Photovoltaic power generation power prediction and electric power system scheduling method and system for realizing photovoltaic power generation power prediction and electric power system scheduling method

Abstract

The invention discloses a photovoltaic power generation power prediction and power system scheduling method and a system for realizing the method. The method comprises four steps of data acquisition and preprocessing, similar day selection and training set construction, GA (Genetic Algorithm)-fuzzy RBF (Radial Basis Function) neural network modeling and power scheduling plan generation. The method comprises the following steps: selecting a meteorological variable with high correlation degree through a Spearman rank correlation coefficient, and constructing a time sequence feature; similar days are utilized to construct training samples, and the generalization ability of the model is improved; a fuzzy RBF network optimized by GA is adopted to enhance the nonlinear modeling precision; and inputting a prediction result into the power system simulation model, and generating an optimal scheduling instruction by adopting dynamic programming. According to the method, the photovoltaic power prediction accuracy and scheduling efficiency can be remarkably improved, the power grid stability is enhanced, and the new energy consumption capability is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation prediction and scheduling, and in particular to a photovoltaic power generation power prediction and power system scheduling method and a system for implementing the method. Background Art

[0002] As the proportion of renewable energy generation in new power systems continues to rise, its inherent intermittent and fluctuating output has become a key constraint on the system's stable operation. In this context, improving the accuracy of renewable energy power forecasting is particularly important. This not only affects the safe and reliable operation of the power grid but also provides a crucial technical foundation for achieving a clean energy transition.

[0003] Photovoltaic power generation is rapidly developing within the renewable energy power generation system. Photovoltaic power generation is affected by environmental factors such as solar radiation intensity, temperature, and cloud movement, resulting in significant randomness and volatility in output power. Traditional physical model predictions rely on precise meteorological data and photovoltaic module parameters, making them difficult to address complex weather changes. Statistical models such as linear regression and support vector machines (SVMs) are less capable of capturing nonlinear relationships and are prone to large errors. As photovoltaic penetration increases, the power grid needs to adjust the output of traditional units in real time to balance power fluctuations. Therefore, a high-precision prediction method, combined with a dynamic scheduling strategy, is urgently needed to improve grid operating efficiency and renewable energy absorption capacity. Summary of the Invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a photovoltaic power generation power prediction and power system scheduling method and a system for implementing the method, specifically a "forecasting-scheduling" integrated framework, which improves the deep learning performance of renewable energy power generation output prediction by analyzing the key influencing factors and output logic of the photovoltaic power generation prediction model, and helps to guide power system dispatching staff to timely adjust the renewable energy power generation and consumption strategy, thereby increasing the new energy consumption capacity of the power system and enhancing the competitiveness of power grid enterprises.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A photovoltaic power generation power forecasting and power system dispatching method, including data collection and preprocessing, similar day selection and training set construction, GA genetic algorithm-fuzzy RBF neural network modeling and optimization, and power system dispatching plan generation;

[0007] The data collection and preprocessing: obtain the historical operation data of the photovoltaic power station, including the solar radiation intensity I r 、Ambient temperature T a , PV module temperature T c , actual output power P sand grid dispatching demand parameters; the input variables were screened by the Spearman rank correlation coefficient, and the solar radiation intensity I that was significantly correlated with the output power was retained. r and ambient temperature T a Or PV module temperature T c As input; perform outlier cleaning and normalization on the data, and construct a time series feature vector;

[0008] The similar day selection and training set construction are as follows: based on the weather type of the day to be predicted, candidate days with the same weather type and similar dates are screened from the historical operation data of the photovoltaic power station; the Euclidean distance between the temperature of the candidate day and the day to be predicted is calculated, and the historical day with the smallest distance is selected as the similar day:

[0009]

[0010] Construct a training sample set, the input variables include the solar radiation intensity I of similar days rh (t), temperature T ah (t) and the predicted temperature T on the day to be predicted af (t);

[0011] The GA genetic algorithm-fuzzy RBF neural network modeling and optimization: construct a fuzzy radial basis function neural network, the input layer receives the input, the fuzzification layer divides the fuzzy subsets by K-means clustering, and the membership function adopts Gaussian distribution; the neural network weights are optimized by genetic algorithm, and the fitness function is the mean square error (MSE) between the predicted power and the actual power; the output layer generates the photovoltaic power prediction value P sf (t);

[0012] The power system dispatch plan generation method comprises the following steps: inputting the photovoltaic power generation power forecast value into the power system simulation model, combining it with the grid load demand, the operation constraints of the thermal power units, and the reserve capacity requirements; generating an optimal dispatch plan using a dynamic programming algorithm, with the objective function being to minimize the operating cost and the photovoltaic power generation power deviation; and outputting dispatch instructions, including the output adjustment of the thermal power units, the renewable energy consumption strategy, and the reserve capacity allocation plan.

