Wind power prediction method and device based on large model
The method converts wind power data into text form and uses a large model with reinforcement learning to enhance prediction accuracy and generalization, addressing the limitations of existing forecasting methods.
Patent Information
- Application Number
- CN202411048830.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-12-27
AI Technical Summary
Current wind power forecasting methods face challenges with low accuracy, instability in prediction, difficulty in handling uncertainties, and low generalization performance due to the limitations of time series analysis, machine learning, and deep learning algorithms.
A wind power prediction method using a large model that converts historical wind power time series data into text form, leveraging a pre-trained model with an embedding layer, feature extraction layers, and output projection layers, and employs reinforcement learning for model optimization.
Improves prediction accuracy and generalization by enhancing the model's understanding and reasoning ability through comprehensive data analysis and feedback-based training, resulting in more precise wind power forecasts.
Smart Images
Figure CN119209462A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of this specification relate to the technical field of load forecasting, and in particular to a method and device for wind power forecasting based on a large model. Background Art
[0002] In recent years, with the transformation of the global energy structure and the increasingly severe climate change problem, wind power, as an important part of renewable energy, has been developing at an accelerating pace. The large-scale grid connection of wind power not only helps to reduce greenhouse gas emissions and improve energy efficiency, but also poses new challenges to the stable operation and load forecasting of the power system. Therefore, the research and application of short-term wind power forecasting technology is particularly important. Short-term wind power forecasting technology refers to the prediction of wind farm output power in the next few hours to days. Since wind power generation has significant randomness and volatility, accurate prediction of wind power is of great significance for the dispatch and management of power grids. Effective wind power forecasting can help power grid operators optimize the allocation of power generation resources, reduce the demand for spare capacity, reduce operating costs, and improve the power grid's ability to absorb renewable energy.
[0003] The current wind power forecasting methods mainly include methods based on time series analysis, methods based on machine learning, and methods based on deep learning. The time series analysis method can capture the trend and periodicity of historical data and is suitable for power forecasting under relatively stable meteorological conditions to a certain extent. However, this method is usually difficult to handle nonlinear and non-stationary data, and often performs poorly for wind power generation affected by multiple factors. Machine learning-based algorithms can automatically learn complex relationships in data and can handle multidimensional data, but they are insufficient for complex environments with variable scales and unstructured feature inputs. The deep learning-based method has a strong feature learning ability, can handle complex nonlinear problems and large-scale data, and can extract advanced features from data at multiple spatiotemporal scales. However, in contrast, the deep learning method needs to continuously adjust the training parameters for different scenarios, has a high risk of overfitting and low generalization performance, resulting in low prediction accuracy. Therefore, there is an urgent need for a wind power forecasting method based on a large model that can solve the problems of low wind power forecasting accuracy, unstable forecasting effect, difficulty in effectively dealing with the uncertainties of wind power load changes, and low generalization performance of deep learning algorithms in the prior art. Summary of the invention
[0004] In view of the above-mentioned problems in the prior art, the purpose of the embodiments of this specification is to provide a wind power prediction method and device based on a large model, so as to solve the problems in the prior art of low wind power prediction accuracy, unstable prediction effect, difficulty in effectively dealing with the uncertain factors of wind power load changes and low generalization performance of deep learning algorithms.
[0005] In order to solve the above technical problems, the specific technical solutions of the embodiments of this specification are as follows:
[0006] On the one hand, an embodiment of this specification provides a wind power prediction method based on a large model, the method comprising:
[0007] Acquire historical wind power time series data of the wind farm to be predicted, wherein the historical wind power time series data includes historical meteorological forecast data, historical actual meteorological data, historical wind turbine operation data, historical wind power load data and historical wind power data;
[0008] Converting the historical wind power time series data into text to obtain a text representation of the historical wind power time series data;
[0009] Generate corresponding description text and forecast demand prompt text according to the historical wind power time series data;
[0010] The text representation, description text and forecast demand prompt text of the historical wind power time series data are input into a pre-trained wind power forecasting model to obtain the wind power forecast value of the wind farm to be forecasted.
[0011] Furthermore, the generating corresponding description text according to the historical wind power time series data includes:
[0012] Filling missing values in the historical wind power time series data using an interpolation method to obtain filled historical wind power time series data;
[0013] Calculating the statistics of the filled historical wind power time series data;
[0014] The statistical quantity is filled into a pre-created description text template to obtain a description text of the historical wind power time series data.
