New energy output prediction method and related equipment

Abstract

The embodiment of the invention provides a new energy output prediction method and related equipment, and belongs to the technical field of renewable energy prediction. The method comprises the following steps: acquiring historical output data, meteorological data and a scheduling plan; preprocessing the historical output data, the meteorological data and the scheduling plan to obtain input data; respectively inputting the input data into a first model and a second model in prediction models to obtain a first prediction result and a second prediction result; determining a target prediction result according to the first prediction result, the first weight, the second prediction result and the second weight; wherein the first weight and the second weight are corrected through the error between the target prediction result and the real output. According to the method, the output is predicted through the original data including the scheduling plan, so that the problem of large prediction deviation can be relieved; and meanwhile, prediction is carried out through multiple models, so that the prediction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of renewable energy prediction, and particularly relates to a new energy output prediction method and related equipment. BACKGROUND

[0002] With the increasing penetration of new energy power generation in the power system, the output fluctuation and uncertainty of new energy pose a severe challenge to the safety and dispatching operation of the power grid. The prediction method of the related technology mainly constructs a model based on meteorological data and output data, ignoring the influence of other factors on the actual value of the output, resulting in large prediction deviation and poor controllability.

[0003] To sum up, the technical problems existing in the related technology need to be improved. SUMMARY

[0004] The main purpose of the embodiments of the present application is to provide an efficient new energy output prediction method and related equipment.

[0005] To achieve the above purpose, one aspect of the embodiments of the present application provides a new energy output prediction method, which comprises: acquiring historical output data, meteorological data and a dispatching plan; preprocessing the historical output data, the meteorological data and the dispatching plan to obtain input data; inputting the input data into a first model and a second model in a prediction model respectively to obtain a first prediction result and a second prediction result; wherein the prediction model comprises at least the first model and the second model; determining a target prediction result according to the first prediction result, a first weight, the second prediction result and a second weight; wherein the first weight and the second weight are corrected by the error between the target prediction result and the actual output. The original data including the dispatching plan is used to predict the output, which can alleviate the problem of large prediction deviation; at the same time, multiple models are used for prediction, which is conducive to improving the prediction accuracy.

[0006] In some embodiments, the target prediction result is corrected by the following steps:

[0007] determining a prediction residual sequence by an autoregressive moving average model;

[0008] determining a correction error value according to the prediction residual sequence, an autoregressive coefficient, a moving average coefficient and white noise interference;

[0009] correcting the target prediction result according to the correction error value and updating the target prediction result.

[0010] In some embodiments, the first weight is corrected by the following steps:

[0011] determining a first prediction error according to the first prediction result and the actual output;

[0012] determining a second prediction error according to the second prediction result and the real output;

[0013] correcting the first weight according to the first prediction error, the second prediction error and a dynamic weight adjustment factor.

[0014] In some embodiments, the prediction model is trained by the following steps:

[0015] obtaining historical output data samples, meteorological data samples and scheduling plan samples;

[0016] preprocessing the historical output data samples, the meteorological data samples and the scheduling plan samples to obtain input data samples;

[0017] inputting the input data samples into a prediction model respectively to obtain target prediction samples;

[0018] constructing a loss function according to the target prediction samples and real results, and training the prediction model according to the loss function to obtain a trained prediction model; wherein the loss function comprises a scheduling deviation loss and a peak-valley ratio deviation loss.

[0019] In some embodiments, the loss function is determined by the following steps:

[0020] determining a scheduling deviation loss according to the target prediction samples and a scheduling target output; the scheduling target output is related to the scheduling plan samples;

[0021] determining a predicted peak output and a predicted valley output according to the target prediction samples;

[0022] determining a predicted peak-valley ratio according to the predicted peak output, the predicted valley output and a predetermined constant;

[0023] determining a peak-valley ratio deviation loss according to the predicted peak-valley ratio and a scheduling peak-valley ratio; the scheduling peak-valley ratio is related to the scheduling plan samples;

[0024] determining a loss function according to the scheduling deviation loss, the peak-valley ratio deviation loss and a regular loss.