[0013] As a further improvement of the present invention, the photovoltaic power generation power prediction and power system scheduling method is characterized in that: the Spearman rank correlation coefficient is the solar radiation intensity correlation coefficient and the temperature correlation coefficient, the screening criteria are the solar radiation intensity correlation coefficient ≥ 0.9, the temperature correlation coefficient ≥ 0.35, and the remaining variables are eliminated when the correlation coefficient is < 0.2; the time series feature vector includes the sliding window mean of the previous 24 hours, the trend term and the periodic component.

[0014] As a further improvement of the present invention, the average value of the sliding window in the first 24 hours is The trend term is: fitting the time series slope through linear regression to characterize the direction of power change; the periodic component is: extracting the daily cycle and seasonal cycle components using Fourier transform.

[0015] As a further improvement of the present invention, in the selection of similar days, the weather type classification is based on the Hottel model, and a sunny day is defined as a daily total radiation intensity ≥85% of the theoretical maximum value and a cloud coverage rate ≤10%; there are the following classification standards for non-sunny days: a cloudy day is defined as a cloud coverage rate of 10% to 70%, and a light intensity fluctuation frequency of >2 times / hour; a rainy day is defined as a cloud coverage rate >70%, or a precipitation probability >60%.

[0016] As a further improvement of the present invention, the photovoltaic power generation prediction and power system scheduling method is characterized in that: when the day to be measured is cloudy, in addition to the Euclidean distance of temperature, an additional light intensity mutation matching index is introduced:

[0017]

[0018] Among them, ΔI r,h (t) is the first-order difference of the radiation intensity of a similar historical day in time period t, which is used to characterize the variation of the radiation intensity in adjacent time periods of the historical day and reflect the instantaneous fluctuation characteristics caused by cloud movement. r,f (t) is the first-order difference of the radiation intensity of the day to be predicted in time period t, N is the total number of sampling points in the evaluation period; S i is the absolute average deviation of the light intensity change pattern between the historical day and the predicted day, that is, the light intensity mutation matching degree. The light intensity mutation matching degree S is selected i The smallest historical day is used as the similarity day.

[0019] As a further improvement of the present invention, in the GA-fuzzy RBF neural network modeling, the number of fuzzy subsets and the total number of fuzzy rules of the fuzzy RBF neural network are determined; the membership product of each rule corresponding to the input variable is multiplied; the population size of the genetic algorithm is specified; and the crossover rate is adaptively adjusted according to the algebra, and the formula is:

[0020]

[0021] Among them, P c is the adaptive crossover rate of the genetic algorithm, g is the current genetic generation, and G is the total generation of the genetic algorithm;

[0022] The genetic algorithm introduces an elite retention strategy, which retains the top 5% of individuals in fitness in each generation and directly enters the next generation.

[0023] As a further improvement of the present invention, in the generation of the power system dispatch plan, the time resolution of the dynamic programming algorithm is established, and the constraints are defined, including the start and stop time of the thermal power unit, the ramp rate limit, and the standby load limit; and the objective function is defined as:

[0024]

[0025] Where t refers to a fixed time interval of 15 minutes, which is used to discretize the scheduling time; the total scheduling period of the prediction is T, t∈{1,2,…,T / Δt}; C gen (t) is the power generation cost in period t, λ is the photovoltaic power generation power deviation penalty coefficient, P grid (t) is the actual grid-connected power at time t, P pred (t) is the predicted grid-connected power at time t; the inclusion of t makes the optimization process time-series-dependent, ensuring that scheduling decisions are continuous across time periods and that dynamic constraints such as ramp rates are met. The real-time update mechanism for scheduling instructions includes receiving the latest forecast data every 15 minutes. If the actual power deviates from the predicted power by more than 10%, the scheduling plan is reoptimized.

