A management method for improving power generation capacity of a photovoltaic power station

Abstract

The present application relates to the technical field of photovoltaic power station intelligent scheduling, and more particularly to a management method for improving the power generation capacity of a photovoltaic power station, which comprises: analyzing power regulation policy texts, meteorological warning data and historical price fluctuation data to generate a three-dimensional dynamic constraint condition set; using an LSTM time sequence model to generate a basic power generation output prediction sequence, superimposing the three-dimensional constraints to generate a policy-adapted dispatchable capacity; screening adjacent reference power stations based on a digital twin model to correct a meteorologically corrected dispatchable capacity; starting in-station compensation when the target power station's power generation output is insufficient; screening auxiliary power stations and generating an incremental power generation instruction set when the compensation is insufficient; using a consortium chain to store the instructions and results; and achieving cross-station collaborative training through federated learning. The present application can effectively improve the power generation capacity of a photovoltaic power station and the efficiency of power grid scheduling, and solves the problems of power generation loss caused by meteorological fluctuations and low cross-station collaboration efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent dispatching technology for photovoltaic power plants, and in particular to a management method for improving the power generation capacity of photovoltaic power plants. Background Technology

[0002] In the early stages, photovoltaic power plant management relied primarily on manual inspections and simple logic controls, enabling only basic start-up and shutdown operations and fault alarms. Managers operated and scheduled power plants based on fixed schedules or simple weather forecasts. Responses to power regulation policies, weather warnings, and electricity price fluctuations were neither timely nor accurate, making it difficult to establish effective dynamic constraints to guide power plant operation. This limited power generation efficiency and the stability of power supply. Furthermore, early data analysis methods failed to deeply mine and accurately predict historical power generation data from photovoltaic power plants, resulting in low accuracy in predicting dispatchable capacity and impacting the scientific and rational nature of grid dispatching.

[0003] With the rapid development of the photovoltaic industry, photovoltaic power plants are gradually increasing their proportion in the energy supply system, placing higher demands on their management technology. The stability of power supply and power generation efficiency have become key concerns, requiring more advanced data processing and analysis technologies to optimize the operation and management of photovoltaic power plants. Some power generation forecasting methods based on simple models have begun to emerge in the market. However, these technologies have significant drawbacks: firstly, they cannot decouple the combined effects of meteorological factors (such as irradiance and temperature) on system efficiency, resulting in prediction error rates generally exceeding 20%; secondly, they lack cross-site collaborative dispatch mechanisms, making it impossible to cope with the grid stability requirements of large-scale clusters.

[0004] Currently, although some solutions attempt to combine digital twin technology or blockchain for evidence storage, technical bottlenecks remain: Existing technologies primarily rely on a single physical model to decouple meteorological deviations, without incorporating historical data from neighboring power plants for comparative analysis. This results in insufficient correction accuracy, leading to inaccurate corrections of meteorological factors and hindering the precise assessment and improvement of power generation capacity. Furthermore, the lack of simultaneous quantification of both ramp-up rate and installed capacity constraints during cross-station dispatching easily triggers grid voltage fluctuations, making it difficult to guarantee grid stability when multiple power plants are coordinating power generation. In existing technologies, policy compliance verification and model optimization lack automated mechanisms, and data sharing among multiple power plants poses a privacy risk. This not only increases management costs but also limits model optimization and widespread adoption.

[0005] Chinese invention CN116526553B discloses a management method and system for improving the power generation capacity of photovoltaic power plants. It analyzes the lost power through modeling and simulation technology and loss algorithm model and generates operation and maintenance plans. However, this invention does not involve a meteorological deviation compensation strategy within the station and lacks cross-station collaborative scheduling logic for generating incremental power generation instructions. It only passively responds to losses through operation and maintenance plans.

