Photovoltaic system generating capacity prediction method, system, device and medium

Abstract

The invention relates to the technical field of photovoltaic generating capacity prediction, in particular to a photovoltaic system generating capacity prediction method, system and device and a medium, and the method comprises the steps: obtaining historical operation data sets of a plurality of photovoltaic power stations, and carrying out the preprocessing; constructing a local prediction model for each photovoltaic power station based on the historical operation data set of each photovoltaic power station; performing cooperative training on the local prediction models of the plurality of photovoltaic power stations by using a federated learning framework; inputting meteorological data and historical operation data collected by a target photovoltaic power station in real time into the local prediction model after cooperative training, fusing state information output by the state estimation sub-model and feature information extracted by the meteorological feature sub-model, and generating an initial power generation prediction sequence in a future time period; and adjusting the initial generating capacity prediction sequence through online dynamic correction based on generating power data acquired by the target photovoltaic power station in real time, and outputting a final generating capacity prediction result. The invention aims to improve the prediction precision of the generating capacity of the photovoltaic system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power generation prediction, in particular to a photovoltaic system power generation prediction method, system, device and medium. BACKGROUND

[0002] With the development of new energy power systems, photovoltaic power generation prediction technology has become a key technology to support grid safety dispatching and market transactions. Existing photovoltaic power generation prediction technology is mostly centered on the correlation modeling of meteorological conditions and power generation, and the mapping relationship from meteorological factors to output power is constructed through neural network algorithms. This kind of technology usually simplifies photovoltaic power stations into static or quasi-static systems, and does not explicitly model the slow performance degradation process generated during the operation of photovoltaic components. At the same time, in view of the deployment characteristics of distributed photovoltaic power stations, existing research has attempted to use a federated learning framework to explore ways to improve model generalization ability while protecting data privacy.

[0003] Due to the data island phenomenon commonly existing in distributed photovoltaic power stations, the training data of a single power station is scarce, and there is a natural conflict between centralized training mode and data privacy protection requirements. Simple local models cannot effectively adapt to the equipment model installation environment and climate regional heterogeneity between different power stations, directly leading to insufficient model generalization ability. At the same time, the existing prediction model has a fixed structure after deployment, lacks a mechanism for rapid self-correction using online real-time power generation data, and is difficult to cope with prediction bias caused by sudden weather changes. It is also limited by the lack of modeling of the slow performance degradation process of photovoltaic components, further restricting the improvement of photovoltaic power generation prediction accuracy. SUMMARY

[0004] In order to improve the prediction accuracy of photovoltaic system power generation, the present application provides a photovoltaic system power generation prediction method, system, device and medium, and the technical scheme adopted is as follows:

[0005] The technical scheme of the first aspect of the present application provides a photovoltaic system power generation prediction method, which comprises:

[0006] Obtaining and preprocessing historical operation data sets of a plurality of photovoltaic power stations; the historical operation data set comprises historical meteorological data and corresponding historical power generation data;

[0007] Based on the historical operation data set of each photovoltaic power station, a local prediction model is constructed for each photovoltaic power station; the local prediction model comprises a state estimation sub-model and a meteorological feature sub-model;

[0008] Collaborative training of the local prediction models of the plurality of photovoltaic power stations is performed using a federated learning framework; wherein the meteorological feature sub-model participates in global knowledge fusion, and the state estimation sub-model is locally trained;

[0009] input the meteorological data and historical operation data of the target photovoltaic power station collected in real time into the local prediction model trained in collaboration, fuse the state information output by the state estimation sub-model and the feature information extracted by the meteorological feature sub-model, and generate an initial power generation prediction sequence of a future period;

[0010] Based on the real-time acquired power generation data of the target photovoltaic power station, the initial power generation prediction sequence is adjusted through online dynamic correction, and the final power generation prediction result is output.

[0011] Further, based on the historical operation data set of each photovoltaic power station, a local prediction model is constructed for each photovoltaic power station, including:

[0012] A state estimation sub-model and a meteorological feature sub-model are constructed for each photovoltaic power station, respectively.

[0013] The state estimation sub-model is configured to receive time series data reflecting the cumulative effect of photovoltaic power station equipment operation.

[0014] The meteorological feature sub-model is configured to receive historical meteorological data.

[0015] Further, the state estimation sub-model is locally trained, including:

[0016] The time series data reflecting the cumulative effect of photovoltaic power station equipment operation is taken as the driving input of the state estimation sub-model.

[0017] The state estimation sub-model uses a recurrent neural network with a gating mechanism to iteratively update its internal hidden state according to the driving input at the current time and the internal hidden state at the previous time.

[0018] The updated internal hidden state is mapped to state information representing the dynamic performance of the equipment at the current time.

[0019] Further, the local prediction models of multiple photovoltaic power stations are collaboratively trained using a federated learning framework, further including:

[0020] The meteorological feature sub-model uses a time series convolutional network to extract hierarchical features from the input meteorological data to obtain deep meteorological features.

[0021] During the collaborative training process of the federated learning framework, the central server receives the meteorological feature sub-model parameters uploaded by each photovoltaic power station based on local data updates.

[0022] The central server fuses the received meteorological feature sub-model parameters according to a preset rule, generates and distributes globally shared meteorological feature sub-model parameters to each photovoltaic power station.

[0023] Further, the state information output by the state estimation sub-model is fused with the feature information extracted by the meteorological feature sub-model to generate an initial power generation prediction sequence for a future time period, including:

[0024] The state information output by the state estimation sub-model is fused with the feature information extracted by the meteorological feature sub-model to generate a fusion feature vector;

[0025] The fusion feature vector is input into an attention fusion layer to calculate attention weights of the state information for each time step feature in the feature information, and the feature information is fused according to the attention weights;

[0026] The fusion feature is input into a prediction output layer to be mapped into a power generation prediction value for a single future time;

[0027] The initial power generation prediction sequence is generated by rolling along a preset time dimension.