[0025] In some embodiments, the method further comprises:

[0026] determining a constraint prediction value according to the target prediction result and a maximum output value of the new energy equipment;

[0027] determining a mean absolute error according to the constraint prediction value, the real output and a sliding time window;

[0028] retrain the prediction model if the average absolute error is greater than an error tolerance threshold.

[0029] In some embodiments, the method further comprises:

[0030] obtaining a scheduling curve;

[0031] feature encoding the scheduling curve, determining a scheduling peak-valley ratio and a scheduling target output.

[0032] To achieve the above object, another aspect of the embodiments of the present application provides a new energy output prediction device, which comprises:

[0033] an obtaining module, configured to obtain historical output data, meteorological data and a scheduling plan;

[0034] a preprocessing module, configured to preprocess the historical output data, the meteorological data and the scheduling plan to obtain input data;

[0035] a prediction module, configured to input the input data into a first model and a second model in a prediction model respectively to obtain a first prediction result and a second prediction result; wherein the prediction model comprises at least the first model and the second model;

[0036] a determining module, configured to determine a target prediction result according to the first prediction result, a first weight, the second prediction result and a second weight; wherein the first weight and the second weight are corrected by an error between the target prediction result and a real output.

[0037] To achieve the above object, another aspect of the embodiments of the present application provides an electronic device, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0038] To achieve the above object, another aspect of the embodiments of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above method.

[0039] To achieve the above object, another aspect of the embodiments of the present application provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the above method.

[0040] The embodiments of the present application at least have the following beneficial effects: the present application provides a new energy output prediction method, device, electronic equipment, storage medium and program product, the method comprises: acquiring historical output data, weather data and scheduling plan; preprocessing the historical output data, the weather data and the scheduling plan to obtain input data; inputting the input data into a first model and a second model in a prediction model respectively to obtain a first prediction result and a second prediction result; wherein the prediction model comprises at least a first model and a second model; determining a target prediction result according to the first prediction result, a first weight, the second prediction result and a second weight; wherein the first weight and the second weight are corrected by the error between the target prediction result and the real output. The present application can alleviate the problem of large prediction deviation by predicting the output based on the original data of the scheduling plan; at the same time, the prediction by multiple models is beneficial to improve the prediction accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a flowchart of a new energy output prediction method provided by the embodiments of the present application;

[0042] Figure 2 is a flowchart of another new energy output prediction method provided by the embodiments of the present application;

[0043] Figure 3 is a flowchart of a new energy power prediction method provided by the embodiments of the present application

[0044] Figure 4 is a structural schematic diagram of a new energy output prediction device provided by the embodiments of the present application;

[0045] Figure 5 is a hardware structural schematic diagram of an electronic equipment provided by the embodiments of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the present application clearer, the following will further describe the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application. When the following description relates to the drawings, the same numbers in different drawings represent the same or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application, but are only examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.

[0047] 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 application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to be limiting of this application.

[0048] With the increasing penetration of new energy power generation in the power system, its output fluctuation and uncertainty pose a serious challenge to the safety and dispatch operation of the power grid. Traditional prediction methods mainly build models based on meteorological data and output data, ignoring the influence of dispatching constraints and dispatching response behavior on the actual value of the output, resulting in large prediction deviation and poor controllability. In recent years, deep learning technologies such as CNN, LSTM, GRU, etc. have shown good performance in new energy output prediction. However, a single model is difficult to balance generalization ability and prediction accuracy, and lacks effective adaptation mechanisms for error feedback and model performance degradation. Therefore, it is urgent to build a new energy multi-dimensional prediction technology considering dispatching plan with integrated modeling, error feedback and adaptive optimization capability to improve prediction accuracy and robustness.

[0049] The new energy output prediction method provided by the embodiments of the application relates to the technical field of renewable energy prediction. The new energy output prediction method provided by the embodiments of the application can be applied to a terminal, can also be applied to a server, and can also be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a vehicle-mounted terminal, and the like, but is not limited thereto; the server end can be configured as an independent physical server, can also be configured as a server cluster or a distributed system composed of multiple physical servers, can also be configured as a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and big data and artificial intelligence platform, and the server can also be a node server in a blockchain network; the software can be an application for implementing the new energy output prediction method, and the like, but is not limited to the above forms.

