to the feedback correction model;

[0161] The actual control power calculation method is:

[0162]

[0163] Where: They are the actual control result and natural power value of station j at time t+Δt. When the optimization result is less than the natural power of the station, the wind and solar stations operate with limited power, and the actual control result is the optimization result; conversely, when the optimization result is greater than the natural power of the station, the wind and solar stations track the maximum power point, and the actual control result is the natural power value.

[0164] Step S6: The feedback correction model calculates the prediction error according to the natural power sent by the monitoring system and the predicted natural power sent by the prediction model, and predicts the power prediction error at the next moment, and performs feedback correction on the multi-scale prediction of the prediction model. The corrected prediction value is used to guide the optimization scheduling of each layer and carry out the active power control of the Shagohuang New Energy Base at the next moment;

[0165] The schematic diagram of the feedback correction model is as follows Figure 2 As shown;

[0166] In step S6, the prediction model is composed of a convolutional neural network, a Transformer model, and a feedforward neural network. The model input is multi-source data, including the NWP data and historical output data of the target station, as well as the NWP data and historical output data of other stations;

[0167] The prediction error calculation formula is:

[0168] e t =P t pre -P t nat ;

[0169] Where: P t pre is the output value of the prediction model before correction;

[0170] The power prediction error at the next moment is predicted by the following formula:

[0171]

[0172] f(x) is the exponential smoothing model;

[0173] The calculation formula for feedback correction of the multi-scale prediction of the prediction model is:

[0174]

[0175] This embodiment highlights the advantages of the present invention by comparing with two classical control methods, which are:

[0176] Fixed ratio control method, which distributes output according to the installed capacity of the station, is calculated as follows:

[0177]

[0178] The power optimization allocation result of station b at the optimization time; is the installed capacity of station b;

[0179] Variable ratio control method, which distributes output according to the power generation capacity of the station, and its calculation method is:

[0180]

[0181] The predicted power correction value at the optimization time for station b;

[0182] The rolling cycle of both control methods is 30 minutes, and the rolling process is constrained by the station climbing rate, specifically:

[0183]

[0184] is the actual power value of station b at the previous moment; is the climbing rate limit of station b;

[0185] In addition, the above two methods also need to compare the optimization results with the natural power and calculate the actual control results.

[0186] In order to quantitatively evaluate the control performance of each method, this embodiment proposes four evaluation indicators, namely:

[0187] Power generation difference:

[0188]

[0189] Power generation reaching the target rate:

[0190]

[0191] Maximum control deviation:

[0192]

[0193] Number of “over-issuances”:

[0194]

[0195] In the four evaluation index formulas: D is the total number of control nodes, and the time resolution of adjacent control nodes is 30 minutes; is the scheduling plan value issued by the scheduling center at the control node d; is the base control output power; Max(x) is the maximum value function; δ b,d is the "over-generation" judgment mark. When the optimization result of station b at control node d is greater than its natural power, it is considered "over-generation"; otherwise, there is no "over-generation" problem. Finally, the number of "over-generation" of each station control node is counted to obtain the total number of "over-generation" of the control method N c .

[0196] As shown in Table 3, the active power control results of different methods are shown.

[0197] Table 3 Control results of different control methods

[0198]

[0199]

[0200] It can be seen that the power generation difference of the method of the present invention is reduced by 93.05% and 60.97% respectively compared with the fixed ratio method and the variable ratio method, and the power generation compliance rate is increased by 10.96% and 1.21% respectively. The new energy consumption capacity is further enhanced, and its average wind and solar abandonment rates are reduced by 2.86% and 0.38% respectively. The coordination control ability is also greatly improved, and the standard deviation of wind and solar abandonment rates is reduced by 2.93% and 1.31% respectively. The control stability of the method of the present invention is also improved. Compared with the other two types of control methods, the maximum control deviation is reduced by 81.22% and 46.88% respectively. The method of the present invention aims to solve the problem of "over-generation" of control instructions to improve control accuracy. For this problem, the number of "over-generation" of the control method of the present invention is greatly reduced, by 231 times and 44 times respectively, and the control effect is significantly improved. In summary, the control method of the present invention is the best in all indicators and has obvious control advantages.