[0026] A system for realizing photovoltaic power generation power prediction and power system scheduling method, comprising:

[0027] Data management module: integrates meteorological data, historical power and grid parameters;

[0028] Prediction engine: performs GA-fuzzy RBF neural network calculations;

[0029] Scheduling Optimization Module: Runs dynamic programming algorithms to generate optimal scheduling plans; supports multi-timescale scheduling, including day-ahead planning and real-time scheduling;

[0030] Interface module: transmits dispatching instructions to the power grid control system, complies with the IEC 61970 standard, and supports the Common Information Model (CIM) data format.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The photovoltaic power generation power prediction and power system dispatching method of the present invention improves the performance of the deep learning model for renewable energy power generation output prediction by analyzing the key influencing factors and output logic of photovoltaic power generation power prediction, which helps to guide power system dispatching staff to timely adjust the renewable energy power generation and consumption strategy, thereby increasing the renewable energy consumption capacity of the power system and enhancing the competitiveness of power grid enterprises.

[0033] The present invention reduces the impact of power fluctuations on the power grid through high-precision prediction, thereby improving the stability of the power grid; optimizing the unit output and standby configuration, reducing fuel consumption and equipment loss, and lowering operating costs; the integrated prediction and scheduling mode can dynamically schedule strategies to maximize photovoltaic utilization, reduce abandoned light, and promote the consumption of renewable energy. The scheduling output is highly compatible: it supports seamless integration with existing SCADA systems and EMS platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 Schematic diagram of the method of the present invention and a system for implementing the method. DETAILED DESCRIPTION

[0036] In order to make the purpose and technical solution of the present invention clearer and easier to understand, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0037] The technical solutions of the present invention will be described clearly and completely below with reference to the accompanying drawings and specific embodiments. It is obvious that the embodiments described are only a portion of the embodiments of the present invention, rather than all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are considered to fall within the scope of the present invention.

[0038] Photovoltaic power generation is rapidly developing within the renewable energy power generation system. Photovoltaic power generation is affected by environmental factors such as solar radiation intensity, temperature, and cloud movement, resulting in significant randomness and volatility in output power. Traditional physical model predictions rely on precise meteorological data and photovoltaic module parameters, making them difficult to address complex and sudden weather changes. Statistical models such as linear regression and support vector machines (SVMs) are less capable of capturing nonlinear relationships and are prone to large errors. As photovoltaic penetration increases, the power grid needs to adjust the output of traditional units in real time to balance power fluctuations. Therefore, a high-precision prediction method, combined with a dynamic scheduling strategy, is urgently needed to improve grid operating efficiency and renewable energy absorption capacity.

[0039] The present invention proposes a new integrated prediction and scheduling model, which can solve the problem that the current typical scheduling method is usually based on fixed prediction values ​​and lacks a dynamic adjustment mechanism, resulting in a high curtailment rate. The prediction and scheduling method provided by the present invention can achieve stable operation and economical scheduling of the power system.

[0040] See also Figure 1 , Figure 1 This is a schematic diagram of the steps of the implementation process of the present invention. Figure 1 As shown, the present invention provides a photovoltaic power generation power prediction and power system scheduling method, which includes four steps: data collection and preprocessing, dynamic matching of similar days, GA genetic algorithm-fuzzy RBF neural network modeling, and power system scheduling plan generation.

[0041] Data collection and preprocessing to obtain historical operating data of photovoltaic power plants, including solar radiation intensity (I r ), ambient temperature (T a ), PV module temperature (T c ), actual output power (P s ) and grid dispatch demand parameters;

[0042] The input variables were screened by the Spearman rank correlation coefficient, and the solar radiation intensity (I r ) and temperature (T a or T c ) as input; perform outlier cleaning and normalization on the data, and construct a time series feature vector;

[0043] Specifically, outlier cleaning was performed on the data, and the 3σ principle was adopted to remove data points that exceeded ±3 times the standard deviation of the mean.

[0044] Specifically, the data normalization process uses Min-Max normalization, and the formula is: And construct the time series feature vector based on the normalized data.

[0045] Similar day selection and training set construction: Based on the weather type of the day to be predicted, candidate days with the same weather type and similar dates are selected from the historical operation data of the PV power station. The Euclidean distance between the temperature of the candidate day and the day to be predicted is calculated, and the historical day with the smallest distance is selected as the similar day:

[0046]

[0047] Specifically, the input variables for the training sample construction include the solar radiation intensity I at time t on a similar day rh (t), temperature T ah (t) and the predicted temperature T on the day to be predicted af (t). The output variable is the actual power P at time t on the similar day sh (t).

