Photovoltaic cluster output optimization method based on scheduling error

Abstract

The invention discloses a photovoltaic cluster output optimization method based on a scheduling error, and relates to the technical field of photovoltaic cluster output optimization, and the method is technically characterized in that a scheduling curve is constructed based on historical data, an SSA-LSTM algorithm is used for solving, and a minimum scheduling target response error is taken as a target; a photovoltaic cluster output sorting index system is established in combination with an ideal output curve under a typical meteorological day, station output sorting is optimized, a maximum output or tracking output mode is adaptively selected, and finally the purposes of tracking a scheduling curve and improving scheduling precision are achieved. The effectiveness of the method provided by the invention is verified through an actual data analysis result, and the method provided by the invention can be used for reasonably distributing the output of each station and improving the photovoltaic response precision, and has important significance on the safe and economic operation of a power system.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic cluster output optimization, and particularly relates to a photovoltaic cluster output optimization method based on scheduling error. Background Art

[0002] Photovoltaic power generation has significant volatility and randomness, and photovoltaic clusters usually adopt the maximum output form during power generation, resulting in frequent light curtailment phenomena. Therefore, scholars at home and abroad have proposed various photovoltaic control strategies, mainly focusing on minimizing environmental hazards, maximizing overall power generation efficiency, or ensuring power consumption.

[0003] Currently, the existing photovoltaic control strategies in the prior art are as follows:

[0004] 1. Prediction is carried out after clustering of photovoltaic clusters based on correlations such as time - space and meteorology.

[0005] 2. Photovoltaic power generation is combined with other power generation methods, such as coordinated control of multiple power generation methods including photovoltaic - storage, photovoltaic - wind - thermal - storage, etc., and control is carried out with the goal of optimal economy or lowest volatility.

[0006] 3. Considering demand - side response, photovoltaic source - load coordinated control is realized by using the form of participation of the whole - section or segmented demand to achieve maximum photovoltaic power consumption.

[0007] 4. A hybrid time - scale reactive power / voltage control method combining day - ahead reactive power / voltage optimization control and real - time autonomous control is proposed.

[0008] 5. Based on the hierarchical distributed model predictive control theory, multi - time - scale optimization control is adopted, and a rolling optimization model with gradually refined time - space scales is established for each layer to reduce the randomness of new - energy power generation and ensure the safety of the power system.

[0009] The above - mentioned prior art conducts photovoltaic control through multi - level modeling of multi - energy systems containing photovoltaic power generation, and adopts different control strategies and coordinated control for different optimization goals to achieve objectives such as suppressing volatility and ensuring power consumption. However, it mainly targets energy forms with a low proportion of photovoltaic output. At present, however, photovoltaic power is developing rapidly, the number of photovoltaic power stations is large, the time - space characteristics between stations vary greatly, and the output characteristics, maintenance status, and interference factors within the stations are also different. It is urgent to coordinate the output of stations within the cluster, track the scheduling plan, suppress volatility, and ensure the safe and stable operation of the power system.

[0010] Therefore, the present invention aims to propose a photovoltaic cluster output optimization method based on scheduling error to solve the above problems. Summary of the Invention

[0011] The purpose of the present invention is to solve the above problems and provide a photovoltaic cluster output optimization method based on scheduling error.

[0012] To achieve the above object, the technical solution of the present invention is as follows:

[0013] The present invention provides a method for optimizing the output of a photovoltaic cluster based on scheduling error. A scheduling curve is constructed based on historical data, and the SSA-LSTM algorithm is used for solution. With the goal of minimizing the scheduling target response error, a photovoltaic cluster output ranking index system is established in combination with the ideal output curve on a typical meteorological day, the output ranking of the power station is optimized, and the maximum output or tracking output mode is adaptively selected. Finally, the scheduling curve is tracked and the scheduling accuracy is improved;

[0014] Specifically, it includes the following steps:

[0015] S1. Construct a scheduling response error based on SSA-LSTM:

[0016] S1-1. A prediction model based on SSA-LSTM;

[0017] S1-2. The LSTM algorithm model;

[0018] S1-3. The scheduling target response error;

[0019] S2. The index system for the output order of the power station:

[0020] S2-1. Construct a typical meteorological output curve;

[0021] S2-2. Construct an index system for the output ranking of the power station;

[0022] S3. Output allocation and optimization constraints:

[0023] S3-1. Optimize the cluster output constraints;

[0024] S3-2. Optimize the output of the power station.