[0006] Therefore, this invention discloses a management method for improving the power generation capacity of photovoltaic power plants. Summary of the Invention

[0007] The purpose of this invention is to address the problems in the existing technology of photovoltaic power plants, such as severe power generation losses due to weather fluctuations, low cross-station coordination efficiency, and difficulty in verifying policy compliance, by proposing a management method to improve the power generation capacity of photovoltaic power plants.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a management method for improving the power generation capacity of a photovoltaic power station, comprising the following steps:

[0009] Step S1: Analyze the power regulation policy text, meteorological early warning data, and historical electricity price fluctuation data to generate a three-dimensional dynamic constraint set;

[0010] Step S2: Based on the mapping relationship between real-time meteorological parameters and power generation output data, a basic power generation output prediction sequence is generated through an LSTM time series model, and a three-dimensional dynamic constraint set is superimposed to generate policy-adaptive dispatchable capacity.

[0011] Step S3: Based on the digital twin model, select neighboring reference power plants, and correct the policy-adapted dispatchable capacity through meteorological deviation decoupling analysis, and output the meteorological-corrected dispatchable capacity.

[0012] Step S4: When the real-time power output of the target power station is lower than 8% of the weather-corrected dispatchable capacity for 15 consecutive minutes, the station's weather deviation compensation strategy is activated, and corresponding compensation control is executed according to the type of weather deviation.

[0013] Step S5: If the power output of the target power station continues to be insufficient after compensation, auxiliary power stations are selected by calculating the available margin of nearby reference power stations, and an incremental power generation instruction set is generated with the goal of maximizing grid stability and power generation efficiency.

[0014] Step S6: The incremental power generation instruction set and execution results are reliably stored through the consortium blockchain, and policy compliance is automatically verified based on smart contracts to generate audit records.

[0015] Step S7: Optimize model parameters based on consortium blockchain audit records, and use a cryptographic differential privacy federated learning mechanism to achieve cross-power station collaborative training.

[0016] The beneficial effects of the technical solution provided by this invention include at least the following:

[0017] This invention combines an LSTM time series model with a three-dimensional dynamic constraint set to accurately predict power generation output and correct it in real time, thereby reducing the prediction error rate.

[0018] This invention uses a digital twin model and a decoupled analysis of meteorological deviations from a nearby reference power station to quickly identify the dominant factors causing temperature and irradiance deviations and implement targeted compensation strategies to reduce power generation losses caused by meteorological fluctuations.

[0019] This invention, through the collaborative scheduling of available margin calculation and multi-objective optimization algorithm, can dynamically allocate incremental power of auxiliary power stations and improve the efficiency of power generation clusters while ensuring grid stability.

[0020] This invention, through a federated learning mechanism with encrypted differential privacy, can achieve collaborative optimization of model parameters while protecting the data privacy of each power station, thereby improving the prediction accuracy of the global model. Attached Figure Description

[0021] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a management method for improving the power generation capacity of a photovoltaic power plant, provided as an embodiment of the present invention. Detailed Implementation

[0023] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a management method for improving the power generation capacity of a photovoltaic power station according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0026] The following description, in conjunction with the accompanying drawings, details a specific scheme for a management method to improve the power generation capacity of a photovoltaic power station provided by the present invention.

[0027] Please see Figure 1 The diagram illustrates a management method for improving the power generation capacity of a photovoltaic power plant according to an embodiment of the present invention. The method includes the following steps:

[0028] Step S1: Analyze the power regulation policy text, meteorological early warning data, and historical electricity price fluctuation data to generate a three-dimensional dynamic constraint set;

[0029] Step S1 further includes the following sub-steps:

[0030] S1-1 uses natural language processing to parse the power regulation policy text, extract peak-valley electricity price periods, carbon emission indicators, and peak-shaving and frequency regulation requirements, and generates a policy signal feature vector.

[0031] S1-2 processes meteorological early warning data through a spatiotemporal data fusion algorithm to generate meteorological impact maps covering each power station;

[0032] S1-3, Construct an electricity price fluctuation prediction model to predict the electricity price fluctuation trend in the future T period based on historical electricity price fluctuation data and policy signal feature vectors;

[0033] S1-4 integrates policy signal feature vectors, meteorological impact maps, and electricity price fluctuation trends to generate a three-dimensional dynamic constraint set including time, space, and policy dimensions.