[0050] The application is operable in a multitude of generic or specific computer system environments or configurations. Examples of well known computing systems, environments, and / or configurations that can be suitable for use with the application include personal computers, server computers, handheld or laptop devices, tablet devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like. The application can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like, that perform particular tasks or implement particular abstract data types. The application can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in local and remote computer storage media including memory storage devices.

[0051] Figure 1 is an optional flowchart of the new energy output prediction method provided by the embodiments of the application, Figure 1 The method in the method can include but is not limited to including steps S100 to S400.

[0052] Step S100, obtaining historical output data, meteorological data and scheduling plan;

[0053] Step S200, preprocessing the historical output data, the meteorological data and the scheduling plan to obtain input data;

[0054] Step S300, inputting the input data into a first model and a second model in a prediction model respectively to obtain a first prediction result and a second prediction result; wherein the prediction model at least includes the first model, the second model;

[0055] Step S400, determining a target prediction result according to the first prediction result, a first weight, the second prediction result and a second weight; wherein the first weight and the second weight are corrected by an error between the target prediction result and a real output.

[0056] The data preprocessing can include normalization processing, feature extraction and the like. The prediction model can include a first model, a second model and a third model, wherein the first model can be established based on a long short-term memory network (LSTM), the second model can be established based on a convolutional neural network (CNN), and the third model can be established based on a gated recurrent unit (GRU). The first prediction result corresponds to the first weight, and the second prediction result corresponds to the second weight.

[0057] In some embodiments, the target prediction result is corrected by the following steps:

[0058] determining a prediction residual sequence by an autoregressive moving average model;

[0059] determining a correction error value according to the prediction residual sequence, the autoregressive coefficient, the moving average coefficient and the white noise disturbance;

[0060] correcting the target prediction result according to the correction error value, and updating the target prediction result.

[0061] determining a correction error value according to the prediction residual sequence, the autoregressive coefficient, the moving average coefficient and the white noise disturbance, including: determining a first value according to a product of the autoregressive coefficient and the prediction residual sequence; determining a second value according to a product of the moving average coefficient and the white noise disturbance; and determining the correction error value according to a sum of the first value, the second value and the white noise disturbance.

[0062] In some embodiments, the first weight is corrected by the following steps:

[0063] determining a first prediction error according to the first prediction result and the actual output;

[0064] determining a second prediction error according to the second prediction result and the actual output;

[0065] correcting the first weight according to the first prediction error, the second prediction error and a dynamic weight adjustment factor.

[0066] Specifically, a first prediction error is determined according to a difference between the first prediction result and the actual output; a second prediction error is determined according to a difference between the second prediction result and the actual output; a third value is determined according to a product of the first prediction error and a dynamic weight adjustment factor; a fourth value is determined according to a product of the second prediction error and the dynamic weight adjustment factor; a fifth value is determined according to a sum of the third value and the fourth value; the first weight is corrected according to a quotient of the third value and the fifth value. The second weight is corrected according to a quotient of the fourth value and the fifth value. The dynamic weight adjustment factor can be set according to actual requirements.

[0067] In some embodiments, the prediction model is trained by the following steps:

[0068] obtaining historical output data samples, meteorological data samples and scheduling plan samples;

[0069] preprocessing the historical output data samples, the meteorological data samples and the scheduling plan samples to obtain input data samples;

[0070] inputting the input data samples into the prediction model respectively to obtain target prediction samples;

[0071] According to the target prediction sample and a true result, a loss function is constructed, and the prediction model is trained according to the loss function, to obtain a trained prediction model; wherein the loss function comprises a scheduling bias loss and a peak-valley ratio bias loss.

[0072] In some embodiments, the loss function is determined by the following steps:

[0073] According to the target prediction sample and a scheduling target output, a scheduling bias loss is determined; the scheduling target output is related to the scheduling plan sample;

[0074] According to the target prediction sample, a predicted peak output and a predicted valley output are determined;

[0075] According to the predicted peak output, the predicted valley output and a predetermined constant, a predicted peak-valley ratio is determined;

[0076] According to the predicted peak-valley ratio and a scheduling peak-valley ratio, a peak-valley ratio bias loss is determined, wherein the scheduling peak-valley ratio is related to the scheduling plan sample;

[0077] According to the scheduling bias loss, the peak-valley ratio bias loss and a regular loss, a loss function is determined.