Claims

1. An active power control method for Shagohuang new energy base, characterized by: The process includes the following steps, which are performed in sequence: Step S1: Modeling of the base wind-solar dispatching layer, cluster sub-clustering allocation layer, intra-cluster station optimization layer, station power tracking layer, feedback correction model and prediction model are performed respectively. The base wind-solar dispatching layer receives the dispatching plan power value issued by the dispatching center, and adopts the prediction control method based on the 30-minute ultra-short-term power prediction information of the wind-solar cluster, with the goal of tracking the planned value of the dispatching center, to optimize the power value of the wind power cluster and photovoltaic cluster in the next 1 hour, and sends the optimization result as the planned power value of the wind power cluster and photovoltaic cluster to the cluster sub-clustering allocation layer; Step S2: The cluster grouping allocation layer receives the planned power values ​​of the wind power cluster and photovoltaic cluster issued by the wind and solar dispatching layer of the base, adopts a distributed prediction control method, uses the 15-minute ultra-short-term power prediction information of the cluster and wind and solar stations to optimize the power allocation of wind and solar stations, and then combines the station dynamic grouping strategy to obtain the planned power value of each group, and sends it to the station optimization layer within the group; Step S3: The station optimization layer within the cluster receives the planned power value of the cluster group issued by the cluster allocation layer, and optimizes the allocation based on the ultra-short-term power prediction result with a resolution of 5 minutes, with the goal of tracking the planned power of the cluster group. The rolling optimization cycle is 5 minutes, and the power value of the wind and solar stations within the cluster in the next 15 minutes is optimized, that is, the power of the stations within the cluster in the next three time sections is optimized each time, but only the optimization result of the first time section is issued as the planned power value of the station to the station power tracking layer; Step S4: The station power tracking layer receives the station planned power value sent by the station optimization layer within the group, and calculates and obtains the station output value based on the ultra-short-term power prediction result with a resolution of 1 minute; Step S5: The monitoring system of the wind and solar power station performs active power control according to the station output value obtained by the station power tracking layer, and records the actual control power and natural power, wherein the actual control power is used to construct the climbing constraint of each layer, and the natural power is sent to the feedback correction model; Step S6: The prediction model is used for multi-scale prediction of natural power, and the input of the prediction model is multi-source data, including numerical weather forecast data and historical output data of the target station, and numerical weather forecast data and historical output data of other stations; The feedback correction model calculates the prediction error based on the natural power sent by the monitoring system of the wind and solar power station and the predicted natural power sent by the prediction model, and predicts the power prediction error at the next moment, performs feedback correction on the multi-scale prediction of the prediction model, obtains the corrected prediction value and uses it to guide the optimization scheduling of each layer, and carry out active power control of the Shagohuang New Energy Base at the next moment.

2. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The method for modeling the base wind-solar scheduling layer in step S1 is as follows: Establish optimization model objective function minJ Base : Where: c is the number of cluster types, in this invention, c=2, representing wind power cluster and photovoltaic cluster respectively; α h,1 , α h,2 To optimize the adaptive weights; λ h Adaptive weights for clusters; The allocated power value for cluster h at the optimization moment; 30-minute ultra-short-term power prediction value for cluster optimization; is the actual power value of the cluster at the previous moment; T is the optimization time domain. The base layer optimizes the cluster power in the base for the next 1 hour each time, with an optimization resolution of 30 minutes, that is, T = 2, and as the k value changes, the time t is optimized forward with a step size of Δt until the optimization control in the time domain is completed, and finally the optimization result when k = 1 is issued; The above model also considers the scheduling plan constraints, cluster output constraints and cluster ramp rate constraints during the rolling optimization process. The specific formula is as follows: Where: To optimize the dispatch plan power value issued by the dispatch center at the moment; is the minimum output of cluster h; Install capacity for the cluster; is the maximum climbing rate of the cluster at a 30-min time scale; After the optimization is completed, the optimization results of the base layer are sent to the cluster distribution layer: Where: The cluster scheduling instruction sent by cluster h to the cluster distribution layer.

3. The active power control method for the Shagohuang new energy base according to claim 2 is characterized in that: The optimized adaptive weight α h,1 , α h,2 The parameter calculation formula is as follows: Where: are the power prediction values ​​of the wind power cluster and the photovoltaic cluster at the optimization time respectively; The cluster adaptive weight λ h Take the wind power cluster weight λ1 and the photovoltaic cluster weight λ2 respectively. The larger the weight value, the greater the adjustment of the cluster output relative to the optimization target. The specific formula is as follows:

4. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The method for clustering and allocation layer modeling in step S1 is as follows: Establish optimization model objective function minJ Clu : Where: The optimal allocation result of the optimal time of field station j in field group i at the cluster layer; θ i is the adaptive weight of the field group; m is the number of clusters in the cluster; n is the number of wind farms or photovoltaic power stations in cluster i; Optimize the 15-minute ultra-short-term power forecast value for cluster-level stations; is the actual power value of the cluster layer station at the previous moment; The above optimization constraints include cluster planning constraints, station output constraints and station ramp rate constraints. The specific formulas are as follows: Where: The minimum output of the station; Installed capacity for the station; It is the climbing rate limit of the station on a 15-minute time scale; After rolling optimization, the optimization allocation results of each cluster in the cluster are calculated according to the dynamic grouping results, and the results are sent to the cluster station optimization layer. The optimization results of each cluster in the cluster are as follows: In the formula, Optimization assignment results for field group i.

5. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The dynamic clustering strategy of the midfield stations in step S2 is specifically as follows: ① Dynamic grouping strategy of wind power clusters: The dynamic clustering strategy of wind farms is based on the ultra-short-term power forecast value of the wind farm from the current moment to the next 15 minutes, with a resolution of 5 minutes. The wind farm group is divided into an up-climbing group, a down-climbing group, a stable group and a fluctuating group by judging the trend factor of the wind farm output power change. The trend factor k j The calculation formula is as follows: in: Where: is the power forecast value of wind farm j in wind farm cluster h1 at time t+nΔt. When n-1=0, it represents the power value of the wind farm at this moment. N wp is the number of future time sections. The ultra-short-term power forecast of 5 minutes within 15 minutes has 3 time sections, namely N wp =3; C(x) is the section trend judgment function; η is the wind farm power fluctuation threshold; Installed capacity for wind farms; ② Photovoltaic cluster dynamic grouping strategy The dynamic clustering of photovoltaic clusters is based on the predictable factor. The clustering is performed according to the ultra-short-term power prediction error 15 minutes before the optimization time. The resolution is 5 minutes and the clusters are divided into strong prediction group, medium prediction group and weak prediction group. The calculation formula of the predictable factor μ is as follows: Where: are the power prediction value and natural power value of PV power station j in PV cluster h2 at time t+(nN)Δt respectively; is the installed capacity of photovoltaic power station; N pv is the number of historical time sections.

6. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The method for modeling the optimization layer of the stations within the cluster in step S1 is as follows: Establish optimization model objective function minJ SubClu : Where: The 5-minute ultra-short-term power prediction value of the inner-layer station j at the optimized time; is the actual power value of the inner-layer station in the group at the previous moment; Optimizing the allocation results of the optimized time for the stations within the group; The optimization constraints of the above model include the station group planning constraints, station output constraints and station ramp rate constraints. The specific formula is as follows: Where: is the lowest output of station j in the group; The installed capacity of the stations within the cluster; It is the climbing rate limit of the station on a 5-minute time scale; Send the optimization allocation results of the inner layer of the group to the power tracking layer of the station: Where: The station dispatching instruction sent from station j within the group to the station power tracking layer.

7. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The method for modeling the field station power tracking layer in step S1 is as follows: Establish optimization model objective function minJ Sta : Where: is the actual power value of station j at the previous moment in the station layer; The optimization allocation result of the station optimization time at the station level; The optimization constraints of the above model include station planning constraints, station ramp rate constraints and station output constraints. The specific formula is as follows: Where: is the 1-minute ultra-short-term wind power forecast value at the optimization time of station j; It is the climbing rate limit of the station on a 1-minute time scale.

8. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The actual control power calculation method in step S5 is: Where: are the actual control results and natural power values ​​of station j at time t+Δt respectively; if the optimization result is less than the natural power of the station, the wind-solar station will operate with limited power, and the actual control result will be the optimization result; conversely, if the optimization result is greater than the natural power of the station, the wind-solar station will track the maximum power point, and the actual control result will be the natural power value.

9. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The prediction model in steps S1 and S6 is composed of a convolutional neural network, a Transformer model and a feedforward neural network.

10. The active power control method for the Shagohuang new energy base according to claim 1 is characterized by: The prediction error calculation formula in step S6 is: Where: P t pre is the output value of the prediction model before correction; e t is the prediction error at time t; is the natural power value at time t; The power prediction error at the next moment is obtained by the following formula Where: f(x) is the exponential smoothing model; e t-sΔt is the prediction error at time t-sΔt; The calculation formula for feedback correction of the multi-scale prediction of the prediction model in step S6 is: Where: is the predicted power correction value at the optimization time; is the power prediction value at the optimization time.