[0048] The GA genetic algorithm-fuzzy RBF neural network modeling and optimization part is mainly divided into building a fuzzy radial basis function neural network and genetic algorithm optimization. In building a fuzzy radial basis function neural network, the input layer receives the input, the fuzzification layer divides the fuzzy subsets by K-means clustering, and the membership function adopts Gaussian distribution; the neural network weights are optimized by genetic algorithm, and the fitness function is the mean square error (MSE) between the predicted power and the actual power; the output layer generates the photovoltaic power prediction value P sf (t);

[0049] Specifically, the input layer of the fuzzy RBF neural network receives the solar radiation intensity I rh (t), temperature T ah (t) and the predicted temperature T on the day to be predicted af (t). Each input is divided into three fuzzy subsets: high, medium, and low. The total number of fuzzy rules is 27; each rule corresponds to the membership product of the input variables. The activation strength α of the kth rule k is the input variable membership μ ij (x i ), that is:

[0050]

[0051] The fuzzification layer is divided into fuzzy subsets by K-means clustering, and the membership function adopts Gaussian distribution;

[0052]

[0053] Among them, x i Refers to the specific value of the i-th input variable after normalization, and x1 refers to the solar radiation intensity I at the corresponding time t rh (t), x2 refers to the ambient temperature Tah(t), and x3 refers to the actual power P at the corresponding time on a similar day sh (t);c ij Refers to the cluster center of the i-th input variable in the j-th fuzzy subset; σ ij is the width of the membership function of the i-th input variable in the j-th fuzzy subset, which is used to control the coverage of the fuzzy subset and determine the decay rate of the membership degree; μ ij Refers to input x i The membership degree of the jth fuzzy subset, where j = 1 corresponds to the low fuzzy subset, j = 2 corresponds to the medium fuzzy subset, and j = 3 corresponds to the high modulus subset.

[0054] Specifically, the fuzzy inference layer of the neural network is designed with weights w1~w 27 The encoding is a real number chromosome, the population size is 300, and the fitness function is the mean square error (MSE) between the predicted power and the actual power, that is,

[0055]

[0056] Among them, MSE refers to the mean square error between the predicted power and the actual power, reflecting the overall accuracy of the model; M is the total number of training samples; P sf,n Refers to the predicted power value of the neural network for the nth sample; P sd,n Refers to the actual measured power value corresponding to the nth sample; Fitness is used to convert the MSE into a maximization problem, that is, mapping the MSE of (0, +∞) to the interval (0, 1], which facilitates the selection operation and avoids the risk of division by zero.

[0057] Specifically, to ensure a balance between global search capability and computational efficiency, the crossover rate is adaptively adjusted algebraically, as follows:

[0058]

[0059] Among them, P c is the adaptive crossover rate of the genetic algorithm, g is the current genetic generation, and G is the total number of generations in the genetic algorithm. A high crossover rate of 0.8 is used initially to promote exploration, and later reduced to 0.5 to refine the search. The algorithm incorporates an elite retention strategy, retaining the top 5% of fitness individuals in each generation to advance directly to the next generation, preventing loss of high-quality genes.

[0060] The power system dispatcher inputs the predicted value of photovoltaic power generation into the power system simulation model, combined with the grid load demand, thermal power unit operating constraints and reserve capacity requirements; uses a dynamic programming algorithm to generate the optimal dispatch plan, with the objective function of minimizing operating costs and power deviation; and outputs dispatch instructions, including thermal power unit output adjustment, renewable energy absorption strategy and reserve capacity allocation plan.

[0061] Specifically, the data preprocessing uses the Spearman rank correlation coefficient method to analyze the specific screening threshold value set as: solar radiation intensity I r Correlation coefficient ≥ 0.9; Temperature T a or T c The correlation coefficient was ≥0.35; other variables (such as wind speed and humidity) were excluded when the correlation coefficient was <0.2.