[0078] According to the target prediction sample and a scheduling target output, a scheduling bias loss is determined, comprising: according to a difference between the target prediction sample and the scheduling target output, a sixth numerical value is determined, and according to a quotient of the sixth numerical value and the scheduling target output, the scheduling bias loss is determined.

[0079] According to the predicted peak output, the predicted valley output and a predetermined constant, a predicted peak-valley ratio is determined, comprising: according to a sum of the predicted valley output and the predetermined constant, a seventh numerical value is determined, and according to a quotient of the predicted peak output and the seventh numerical value, the predicted peak-valley ratio is determined; and according to a difference between the predicted peak-valley ratio and a scheduling peak-valley ratio, a peak-valley ratio bias loss is determined.

[0080] In some embodiments, the method further comprises:

[0081] According to the target prediction result and a maximum output value of the new energy equipment, a constraint prediction value is determined;

[0082] According to the constraint prediction value, the true output and a sliding time window, an average absolute error is determined;

[0083] If the average absolute error is greater than an error tolerance threshold, the prediction model is retrained.

[0084] According to the target prediction result and the maximum output value of the new energy equipment, a constraint prediction value is determined, including: according to the minimum value between the target prediction result and the maximum output value of the new energy equipment, an eighth value is determined, and according to the maximum value between the eighth value and a preset comparison value, the constraint prediction value is determined. The preset comparison value can be set according to actual needs, and for example, the preset comparison value can be 0.

[0085] In some embodiments, the method further comprises:

[0086] Obtaining a scheduling curve;

[0087] Feature encoding is performed on the scheduling curve to determine a scheduling peak-valley ratio and a scheduling target output.

[0088] In the following, the method provided by the present application is described and explained in detail in combination with specific application examples:

[0089] It should be noted that in the following examples, the formulas given are exemplary examples, and the present application does not limit the calculation formulas between the features.

[0090] The present application relates to the technical field of new energy power generation prediction and intelligent scheduling, in particular to a multi-dimensional output prediction method that fuses scheduling plan information, meteorological variables and historical power generation data, which is suitable for short-term and ultra-short-term prediction scenarios of renewable energy such as wind power and photovoltaic power.

[0091] In order to overcome the problems of poor stability and easy error accumulation of the prediction model in the prior art, the present application proposes a new energy multi-dimensional prediction technology considering scheduling plan, including: original data normalization, scheduling data structuring, multi-feature input construction, deep model modeling (CNN, LSTM, GRU), dynamic integrated prediction, error correction, scheduling constraint embedding and model self-updating mechanism. The method has strong adaptability and prediction accuracy.

[0092] The technical solution of the present application is as follows:

[0093] According to a first aspect of an embodiment of the present application, with reference to Figure 2 , a self-adaptive new energy output prediction method based on an integrated deep learning algorithm is provided, including:

[0094] Original data preprocessing, constructing an input vector for model training, and feature extraction combining historical output and meteorological information;

[0095] Constructing a single prediction model, specifically including a long short-term memory network (LSTM) state transition mechanism, a convolutional neural network (CNN) for extracting local features and a gated recurrent unit (GRU) structure;

[0096] Multi-model dynamic weighted integrated prediction;

[0097] Error modeling and correction;

[0098] Scheduling target constraints and tracking;

[0099] Result boundary control;

[0100] The adaptive new energy power output prediction method based on the above-mentioned integrated deep learning algorithm not only has adaptive adjustment capability and deployment flexibility, but is also applicable to a variety of new energy scenarios, and can significantly improve the accuracy and stability of power output prediction.

[0101] Preferably, the original data is preprocessed to input a feature vector. This includes normalizing the original new energy output sequence x(t):

[0102]

[0103] Where, x norm (t) represents the normalized new energy processing sequence; x(t) represents the original historical power output value at time t. min x max This represents the minimum and maximum values ​​of the output data.

[0104] Multidimensional input feature construction:

[0105] X(t)={x(t-1),x(t-2),...,x(tn),w(t),s(t),r(t),D t} (2)

[0106] Where X(t) represents the multidimensional input feature vector at time t, x(ti) represents the historical output data lagged by k time steps, w(t) represents meteorological data such as wind speed at time t, s(t) represents solar radiation data at time t, and r(t) represents rainfall or other external disturbance factors at time t; D t This indicates the planned scheduling value. Specifically, the meteorological data in this application includes data such as wind speed, solar radiation, and rainfall.