[0025] Furthermore, step S1-1 is applied to optimize the regularization rate, learning rate, and number of hidden layer units of the LSTM, thereby improving the overall operation speed and prediction accuracy; specifically:

[0026] Let the discoverer not converge to the origin first, and change from jumping to the optimal position to moving towards the optimal position. Its update rule is as follows:

[0027] (1) Establish a discoverer position update rule in combination with the sine-cosine algorithm:

[0028]

[0029] Among them, t represents the current iteration number, DT represents the total iteration number, X a,b (t) represents the position information value of the a-th sparrow at the b-th dimension and iteration number t, Xbest (t) is the current global optimal position;

[0030] r2 and r3 are random numbers obeying the normal distribution;

[0031] ω1 and ω2 are weight coefficients, taking 0.95 and 0.5 respectively;

[0032] R (R ∈ [0, 1]) represents the warning value, and ST (ST ∈ [0.5, 1]) represents the safety value;

[0033] (2) Update of the joiner's position:

[0034]

[0035] Among them, Q is a random number obeying the normal distribution, and L represents a j-dimensional vector with all internal elements being 1; X worst (t) is the position with the worst current global fitness; when When, the joiner follows the discoverer to move; when When, the joiner gets rid of the current position;

[0036] (3) Update of the scout's position:

[0037]

[0038] Among them, β, K ∈ randn(0, 1) are used for step size control; f i is the fitness of the current individual; f b is the current global best fitness.

[0039] Furthermore, step S1-2 is specifically: according to the improved neural network from left to right f t 、i t 、o t respectively represent the forget gate, input gate, output gate, and C t represents the current state; then there is:

[0040]

[0041] In the formula: σ is the sigmoid function, and the output of sigmoid determines whether the information should be saved; W f 、W i 、W C 、W o are weight matrices; b f 、b i 、b C 、b o are bias parameter matrices; x t 、h t-1respectively represent the input at time t, h t is the output, C t represents the current state.

[0042] Furthermore, step S1-3 is specifically as follows:

[0043] Construct a training feature set with environmental temperature, light intensity, environmental humidity, weather conditions, wind speed, and air pressure factors as features. The output is the intra-day prediction scheduling value of the photovoltaic cluster, then there is the following:

[0044]

[0045] where: t = (1, 2,...T) is the sampling time within the T interval; k = (1, 2,..., m) is the number of stations within the photovoltaic cluster; X k,t is the input feature at time t; environmental temperature T(T1, T2,..., T t ), light intensity SA(S1, S2,..., S t ), environmental humidity H(H1, H2,..., H t ), weather conditions W(W1, W2,..., W t ), wind speed D(D1, D2,..., D t ), air pressure F(F1, F2,..., F t ); is the scheduling target; the subscript i represents the i-th meteorological condition, t represents the time point, k represents the k-th station in the cluster, C represents the overall cluster; the superscript dis represents the scheduling output, real represents the actual output, opt represents the optimized output, for represents the predicted output value, ideal represents the ideal output, C represents the overall cluster, use represents the maximum available power;

[0046] At this time, the total scheduling error can be written as:

[0047]

[0048] The scheduling error of each station can be written as:

[0049]

[0050] Let be the scheduling margin of the k-th station, that is, the difference between the ideal value and the scheduling value of the output at time t, describing the output margin of the k-th station after receiving of the scheduling; it can be written as:

[0051]

[0052] Furthermore, step S2-1 for constructing the typical meteorological output curve is specifically as follows:

[0053] By selecting the theoretical output values of each PV power station in the cluster on different typical days as reference values, four typical meteorological characteristics, namely sunny type, cloudy type, shower type, and overcast type, are obtained. The PV output data is clustered under four climate environment categories respectively using the principle of minimizing the maximum difference and maximizing the minimum similarity to obtain the typical daily output curve of the station that can represent this type of meteorological characteristic, which can be written as:

[0054]

[0055] Among them, is the typical output value at the nth moment in the basic typical day curve under the ith meteorological condition.

[0056] Furthermore, the specific construction of the station output sorting index system in step S2-2 is as follows:

[0057] A. Controllability and measurability:

[0058] Give the controllability and measurability α t Characterize whether the historical data of the station is complete. In addition, the controllability and measurability also characterize whether the station has the ability to generate electricity according to the dispatching value in history. Then there is:

[0059]

[0060] Among them, t = (1, 2....T) is the sampling point of the PV power station data within the previous 5 minutes, and P t dis is the station output dispatching value, and P t real is the actual output value of the station; the closer the controllability and measurability are to 1, the more controlled the station is. On the contrary, it means that the station may be in a state where it cannot be regulated in real time. When the value is 0, it means that the station was in a failure or maintenance state in history;

[0061] B. Cosine similarity:

[0062] In any meteorological type, the closer the actual curve P t real of the station is to the ideal output curve P t ideal the more it meets the dispatching requirements, which is the most core index for optimal control. The cosine similarity β t is used to represent the similarity between the actual output value and the ideal output value within the previous 5 minutes at time t, and T is the number of sampling points within the previous 5 minutes at time t; then:

[0063]

[0064] According to the above formula, the closer the similarity is to 1, the more similar the actual value at time t and the predicted value at time t are to the theoretical output curve before and after time t, indicating a higher system following performance;

[0065] C. Prediction accuracy:

[0066] The predicted value is the mid-value of the power output of the PV power station when it receives dispatching and outputs according to the demand. The prediction accuracy describes the similarity between the predicted power output value and the actual power output value. The higher the prediction accuracy, the more substantial the actual power output of the power station, and the more it can meet the dispatching requirements. Therefore:

[0067]

[0068] D. Output compliance:

[0069] Under the same meteorological conditions, evaluate the output size of the power station, that is, the output compliance γ t , that is:

[0070]

[0071] When the similarity is close to 1, when the output deviation γ t is closer to 0, it means that the output of the power station is more similar to the ideal output. The greater the deviation, the worse the following performance of the power station, the lower the accuracy of the prediction result, and the lower the dispatchability;

[0072] The above indicators are all values between [0 - 1]. From the physical meanings of each indicator, it can be concluded that the closer each indicator is to 1, the more ideal the output of the power station. Therefore, construct the following sorting indicator lin:

[0073]

[0074] Describe as follows:

[0075] (1) When the lin values are different, output according to the size of the lin value;

[0076] (2) When the lin values are the same, sort according to the size order of β t , γ t , α t respectively.

[0077] Furthermore, the optimization of the cluster output constraint in step S3-1 is specifically:

[0078] To ensure the effectiveness and real-time performance of the dispatching tracking, set the rolling sorting time to 5 minutes, that is, perform a sorting every 5 minutes according to the dispatching tracking error information;

[0079] When the cluster scheduling tracking error is less than or equal to the margin of the first station in the ranking, changing the output of the first station can meet the cluster scheduling tracking requirements. At this time, the first station operates in a following mode, coordinates its own output, and makes the scheduling error approach 0, which is called the following mode;

[0080] When the total cluster scheduling tracking error is greater than the scheduling margin of the first station in the ranking, at this time, the first station operates at the maximum output that can be achieved, introduces the second output station, and compares the remaining scheduling error with the scheduling margin of the second station, and performs a loop, which is called the maximum output mode; Thus, the objective function Q can be set as:

[0081]

[0082] Its constraint conditions include the change rate constraint and the cluster tracking optimization range constraint, and can be written as:

[0083]

[0084] Furthermore, the optimization of the station output in step S3-2 is specifically as follows:

[0085] When the cluster scheduling error function is greater than the scheduling margin of the station, the station operates at the maximum output that can be achieved; when working in the following mode, the output is carried out in the order of the pre-ranked stations; the output mode is judged and adjusted every 1 minute;

[0086] Then, when the station adopts the maximum output mode, the objective function can be written as:

[0087]

[0088] Where t=(1,2,...,T1), which is the sampling time within 1 minute;

[0089] When the station adopts the following mode, the minimum difference between the optimized output value of this station and the remaining scheduling target of the cluster can be used as the objective function, and can be written as:

[0090]

[0091] Where m1 is the number of stations that have been ranked before this station participates in the output;

[0092] In addition, the station also needs to meet the photovoltaic output limit constraint and the output fluctuation constraint, as follows:

[0093]

[0094] Compared with the existing technology, the beneficial effects of this solution:

[0095] 1. The present invention constructs a scheduling curve based on historical data, uses an improved SSA-LSTM algorithm for solution, aims to minimize the response error of the scheduling target, establishes a sorting index system for the output of a photovoltaic cluster in combination with the ideal output curve under typical meteorological days, optimizes the output sorting of the power station, adaptively selects the maximum output or tracking output mode, and finally achieves the goal of tracking the scheduling curve and improving the scheduling accuracy.