[0028] Further, the initial power generation prediction sequence is adjusted by online dynamic correction, including:

[0029] A state space model is constructed with prediction errors for multiple future time as state variables;

[0030] Real-time power generation data of the target photovoltaic power station is obtained, and a prediction error observation value for a current time is calculated;

[0031] According to the state space model, a Kalman filtering algorithm is used to update the state estimation of the prediction error for a future time by using the prediction error observation value;

[0032] According to the updated state estimation of the prediction error for the future time, the prediction value corresponding to the initial power generation prediction sequence is compensated to obtain a corrected power generation prediction result.

[0033] Further, the method further includes:

[0034] Based on the historical prediction error of the local prediction model, a conditional probability distribution of the prediction error is determined;

[0035] Based on the conditional probability distribution, the statistical characteristics of the historical correction error are analyzed and a prediction interval is constructed, the boundaries of the prediction interval being determined by the product of the prediction value, a standard normal distribution quantile and a prediction standard deviation;

[0036] According to the prediction interval, prediction uncertainty information corresponding to the final power generation prediction result is generated.

[0037] The technical scheme of the second aspect of the present application provides a photovoltaic system power generation prediction system for executing the photovoltaic system power generation prediction method described in the technical scheme of the first application of the present application, and the system includes:

[0038] The data processing module is configured to acquire and preprocess historical operation datasets from multiple photovoltaic power plants; the historical operation datasets include historical meteorological data and corresponding historical power generation data.

[0039] The local model building module is configured to build a local prediction model for each photovoltaic power station based on the historical operation dataset of each photovoltaic power station; the local prediction model includes a state estimation sub-model and a meteorological feature sub-model.

[0040] The collaborative training module is configured to use a federated learning framework to collaboratively train the local prediction models of multiple photovoltaic power plants; among them, the meteorological feature sub-model participates in global knowledge fusion, and the state estimation sub-model is trained locally.

[0041] The initial prediction generation module is configured to input real-time meteorological data and historical operation data collected from the target photovoltaic power station into the locally predicted model after collaborative training, and to integrate the state information output by the state estimation sub-model with the feature information extracted by the meteorological feature sub-model to generate an initial power generation prediction sequence for future periods.

[0042] The online correction output module is configured to adjust the initial power generation prediction sequence based on the real-time power generation data of the target photovoltaic power plant through online dynamic correction, and output the final power generation prediction result.

[0043] The third aspect of the present invention provides an electronic device, comprising:

[0044] At least one processor; and,

[0045] A memory communicatively connected to the at least one processor; wherein,

[0046] The memory stores instructions that can be executed by the at least one processor, which, when executed, enable the at least one processor to perform the photovoltaic system power generation prediction method as described in the first aspect of the present invention.

[0047] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the photovoltaic system power generation prediction method described in the first aspect of the present invention.

[0048] The present invention has the following beneficial effects:

[0049] The photovoltaic system power generation prediction method of this invention constructs a local prediction model that integrates a state estimation sub-model and a meteorological feature sub-model. This model can specifically capture the impact of equipment-related states and meteorological factors on power generation. Then, a federated learning framework is used for collaborative training. The meteorological feature sub-model acquires general meteorological power generation correlation patterns across power plants through global knowledge fusion, while the state estimation sub-model adapts to the specific characteristics of a single power plant through local training. This balances global commonalities with local specificities, enhancing the model's representation and generalization capabilities across different climate regions, equipment models, and installation environments. Newly built power plants only require a short period of historical data to achieve the prediction accuracy of independently trained models. Real-time data from the target photovoltaic power plant is input into the collaboratively trained local prediction model, and the output information of the two sub-models is fused. The resulting initial power generation prediction sequence is more closely aligned with the actual power generation scenario. Finally, based on real-time power generation data, the initial prediction sequence is adjusted online through dynamic correction, which can promptly correct prediction deviations caused by various factors and suppress the accumulation and propagation of prediction errors. Attached Figure Description

[0050] 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.

[0051] Figure 1 This is a flowchart of a photovoltaic system power generation prediction method provided by an exemplary embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of the structure of a photovoltaic system power generation prediction system provided in an exemplary embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the structure of an application embodiment of the electronic device of the present invention. Detailed Implementation

[0054] 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 photovoltaic system power generation prediction method, system, device, and medium proposed 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.

[0055] 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.

[0056] The following description, in conjunction with the accompanying drawings, details a specific scheme for a photovoltaic system power generation prediction method, system, equipment, and medium provided by the present invention.

[0057] Please see Figure 1 The diagram illustrates a flowchart of a photovoltaic system power generation prediction method according to an embodiment of the present invention, the method comprising:

[0058] Step S100: Obtain historical operation datasets from multiple photovoltaic power plants and preprocess them; the historical operation datasets include historical meteorological data and corresponding historical power generation data; for each photovoltaic power plant, collect continuous historical operation data, with a time span covering different seasons, weather types, and equipment operation stages, and the collection time step can be set according to the actual scenario; for the first A photovoltaic power station, whose dataset is defined as a collection of An ordered set of pairs at a given time point:

[0059]

[0060] In the formula, Indicates the first Local historical datasets of photovoltaic power plants; It is The real-valued eigenvector of dimension represents the th dimension. A photovoltaic power station in All input features collected at any given time; It is a real-valued scalar, representing the first... A photovoltaic power station in The actual power generation at any given time is specifically the active power output of the inverter, normalized to the range of 0 to 1. The feature vector includes: meteorological features: total horizontal irradiance, direct normal irradiance, ambient temperature, wind speed, relative humidity, etc.; temporal features: year-day, hour, day of the week unique thermal encoding, holiday markings, etc.; spatial and equipment features: latitude and longitude of the power station, component installation tilt angle and azimuth angle, etc.