[0034] It should be noted that when performing semantic understanding and feature extraction on power regulation policy texts using natural language processing technology, a BERT pre-trained model is specifically used to encode word vectors in the policy texts. Named entity recognition technology is used to identify key information such as peak-valley electricity price periods, carbon emission indicators, and peak-shaving and frequency regulation requirements. Finally, a policy signal feature vector containing the core elements of the policy is generated. The policy signal feature vector is coupled with the subsequent electricity price prediction model through an attention mechanism to realize the dynamic guidance of the policy signal on dispatching decisions.

[0035] The steps for processing meteorological early warning data using a spatiotemporal data fusion algorithm to generate meteorological impact maps covering various power stations include:

[0036] The inverse distance weighting method was used to perform weighted interpolation on the irradiance, temperature, and wind speed data of Solargis meteorological sources to construct a regional meteorological field matrix;

[0037] Temporal correlation of meteorological sequences is extracted by sliding window, and short-term fluctuation trends are captured by LSTM.

[0038] Using the power station's geographical location as nodes and meteorological similarity as edge weights, a visualized meteorological impact map is generated, which intuitively reflects the meteorological correlation within the region.

[0039] The specific steps for predicting electricity price fluctuation trends by constructing an LSTM time series model include:

[0040] Historical electricity price data and policy signal feature vectors are concatenated and standardized to the [0,1] interval using Min-Max;

[0041] The nonlinear fluctuation pattern of electricity price series is learned by using an LSTM time series model, and the policy signal feature vector is weighted and incorporated into the LSTM hidden layer through an attention mechanism.

[0042] Based on the time step of the power grid dispatch cycle adjustment model, short-term forecasts (T≤4 hours) focus on high-frequency fluctuation characteristics, while long-term forecasts (T>24 hours) combine trend guidance from policy signals.

[0043] The steps for generating a three-dimensional constraint set include:

[0044] Z-Score standardization was applied to the policy feature vector (numerical), meteorological impact map (spatial matrix), and electricity price fluctuation trend (time series).

[0045] In the time dimension, time-of-use power economy ranges are generated based on electricity price fluctuation trends;

[0046] In the spatial dimension, the regional power reduction coefficient is set based on the meteorological impact map;

[0047] The policy dimension, generated based on policy signal feature vectors, includes:

[0048] Power output range is set based on peak and off-peak electricity pricing periods;

[0049] Allocation of power plant-level carbon quotas based on carbon emission indicators;

[0050] Determine the ramp rate threshold based on peak shaving and frequency modulation requirements;

[0051] Constraints from all dimensions are integrated into a three-dimensional constraint matrix, and weighted summation is used to generate the final constraints.

[0052] Step S2: Based on the mapping relationship between real-time meteorological parameters and power generation output data, a basic power generation output prediction sequence is generated through an LSTM time series model, and a three-dimensional dynamic constraint set is superimposed to generate policy-adaptive dispatchable capacity.

[0053] Step S2 further includes the following sub-steps:

[0054] S2-1 collects real-time power generation output, historical power generation output sequence, real-time meteorological parameters and equipment status data of each power station;

[0055] S2-2 uses an LSTM time series model to learn the mapping relationship between real-time meteorological parameters and real-time power generation output, and outputs a basic power generation output prediction sequence.

[0056] S2-3, superimposed three-dimensional dynamic constraint condition set to generate policy-adaptive schedulable capacity.

[0057] It should be noted that the historical power generation output sequence collects minute-level power generation output data for each power station for at least 3 years, covering all operating conditions such as normal operation and fault maintenance.

[0058] Real-time meteorological parameters are obtained simultaneously, including indicators such as irradiance, component temperature, wind speed, and humidity. Data sources include on-site weather stations and third-party meteorological services.

[0059] Equipment status data, including inverter efficiency, photovoltaic module temperature coefficient, and bracket tilt angle, is collected and transmitted in real time through the SCADA system.

[0060] The collected data is preprocessed, including outlier filtering, normalization, and time-series alignment.

[0061] The LSTM timing model architecture design includes:

[0062] The input layer has 12 feature dimensions (including 3 dimensions of historical power, 6 dimensions of meteorological parameters, and 3 dimensions of equipment status), and the time step is set to 48 (corresponding to 12 hours of historical data).