[0096] 2. The effectiveness of the method proposed by the present invention is verified by the results of actual data analysis. The method of the present invention can reasonably distribute the output of each power station, improve the response accuracy of the photovoltaic, and is of great significance to the safe and economic operation of the power system. Description of the Drawings

[0097] Figure 1 is the optimization solution process of the cluster output sorting based on the scheduling error function in the embodiment of the present invention;

[0098] Figure 2 is the basic neuron of the LSTM network in the embodiment of the present invention;

[0099] Figure 3 is the output curve of a power station on different dates under the same meteorological conditions in the embodiment of the present invention;

[0100] Figure 4 is the total output curve and the scheduling curve of the photovoltaic cluster under a certain typical day in the embodiment of the present invention;

[0101] Figure 5 is the calculation results of controllability and measurability, cosine similarity, prediction accuracy, and output deviation in the embodiment of the present invention;

[0102] Figure 6 is the total scheduling error curve of the photovoltaic cluster and the scheduling margin curve of a power station under a certain typical day in the embodiment of the present invention;

[0103] Figure 7 is the output curve of each power station after optimization in the embodiment of the present invention;

[0104] Figure 8 is the total scheduling error after optimization in the embodiment of the present invention;

[0105] Figure 9 is the comparison of the scheduling curve, dynamic grouping control, and LSTM prediction curve in the embodiment of the present invention. Detailed Embodiment

[0106] To enable those skilled in the art to better understand the solution of the present invention, the technical solution of the present invention will be further described in detail below in conjunction with the embodiments and drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the scope of protection of the present invention.

[0107] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below in conjunction with the embodiments.

[0108] Embodiment:

[0109] The solution provided in the embodiment of the present invention is as described in the above invention content, providing a method for optimizing the output of a photovoltaic cluster based on scheduling error, and its process is as Figure 1 shown. The method is as follows: constructing a scheduling curve based on historical data, using an improved SSA-LSTM algorithm for solution, aiming at minimizing the scheduling target response error, establishing a photovoltaic cluster output sorting index system in combination with the ideal output curve under typical meteorological days, optimizing the output sorting of the power station, and adaptively selecting the maximum output or tracking output mode, and finally achieving the goal of tracking the scheduling curve and improving the scheduling accuracy.

[0110] Specifically, it includes the following steps:

[0111] S1. Constructing a scheduling response error based on SSA-LSTM:

[0112] S1-1. A prediction model based on SSA-LSTM; applied to optimize the regularization rate, learning rate, and number of hidden layer units of LSTM, thereby improving the overall operation speed and prediction accuracy; specifically:

[0113] Let the discoverer not converge to the origin first, and change from jumping to the optimal position to moving towards the optimal position, and its update rule is as follows:

[0114] (1) Establishing a discoverer position update rule in combination with the sine-cosine algorithm:

[0115]

[0116] Among them, t represents the current iteration number, DT represents the total iteration number, X a,b (t) represents the position information value of the a-th sparrow at the b-th dimension and iteration number t, and X best (t) is the current global optimal position;

[0117] r2, r3 are random numbers obeying the normal distribution;

[0118] ω1 and ω2 are weight coefficients, taking 0.95 and 0.5 respectively;

[0119] R (R ∈ [0, 1]) represents the warning value, and ST (ST ∈ [0.5, 1]) represents the safety value;

[0120] (2) Update of the joiner's position:

[0121]

[0122] Among them, Q is a random number subject to a normal distribution, and L represents a j-dimensional vector with all internal elements being 1; X worst (t) is the position with the worst current global fitness; when , the joiner follows the discoverer to move; when , the joiner gets rid of the current position;

[0123] (3) Update of the scout's position:

[0124]

[0125] Among them, β, K ∈ randn(0, 1) are used for step size control; f i is the fitness of the current individual; f b is the current global best fitness.

[0126] S1-2, LSTM algorithm model; LSTM is an improved neural network with continuous feedback connections, and its structure Figure 2 is shown as follows. LSTM uses the logical control of the gate structure to determine whether data is updated or discarded, reduces the problem of gradient disappearance or explosion, enhances the problem of long-term time series processing, enables the network to converge faster, and is widely used in fields such as photovoltaic, wind power output prediction, and load prediction. As Figure 2 shown, from left to right, f t , i t , o t respectively represent the forget gate, input gate, output gate, and C t represents the current state; then there are:

[0127]

[0128] In the formula: σ is the sigmoid function, and the output of sigmoid determines whether information should be saved; W f , W i , W C , W o are weight matrices; b f , b i , b C , bo is the bias parameter matrix; x t , h t-1 represent the input at time t and h t is the output, and C t represents the current state.