[0061] In some embodiments, to improve the model's robustness to data noise and input uncertainty, adversarial feature enhancement techniques are introduced for the original feature vector. Generate its enhanced version The calculation formula is as follows:

[0062]

[0063] In the formula, Let represent the enhanced eigenvector of the k-th power station at time t; This represents the Hadamard product, which is the element-wise multiplication of two vectors. It is A dimensional binary mask vector, where each element independently carries a predefined probability. It takes the value 0, or with probability The value is 1; It is a with A noise vector of the same dimension, each element of which independently varies from a mean of 0 and a variance of 0. Sampling in a Gaussian distribution, i.e. ,in It is the identity matrix; It is a hyperparameter for controlling noise intensity.

[0064] Step S200: Based on the historical operation dataset of each photovoltaic power station, construct a local prediction model for each photovoltaic power station; the local prediction model includes a state estimation sub-model and a meteorological feature sub-model;

[0065] Step S200 specifically includes:

[0066] Step S210: Construct a state estimation sub-model and a meteorological feature sub-model for each photovoltaic power station. The model adopts a parallel processing dual-branch architecture, consisting of two sub-models that are independent of each other in terms of function and parameters, which are used to process different types of input information and extract corresponding features respectively.

[0067] In some embodiments, the state estimation sub-model is configured as a sequence modeling network whose function is to learn and quantify the slow time-varying trend of photovoltaic power plant power generation performance over operating time. A gated recurrent neural network, such as a gated recurrent unit, can be used as its structure.

[0068] In some embodiments, the meteorological feature sub-model is configured as a deep feature extraction network, whose function is to capture local and global patterns highly correlated with instantaneous fluctuations in power generation from complex meteorological time series. A one-dimensional temporal convolutional network can be selected as its structure, utilizing the sliding of its convolutional kernels in the time dimension to extract meteorological features in layers.

[0069] Step S220: Configure the state estimation sub-model to receive time-series data reflecting the cumulative effects of photovoltaic power plant equipment operation; configure the input of this sub-model to be time-series data reflecting the cumulative effects of photovoltaic power plant equipment operation. The driving input is the cumulative operating time sequence of the power plant, denoted as... superscript Indicates the first Power station, subscript This represents the time index. To enhance the expressive power of the input information, the runtime can be combined with derived features such as the smoothed historical average performance degradation rate to form the driving input vector. The driving input vector is a sub-vector representing the cumulative effect of device operation within the enhanced feature vector; the state estimation sub-model, through its recurrent neural network structure, maps the driving input sequence into a continuous, low-dimensional hidden state vector. This vector encodes the cutoff time. The overall performance status of the equipment. The calculation process is a recursive estimation:

[0070] set up For the model at the previous time step The implicit state at the current moment; The model receives the driving input vector. The implicit state at the current moment is calculated and updated through the internal gating mechanism of the gated loop unit. , can be represented as:

[0071]

[0072] in, This represents the dynamic state vector of the k-th power station at time t, which characterizes the current health status of the equipment. This represents the dynamic state vector of the k-th power station at time t-1, providing historical dependency information for updating the current state; This represents the driving input vector of the k-th power station at time t, which includes the time interval and cumulative runtime; The parameter is The forward propagation function of the gated recurrent unit network, based on the new input and past state This determines the method and content of state updates; then the calculated implicit state is used to determine the method and content of state updates. A linear transformation is performed through a fully connected layer, and the output is the final state information.

[0073] Step S230: Configure the meteorological feature sub-model to receive historical meteorological data. The input of the sub-model is the historical meteorological data preprocessed in step S100; each row of the input matrix corresponds to the same time point, and each column represents a different meteorological variable; the temporal convolutional network ensures the rationality of the temporal data through causal convolution to avoid future information leakage; local temporal correlations of a single feature dimension are extracted through shallow convolution, such as short-term irradiance change trends; subsequent layers expand the receptive field by increasing the hole rate, and gradually extract multi-scale, global features, such as daytime irradiance peak features and seasonal climate correlation features; after each convolution, a layer normalization operation is applied to accelerate model convergence, improve the stability and robustness of feature extraction, and finally obtain deep meteorological features that can characterize the potential nonlinear correlation between input features and power generation.

[0074] Based on the standardized and robust data preprocessed in step S100, this embodiment constructs functionally differentiated and synergistically complementary dual-branch local prediction models for each photovoltaic power station. The state estimation sub-model, through precise adaptation of the GRU network to the time-series data of cumulative effects of equipment operation, achieves explicit modeling of the dynamic health status of the equipment, effectively capturing slowly time-varying internal influencing factors such as component aging. The meteorological feature sub-model, through deep fusion of the TCN network and multi-dimensional input features, fully explores the deep correlation between meteorology, time, space, and power generation, accurately capturing the complex influence of the external environment. The independent construction of the dual-branch models and the differentiated configuration of personalized and shared parameters ensure accurate adaptation to the characteristics of local power stations and provide a foundation for subsequent cross-current attention fusion and asynchronous personalized federated collaborative training.