[0062] Specifically, the time series feature construction method is to perform a sliding window analysis on the input variables with a window length of 24 hours and extract the following features: Sliding window mean:

[0063] Specifically, the trend term is characterized by fitting the slope of the time series through linear regression to represent the direction of power change. The periodic component is extracted using Fourier transforms to capture the 24-hour daily cycle and the 365-day seasonal cycle. This enhances the model's ability to capture temporal patterns and reduces forecast bias caused by sudden weather changes.

[0064] Specifically, based on the Hottel model, a clear day is defined as one with a daily total radiation intensity ≥ 85% of the theoretical maximum and cloud cover ≤ 10%. Daily total radiation is calculated based on geographic location and season, while cloud cover is determined using meteorological satellite data or ground observations. Non-sunny days are classified using the following criteria: a cloudy day is defined as one with a cloud cover of 10% to 70% and a light intensity fluctuation frequency of more than two times per hour; a rainy day is defined as one with a cloud cover of more than 70% or a probability of precipitation greater than 60%.

[0065] Specifically, the day to be measured is cloudy. In addition to the temperature Euclidean distance, an additional light intensity mutation matching index is introduced:

[0066]

[0067] Among them, ΔI r,h (t) The first-order difference of the radiation intensity of a historically similar day in time period t, i.e. ΔI r,h (t) = I r,h (t+1)-I r,h (t), which is used to characterize the variation of radiation intensity between adjacent time periods of historical days and reflects the instantaneous fluctuation characteristics caused by cloud movement, ΔI r,f (t) is the first-order difference of the radiation intensity of the day to be predicted in time period t, that is, ΔI r,f (t) = I r,f (t+1)-I r,f (t), N is the total number of sampling points during the evaluation period, and it is necessary to ensure that the historical day and the predicted day are aligned. Select the light intensity mutation matching degree S i The smallest historical day is used as the similarity day.

[0068] Specifically, in the power system dispatch plan generation part, the time resolution of the dynamic programming algorithm is established to be 15 minutes, consistent with the output of the prediction model; the constraints are set, including the start and stop time of the thermal power units to avoid frequent start and stop losses, the ramp rate limit is ≤±5% rated power / minute, and to meet the N-1 safety criterion, the reserve capacity is limited to ≥ 10% of the maximum load. The objective function is defined as:

[0069]

[0070] Where t refers to a fixed time interval of 15 minutes, which is used to discretize the scheduling time. The total scheduling period for prediction is T, t∈{1,2,…,T / Δt}. gen (t) is the power generation cost at time t, λ is the photovoltaic power generation power deviation penalty coefficient, P grid (t) is the actual grid-connected power at time t, P pred(t) is the predicted grid-connected power at time t; the inclusion of t makes the optimization process time-series-dependent, ensuring that scheduling decisions are continuous across time periods and that dynamic constraints such as ramp rates are met. The real-time update mechanism for dispatch instructions includes receiving the latest forecast data every 15 minutes. If the actual power deviates from the predicted power value by more than 10%, the following actions are immediately triggered: the current dispatch cycle is interrupted, the optimization algorithm is re-run based on the latest data, and correction instructions are pushed to the grid control system.

[0071] The photovoltaic power generation power prediction method and power system dispatching system is an integrated "forecasting-dispatching" framework.

[0072] The system includes:

[0073] Data management module: integrates meteorological data, historical power and grid parameters.

[0074] Specifically, data source integration includes: meteorological data (irradiance, temperature, cloud cover), historical power curves, and grid parameters (load demand, unit capacity); database architecture: using a time series database (such as InfluxDB) to support high-speed reading and writing and complex queries.

[0075] The prediction engine is divided into hardware configuration and software implementation. The hardware uses GPU accelerated computing nodes and supports parallel genetic algorithm operations. The software builds a fuzzy RBF neural network based on the TensorFlow framework and is trained based on one year of scheduling and weather data.

[0076] Scheduling Optimization Module: This module runs dynamic programming algorithms to generate optimal scheduling plans. It supports multi-timescale scheduling, including day-ahead scheduling with 24-hour resolution and real-time scheduling with 15-minute resolution. The algorithm library integrates dynamic programming, mixed-integer linear programming (MILP), and heuristic algorithms.

[0077] Interface module: transmits dispatch instructions to the power grid control system, complies with the IEC 61970 standard, and supports the CIM (Common Information Model) data format.