[0129] S1-3, the scheduling target response error; taking environmental temperature, light intensity, environmental humidity, weather conditions, wind speed, and air pressure factors as features to construct a training feature set, and the output is the intra-day prediction scheduling value of the photovoltaic cluster, then there is the following:

[0130]

[0131] Among them: t = (1, 2,... T) is the sampling time within the T interval; k = (1, 2,..., m) is the number of internal stations in the photovoltaic cluster; X k,t is the input feature at time t; environmental temperature T(T1, T2,..., T t ), light intensity SA(S1, S2,..., S t ), environmental humidity H(H1, H2,..., H t ), weather conditions W(W1, W2,..., W t ), wind speed D(D1, D2,..., D t ), air pressure F(F1, F2,..., F t ); is the scheduling target; the subscript i represents the i-th meteorological condition, t represents the time point, k represents the k-th station in the cluster, C represents the overall cluster; the superscript dis represents the scheduling output, real represents the actual output, opt represents the optimized output, for represents the predicted output value, ideal represents the ideal output, C represents the overall cluster, and use represents the maximum available power;

[0132] At this time, the total scheduling error can be written as:

[0133]

[0134] The scheduling error of each station can be written as:

[0135]

[0136] Let be the scheduling margin of the k-th station, that is, the difference between the ideal value and the scheduling value of the output at time t, describing the output margin of the k-th station after receiving the scheduling; it can be written as:

[0137]

[0138] S2, the station output sequence index system:

[0139] S2-1. Construct a typical meteorological output curve. The meteorological conditions at the location of the power station are the most significant factors affecting the output. Under different meteorological conditions, the photovoltaic output varies significantly, but under the same meteorological conditions, the photovoltaic output curves are relatively similar. For example, Figure 3 as shown; by selecting the theoretical output values of each photovoltaic power station in the cluster on different typical days as reference values, four typical meteorological characteristics, namely sunny type, cloudy type, shower type, and overcast type, are obtained; the photovoltaic output data is clustered under four climate environment categories respectively using the principle of minimizing the maximum difference and maximizing the similarity to obtain the single-day typical output curve of the power station representing this type of meteorological characteristic, which can be written as:

[0140]

[0141] where, is the typical output value at the nth moment in the basic typical day curve under the ith meteorological condition.

[0142] S2-2. Construct an index system for ranking the power station output; A. Controllability and measurability:

[0143] Give the controllability and measurability α t to characterize whether the historical data of the power station is complete. In addition, the controllability and measurability also characterize whether the power station has the ability to generate electricity according to the dispatching value in history. Then there is:

[0144]

[0145] where t=(1,2....T) is the sampling point of the photovoltaic power station data within the previous 5 minutes, P t dis is the dispatching value of the power station output, and P t real is the actual output value of the power station; the closer the controllability and measurability are to 1, the more controlled the power station is. On the contrary, it means that the power station may be in a state where it cannot be regulated in real time. When the value is 0, it means that the power station was in a faulty or overhaul state in history;

[0146] B. Cosine similarity:

[0147] In any meteorological type, the closer the actual curve P t real of the power station is to the ideal output curve P t ideal the more it meets the dispatching requirements. It is the most core index for optimal control. The cosine similarity β t is used to represent the similarity between the actual output value and the ideal output value within the previous 5 minutes at time t, and T is the number of sampling points within the previous 5 minutes at time t; then:

[0148]

[0149] According to the above formula, the closer the similarity is to 1, it indicates that the use of the actual value at time t and the predicted value at time t before and after time t is more similar to the theoretical output curve, representing a higher system following ability;

[0150] C. Prediction accuracy:

[0151] The predicted value is the mid-value of the photovoltaic power station's acceptance of dispatching and output according to requirements. The prediction accuracy describes the similarity between the predicted output value and the actual output value. The higher the prediction accuracy, the more remarkable the actual output of the power station, and the more it can meet the dispatching requirements. Then:

[0152]

[0153] D. Output compliance:

[0154] Under the same meteorological conditions, evaluate the output size of the power station, that is, the output compliance γ t , that is:

[0155]

[0156] When the similarity is close to 1, when the output deviation γ t is closer to 0, it indicates that the output of the power station is more similar to the ideal output. The greater the deviation, the worse the following ability of the power station, the lower the accuracy of the prediction result, and the lower the dispatchability;

[0157] The above indicators are all values between [0 - 1]. From the physical meanings of each indicator, it can be obtained that the closer each indicator is to 1, the more ideal the output of the power station. Therefore, construct the following sorting indicator lin:

[0158]

[0159] Describe as follows:

[0160] (1) When the lin values are different, output according to the size of the lin values;

[0161] (2) When the lin values are the same, sort according to the size order of β t , γ t , α t in turn.