[0015] Furthermore, the text conversion of the historical wind power time series data to obtain a text representation of the historical wind power time series data includes:
[0016] Normalizing the historical wind power time series data to obtain normalized historical wind power time series data;
[0017] Dividing the normalized historical wind power time series data into a number of time series data segments of equal length;
[0018] Mapping the numerical value contained in each of the time series data segments to a corresponding word in a preset vocabulary set to obtain a vocabulary set for each of the time series data segments;
[0019] A text representation of each of the time series data segments is obtained according to the vocabulary set of each of the time series data segments.
[0020] Furthermore, the wind power prediction model includes an embedding layer, a feature extraction layer, a fully connected layer and an output projection layer;
[0021] The embedding layer is used to capture the context information of the historical wind power time series data text representation, description text and forecast demand prompt text to obtain the corresponding word embedding vector;
[0022] The feature extraction layer is used to extract features from the word embedding vector to obtain a feature vector representation;
[0023] The fully connected layer is used to perform time series regression prediction based on the feature vector representation to obtain an output text embedding vector;
[0024] The output projection layer is used to flatten and project the output text embedding vector to obtain a wind power prediction value.
[0025] Furthermore, the feature extraction layer includes a temporal convolutional network layer, a proxy attention layer and a pooling layer; the feature extraction of the word embedding vector to obtain a feature vector representation includes:
[0026] Inputting the word embedding vector into the temporal convolutional network layer for feature extraction to obtain a first feature representation vector;
[0027] Inputting the first feature representation vector into the proxy attention layer to obtain a weighted first feature representation vector;
[0028] The weighted first feature representation vector is mapped to a low-dimensional feature space using the pooling layer to obtain a second feature representation vector.
[0029] Furthermore, the training process of the wind power prediction model includes:
[0030] Acquire historical wind power time series data and actual wind power values corresponding to the historical wind power time series data;
[0031] Converting the historical wind power time series data into text to obtain a text representation of the historical wind power time series data;
[0032] Generate corresponding description text and forecast demand prompt text according to the historical wind power time series data;
[0033] Inputting the text representation of the historical wind power time series data, the description text and the forecast demand prompt text into the embedding layer to obtain a word embedding vector;
[0034] Inputting the word embedding vector into the feature extraction layer to obtain a feature vector representation;
[0035] Input the feature vector representation into the fully connected layer to obtain an output text embedding vector;
[0036] Inputting the output text embedding vector into the output projection layer to obtain a wind power prediction value;
[0037] Comparing the actual wind power value with the predicted wind power value, iteratively optimizing the wind power prediction model based on the comparison result and a preset loss function until a preset iteration termination condition is met, thereby obtaining a fine-tuned wind power prediction model;
[0038] Inputting the wind power prediction value into a pre-trained reward model to perform quality assessment on the wind power prediction value to obtain a reward value;
[0039] The fine-tuned wind power prediction model parameters are iteratively optimized using a reinforcement learning algorithm according to the reward value until the model performance reaches a preset condition, thereby obtaining a trained wind power prediction model.
[0040] On the other hand, the embodiments of this specification also provide a wind power prediction device based on a large model, the device comprising:
[0041] An acquisition module is used to acquire historical wind power time series data of the wind farm to be predicted, wherein the historical wind power time series data includes historical meteorological forecast data, historical actual meteorological data, historical wind turbine operation data, historical wind power load data and historical wind power data;
[0042] A conversion module, used for performing text conversion on the historical wind power time series data to obtain a text representation of the historical wind power time series data;
[0043] A generation module, used to generate corresponding description text and forecast demand prompt text according to the historical wind power time series data;
[0044] The prediction module is used to input the text representation of the historical wind power time series data, the description text and the prediction demand prompt text into a pre-trained wind power prediction model to obtain the wind power prediction value of the wind farm to be predicted.
[0045] On the other hand, an embodiment of the present specification further provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein when the computer program is executed by the processor, the instructions of any one of the above methods are executed.
[0046] On the other hand, an embodiment of the present specification further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor of a computer device, the computer program executes instructions of any one of the above-described methods.
[0047] On the other hand, the embodiments of this specification further provide a computer program product, which, when executed by a processor of a computer device, executes instructions of any one of the above methods.