[0063] Hidden layers, each containing 128 LSTM units, with tanh activation function, and Dropout (ratio 0.2) between layers to prevent overfitting;

[0064] The output layer, a fully connected layer, is mapped to the predicted power value for a future time period T (T∈[24,168] ​​hours), with an output dimension of T×1.

[0065] The loss function uses a combination of mean squared error and mean absolute error, expressed by the following formula:

[0066] Using the Adam optimizer (learning rate 0.001, β1=0.9, β2=0.999), the batch size is set to 64, and the training period is set to 200 rounds.

[0067] The logic for superimposing a set of three-dimensional dynamic constraints includes:

[0068] Policy-based adjustments are made, adjusting the upper and lower limits of power based on the characteristic vector of policy signals.

[0069] Meteorological dimension correction: Based on meteorological impact maps, power restrictions are relaxed for power plants in high-irradiance areas, while derating constraints are set for shaded areas.

[0070] The electricity price dimension is adjusted, and combined with the electricity price forecast trend, a "priority for energy storage charging" constraint is generated during the off-peak electricity price period, while the power limit is increased during the peak period.

[0071] The formula for generating the final policy-adaptive schedulable capacity is expressed as follows:

[0072]

[0073] in, Indicates the dynamic correction factor; This indicates policy-adaptive schedulable capacity; This represents the base generation output prediction value output by the LSTM time series model; and These represent the minimum and maximum policy adaptation capacities, respectively. This represents a truncation function that restricts intermediate values ​​to a certain range. Interval.

[0074] Step S3: Based on the digital twin model, select neighboring reference power plants, and correct the policy-adapted dispatchable capacity through meteorological deviation decoupling analysis, and output the meteorological-corrected dispatchable capacity.

[0075] Step S3 further includes the following sub-steps:

[0076] S3-1: Construct a digital twin model of the target power station, synchronize photovoltaic module temperature, irradiance and inverter efficiency, and predict the photovoltaic module temperature and irradiance baseline values ​​of the target power station for the next 15 minutes based on the meteorological correlation of neighboring power stations.

[0077] S3-2, Based on the digital twin model, screen neighboring power stations with a meteorological similarity of ≥90% to the target power station and a distance of ≤5km to obtain neighboring reference power stations;

[0078] S3-3 calculates the meteorological deviation impact factor of photovoltaic module temperature and irradiance using the Euclidean distance algorithm, outputs the comprehensive loss correction coefficient, and corrects the policy-adapted schedulable capacity based on the comprehensive loss correction coefficient, outputting the meteorological corrected schedulable capacity.

[0079] It should be noted that the specific steps for constructing a digital twin model of the target power plant are as follows:

[0080] The layout of the photovoltaic array is obtained by 3D laser scanning, and the physical topology of the module-support-inverter is constructed.

[0081] An equivalent circuit model of the photovoltaic module is established, and the inverter adopts a space vector control model in the dq coordinate system.

[0082] Based on the Fourier heat conduction equation and fluid dynamics model, the temperature field distribution of the component is simulated.

[0083] The temperature of the photovoltaic module is collected by an infrared thermal imager (accuracy ±0.5℃) or an embedded temperature sensor on the module surface.

[0084] Irradiance is measured using a photosynthetically active radiation sensor (accuracy ±5W / m²), which is deployed at the center of the component array.

[0085] Inverter efficiency is calculated in real time by collecting DC-side input power and AC-side output power through the SCADA system.

[0086] The synchronization error of the collected data is controlled within ≤1.5% to ensure the real-time consistency between the model and the physical power station.

[0087] The photovoltaic module temperature reference value refers to the theoretical temperature value generated by the digital twin model through the heat conduction equation based on the meteorological data of the target power plant for the same period in the past three years and the historical correlation data of the neighboring reference power plant. It reflects the ideal temperature level that the module should reach under the meteorological conditions.

[0088] The irradiance baseline value refers to the theoretical irradiance value predicted by the digital twin model by combining the historical irradiance time series characteristics, geographical latitude and atmospheric transparency parameters. It is generated synchronously with the temperature baseline value and serves as a benchmark for evaluating real-time irradiance deviation.