[0162] S3. Output allocation and optimization constraints:

[0163] S3-1. Optimization of cluster output constraints; To ensure the effectiveness and real-time nature of dispatching tracking, set the rolling sorting time to 5 minutes, that is, perform sorting every 5 minutes according to the dispatching tracking error information;

[0164] When the cluster scheduling tracking error is less than or equal to the margin of the first station, changing the output of the first station can meet the cluster scheduling tracking requirements. At this time, the first station operates in a following mode, coordinating its own output to make the scheduling error close to 0, which is called the following mode;

[0165] When the total cluster scheduling tracking error is greater than the scheduling margin of the first station in the sorting, at this time, the first station operates at the maximum output it can reach, introduces the second output station, and compares the remaining scheduling error with the scheduling margin of the second station, and conducts a loop, which is called the maximum output mode; Thus, the objective function Q can be set as:

[0166]

[0167] Its constraint conditions include the change rate constraint and the cluster tracking optimization range constraint, and can be written as:

[0168]

[0169] S3-2. Optimization of station output; when the cluster scheduling error function is greater than the scheduling margin of the station, the station operates at the maximum output it can reach; when working in the following mode, the output is carried out in the order of the pre-sorted stations; the output mode is judged and adjusted every 1 minute;

[0170] Then, when the station adopts the maximum output mode, the objective function can be written as:

[0171]

[0172] where t=(1,2,...,T1), which is the sampling time within 1 minute;

[0173] When the station adopts the following mode, the minimum difference between the optimized output value of this station and the remaining scheduling target of the cluster can be used as the objective function, and can be written as:

[0174]

[0175] where m1 is the number of stations that have been sorted before this station participates in the output;

[0176] In addition, the station also needs to meet the photovoltaic output limit constraint and the output fluctuation constraint, as shown below:

[0177]

[0178] The following is the numerical example verification of the embodiment of the present invention

[0179] Select the active power output data of a certain photovoltaic cluster in Guizhou in 2020 and the corresponding weather data for verification. Ten power stations in the cluster are selected this time, with a capacity between 60MW and 120MW, and each power station is controllable and measurable. Affected by factors such as the operation status and maintenance of the power station itself, not all power stations can reach the optimal output on the same typical meteorological day, and the typical meteorological day curve is not smooth. The total cluster output obtained by superimposing it and the scheduling curve predicted by SSA-LSTM are as Figure 4 shown.

[0180] By calculating the controllability and measurability, cosine similarity, prediction accuracy, and output compliance indicators of photovoltaic power stations within each time scale in a rolling manner, the calculation results of controllability and measurability, cosine similarity, prediction accuracy, and output deviation can be obtained as Figure 5 shown.

[0181] It can be seen that the sorting indicators calculated for each moment of the power station within a fixed time period change, representing the changes in the output of the power station at different moments. The cluster scheduling error and the scheduling margin of a certain power station within the same time period are shown in the following figure. Within the first 23 sampling points, the grid demand is relatively low at this time, and the power station needs to adopt a tracking mode to meet the grid connection demand; afterwards, the grid demand increases, and the output of only this power station cannot fill the scheduling error. Therefore, it is necessary to reasonably arrange the output of the power stations within the cluster to improve the scheduling response accuracy.

[0182] As Figure 6 shown, the total scheduling error curve of the photovoltaic cluster and the scheduling margin curve of a certain power station on a certain typical day, and the final optimized output result is as Figure 7 shown. After optimization, the top 9 power stations in the ranking all adopt the maximum output mode, and the last-ranked power station 8 adopts the following mode for output. The total scheduling error after optimization is as Figure 8 shown. Compared with the scheduling error in Figure 6 , the total scheduling error after optimization is significantly reduced. It shows that the method of the present invention can greatly improve the scheduling tracking accuracy.

[0183] Compared with the output mode of the photovoltaic cluster with a fixed proportion of output, the scheduling curve of the photovoltaic cluster before optimization and the output scheduling curve of the photovoltaic cluster after optimization are as Figure 9 shown. It can be seen that the output of the cluster after optimization by the present invention can well track the scheduling curve of the photovoltaic cluster. During the entire scheduling tracking process, the active power output of the method of the present invention is relatively stable and there is no obvious fluctuation. It performs better in the second half compared with the fixed proportion control method, and also maintains good output and smoothness during the peak period, indicating that the method proposed by the present invention increases the predictability during this time period.

[0184] Conclusion:

[0185] Aiming at the significant volatility and randomness problems existing in the active power output of photovoltaic clusters, this invention constructs a photovoltaic output scheduling curve by using SSA-LSTM, establishes a sorting index for the output of power stations, and optimizes and coordinates the output of each power station in the cluster with the goal of minimizing the scheduling error. The following conclusions are obtained from the experiments:

[0186] To meet the requirements of the power system, the minimum scheduling error can be used as the optimization goal to coordinate the output of the cluster and improve the scheduling accuracy.

[0187] Different from the optimization of distributed photovoltaic output, there are many power stations in the cluster and the output is uneven. Sorting the output of power stations in a large-scale photovoltaic cluster can coordinate and balance the output between power stations and suppress the volatility of the cluster output.

[0188] Meteorological factors have a great influence on photovoltaic output. The coordinated scheduling can be carried out according to the typical meteorological day curve to reduce the randomness of photovoltaic output.