[0048] One or more technical solutions provided by some embodiments of this specification have at least the following technical effects:
[0049] The embodiments of this specification obtain a variety of historical wind power time series data of the wind farm to be predicted, comprehensively consider the various factors that affect the wind power prediction, thereby providing comprehensive data support for wind power prediction, and secondly convert the historical wind power time series data into a text form that can be processed by the large model, so that the large model can use its own learning and reasoning mode to analyze the historical wind power time series data, and then input the time series text representation, description text and prediction demand prompt text into the pre-trained wind power prediction large model to improve the large model's understanding and reasoning ability of the internal laws and characteristics of the time series data, thereby improving the large model's prediction ability, and then improving the accuracy of wind power prediction.
[0050] The above description is only an overview of the technical solutions of some embodiments of this specification. In order to more clearly understand the technical means of some embodiments of this specification, they can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiments of this specification more obvious and easy to understand, the following specifically cites the preferred embodiments and describes them in detail in conjunction with the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0052] Figure 1 A schematic flow chart of a wind power prediction method based on a large model in some embodiments of this specification is shown;
[0053] Figure 2 A schematic diagram of a process of converting the historical wind power time series data into text to obtain a text representation of the historical wind power time series data in some embodiments of this specification is shown;
[0054] Figure 3 A schematic diagram of a process of generating corresponding description text according to the historical wind power time series data in some embodiments of this specification is shown;
[0055] Figure 4A schematic diagram of a process of extracting features from the word embedding vector to obtain a feature vector representation is shown in some embodiments of this specification;
[0056] Figure 5 A schematic diagram showing a flow chart of a training process of a large wind power prediction model in some embodiments of this specification;
[0057] Figure 6 A schematic diagram of the structure of a wind power prediction device based on a large model in some embodiments of this specification is shown;
[0058] Figure 7 A schematic diagram of the structure of a computer device in this specification is shown.
[0059] Description of the accompanying symbols:
[0060] 601. Get module;
[0061] 602, conversion module;
[0062] 603, generate module;
[0063] 604, prediction module;
[0064] 702. Computer equipment;
[0065] 704, processor;
[0066] 706. Memory;
[0067] 708, driving mechanism;
[0068] 710, input / output module;
[0069] 712. Input devices;
[0070] 714. Output devices;
[0071] 716. Presentation equipment;
[0072] 718. Graphical user interface;
[0073] 720, network interface;
[0074] 722. Communication link;
[0075] 724. Communication bus. DETAILED DESCRIPTION
[0076] The following will be combined with the drawings in the embodiments of this specification to clearly and completely describe the technical solutions in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.
[0077] In order to solve the above problems, the embodiments of this specification provide a wind power prediction method based on a large model. Figure 1 This is a flow chart of a wind power prediction method based on a large model provided in the embodiment of this specification. This specification provides method operation steps as described in the embodiment or flow chart, but more or fewer operation steps may be included based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the order of many steps and does not represent the only execution order. When the system or device product is executed in practice, it can be executed in the order of the method shown in the embodiment or the drawings or in parallel.
[0078] It should be noted that the terms "first", "second", etc. in this specification and claims and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this specification described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.
[0079] Reference Figure 1 As shown, the embodiment of this specification provides a wind power prediction method based on a large model, the method comprising:
[0080] S101: Acquire historical wind power time series data of the wind farm to be predicted, wherein the historical wind power time series data includes historical meteorological forecast data, historical actual meteorological data, historical wind turbine operation data, historical wind power load data and historical wind power data;
[0081] S102: Convert the historical wind power time series data into text to obtain a text representation of the historical wind power time series data;
[0082] S103: generating corresponding description text and forecast demand prompt text according to the historical wind power time series data;
[0083] S104: Inputting the text representation of the historical wind power time series data, the description text and the prediction demand prompt text into a pre-trained wind power prediction model to obtain the wind power prediction value of the wind farm to be predicted.
[0084] The embodiments of this specification obtain a variety of historical wind power time series data of the wind farm to be predicted, comprehensively consider the various factors that affect the wind power prediction, thereby providing comprehensive data support for wind power prediction, and secondly convert the historical wind power time series data into a text form that can be processed by the large model, so that the large model can use its own learning and reasoning mode to analyze the historical wind power time series data, and then input the time series text representation, description text and prediction demand prompt text into the pre-trained wind power prediction large model to improve the large model's understanding and reasoning ability of the internal laws and characteristics of the time series data, thereby improving the large model's prediction ability, and then improving the accuracy of wind power prediction.