[0078] To address grid connection stability issues caused by insufficient photovoltaic power forecasting accuracy, this paper proposes a hybrid forecasting model based on genetic algorithm (GA) optimization and fuzzy radial basis function (RBF) neural networks. By integrating data-driven modeling with intelligent optimization algorithms, the model significantly improves the accuracy of photovoltaic output forecasts, providing reliable data support for grid scheduling.

[0079] In model construction, solar radiation intensity and temperature were first selected as key input variables based on correlation analysis, and a dynamic matching strategy for similar days was incorporated to optimize the quality of training samples. A fuzzy RBF neural network structure was further designed, using K-means clustering to adaptively partition the input space. A genetic algorithm was then used to globally optimize network weights, effectively overcoming the vulnerability of traditional neural networks to local optima. This hybrid model accurately captures the nonlinear relationship between photovoltaic output and meteorological factors, making it particularly well-suited for forecasting complex weather scenarios such as cloudy and rainy weather.

[0080] In grid-connected applications, highly accurate PV power forecasts can directly inform power system scheduling, significantly reducing redundant backup capacity and frequent unit adjustments caused by forecast bias. Furthermore, the time-series power curves generated by the model can be used as input for grid stability analysis, assisting in assessing system operational risks under high PV integration. By improving forecast reliability, this method mitigates the impact of PV fluctuations on the grid at its source, providing technical support for the efficient integration of renewable energy in new power systems.

[0081] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0082] It should be understood that the above description is for illustrative purposes only and is not intended to be limiting. Many embodiments and many applications beyond the examples provided will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of the present teachings should not be determined with reference to the above description, but rather with reference to the preceding claims and the full scope of equivalents to which such claims are entitled. For the purpose of completeness, all articles and references, including disclosures of patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein from the preceding claims is not a disclaimer of such subject matter, nor should it be considered that the applicants did not consider such subject matter to be part of the disclosed inventive subject matter.

[0083] The above are only preferred embodiments of the present invention and do not limit the present invention in any way. Any simple modifications, changes and equivalent structural changes made to the above embodiments based on the technical essence of the present invention are still within the scope of protection of the technical solution of the present invention.

[0084] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the same. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art may still modify or make equivalent substitutions for the specific implementation schemes of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention are within the scope of protection of the claims of the present invention.

Claims

1. A photovoltaic power generation power prediction and power system scheduling method, characterized by: Including data collection and preprocessing, similar day selection and training set construction, GA genetic algorithm-fuzzy RBF neural network modeling and optimization, and power system dispatch plan generation; The data collection and preprocessing: obtain the historical operation data of the photovoltaic power station, including the solar radiation intensity I r 、Ambient temperature T a , PV module temperature T c , actual output power P s and grid dispatching demand parameters; the input variables were screened by the Spearman rank correlation coefficient, and the solar radiation intensity I that was significantly correlated with the output power was retained. r and ambient temperature T a Or PV module temperature T c As input; perform outlier cleaning and normalization on the data, and construct a time series feature vector; The similar day selection and training set construction are as follows: based on the weather type of the day to be predicted, candidate days with the same weather type and similar dates are screened from the historical operation data of the photovoltaic power station; the Euclidean distance between the temperature of the candidate day and the day to be predicted is calculated, and the historical day with the smallest distance is selected as the similar day: Construct a training sample set, the input variables include the solar radiation intensity I of similar days rh (t), temperature T ah (t) and the predicted temperature T on the day to be predicted af (t); The GA genetic algorithm-fuzzy RBF neural network modeling and optimization: construct a fuzzy radial basis function neural network, the input layer receives the input, the fuzzification layer divides the fuzzy subsets by K-means clustering, and the membership function adopts Gaussian distribution; the neural network weights are optimized by genetic algorithm, and the fitness function is the mean square error (MSE) between the predicted power and the actual power; the output layer generates the photovoltaic power prediction value P sf (t); The power system dispatch plan generation method comprises the following steps: inputting the photovoltaic power generation power forecast value into the power system simulation model, combining it with the grid load demand, the operation constraints of the thermal power units, and the reserve capacity requirements; generating an optimal dispatch plan using a dynamic programming algorithm, with the objective function being to minimize the operating cost and the photovoltaic power generation power deviation; and outputting dispatch instructions, including the output adjustment of the thermal power units, the renewable energy consumption strategy, and the reserve capacity allocation plan.