[0075] Step S300: Collaboratively train the local prediction models of multiple photovoltaic power plants using a federated learning framework; wherein, the meteorological feature sub-model participates in global knowledge fusion, and the state estimation sub-model is trained locally;

[0076] Step S300 specifically includes:

[0077] Step S310: Use time-series data reflecting the cumulative effects of photovoltaic power plant equipment operation as the driving input to the state estimation sub-model; this step prepares the specific input data for the state estimation sub-model, which aims to quantify the time-varying cumulative effects of power plant equipment performance, rather than short-term weather fluctuations. For the k-th power plant at time t, its driving input vector... It consists of two core elements:

[0078] Time increment This represents the physical time interval between the current moment and the previous modeling moment, in units such as hours. It encodes the basic time scale of state evolution;

[0079] Cumulative operating age , representing the cumulative operating time of the power plant from commissioning to time t, is usually converted into equivalent full-load hours. It is a key indicator reflecting irreversible performance degradation such as component aging and material fatigue. Equivalent full-load hours refer to the ratio of the actual power generation of the power plant to its rated power, used to quantify the cumulative effect of the actual operating load of the equipment. Therefore, the driving input vector is constructed as follows: This vector serves as the external driving signal for the state evolution equation, guiding the model to learn time-related performance degradation patterns.

[0080] Step S320: The state estimation sub-model employs a gated recurrent neural network to iteratively update its internal hidden state based on the current driving input and the previous hidden state. This step describes how the state estimation sub-model utilizes its gated recurrent neural network structure to recursively update the internal hidden state characterizing the health of photovoltaic power station equipment. First, the model receives driving input data from the previous step, which encodes the cumulative time effect of equipment operation. Simultaneously, the model calls upon its internal hidden state calculated and stored in the previous time step as historical memory. The model's computational unit is configured to include two types of gates: update gates and reset gates. The update gate is responsible for evaluating the importance of the current new input information and determining what proportion of the historical state from the previous time step should be retained in the current state. The reset gate is responsible for determining how much information in the historical state from the previous time step is irrelevant to the calculation of the current new state and should therefore be temporarily ignored. Then, the model uses the output of the reset gate to filter and adjust the historical states from the previous time step. The filtered historical information is then fused with the driving input at the current time step, and a candidate latent state reflecting the possible current state is generated through a nonlinear transformation function. Finally, the model uses the output of the update gate as weights to dynamically weight and fuse the old latent state from the previous time step with the currently generated candidate latent state. The result of this weighted fusion is the final internal latent state updated at the current time step.

[0081] Step S330: Map the updated internal hidden state to state information representing the current dynamics of device performance. Specifically, the input is the updated internal hidden state output in step S320, which is a dense data vector containing complex dynamics of device performance. The hidden state vector is fed into a transformation layer for processing. The transformation layer uses a learnable linear transformation to map the high-dimensional input vector to another feature space. The purpose is to combine and reduce the dimensionality of the information contained in the original hidden state, extracting the most core feature patterns directly related to power generation performance. Finally, the transformation layer outputs a new vector with a more streamlined dimension, which is the final state information.

[0082] In some embodiments, the collaborative training of local prediction models for multiple photovoltaic power plants using a federated learning framework further includes:

[0083] Step S340: The meteorological feature sub-model uses a temporal convolutional network to perform hierarchical feature extraction on the input meteorological data to obtain deep meteorological features. The model receives preprocessed meteorological data and uses the temporal convolutional network as the core feature extractor. The temporal convolutional network consists of multiple stacked layers. At each layer, the temporal convolutional network uses a set of learnable local feature detectors to perform sliding scans on the time axis of the input data. With each sliding scan, the detectors perceive and calculate the combination of meteorological variables within a local time window, capturing local correlation patterns such as rapid increases in irradiance and sustained flat temperatures within that window. Then, the temporal convolutional network selectively compresses or downsamples the feature sequence in the time dimension through pooling operations. While retaining the most critical information, it gradually expands the model's perception range of the input data's time span, thereby enabling it to capture meteorological evolution patterns at different time scales, ranging from minutes to hours. Finally, after processing through convolution, activation, and pooling layers, the original meteorological data is gradually transformed into a series of highly abstract and condensed feature maps. Ultimately, these features are further integrated to form a stable deep meteorological feature vector. This vector comprehensively encodes various complex, nonlinear spatiotemporal patterns in historical meteorological sequences that are closely related to future power generation fluctuations, providing crucial environmental context information for the final prediction.

[0084] Step S350: During the collaborative training process of the federated learning framework, the central server receives meteorological feature sub-model parameters updated based on local data uploaded by each photovoltaic power station; in the federated learning... During the training round, each participating power station Use its local dataset The local model is trained, and the update amount of the meteorological feature sub-model parameters is calculated. Subsequently, each power station only updates these parameters. and the size of its local data Uploaded to the central server. Parameters of the state estimation sub-model. It is always kept locally and not uploaded, thus protecting the unique performance degradation data and patterns of each power station. This represents the update amount of the meteorological feature sub-model parameters for the k-th client in the r-th round of communication;

[0085] Step S360: The central server fuses the received meteorological feature sub-model parameters according to preset rules, generates and distributes globally shared meteorological feature sub-model parameters to each photovoltaic power station. The central server performs asynchronous personalized aggregation, setting... For the first After successfully uploading the updated collection of power stations, the server merges the data using the following parameters:

[0086]

[0087] In the formula, These are the globally shared parameters for the (r+1)th round of communication, i.e., the parameters of the fused meteorological feature sub-model. This is the set of clients that successfully uploaded parameter updates during the r-th round of communication. It is the first The parameters of the meteorological feature sub-models are shared globally. It is the parameter update delay round number for the k-th client in the r-th round of communication; This represents the latency decay weight for the k-th client; It is the total number of client samples participating in the aggregation in the r-th round; the new global parameter obtained from the aggregation. The data is then distributed back to all power plants to update their local meteorological feature sub-models. This process is iterative, allowing the meteorological feature sub-models to learn general meteorological-power generation mapping knowledge across power plants, while the state estimation sub-models continue to undergo localized training.