[0085] In some embodiments, the historical meteorological forecast data includes forecast data such as wind speed, wind direction, temperature and humidity, and the historical meteorological actual data refers to actual meteorological data including wind speed, wind direction, temperature and humidity, wherein wind speed refers to the speed of air flow per unit time, usually in meters per second (m / s). Wind speed is the most important meteorological factor affecting wind power generation. Generally speaking, the higher the wind speed, the greater the output power of the wind turbine. When the wind speed is between 4 meters per second and 25 meters per second, the wind power generation is the highest. When the wind speed is lower than 4 meters per second, the wind turbine will not start; when the wind speed is higher than 25 meters per second, the wind turbine will automatically shut down to avoid damage. The stability of wind direction has an important influence on wind power generation. Stable wind direction helps the impeller of the wind turbine to maintain the optimal angle, so that wind energy can be more fully utilized, while frequent changes in wind direction may cause the impeller of the wind turbine to need to adjust the angle continuously, thereby affecting the power generation efficiency. Changes in temperature can affect the material properties and mechanical properties of the blades of the wind turbine. As the temperature rises, the deflection and stiffness of the blades will also change, which may cause the efficiency of the wind turbine to decrease. Humidity changes will affect the self-cleaning performance of wind turbine blades. For example, under high humidity conditions, water accumulation on the blade surface may lead to a decrease in power conversion efficiency.
[0086] Historical wind turbine operation data include wind turbine operation parameters, capacity factor, wind turbine current and voltage data, and wind turbine fault and shutdown data. The wind turbine operation parameters include torque, speed, and wind axial force. When the wind turbine is running, the blades will generate torque under the action of wind force. The magnitude of the torque reflects the magnitude of the wind force on the wind turbine, and thus directly affects the output of wind power. The generator speed is the speed at which the generator in the wind turbine rotates, usually in revolutions per minute (r / m). The speed determines the relative speed between the wind turbine blades and the air, which in turn affects the capture and conversion efficiency of wind energy. The wind axial force is the force perpendicular to the axis of the wind turbine that the wind turbine is subjected to during operation. Its magnitude also reflects the magnitude and stability of the wind force on the wind turbine. The capacity factor refers to the ratio of the actual power generation of the wind turbine to the theoretical maximum power generation, which reflects the utilization rate and reliability of the wind turbine. The higher the capacity factor, the higher the operating efficiency of the wind turbine, and the more fully it can use wind energy to generate electricity. The wind turbine current and voltage data are data that reflect whether the wind turbine can operate normally. Failures and downtime of wind turbines will affect the output of wind power. Frequent failures and downtime may cause unstable or decreased wind power output.
[0087] Historical wind power load data includes wind turbine load data, blade load data, load factor and other data. The wind turbine load data refers to the load borne by the wind turbine during operation, including resistance, inertia force, hinge force and other factors. The size of the wind turbine load directly determines the energy required by the wind turbine during operation, which in turn affects the output of wind power. The increase in load usually means that more power is needed to maintain the normal operation of the wind turbine. As the component most susceptible to wind load in the wind turbine system, the force condition of the blade directly affects the strength and reliability of the blade. The blade load calculation needs to take into account factors such as wind speed, wind direction, blade angle, and analyze in combination with the structural characteristics of the blade. The load factor refers to the ratio of the actual output power of the wind turbine to the rated power. The load factor reflects the average operating status and utilization rate of the wind turbine over a period of time. The higher the load factor, the closer the average output power of the wind turbine is to the rated power and the higher the operating efficiency.
[0088] It can be seen that wind power is affected by many factors, and some of these factors are unstable, and there will be large differences in different scenarios, such as meteorological factors. Therefore, when using deep learning and other methods to predict wind power, it is necessary to train different prediction models for different scenarios, and the generalization performance is poor. As an intelligent model, the big model has demonstrated strong transfer learning capabilities in many fields. For each task, there is no need to start training from scratch. The generalization performance is high, and the big model can use its own learning and reasoning capabilities to understand and reason about the input data. Compared with deep learning, it can give more accurate predictions. Therefore, in order to improve the generalization and accuracy of wind power prediction, the embodiment of this specification predicts wind power based on the big model. Since the big model can only process text-based language tasks, it is necessary to convert continuous time series data into a text form that the big model can understand.
[0089] In some embodiments, reference Figure 2 , the text conversion of the historical wind power time series data to obtain a text representation of the historical wind power time series data includes:
[0090] S201: normalizing the historical wind power time series data to obtain normalized historical wind power time series data.