[0089] The screening process for nearby reference power plants includes:

[0090] The coordinates of the target power station are obtained through the GIS system, and all power stations within a 5km radius are searched to form a candidate power station list.

[0091] Extract minute-level meteorological data (irradiance and temperature as core parameters) of the past 3 hours from the candidate power station and the target power station, and calculate the comprehensive similarity.

[0092] Power plants with a meteorological similarity of ≥90% among the candidate power plants will be retained as nearby reference power plants;

[0093] The meteorological deviation impact factor is calculated by extracting real-time temperature and irradiance from the target power station and a nearby reference power station to construct a 2D feature vector; the meteorological difference between the target power station and the nearby reference power station is quantified using Euclidean distance, and the comprehensive loss correction coefficient is calculated using the following formula:

[0094]

[0095] in, represents the overall loss correction coefficient; k represents the proportional coefficient, which is fitted using historical data. This represents the normalized Euclidean distance.

[0096] Based on the measured impact of photovoltaic module temperature and irradiance deviations on power generation efficiency, a comprehensive loss correction coefficient is generated and the policy-adaptive dispatchable capacity is updated to obtain the weather-corrected dispatchable capacity, expressed by the formula:

[0097] in, Indicates weather-corrected dispatchable capacity; This indicates policy-adaptive schedulable capacity; This represents the overall loss correction factor.

[0098] The execution logic for policy-adapted schedulable capacity adjustments includes:

[0099] If there is only one nearby reference power station, use directly. Adjusting policies to adapt to schedulable capacity:

[0100] If there are N neighboring reference power stations, a weighted average is used (the weight is the similarity between each neighboring reference power station and the target power station).

[0101] Step S4: When the real-time power output of the target power station is lower than 8% of the weather-corrected dispatchable capacity for 15 consecutive minutes, the station's weather deviation compensation strategy is activated, and corresponding compensation control is executed according to the type of weather deviation.

[0102] Step S4 further includes the following sub-steps:

[0103] S4-1, when the real-time power output of the target power station is lower than 8% of the weather-corrected dispatchable capacity for 15 consecutive minutes, the station compensation is triggered;

[0104] S4-2, Implement compensation strategies based on the type of meteorological deviation, including:

[0105] When temperature deviation is dominant, adjust the reactive power of the inverter;

[0106] When irradiance deviation is the dominant factor, adjust the tilt angle of the photovoltaic modules.

[0107] It should be noted that the temperature sensitivity coefficient is calculated separately. Sensitivity coefficient of irradiance The influence of temperature B and irradiance G on the output is quantified by the following formula:

[0108]

[0109]

[0110] in, This indicates the deviation between real-time output and weather-corrected dispatchable capacity; This indicates the deviation of the temperature from the reference temperature value of the photovoltaic module; This indicates the deviation of the irradiance from the irradiance reference value.

[0111] When temperature deviation is dominant, the increase in component temperature will lead to a decrease in open-circuit voltage. If the grid connection point voltage is too low, the inverter may limit the output power due to low voltage ride-through. The voltage loss can be compensated by adjusting the reactive power of the inverter, maintaining the stability of the system voltage, and indirectly increasing the power generation output.

[0112] When irradiance deviation is dominant, there is insufficient irradiance. By adjusting the tilt angle of the modules, the photovoltaic panels can be made more perpendicular to the direction of sunlight, thereby increasing the amount of irradiance received and improving the power generation.

[0113] Step S5: If the power output of the target power station continues to be insufficient after compensation, auxiliary power stations are selected by calculating the available margin of nearby reference power stations, and an incremental power generation instruction set is generated with the goal of maximizing grid stability and power generation efficiency.

[0114] Step S5 further includes the following sub-steps:

[0115] S5-1 If the power generation output of the target power station is still insufficient after compensation, calculate the available margin of the neighboring reference power stations and select the power stations that satisfy M>0 among the neighboring reference power stations as auxiliary power stations.