[0088] This embodiment, based on the standardized data in step S100 and the dual-branch model architecture in S200, achieves accurate modeling of the cumulative effect of equipment operation and internal dynamic state through local iterative training of the state estimation sub-model, fully adapting to the personalized characteristics of each photovoltaic power station. Through TCN hierarchical feature extraction and federated collaborative training of the meteorological feature sub-model, it efficiently integrates the general "meteorological-power generation" mapping knowledge across power stations while protecting data privacy, improving the model's generalization ability and effectively alleviating the problem of scarce training data caused by data silos. The delay-tolerant weighted aggregation strategy and parameter update transmission mechanism take into account both training stability and communication efficiency, enabling newly built power stations or power stations with scarce data to quickly obtain high-quality model parameters. This embodiment achieves the organic unity of personalized adaptation and global knowledge sharing, providing core model support for subsequent power generation prediction that combines equipment state awareness and strong generalization, fundamentally improving the robustness and accuracy potential of the prediction model.

[0089] Step S400: Input the real-time meteorological data and historical operation data collected by the target photovoltaic power station into the locally predicted model after collaborative training, and fuse the state information output by the state estimation sub-model with the feature information extracted by the meteorological feature sub-model to generate the initial power generation prediction sequence for future periods.

[0090] Step S400 specifically includes:

[0091] Step S410: The state information output by the state estimation sub-model is fused with the feature information extracted by the meteorological feature sub-model to generate a fused feature vector. On one hand, the state estimation sub-model outputs a state information vector representing the current dynamic performance of the equipment based on the historical accumulated operating data of the target power plant. On the other hand, the meteorological feature sub-model performs in-depth analysis on real-time and historical meteorological data to extract a deep meteorological feature vector containing complex spatiotemporal patterns. Then, these two vectors, originating from different information sources and possessing different physical meanings, are concatenated to form a fused feature vector containing more comprehensive information, providing a unified input basis for subsequent refined predictions.

[0092] Step S420: Input the fused feature vector into the attention fusion layer, calculate the attention weights of the state information on the features at each time step in the feature information, and fuse the feature information according to the attention weights; in order to ensure that the prediction can accurately reflect the specific impact of the current equipment status on the utilization efficiency of meteorological conditions, this step introduces an attention fusion mechanism. The specific implementation process is as follows:

[0093] By using a pre-set learnable projection matrix, the device state vector and deep meteorological feature vector are processed respectively, and the two are mapped into query vector and key vector with consistent dimensions.

[0094] Calculate the inner product of the query vector and the key vector to quantify the correlation between the two types of features, and scale it using a specific scale factor to avoid the gradient vanishing problem in subsequent normalization operations due to an excessively large inner product result.

[0095] The scaled result is normalized to obtain the attention weight vector. Each element in the weight vector corresponds to the importance of the meteorological feature at different time steps, and the sum of all elements is 1, thus realizing the probabilistic allocation of weights.

[0096] Finally, the attention weight vector is weighted and summed with the meteorological feature value vector processed by another projection matrix to obtain the context feature vector, which highlights the meteorological feature information that is strongly correlated with the current equipment status.

[0097] Step S430: Input the fused features into the prediction output layer and map them to the predicted power generation value at a single future moment. In specific implementation, the device state vector and the context feature vector after attention fusion are first concatenated again in a fixed order to form a high-dimensional fused feature vector. This vector contains both the device dynamic state information and the key meteorological feature information after screening, thus fully preserving the correlation of the core factors affecting power generation.

[0098] Then, the high-dimensional fused feature vector is input into a preset fully connected layer. Through the linear transformation of the fully connected layer, the high-dimensional features are compressed into one-dimensional values. Then, the system deviation that may be generated during the linear mapping process is compensated by the bias term, and finally the power generation prediction value at a single future moment is obtained.

[0099] Step S440: Generate an initial power generation prediction sequence along a preset time dimension; to generate a complete initial power generation prediction sequence for a future time period (e.g., the next 24 hours), a rolling prediction method is required. Suppose we need to generate a sequence from... arrive The sequence of time points. First, using current real historical operational data and meteorological data, the first predicted value is obtained by executing steps S410 to S430. For subsequent time steps Since future real weather data is unknown, weather forecast data must be used as input to the weather feature sub-model; simultaneously, the state estimation sub-model is updated recursively based on its internal state. At this point, historical forecast values ​​may be used as partial input or to update the model's internal state. The fusion and mapping process from steps S410 to S430 is repeated to sequentially obtain... Finally, these individual forecast values ​​arranged in chronological order are combined to form the initial power generation forecast sequence. .

[0100] This embodiment achieves hierarchical fusion of state information and meteorological features. First, it preserves complete information through vector concatenation, and then uses an attention mechanism to achieve adaptive modulation of equipment state to meteorological features, highlighting key influencing factors. It uses a fully connected layer to achieve accurate mapping of high-dimensional features to power at a single moment, and then generates a continuous prediction sequence through rolling time steps. This fully utilizes the dynamic state of the equipment and the common meteorological knowledge across power plants, and adapts to the temporal evolution of power generation. The generated initial power generation prediction sequence has both physical interpretability and temporal coherence.

[0101] Step S500: Based on the real-time power generation data of the target photovoltaic power station, adjust the initial power generation prediction sequence through online dynamic correction, and output the final power generation prediction result;

[0102] Step S500 specifically includes:

[0103] Step S510: Construct a state-space model with prediction errors at multiple future times as state variables;

[0104] In practice, state variables are first defined, and the prediction errors of each time step within the preset time period are integrated into a state vector. Each error corresponds to the difference between the predicted value and the actual value at the next time step.

[0105] A state transition equation is constructed to describe the evolution of the prediction error over time. The error is assumed to decay over time. A specific matrix is ​​used to characterize the influence of the previous state on the current state. Unmodeled random disturbance factors are also considered, and this uncertainty is represented by a noise term.