[0091] Given a historical time series X∈R N×T , which means that it consists of N one-dimensional variables and T time steps. For each X (i) , reversible instance normalization (Reversible Instance Normalization, RevIN) can be used to normalize it and turn it into a distribution with zero mean and unit standard deviation to reduce the drift of time series distribution, improve the stability of time series data, and thus improve prediction accuracy.
[0092] S202: Divide the normalized historical wind power time series data into a number of time series data segments of equal length.
[0093] After normalization, X (i) Divide into several time series data segments of equal length. Assume that the length of the data segment is L p , then the number of data segments is Where T is the time step, L p is the data segment length, P is the number of data segments, and S is the number of horizontal sliding steps. By dividing the historical wind power time series data into several time series data segments of equal length, local information can be aggregated into each data segment, thereby better retaining local semantic information.
[0094] S203: Map the numerical values contained in each of the time series data segments to corresponding words in a preset vocabulary set to obtain the vocabulary set of each of the time series data segments.
[0095] The preset vocabulary set is usually set according to the data characteristics and task requirements. Specifically, according to the characteristics and scope of the time series data, select the vocabulary that can cover these characteristics. For example, if the time series data contains positive and negative numbers, then the vocabulary set should contain vocabulary for positive and negative numbers. According to the specific time series prediction task, select the vocabulary related to the task, such as vocabulary that describes the trend, periodicity, seasonality, etc. of the time series. By building and using the preset vocabulary set, continuous time series data can be converted into discrete, recognizable symbol sequences, allowing large models to process and analyze time series data.
[0096] S204: Obtaining a text representation of each of the time series data segments according to the vocabulary set of each of the time series data segments.
[0097] For example, based on the vocabulary set, time series data can be described as a text representation similar to "a short-term rise followed by a steady decline". In the above manner, time series data can be effectively converted into natural language input, thereby mapping the input time series to a language task, allowing the large model to leverage its capabilities in natural language processing to perform time series prediction tasks.
[0098] In order to further improve the reasoning ability of the large model for time series data, in some embodiments, refer to Figure 3 , the generating of corresponding description text according to the historical wind power time series data includes:
[0099] S301: Fill missing values in the historical wind power time series data using an interpolation method to obtain filled historical wind power time series data.
[0100] By filling missing values in the original historical wind power time series data, the integrity of the input data can be guaranteed, thereby generating a more accurate description text.
[0101] S302: Calculate the statistics of the filled historical wind power time series data.
[0102] Specifically, the statistics include mean, standard deviation, maximum value, minimum value, median, etc. The above statistics can reflect the characteristics of time series data from different angles.
[0103] S303: Fill the statistical quantity into a pre-created description text template to obtain a description text of the historical wind power time series data.
[0104] Specifically, the description text template may include placeholders, so that the placeholders are replaced by actual statistics in the subsequent filling steps to generate description text and prediction demand prompt text. In some other embodiments, a pre-set regular expression can also be used as a description text template. For example, assuming that the future temperature changes in a certain area are to be predicted, the generated description text may be as follows: "The following is a time series data about the temperature changes in a certain place, where the average temperature of the place is XX, the highest / lowest temperature is XX respectively, the average wind speed / humidity is XX respectively, and the temperature trend is XX. Please predict the temperature trend of the place in the next few days." Therefore, by adding additional description text and prediction demand prompt text to the input time series, the semantic information of the input time series is enriched, so that the large model can better understand the background and target tasks of the time series data, thereby improving the accuracy of the prediction.
[0105] In some embodiments, the wind power prediction large model includes an embedding layer, a feature extraction layer, a fully connected layer and an output projection layer, wherein the embedding layer is used to capture the contextual information of the text representation of the historical wind power time series data, the description text and the forecast demand prompt text to obtain the corresponding word embedding vector. In some embodiments, the BERT model can be used to capture the contextual information in the text data to obtain the corresponding word embedding vector. The feature extraction layer is used to extract features from the word embedding vector to obtain a feature vector representation; the fully connected layer is used to perform time series regression prediction based on the feature vector representation to obtain an output text embedding vector; the output projection layer is used to flatten and project the output text embedding vector to obtain a wind power forecast value. Since the output of the large model is a high-dimensional vector, which contains the model's prediction information for the time series, in order to convert the high-dimensional vector into a prediction result that is easier to understand and use, it is necessary to use the output projection layer to map the high-dimensional vector output by the large model to the target output space through a series of linear transformations. Specifically, the conversion can be achieved by setting a mapping function.