[0116] The available margin for a nearby reference power station is calculated using the following formula:

[0117]

[0118] Where M represents the available margin; Indicates the rated installed capacity of a nearby reference power station; This indicates the real-time power generation output of a nearby reference power station; Indicates the ramp rate limit of the nearest reference power station; Indicates the scheduling period;

[0119] S5-2, under the conditions of voltage fluctuation ≤5%, photovoltaic system power generation efficiency ≥80% and policy constraints, construct a multi-objective optimization algorithm with the goal of maximizing grid stability and power generation efficiency, and allocate incremental power based on the multi-objective optimization algorithm;

[0120] S5-3 generates a structured incremental generation instruction set, including auxiliary power station ID, incremental power, and scheduling period. The scheduling period is the specific execution time interval, with a duration equal to [missing information]. .

[0121] It should be noted that the incremental generation instruction set refers to a structured instruction set that includes auxiliary power station ID, incremental power, and scheduling time period. It is encoded in JSON format to ensure cross-system compatibility. The instructions are sent to the auxiliary power station via the MQTT protocol and use QoS level 2 to ensure transmission reliability.

[0122] Step S6: The incremental power generation instruction set and execution results are reliably stored through the consortium blockchain, and policy compliance is automatically verified based on smart contracts to generate audit records.

[0123] S6-1 uses the SHA-256 algorithm to generate a hash digest of the incremental power generation instruction set and execution results, adds a timestamp and digital signature, and stores the evidence through the consortium blockchain;

[0124] S6-2 verifies the execution result through smart contracts and generates audit records, including:

[0125] When the self-consumption rate is ≥70%, it is marked as meeting the settlement conditions for the highest electricity price tier;

[0126] When carbon emissions exceed policy targets, an over-limit alarm is triggered and trading is frozen.

[0127] It should be noted that the objects of evidence storage in the consortium blockchain include incremental power generation instruction sets and execution results uploaded by auxiliary power plants. A SHA-256 hash digest is generated for the incremental power generation instruction sets and execution results; a trusted timestamp (accurate to milliseconds) and the digital signature of the instruction issuer are appended; the information is uploaded to the consortium blockchain (nodes include power grid companies, photovoltaic power plants, and regulatory agencies, which require authorization to join). After consensus among the nodes, evidence storage is completed, and the data is written into blocks to form a chain structure, ensuring immutability.

[0128] Policy compliance rules (self-consumption rate, carbon emission requirements) are pre-coded into smart contracts and deployed to the consortium blockchain; after the execution results are stored, the contract automatically triggers verification and audit records; the verification results generate structured records (including instruction ID, verification results, and timestamps) and are stored in the consortium blockchain.

[0129] Step S7: Optimize model parameters based on consortium blockchain audit records, and use a cryptographic differential privacy federated learning mechanism to achieve cross-power station collaborative training;

[0130] Step S7 further includes the following sub-steps:

[0131] S7-1 calculates the actual execution deviation rate of policy dimension constraints based on audit records stored in the consortium blockchain, and updates the policy library weight coefficients through an adaptive weighting algorithm.

[0132] S7-2, each power station uses the Paillier algorithm to encrypt the gradients of the LSTM time series model and the digital twin model, and adds Gaussian difference privacy noise to the encrypted gradients;

[0133] S7-3 uses the FedProx algorithm in the cloud to aggregate encryption parameters of each power station, combines zero-knowledge proof to verify the trustworthiness of nodes, and generates a global optimization model.

[0134] S7-4 distributes optimized model parameters through the quantum key distribution channel to perform periodic or triggered model updates.

[0135] It should be noted that the adaptive weighting algorithm logic includes:

[0136] For policy dimensions with higher deviation rates, the weight coefficients are increased to make the model pay more attention to the constraints of that dimension.

[0137] For dimensions with a bias rate of ≤5%, the weights are maintained or finely adjusted to reduce invalid calculations.

[0138] The updated weight coefficients are used to generate the three-dimensional dynamic constraint set.