[0106] The observation equation is constructed because, in practice, only the actual power generation data at the current moment can be obtained, and only the error corresponding to the current moment in the state vector can be observed. Therefore, the directly observable scalar error is extracted from the high-dimensional state vector through the observation matrix, and random error terms in the observation process are introduced to ensure the rationality of the observation model.

[0107] Step S520: Obtain the real-time power generation data of the target photovoltaic power station and calculate the prediction error observation value at the current moment;

[0108] In practice, firstly, real-time power generation data is collected at the current moment through the monitoring system of the target photovoltaic power station. Then, from the initial power generation prediction sequence generated in step S400, the initial prediction value corresponding to the current real-time data is found, which is the previously predicted power generation value at the current moment. Finally, the prediction error observation value at the current moment is obtained by subtracting the corresponding initial prediction value from the actual power generation data collected in real time. This observation value directly reflects the deviation between the initial prediction and the actual situation, providing input for the subsequent state update of the Kalman filter.

[0109] Step S530: Based on the state-space model, the Kalman filter algorithm is used to update the state estimate of the prediction error at future time points using the prediction error observations; this step uses the Kalman filter algorithm to optimally fuse model predictions with the latest observations and dynamically update the error estimates for all future time points.

[0110] In specific implementation, based on the state transition equation constructed in step S510, the prior estimate of the error state at the current time is calculated using the posterior estimate of the error state obtained at the previous time step. At the same time, the covariance matrix of the prior estimate at the current time step is updated by combining the state transition matrix, the error covariance matrix at the previous time step, and the statistical characteristics of the process noise, thereby quantifying the degree of uncertainty of the prior estimate.

[0111] Then, based on the covariance matrix of the prior estimate, the observation matrix, and the statistical characteristics of the observation noise, the Kalman gain is calculated. This gain is used to balance the reliability of the prior estimate and the observation values; the magnitude of the gain is related to the uncertainty of the prior estimate and the reliability of the observations. Using the deviation between the current error observation and the prior estimate, the prior estimate is corrected using the Kalman gain to obtain the posterior estimate of the error state at the current time. Finally, the covariance matrix of the posterior estimate of the error state is updated, quantifying the uncertainty of the corrected state estimate and providing data for the next round of filtering iterations.

[0112] Step S540: Based on the updated state estimate of the future time prediction error, compensate the corresponding prediction values ​​in the initial power generation prediction sequence to obtain the corrected power generation prediction result. Specifically, based on the optimized future time prediction error estimates obtained after step S530, compensate the initial power generation prediction sequence point by point.

[0113] In some embodiments, the predicted value at each future time point in the initial prediction sequence is added to the corresponding updated prediction error estimate. Since the error estimate can be positive or negative, this operation performs an adaptive bias correction on the original prediction. The final output is a sequence of power generation prediction values ​​after this dynamic compensation, which serves as the final, optimized prediction result for the current time point.

[0114] In some embodiments, the photovoltaic system power generation prediction method further includes:

[0115] Step S550: Based on the historical prediction errors of the local prediction model, determine the conditional probability distribution of the prediction errors; in specific implementation, first collect the historical prediction error data generated by the local prediction model after collaborative training, including the difference between the initial prediction value of each historical prediction and the corresponding real-time actual power generation, to form a complete historical error dataset.

[0116] This embodiment assumes that the prediction error follows a normal distribution with a mean of 0, while the variance is not a fixed value but is determined by the dynamic state of the equipment output by the state estimation sub-model. That is, the fluctuation of the prediction error will vary depending on the equipment state. By using historical error data and the dynamic state of the equipment at the corresponding time, a shallow neural network is trained as a mapping model. Based on the current dynamic state of the equipment, the corresponding prediction error variance is output, realizing the dynamic adjustment of the variance with the equipment state, so that the error distribution is more in line with the actual situation.

[0117] Step S560: Analyze the statistical characteristics of historical correction errors based on conditional probability distribution and construct a prediction interval. The boundary of the prediction interval is determined by the product of the predicted value, the quantile of the standard normal distribution, and the prediction standard deviation. Based on the normal distribution determined in step S550, analyze the statistical characteristics of historical correction errors and obtain the prediction standard deviation corresponding to each future time step. This standard deviation is the square root of the error variance and directly reflects the fluctuation range of the predicted value.

[0118] Based on the preset confidence level, the corresponding quantiles of the standard normal distribution are queried. The value of the quantile is positively correlated with the confidence level; the higher the confidence level, the larger the quantile value. The final prediction interval is constructed with the lower bound being the corrected prediction value minus the product of the quantile and the corresponding prediction standard deviation, and the upper bound being the corrected prediction value plus the product of the quantile and the corresponding prediction standard deviation. This interval can ensure that the actual power generation falls within the preset confidence level, thereby quantifying the prediction uncertainty.

[0119] Step S570: Generate prediction uncertainty information corresponding to the final power generation prediction result based on the prediction interval.

[0120] In some embodiments, the prediction interval for each future time step constructed in step S560 is matched one-to-one with the corrected power generation prediction value for the corresponding time step obtained in step S540, forming paired data of "point prediction value - prediction interval" for each time step. Then, the paired data is formatted to annotate the core information of each future time step, including the corrected point prediction value, the lower bound of the prediction interval, the upper bound of the prediction interval, and the corresponding confidence level. Finally, the formatted information is integrated into a prediction uncertainty report, which can intuitively reflect the prediction results and possible fluctuation range of power generation at each future time, providing a comprehensive decision-making reference for subsequent application scenarios such as grid dispatch.