[0106] Specifically, the feature extraction layer includes a temporal convolutional network layer, a proxy attention layer, and a pooling layer. Figure 4 , the feature extraction of the word embedding vector to obtain a feature vector representation includes:
[0107] S401: Input the word embedding vector into the temporal convolutional network layer for feature extraction to obtain a first feature representation vector;
[0108] S402: Inputting the first feature representation vector into the proxy attention layer to obtain a weighted first feature representation vector;
[0109] S403: Mapping the weighted first feature representation vector to a low-dimensional feature space using the pooling layer to obtain a second feature representation vector.
[0110] Through causal convolution and dilated convolution, the temporal convolutional network can capture long-distance sequence dependencies without increasing the number of parameters and computational complexity, solving the gradient explosion and gradient vanishing problems of traditional RNN. Compared with the improved RNN models LSTM and GRU, it is simpler and more efficient. By weighting the extracted feature representation vectors through the proxy attention mechanism, the large model can pay more attention to the key information of the input data, which helps the model to capture and understand key information more accurately, thereby improving the prediction performance of the large model, reducing the amount of calculation, and improving the model operation efficiency.
[0111] In some embodiments, reference Figure 5 The training process of the wind power prediction model includes:
[0112] S501: Acquire historical wind power time series data and actual wind power values corresponding to the historical wind power time series data;
[0113] S502: Convert the historical wind power time series data into text to obtain a text representation of the historical wind power time series data;
[0114] S503: Generate corresponding description text and forecast demand prompt text according to the historical wind power time series data;
[0115] S504: inputting the text representation of the historical wind power time series data, the description text and the forecast demand prompt text into the embedding layer to obtain a word embedding vector;
[0116] S505: Input the word embedding vector into the feature extraction layer to obtain a feature vector representation;
[0117] S506: Input the feature vector representation into the fully connected layer to obtain an output text embedding vector;
[0118] S507: inputting the output text embedding vector into the output projection layer to obtain a wind power prediction value;
[0119] S508: comparing the actual wind power value with the predicted wind power value, iteratively optimizing the wind power prediction model based on the comparison result and a preset loss function until a preset iteration termination condition is met, and obtaining a fine-tuned wind power prediction model;
[0120] S509: inputting the wind power prediction value into a pre-trained reward model to perform quality assessment on the wind power prediction value to obtain a reward value;
[0121] S510: Iteratively optimizing the fine-tuned wind power prediction model parameters using a reinforcement learning algorithm according to the reward value until the model performance reaches a preset condition, thereby obtaining a trained wind power prediction model.
[0122] Since the large model has large-scale parameters and complex computing structure, it is necessary to learn complex patterns and features by training massive data. In order to further improve the performance, generalization ability and training efficiency of the model, feedback-based reinforcement learning (RLHF) can be used to train the large model. Specifically, firstly, the parameters of the large model are iteratively optimized based on the predicted results and the actual results using the preset loss function to obtain the fine-tuned large model. Secondly, it is necessary to train a reward model to evaluate the output quality of the large model and give the corresponding reward value. The reward model is trained by collecting the output values generated by the large model and the corresponding human feedback (such as ratings, preferences, etc.) data. In some embodiments, the reward model can be a neural network that receives the output of the large model as input and outputs a reward value representing the output quality. The training of the reward model can adopt a supervised learning method to optimize the model parameters by minimizing the difference between the predicted reward value and the human feedback reward value. After the reward model training is completed, the large model is iteratively optimized based on the feedback of the reward model based on the reinforcement learning algorithm until the large model reaches the preset target or the training reaches the maximum number of iterations. In each iteration, the large model outputs a prediction value, and the quality of the output prediction value is evaluated through the reward model to obtain the reward value corresponding to the output prediction value. The reinforcement learning algorithm (such as the policy gradient algorithm) is used to update the large model parameters according to the reward value, so that the large model is more inclined to generate high-reward outputs, thereby improving the prediction accuracy of the large model.
[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. And the acquisition, storage, use, and processing of data in the technical solutions described in the embodiments of this application are in compliance with relevant regulations.