[0107] The scheduling curve represents the planned target output and its rate of change, including scheduling-related characteristics such as peak-to-valley ratio requirements. Its purpose in renewable energy output prediction models is to help ensure that the predicted results align with the scheduling plan by tracking and constraining the scheduling targets. For example, during model training and prediction, the features of the scheduling curve (such as the target output and rate of change) are encoded and combined with other input data (such as meteorological data and historical output) as one of the model's input features. In this way, the scheduling curve's impact on the model manifests as the control of deviations in the prediction results and the optimization of compliance with the scheduling targets.

[0108] The relationship between the scheduling curve and the model is embodied in the following aspects: first, target tracking. The model adjusts the prediction result to meet the scheduling requirements through the deviation of the scheduling plan. Second, loss function embedding. The scheduling target deviation and peak-valley ratio constraint are introduced into the loss function to ensure that the prediction value can meet the peak-valley ratio requirement of the scheduling. Third, adaptive optimization. The model dynamically adjusts through the combination of error correction mechanism and scheduling target constraint, thereby optimizing the scheduling compatibility of the prediction result.

[0109] The scheduling curve is not the prediction result of the model, but the target output plan set in advance by the system operator (such as the power grid dispatching center or the station control system), which has priori and is used for reference and tracking by the prediction model. The scheduling curve is embedded as an input feature and a loss function constraint in this patent to improve the schedulability and practicality of the prediction result.

[0110] Preferably, the scheduling curve is feature-encoded.

[0111] Dt=[P plan (t),ΔP(t),R pv (t)] (3)

[0112] Where P plan (t) represents the target output of the scheduling plan at time t, ΔP(t)=P plan (t)-P plan (t-1) represents the scheduling plan change rate. R pv (t) represents the peak-valley ratio requirement of the scheduling.

[0113] Preferably, a sub-prediction model is constructed. It includes constructing a convolutional neural network (CNN) model, performing convolution operation on the input sequence, and extracting local temporal patterns:

[0114]

[0115] Where h i represents the output of the i-th convolution unit; x j represents the j-th input feature; k ij represents the convolution kernel parameter (weight) of CNN, b i represents the bias term, and ReLU represents the linear rectification activation function max(0,x).

[0116] LSTM is established to capture long-term dependencies:

[0117] f i =σ(W f [h t-1 ,x t ]+b f ) (5)

[0118] it = σ(W i [h t-1 , x t ] + b i ) (6)

[0119] o t = σ(W o [h t-1 , x t ] + b o ) (7)

[0120]

[0121] h t = o t · tanh(c t ) (10)

[0122] where f t represents the forget gate value; i t represents the input gate value; o t represents the output gate value;

[0123] c t represents the memory cell state at the current time; c t-1 represents the memory cell state at the previous time; represents the candidate memory content; h t represents the hidden layer output at the current time; h t-1 represents the hidden layer output at the previous time; x t represents the input feature vector at the current time; W f , W i , W o , and W c represent the weight matrices (input part) that control the calculation of the input gate, the forget gate, the output gate, and the memory cell in the LSTM, each of which is multiplied by the input x t at the current time and the hidden state h t-1 at the previous time, thereby determining the state update at the current time. These weight matrices control the calculation of the input gate, the forget gate, the output gate, and the candidate memory content, which are learned through the training process and are not fixed parameters. At the beginning of training, the weight matrices are usually initialized to small random values or use a predefined initialization method. Each weight matrix is multiplied by the input at the current time and the hidden state at the previous time, thereby determining the state update at the current time. The model adjusts these weight matrices through backpropagation of training data, so that the network can capture long-term dependencies when processing sequence data; b f , b o , b i , and b cdenotes a bias vector; σ(·) denotes a Sigmoid function for gating mechanism; tanh(x) denotes a hyperbolic tangent function.