[0121] This embodiment achieves online dynamic correction of the initial prediction sequence using real-time power generation data by constructing a state-space model and Kalman filtering algorithm, thereby improving the real-time accuracy and dynamic adaptability of power generation prediction. Simultaneously, by mining the statistical characteristics of historical prediction errors, a prediction interval based on conditional Gaussian distribution is constructed, generating a complete result containing point prediction values ​​and uncertainty ranges. This avoids the information limitations of traditional deterministic prediction and provides a quantitative risk reference for grid dispatch. It enhances the practicality and reliability of the prediction results from two dimensions: real-time correction and uncertainty quantification. This solves the technical problem that meteorological uncertainties and time-varying equipment states reduce prediction accuracy in photovoltaic forecasting.

[0122] In summary, the photovoltaic system power generation prediction method of this invention constructs a local prediction model that integrates a state estimation sub-model and a meteorological feature sub-model. This model can specifically capture the impact of equipment-related states and meteorological factors on power generation. Then, a federated learning framework is used for collaborative training. The meteorological feature sub-model acquires general meteorological power generation correlation patterns across power plants through global knowledge fusion, while the state estimation sub-model adapts to the specific characteristics of a single power plant through local training. This balances global commonalities with local specificities, enhancing the model's representational ability and generalization of different climate regions, equipment models, and installation environments. Newly built power plants only require a short period of historical data to achieve the prediction accuracy of independently trained models. Real-time data from the target photovoltaic power plant is input into the collaboratively trained local prediction model, and the output information of the two sub-models is fused. The resulting initial power generation prediction sequence is more closely aligned with the actual power generation scenario. Finally, based on real-time power generation data, the initial prediction sequence is adjusted online through dynamic correction, which can promptly correct prediction deviations caused by various factors and suppress the accumulation and propagation of prediction errors.

[0123] Please see Figure 2 The diagram illustrates a structural schematic of a photovoltaic system power generation prediction system according to an embodiment of the present invention, the system comprising:

[0124] The data processing module is configured to acquire and preprocess historical operation datasets from multiple photovoltaic power plants; the historical operation datasets include historical meteorological data and corresponding historical power generation data.

[0125] The local model building module is configured to build a local prediction model for each photovoltaic power station based on the historical operation dataset of each photovoltaic power station; the local prediction model includes a state estimation sub-model and a meteorological feature sub-model.

[0126] The collaborative training module is configured to use a federated learning framework to collaboratively train the local prediction models of multiple photovoltaic power plants; among them, the meteorological feature sub-model participates in global knowledge fusion, and the state estimation sub-model is trained locally.

[0127] The initial prediction generation module is configured to input real-time meteorological data and historical operation data collected from the target photovoltaic power station into the locally predicted model after collaborative training, and to integrate the state information output by the state estimation sub-model with the feature information extracted by the meteorological feature sub-model to generate an initial power generation prediction sequence for future periods.

[0128] The online correction output module is configured to adjust the initial power generation prediction sequence based on the real-time power generation data of the target photovoltaic power plant through online dynamic correction, and output the final power generation prediction result.

[0129] In addition, embodiments of the present invention also provide an electronic device, including:

[0130] At least one processor; and,

[0131] A memory communicatively connected to the at least one processor; wherein,

[0132] The memory stores instructions that can be executed by the at least one processor, which, when executed, enable the at least one processor to perform the photovoltaic system power generation prediction method as described in the first aspect of the present invention.

[0133] Figure 3 This is a schematic diagram of the structure of an application embodiment of the electronic device of the present invention. Below, refer to... Figure 3 This describes an electronic device according to embodiments of the present invention. The electronic device may be either or both of a first device and a second device, or a standalone device independent of them, which may communicate with the first device and the second device to receive acquired input signals from them.

[0134] like Figure 3 As shown, the electronic device includes one or more processors and a memory. The processor may be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the photovoltaic system power generation prediction method and / or other desired functions described in the various embodiments of the present invention.

[0135] In one example, the electronic device may further include input and output devices, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown). Furthermore, the input device may include, for example, a keyboard, a mouse, etc. The output device can output various information to the outside, including determined distance information, direction information, etc. The output device may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0136] Of course, for the sake of simplicity, Figure 3Only some of the components of the electronic device relevant to the present invention are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device may include any other suitable components depending on the specific application.

[0137] In addition to the methods and devices described above, embodiments of the present invention may also be computer program products, wherein a program for implementing a photovoltaic system power generation prediction method is stored on a computer-readable storage medium, and the program for implementing the photovoltaic system power generation prediction method is executed by a processor to implement the steps of the photovoltaic system power generation prediction method described in various embodiments of the present invention.

[0138] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of the present invention. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0139] Furthermore, embodiments of the present invention may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps in the photovoltaic system power generation prediction method according to various embodiments of the present invention as described in the foregoing portion of this specification.

[0140] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0141] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0142] The basic principles of the present invention have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in the present invention are merely examples and not limitations, and should not be considered as essential features of each embodiment of the present invention. Furthermore, the specific details of the invention described above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the present invention to the necessity of employing the specific details described above.

[0143] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they largely correspond to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0144] The block diagrams of devices, apparatuses, devices, and systems involved in this invention are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0145] The methods and apparatus of the present invention may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of the present invention are not limited to the order specifically described above unless otherwise specifically stated. Furthermore, in some embodiments, the present invention may also be implemented as a program recorded on a recording medium, the program comprising machine-readable instructions for implementing the methods according to the present invention. Thus, the present invention also covers recording media storing programs for performing the methods according to the present invention.

[0146] It should also be noted that in the apparatus, device, and method of the present invention, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of the present invention.

[0147] The above description of aspects of the invention is provided to enable any person skilled in the art to make or use the invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the invention. Therefore, the invention is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features of the invention herein.