[0124] Based on the above-mentioned method for predicting wind power based on a large model, the embodiment of this specification also provides a wind power prediction device based on a large model. The device may include a system (including a distributed system), software (application), module, component, server, client, etc. using the method described in the embodiment of this specification and a device combined with necessary implementation hardware. Based on the same innovative concept, the device in one or more embodiments provided in the embodiment of this specification is as described in the following embodiments. Since the implementation scheme and method of the device to solve the problem are similar, the implementation of the specific device in the embodiment of this specification can refer to the implementation of the aforementioned method, and the repetitions will not be repeated. As used below, the term "unit" or "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.
[0125] Specifically, Figure 6 This is a schematic diagram of the module structure of an embodiment of a wind power prediction device based on a large model provided in the embodiment of this specification, referring to Figure 6 As shown, a wind power prediction device based on a large model provided in an embodiment of this specification includes:
[0126] An acquisition module 601 is used to acquire historical wind power time series data of a wind farm to be predicted, wherein the historical wind power time series data includes historical meteorological forecast data, historical actual meteorological data, historical wind turbine operation data, historical wind power load data and historical wind power data;
[0127] A conversion module 602 is used to perform text conversion on the historical wind power time series data to obtain a text representation of the historical wind power time series data;
[0128] A generating module 603 is used to generate corresponding description text and forecast demand prompt text according to the historical wind power time series data;
[0129] The prediction module 604 is used to input the text representation of the historical wind power time series data, the description text and the prediction demand prompt text into a pre-trained wind power prediction model to obtain the wind power prediction value of the wind farm to be predicted.
[0130] The beneficial effects obtained by the device provided in the embodiments of this specification are consistent with the beneficial effects obtained by the above method and will not be repeated here.
[0131] Reference Figure 7As shown, based on the above-mentioned large-model-based wind power prediction method, a computer device 702 is also provided in an embodiment of this specification, wherein the above-mentioned method runs on the computer device 702. The computer device 702 may include one or more processors 704, such as one or more central processing units (CPUs), each of which may implement one or more hardware threads. The computer device 702 may also include any memory 706, which is used to store any kind of information such as code, settings, data, etc. Non-limitingly, for example, the memory 706 may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of the computer device 702. In one case, when the processor 704 executes an associated instruction stored in any memory or combination of memories, the computer device 702 may perform any operation of the associated instruction. The computer device 702 also includes one or more drive mechanisms 708 for interacting with any storage, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.
[0132] The computer device 702 may also include an input / output module 710 (I / O) for receiving various inputs (via input devices 712) and for providing various outputs (via output devices 714). A specific output mechanism may include a presentation device 716 and an associated graphical user interface (GUI) 718. In other embodiments, the input / output module 710 (I / O), the input device 712, and the output device 714 may not be included, and the computer device 702 may be used as a computer device in a network. The computer device 702 may also include one or more network interfaces 720 for exchanging data with other devices via one or more communication links 722. One or more communication buses 724 couple the components described above together.
[0133] The communication link 722 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 722 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.
[0134] Corresponding to Figures 1 to 5 In addition to the method shown, an embodiment of the specification also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed.
[0135] The embodiment of the present specification also provides a computer-readable instruction, wherein when the processor executes the instruction, the program therein causes the processor to execute the following Figures 1 to 5 The method shown.
[0136] The embodiments of the present specification also provide a computer program product, including at least one instruction or at least one program, wherein the at least one instruction or the at least one program is loaded and executed by a processor to implement the following Figures 1 to 5 The method shown.
[0137] It should be understood that in the various embodiments of this specification, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this specification.
[0138] It should also be understood that in the embodiments of this specification, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this specification generally indicates that the associated objects before and after are in an "or" relationship.
[0139] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this specification can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this specification.
[0140] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0141] In the several embodiments provided in this specification, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.
[0142] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this specification.
[0143] In addition, each functional unit in each embodiment of this specification may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0144] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this specification is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of this specification. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0145] Specific embodiments are used in this specification to illustrate the principles and implementation methods of this specification. The description of the above embodiments is only used to help understand the methods and core ideas of this specification. At the same time, for those skilled in the art, according to the ideas of this specification, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on this specification.
Claims
1. A wind power prediction method based on a large model, characterized in that: The method comprises: Acquire historical wind power time series data of the wind farm to be predicted, wherein the historical wind power time series data includes historical meteorological forecast data, historical actual meteorological data, historical wind turbine operation data, historical wind power load data and historical wind power data; Converting the historical wind power time series data into text to obtain a text representation of the historical wind power time series data; Generate corresponding description text and forecast demand prompt text according to the historical wind power time series data; The text representation, description text and forecast demand prompt text of the historical wind power time series data are input into a pre-trained wind power forecasting model to obtain the wind power forecast value of the wind farm to be forecasted.