2. The photovoltaic power generation power prediction and power system scheduling method according to claim 1, characterized in that: The Spearman rank correlation coefficient is the correlation coefficient of solar radiation intensity and temperature. The screening criteria are solar radiation intensity correlation coefficient ≥ 0.9, temperature correlation coefficient ≥ 0.35, and other variables are excluded when the correlation coefficient is < 0.2; The time series feature vector includes the sliding window mean of the previous 24 hours, a trend term, and a periodic component.

3. The photovoltaic power generation power prediction and power system scheduling method according to claim 2, characterized in that: The sliding window average of the previous 24 hours The trend term is: fitting the time series slope through linear regression to characterize the direction of power change; the periodic component is: extracting the daily cycle and seasonal cycle components using Fourier transform.

4. The photovoltaic power generation power prediction and power system scheduling method according to claim 1, characterized in that: In the selection of similar days, the weather type classification is based on the Hottel model. A sunny day is defined as a day with a total daily radiation intensity ≥85% of the theoretical maximum and a cloud cover rate ≤10%. For non-sunny days, the following classification standards are used: a cloudy day is defined as a cloud cover rate of 10% to 70% and a light intensity fluctuation frequency of >2 times / hour; a rainy day is defined as a cloud cover rate >70% or a precipitation probability >60%.

5. The photovoltaic power generation power prediction and power system scheduling method according to claim 4, characterized in that: The day to be measured is cloudy. In addition to the Euclidean distance of temperature, an additional light intensity mutation matching index is introduced: where ΔI r,h (t) is the first-order difference of the radiation intensity of a similar historical day in time period t, which is used to characterize the variation of the radiation intensity between adjacent time periods of a historical day and reflect the instantaneous fluctuation characteristics caused by cloud movement. r,f (t) is the first-order difference of the radiation intensity of the day to be predicted in time period t, N is the total number of sampling points in the evaluation period; S i is the absolute average deviation of the light intensity change pattern between the historical day and the predicted day, that is, the light intensity mutation matching degree. The light intensity mutation matching degree S is selected i The smallest historical day is used as the similarity day.

6. The photovoltaic power generation power prediction and power system scheduling method according to claim 1, characterized in that: In the GA-fuzzy RBF neural network modeling, the number of fuzzy subsets and the total number of fuzzy rules of the fuzzy RBF neural network are determined; the membership product of each rule corresponding to the input variable is multiplied; the population size of the genetic algorithm is specified; and the crossover rate is adaptively adjusted according to the algebra. The formula is: Among them, P c is the adaptive crossover rate of the genetic algorithm, g is the current genetic generation, and G is the total generation of the genetic algorithm; The genetic algorithm introduces an elite retention strategy, which retains the top 5% of individuals in fitness in each generation and directly enters the next generation.

7. The photovoltaic power generation power prediction and power system scheduling method according to claim 1, characterized in that: In the generation of the power system dispatch plan, the time resolution of the dynamic programming algorithm is established, and the constraints are defined, including the start and stop time of the thermal power units, the ramp rate limit, and the standby load limit; the objective function is defined as: Where t refers to a fixed time interval of 15 minutes, which is used to discretize the scheduling time; the total scheduling period of the prediction is T, t∈{1,2,…,T / Δt}; C gen (t) is the power generation cost in period t, λ is the photovoltaic power generation power deviation penalty coefficient, P grid (t) is the actual grid-connected power at time t, P pred (t) is the predicted grid-connected power at time t; the introduction of t makes the optimization process time-series-related, ensuring that scheduling decisions in each time period are continuous and meet dynamic constraints such as ramp rate; the real-time update mechanism for scheduling instructions includes receiving the latest data forecast every 15 minutes. If the actual power deviates from the power forecast value by more than 10%, the scheduling plan is triggered to be re-optimized.

8. A system for implementing the photovoltaic power generation power prediction and power system scheduling method according to any one of claims 1 to 7, characterized in that: The system includes: Data management module: integrates meteorological data, historical power and grid parameters; Prediction engine: performs GA-fuzzy RBF neural network calculations; Scheduling optimization module: runs dynamic programming algorithm to generate the optimal scheduling plan; Supports multi-timescale scheduling, including day-ahead planning and real-time scheduling; Interface module: transmits dispatching instructions to the power grid control system, complies with the IEC 61970 standard, and supports the Common Information Model (CIM) data format.

Images