[0148] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the invention to the forms described herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for predicting the power generation of a photovoltaic system, characterized in that, The method includes: Acquire and preprocess historical operation datasets from multiple photovoltaic power plants; the historical operation datasets include historical meteorological data and corresponding historical power generation data. Based on the historical operation datasets of each photovoltaic power station, a local prediction model is constructed for each photovoltaic power station; the local prediction model includes a state estimation sub-model and a meteorological feature sub-model. A federated learning framework is used to collaboratively train local prediction models for multiple photovoltaic power plants; among them, the meteorological feature sub-model participates in global knowledge fusion, while the state estimation sub-model is trained locally. The meteorological data and historical operation data collected in real time from the target photovoltaic power station are input into the locally predicted model after collaborative training. The state information output by the state estimation sub-model and the feature information extracted by the meteorological feature sub-model are fused to generate the initial power generation prediction sequence for future periods. Based on the real-time power generation data of the target photovoltaic power plant, the initial power generation prediction sequence is adjusted through online dynamic correction, and the final power generation prediction result is output.

2. The photovoltaic system power generation prediction method as described in claim 1, characterized in that, Based on the historical operation datasets of each photovoltaic power station, local prediction models are constructed for each photovoltaic power station, including: For each photovoltaic power station, a state estimation sub-model and a meteorological characteristic sub-model are constructed respectively; Configure the state estimation sub-model to receive time-series data reflecting the cumulative effects of photovoltaic power plant equipment operation; Configure the meteorological feature sub-model to receive historical meteorological data.

3. The photovoltaic system power generation prediction method as described in claim 2, characterized in that, The state estimation sub-model is trained locally, including: Time series data reflecting the cumulative effects of photovoltaic power plant equipment operation are used as the driving input for the state estimation sub-model; The state estimation sub-model employs a recurrent neural network with a gating mechanism, which iteratively updates its internal hidden state based on the driving input at the current moment and the internal hidden state at the previous moment. The updated internal implicit state is mapped to state information that characterizes the dynamic performance of the device at the current moment.

4. The photovoltaic system power generation prediction method as described in claim 2, characterized in that, The collaborative training of local prediction models for multiple photovoltaic power plants using a federated learning framework also includes: The meteorological feature sub-model uses a temporal convolutional network to extract layered features from the input meteorological data, thereby obtaining deep meteorological features; During the collaborative training process of the federated learning framework, the central server receives meteorological feature sub-model parameters uploaded by each photovoltaic power station based on local data updates. The central server fuses the received meteorological feature sub-model parameters according to preset rules, generates and distributes globally shared meteorological feature sub-model parameters to each photovoltaic power station.

5. The photovoltaic system power generation prediction method as described in claim 1, characterized in that, By fusing the state information output by the state estimation sub-model with the feature information extracted by the meteorological feature sub-model, an initial power generation prediction sequence for future periods is generated, including: The state information output by the state estimation sub-model is fused with the feature information extracted by the meteorological feature sub-model to generate a fused feature vector; The fused feature vector is input into the attention fusion layer, the attention weights of the state information to the features at each time step in the feature information are calculated, and the feature information is fused according to the attention weights. The fused features are input to the prediction output layer and mapped to the predicted power generation value at a single future moment. The initial power generation prediction sequence is generated by rolling along a preset time dimension.

6. The photovoltaic system power generation prediction method as described in claim 1, characterized in that, The initial power generation forecast sequence is adjusted through online dynamic correction, including: Construct a state-space model with the prediction error at multiple future times as the state variable; Obtain real-time power generation data of the target photovoltaic power station and calculate the prediction error observation value at the current moment; Based on the state-space model, the Kalman filter algorithm is used to update the state estimate of the prediction error at future time steps using the prediction error observations; Based on the updated state estimate of the future time prediction error, the corresponding prediction values ​​in the initial power generation prediction sequence are compensated to obtain the corrected power generation prediction results.

7. The photovoltaic system power generation prediction method according to any one of claims 1 to 6, characterized in that, The method further includes: Based on the historical prediction errors of the local prediction model, the conditional probability distribution of the prediction errors is determined; The statistical characteristics of historical correction errors are analyzed based on conditional probability distribution, and a prediction interval is constructed. The boundary of the prediction interval is determined by the product of the predicted value, the standard normal distribution quantile, and the prediction standard deviation. The prediction uncertainty information corresponding to the final power generation prediction result is generated based on the prediction interval.

8. A photovoltaic system power generation prediction system, characterized in that, The system is used to perform the photovoltaic system power generation prediction method according to any one of claims 1 to 7, the system comprising: The data processing module is configured to acquire and preprocess historical operation datasets from multiple photovoltaic power plants; the historical operation datasets include historical meteorological data and corresponding historical power generation data. The local model building module is configured to build a local prediction model for each photovoltaic power station based on the historical operation dataset of each photovoltaic power station; the local prediction model includes a state estimation sub-model and a meteorological feature sub-model. The collaborative training module is configured to use a federated learning framework to collaboratively train the local prediction models of multiple photovoltaic power plants; among them, the meteorological feature sub-model participates in global knowledge fusion, and the state estimation sub-model is trained locally. The initial prediction generation module is configured to input real-time meteorological data and historical operation data collected from the target photovoltaic power station into the locally predicted model after collaborative training, and to integrate the state information output by the state estimation sub-model with the feature information extracted by the meteorological feature sub-model to generate an initial power generation prediction sequence for future periods. The online correction output module is configured to adjust the initial power generation prediction sequence based on the real-time power generation data of the target photovoltaic power plant through online dynamic correction, and output the final power generation prediction result.

9. An electronic device, characterized in that, include: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the photovoltaic system power generation prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the photovoltaic system power generation prediction method according to any one of claims 1 to 7.

Images