[0189] SSA has good local optimization functions and can be used to improve the hyperparameters of algorithms such as LSTM to increase the convergence speed and accuracy of the algorithms.

[0190] The above specific embodiments are only explanations of the present invention and are not limitations thereof. After reading this specification, those skilled in the art can make modifications to these embodiments without creative contributions as needed, but as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A photovoltaic cluster output optimization method based on scheduling error, characterized in that: Based on historical data, the dispatch curve is constructed and solved using the SSA-LSTM algorithm. The goal is to minimize the dispatch target response error. Combined with the ideal output curve on a typical meteorological day, a photovoltaic cluster output ranking index system is established to optimize the station output ranking, and the maximum output or tracking output mode is adaptively selected, ultimately achieving the goal of tracking the dispatch curve and improving the dispatch accuracy. The specific steps include: S1. Constructing scheduling response error based on SSA-LSTM: S1-1, prediction model based on SSA-LSTM; S1-2, LSTM algorithm model; S1-3, scheduling target response error; S2. Station output sequence indicator system: S2-1 constructs a typical meteorological output curve; S2-2. Construct a station output ranking index system; S3. Output distribution and optimization constraints: S3-1, cluster output constraint optimization; S3-2. Optimization of station output.

2. The photovoltaic cluster output optimization method based on scheduling error as described in claim 1, characterized in that: Step S1-1 is used to optimize the regularization rate, learning rate, and number of hidden layer units of LSTM, thereby improving the overall operation speed and prediction accuracy; specifically: Let the finder not converge to the origin first, and change from jumping to the optimal position to moving to the optimal position. The update rule is as follows: (1) Combine the sine and cosine algorithms to establish the discoverer position update rule: Among them, t represents the current iteration number, DT represents the total number of iterations, and X a,b (t) represents the position information value of the a-th sparrow at the b-th dimension when the iteration number is t, and X best (t) is the current global optimal position; r2 and r3 are random numbers that follow a normal distribution; ω1 and ω2 are weighting coefficients, taking 0.95 and 0.5 respectively; R(R∈[0,1]) represents the warning value, and ST(ST∈[0.5,1]) represents the safety value; (2) Joiner location update: where Q is a random number following a normal distribution, L represents a j-dimensional vector with all internal elements being 1; X worst (t) is the position with the worst current global fitness; when occurs, the joiner follows the discoverer to move; when occurs, the joiner breaks away from the current position; (3) Scout position update: where β, K ∈ randn(0, 1) are used for step size control; f i is the fitness of the current individual; f b is the fitness of the current global best.

3. The photovoltaic cluster output optimization method based on scheduling error as described in claim 1, characterized in that: the steps S1-2 specifically is: According to the improved neural network from left to right f t , i t , o t respectively represent the forget gate, the input gate, and the output gate, and C t represents the current state; then there is: where: σ is the sigmoid function, and the output of sigmoid determines whether the information should be saved; W f 、W i 、W C 、W o are weight matrices; b f 、b i 、b C 、b o are bias parameter matrices; x t 、h t-1 represent the inputs at time t respectively, h t is the output, and C t represents the current state.

4. The photovoltaic cluster output optimization method based on scheduling error as described in claim 1, characterized in that: the steps S1-3 is specifically: The ambient temperature, light intensity, ambient humidity, weather conditions, wind speed and air pressure factors are used as features to construct a training feature set, and the output is the intraday forecast dispatch value of the photovoltaic cluster, which is as follows: where: t = (1, 2,... T) is the sampling time within the T interval; k = (1, 2,..., m) is the number of internal stations in the photovoltaic cluster; X k,t is the input feature at time t; environmental temperature T (T1, T2,..., T t ), light intensity SA (S1, S2,..., S t ), environmental humidity H (H1, H2,..., H t ), weather condition W (W1, W2,..., W t ), wind speed D (D1, D2,..., D t ), air pressure F (F1, F2,..., F t ); is the scheduling objective; the subscript i represents the i-th meteorological condition, t represents the time point, k represents the k-th station in the cluster, C represents the overall cluster; the superscript dis represents the scheduling output, real represents the actual output, opt represents the optimized output, for represents the predicted output value, ideal represents the ideal output, C represents the overall cluster, use represents the maximum available power; At this time, the total scheduling error can be written as: The dispatch error of each station can be written as: Let be the dispatching margin of substation k, that is, the difference between the ideal value and the dispatching value of the output at time t, describing the output margin of substation k after receiving dispatching; it can be written as:

5. The photovoltaic cluster output power optimization method based on scheduling error as claimed in claim 1, wherein: Step S2-1 constructs a typical meteorological output curve as follows: By selecting the theoretical output values ​​of each photovoltaic power station in the cluster on different typical days as reference values, four typical meteorological characteristics are obtained: sunny type, cloudy type, shower type, and cloudy type; The photovoltaic output data were clustered using the principle of maximum difference and minimum similarity under four climate environment categories, and the typical daily output curve of the station that can represent the meteorological characteristics of this type was obtained, which can be written as: Among them, is the typical output value at the nth moment in the basic typical daily curve under the ith meteorological condition.