2. The method according to claim 1, characterized in that The generating corresponding description text according to the historical wind power time series data includes: Filling missing values in the historical wind power time series data using an interpolation method to obtain filled historical wind power time series data; Calculating the statistics of the filled historical wind power time series data; The statistical quantity is filled into a pre-created description text template to obtain a description text of the historical wind power time series data.
3. The method according to claim 1, characterized in that: The converting the historical wind power time series data into text to obtain a text representation of the historical wind power time series data includes: Normalizing the historical wind power time series data to obtain normalized historical wind power time series data; Dividing the normalized historical wind power time series data into a number of time series data segments of equal length; Mapping the numerical value contained in each of the time series data segments to a corresponding word in a preset vocabulary set to obtain a vocabulary set for each of the time series data segments; A text representation of each of the time series data segments is obtained according to the vocabulary set of each of the time series data segments.
4. The method according to claim 1, characterized in that The wind power prediction model includes an embedding layer, a feature extraction layer, a fully connected layer and an output projection layer; The embedding layer is used to capture the context information of the historical wind power time series data text representation, description text and forecast demand prompt text to obtain the corresponding word embedding vector; The feature extraction layer is used to extract features from the word embedding vector to obtain a feature vector representation; The fully connected layer is used to perform time series regression prediction based on the feature vector representation to obtain an output text embedding vector; The output projection layer is used to flatten and project the output text embedding vector to obtain a wind power prediction value.
5. The method according to claim 4, characterized in that The feature extraction layer includes a temporal convolutional network layer, a proxy attention layer and a pooling layer; the feature extraction of the word embedding vector to obtain a feature vector representation includes: Inputting the word embedding vector into the temporal convolutional network layer for feature extraction to obtain a first feature representation vector; Inputting the first feature representation vector into the proxy attention layer to obtain a weighted first feature representation vector; The weighted first feature representation vector is mapped to a low-dimensional feature space using the pooling layer to obtain a second feature representation vector.
6. The method according to claim 4, characterized in that The training process of the wind power prediction model includes: Acquire historical wind power time series data and actual wind power values corresponding to the historical wind power time series data; Converting the historical wind power time series data into text to obtain a text representation of the historical wind power time series data; Generate corresponding description text and forecast demand prompt text according to the historical wind power time series data; Inputting the text representation of the historical wind power time series data, the description text and the forecast demand prompt text into the embedding layer to obtain a word embedding vector; Inputting the word embedding vector into the feature extraction layer to obtain a feature vector representation; Input the feature vector representation into the fully connected layer to obtain an output text embedding vector; Inputting the output text embedding vector into the output projection layer to obtain a wind power prediction value; Comparing the actual wind power value with the predicted wind power value, iteratively optimizing the wind power prediction model based on the comparison result and a preset loss function until a preset iteration termination condition is met, thereby obtaining a fine-tuned wind power prediction model; Inputting the wind power prediction value into a pre-trained reward model to perform quality assessment on the wind power prediction value to obtain a reward value; The fine-tuned wind power prediction model parameters are iteratively optimized using a reinforcement learning algorithm according to the reward value until the model performance reaches a preset condition, thereby obtaining a trained wind power prediction model.
7. A wind power prediction device based on a large model, characterized in that: The device comprises: An acquisition module is used to acquire historical wind power time series data of the wind farm to be predicted, wherein the historical wind power time series data includes historical meteorological forecast data, historical actual meteorological data, historical wind turbine operation data, historical wind power load data and historical wind power data; A conversion module, used for performing text conversion on the historical wind power time series data to obtain a text representation of the historical wind power time series data; A generation module, used to generate corresponding description text and forecast demand prompt text according to the historical wind power time series data; The prediction module is used to input the text representation of the historical wind power time series data, the description text and the prediction demand prompt text into a pre-trained wind power prediction model to obtain the wind power prediction value of the wind farm to be predicted.
8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.
10. A computer program product, characterized in that The method comprises at least one instruction or at least one program, wherein the at least one instruction or the at least one program is loaded and executed by a processor to implement the method according to any one of claims 1 to 6.
Citation Information
Cited By
Offshore photovoltaic power station generating capacity prediction method and device, electronic equipment and medium
CN120297471A