[0139] The policy library weight coefficient is a parameter that measures the impact of policy dimension constraints (such as carbon emission targets, peak-valley electricity pricing periods, and peak-shaving and frequency regulation requirements) on dispatch decisions. Its value ranges from [0,1] and sums to 1. The initial value is set based on the priority of policy documents and is subsequently dynamically updated using an adaptive weighting algorithm.

[0140] After local training at each power station, the gradient parameters of the LSTM model (power generation prediction) and the digital twin model (weather correction) are extracted. The Paillier homomorphic encryption algorithm is used to encrypt the gradients. The gradients can be calculated directly in the encrypted state without decryption, ensuring that the gradient data is not leaked.

[0141] Gaussian difference privacy noise is added to the encryption gradient with a privacy budget of ε=1.0 to ensure data privacy is not compromised. The noise formula is expressed as:

[0142]

[0143] in, The gradient after adding noise is represented by g; g represents the original gradient. This represents Gaussian noise with a mean of 0. σ represents the variance, and σ controls the noise level; the larger σ is, the stronger the noise. I represents the identity matrix.

[0144] The cloud receives encrypted gradients from each power station and uses the FedProx algorithm (which adds a proximal term to the traditional federated average) to solve the problem of large differences in data distribution among power stations and reduce aggregation bias.

[0145] Zero-knowledge proofs are used to verify the trustworthiness of nodes. Each power station needs to prove that the gradient encryption and noise addition comply with the rules and that the original data does not need to be leaked. After the cloud verification is passed, the gradient of the node is included in the aggregation.

[0146] Quantum key distribution is used to generate encryption keys, which are used to encrypt global model parameters; the keys are updated in real time (once per hour) to ensure absolute security during parameter transmission.

[0147] Periodic updates are performed weekly by default (matching the power grid dispatch cycle); triggered updates are triggered immediately when the policy implementation deviation rate is ≥10% for three consecutive times.

[0148] In this way, a management method to improve the power generation capacity of photovoltaic power plants can be achieved.

[0149] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A management method for improving the power generation capacity of a photovoltaic power station, characterized in that, The method includes: Step S1: Analyze the power regulation policy text, meteorological early warning data, and historical electricity price fluctuation data to generate a three-dimensional dynamic constraint set; Step S2: Based on the mapping relationship between real-time meteorological parameters and power generation output data, a basic power generation output prediction sequence is generated through an LSTM time series model, and a three-dimensional dynamic constraint set is superimposed to generate policy-adaptive dispatchable capacity. Step S3: Based on the digital twin model, select neighboring reference power plants, and correct the policy-adapted dispatchable capacity through meteorological deviation decoupling analysis, and output the meteorological-corrected dispatchable capacity. Step S4: When the real-time power output of the target power station is lower than 8% of the weather-corrected dispatchable capacity for 15 consecutive minutes, the station's weather deviation compensation strategy is activated, and corresponding compensation control is executed according to the type of weather deviation. Step S5: If the power output of the target power station continues to be insufficient after compensation, auxiliary power stations are selected by calculating the available margin of nearby reference power stations, and an incremental power generation instruction set is generated with the goal of maximizing grid stability and power generation efficiency. Step S6: The incremental power generation instruction set and execution results are reliably stored through the consortium blockchain, and policy compliance is automatically verified based on smart contracts to generate audit records. Step S7: Optimize model parameters based on consortium blockchain audit records, and use a cryptographic differential privacy federated learning mechanism to achieve cross-power station collaborative training; Step S1 further includes the following sub-steps: S1-1 uses natural language processing to parse the power regulation policy text, extract peak-valley electricity price periods, carbon emission indicators, and peak-shaving and frequency regulation requirements, and generates a policy signal feature vector. S1-2 processes meteorological early warning data through a spatiotemporal data fusion algorithm to generate meteorological impact maps covering each power station; S1-3, Construct an electricity price fluctuation prediction model to predict the electricity price fluctuation trend in the future T period based on historical electricity price fluctuation data and policy signal feature vectors; S1-4, integrate the policy signal feature vector, meteorological impact map and electricity price fluctuation trend to generate a three-dimensional dynamic constraint set including time dimension, spatial dimension and policy dimension; Step S2 further includes the following sub-steps: S2-1 collects real-time power generation output, historical power generation output sequence, real-time meteorological parameters and equipment status data of each power station; S2-2, learn the mapping relationship between the real-time meteorological parameters and real-time power generation output through the LSTM time series model, and output the basic power generation output prediction sequence; S2-3, superimposed three-dimensional dynamic constraint condition set to generate policy-adaptive schedulable capacity; Step S3 further includes the following sub-steps: S3-1: Construct a digital twin model of the target power station, synchronize photovoltaic module temperature, irradiance and inverter efficiency, and predict the photovoltaic module temperature and irradiance baseline values ​​of the target power station for the next 15 minutes based on the meteorological correlation of neighboring power stations. S3-2, Based on the digital twin model, screen neighboring power stations with a meteorological similarity of ≥90% to the target power station and a distance of ≤5km to obtain neighboring reference power stations; S3-3 calculates the meteorological deviation impact factor of photovoltaic module temperature and irradiance using the Euclidean distance algorithm, outputs the comprehensive loss correction coefficient, and corrects the policy-adapted schedulable capacity based on the comprehensive loss correction coefficient, outputting the meteorological corrected schedulable capacity.