6. The photovoltaic cluster output optimization method based on scheduling error according to claim 1, characterized in that: Step S2-2 constructs the station output ranking index system as follows: A. Controllable and measurable: Give the controllability and measurability α t Characterize whether the historical data of the station is complete. In addition, the controllability and measurability also characterize whether the station has the ability to generate electricity according to the dispatching value in history. Then, we have: Among them, t=(1, 2....T) are the sampling points of the photovoltaic power station data within the first 5 minutes, and P t dis is the scheduling value of the station output, and P t real is the actual output value of the station; the closer the controllability and measurability are to 1, the more controlled the station is, otherwise it means that the station may be in a state where real-time regulation is impossible. When the value is 0, it means that the station has been in a fault or maintenance state in history; B. Cosine similarity: Among any meteorological types, the actual curve P of the power station t real The closer it is to the ideal output curve P t ideal the more it meets the dispatching requirements, which is the most core index of optimal control. Using the cosine similarity β t represents the similarity between the actual output value and the ideal output value within the previous 5 minutes at time t, and T is the number of sampling points within the previous 5 minutes at time t; then: According to the above formula, the closer the similarity is to 1, the more similar the actual value at time t and the predicted value at time t are to the theoretical output curve before and after time t, which means the system has higher followability; C. Prediction accuracy: The predicted value is the mid-range value of the photovoltaic station that accepts dispatch and outputs according to demand. The prediction accuracy describes the similarity between the predicted output value and the actual output value. The higher the prediction accuracy, the more impressive the actual output of the station is, and it can better meet the dispatch demand. Then: D. Output compliance: Evaluate the output of the power station under the same meteorological conditions, that is, the output compliance γ t , namely: When the similarity is close to 1, when the output deviation γ t is closer to 0, it indicates that the output of the station is more similar to the ideal output. The greater the deviation, the worse the followability of the station, the lower the accuracy of the prediction result, and the lower the dispatchability; The above indicators are all values ​​between [0-1]. From the physical meaning of each indicator, it can be concluded that the closer each indicator is to the output of station 1, the more ideal it is. Therefore, the following ranking indicator lin is constructed: The description is as follows: (1) When the lin value is different, the output is based on the lin value; (2) When the lin values are the same, sort them in the order of the magnitudes of β t , γ t , α t respectively.

7. The photovoltaic cluster output optimization method based on scheduling error according to claim 1, wherein: The cluster output constraint optimization in step S3-1 is specifically as follows: To ensure the effectiveness and real-time performance of scheduling tracking, the time for rolling sorting is set to 5 minutes, that is, sorting is performed every 5 minutes according to the scheduling tracking error information; When the cluster scheduling tracking error is less than or equal to the margin of the first station, changing the output of the first station can meet the cluster scheduling tracking requirements. At this time, the first station operates in a following mode, coordinates its own output, and makes the scheduling error approach 0, which is called the following mode; When the total cluster scheduling tracking error is greater than the scheduling margin of the first station in the sorting, at this time, the first station operates at the maximum output that can be achieved, introduces the second output station, and compares the remaining scheduling error with the scheduling margin of the second station, and performs a loop, which is called the maximum output mode; Thus, the objective function Q can be set as: Its constraint conditions include the rate of change constraint and the cluster tracking optimization range constraint, and can be written as:

8. The photovoltaic cluster output optimization method based on scheduling error according to claim 1, characterized in that: The optimization of the station output in step S3-2 is specifically as follows: When the cluster scheduling error function is greater than the scheduling margin of the station, the station operates at the maximum output that can be achieved; when working in the following mode, the output is carried out in the pre-sorted order of the stations; the output mode is judged and adjusted every 1 minute; Then, when the station adopts the maximum output mode, the objective function can be written as: Where t = (1, 2,..., T1), which is the sampling time within 1 minute; When the station adopts the following mode, the minimum difference between the optimized output value of this station and the remaining scheduling target of the cluster can be used as the objective function, and can be written as: Where m1 is the number of stations that have been sorted before this station participates in the output; In addition, the station also needs to meet the photovoltaic output limit constraint and the output fluctuation constraint, as follows:

Images