2. The management method for improving the power generation capacity of a photovoltaic power station according to claim 1, characterized in that: Step S4 further includes the following sub-steps: S4-1, when the real-time power output of the target power station is lower than 8% of the weather-corrected dispatchable capacity for 15 consecutive minutes, the station compensation is triggered; S4-2, Implement compensation strategies based on the type of meteorological deviation, including: When temperature deviation is dominant, adjust the reactive power of the inverter; When irradiance deviation is the dominant factor, adjust the tilt angle of the photovoltaic modules.

3. The management method for improving the power generation capacity of a photovoltaic power station according to claim 1, characterized in that: Step S5 further includes the following sub-steps: S5-1 If the power generation output of the target power station is still insufficient after compensation, calculate the available margin of each power station and select the power station that satisfies M>0 among the nearby reference power stations as the auxiliary power station. The available margin for a nearby reference power station is calculated using the following formula: Where M represents the available margin; Indicates the rated installed capacity of a nearby reference power station; This indicates the real-time power generation output of a nearby reference power station; Indicates the ramp rate limit of a nearby reference power station; Indicates the scheduling period; S5-2, Under the conditions of voltage fluctuation ≤5%, photovoltaic system power generation efficiency ≥80% and policy constraints, a multi-objective optimization algorithm is constructed with the goal of maximizing grid stability and power generation efficiency, and incremental power is allocated based on the multi-objective optimization algorithm; S5-3, Generate a structured incremental generation instruction set, including auxiliary power station ID, incremental power, and scheduling period, wherein the scheduling period is a specific execution time interval with a duration equal to... .

4. The management method for improving the power generation capacity of a photovoltaic power station according to claim 1, characterized in that: Step S6 further includes the following sub-steps: S6-1 uses the SHA-256 algorithm to generate a hash digest of the incremental power generation instruction set and execution results, adds a timestamp and digital signature, and stores the evidence through the consortium blockchain; S6-2 verifies the execution result through smart contracts and generates audit records, including: When the self-consumption rate is ≥70%, it is marked as meeting the settlement conditions for the highest electricity price tier; When carbon emissions exceed policy targets, an over-limit alarm is triggered and trading is frozen.

5. The management method for improving the power generation capacity of a photovoltaic power station according to claim 1, characterized in that: Step S7 further includes the following sub-steps: S7-1 calculates the actual execution deviation rate of policy dimension constraints based on audit records stored in the consortium blockchain, and updates the policy library weight coefficients through an adaptive weighting algorithm. S7-2, each power station uses the Paillier algorithm to encrypt the gradients of the LSTM time series model and the digital twin model, and adds Gaussian difference privacy noise to the encrypted gradients; S7-3 uses the FedProx algorithm in the cloud to aggregate encryption parameters of each power station, combines zero-knowledge proof to verify the trustworthiness of nodes, and generates a global optimization model. S7-4 distributes optimized model parameters through the quantum key distribution channel to perform periodic or triggered model updates.

Images