[0124] Constructing GRU model to improve training efficiency and robustness:

[0125] z t = σ(W z x t + U z h t-1 ) (11)

[0126] r t = σ(W r x r + U r h t-1 ) (12)

[0127]

[0128] wherein z t denotes an update gate; r t denotes a reset gate; denotes a candidate hidden state; h t denotes a current time hidden state; W z , W r , W h denote input weight matrices; U z , U r , U h denote hidden layer weight matrices; and denotes a Hadamard product (element-wise multiplication).

[0129] Preferably, it further comprises weighted ensemble prediction on multiple models:

[0130]

[0131] wherein, denotes the prediction output of the i-th model at time t; ω i (t) denotes the dynamic weight of the i-th model; denotes the final prediction after correction; and M denotes the total number of ensemble models. Multiple models will generate multiple prediction results based on the same input data during the entire prediction process. Next, the prediction results of these different models will be combined by adjusting the dynamic weight, so as to obtain the final comprehensive prediction result.

[0132] Preferably, it further comprises error modeling (ARMA) and error correction. The purpose of error modeling is to identify and compensate for systematic bias and random error, correct the preliminary prediction result of the model, and make the final prediction value more accurate. Specifically, the error modeling and error correction are as follows:

[0133]

[0134] represents the weighted fused preliminary output prediction value; a represents a dynamic weight adjustment factor; e i (t) represents the prediction error of the i-th model at time t; ΔP(t) represents the prediction residual sequence at time t; represents the error value predicted by the error correction module; φ k represents the AR model autoregressive coefficient; θ j represents the MA model moving average coefficient; e(t) represents the white noise disturbance term. w is a coefficient dynamically adjusted in the error correction process, used to weight the corrected prediction result, helping to improve the final prediction accuracy.

[0135] The dynamic weight adjustment factor is used to determine the weight adjustment in formula 15. Specifically, after multiple prediction models generate prediction results, the prediction value of each model will be weighted according to its performance. The role of the dynamic weight adjustment factor is to dynamically adjust the weight ai of each model according to its performance (such as error or accuracy) at time step t. The purpose of dynamically adjusting these weights is to optimize the prediction result, so that the final comprehensive prediction is as accurate as possible.

[0136] To further improve the executability of the prediction result and the consistency of the scheduling, a scheduling target deviation metric and a target tracking loss control module are introduced in the present application. This module is mainly used to model the deviation between the prediction value and the scheduling plan curve, and is embedded in the overall loss function through a regularization mechanism to realize the compatibility of peak-to-valley ratio, load response and other characteristic constraints. Specifically, the scheduling execution logic is converted into mathematical constraints and embedded in model training, so that the new energy output prediction result is not only accurate, but also better meets the scheduling execution needs of the power system. The function is to make the prediction result more consistent with the requirements of the scheduling plan, improve the scheduling compatibility, and facilitate execution; control the deviation of the prediction curve and the scheduling target. Enhance the smoothness and stability of the prediction curve to avoid sharp fluctuations or deviation from the scheduling requirements; introduce constraints on the peak-to-valley ratio to enhance the scheduling adaptability and operability of the prediction model, and meet the responsiveness requirements of the power grid, such as peak clipping and valley filling, load balancing, etc. The specific implementation is to embed the loss function through the scheduling target prediction deviation metric and the peak-to-valley ratio constraint to improve the prediction accuracy and scheduling executability.

[0137] Preferably, the scheduling target prediction deviation is:

[0138]

[0139] wherein, E plan (t) represents the relative deviation of the prediction value and the scheduling plan; represents the prediction value output by the model; Pplan (t) denotes the target output of the scheduling plan at time t.

[0140] Peak-to-valley ratio constraint term (scheduling compatibility index):

[0141]

[0142] wherein, denotes the model-predicted peak / valley output, denotes the peak-to-valley ratio required by the scheduling, ∈ denotes a minimum constant to prevent division by zero, i.e., a predetermined constant.

[0143] Complete loss function combination (including scheduling deviation penalty):

[0144] L total = λ1·L pred + λ2·L track + λ3·R pv (21)

[0145] wherein, L total denotes the total loss function, L pred denotes the basic prediction loss, L track denotes the loss related to scheduling deviation, R pv denotes the peak-to-valley ratio deviation penalty, λ1, λ2, λ3 denote the weights of each sub-term.

[0146] Preferably, it also includes a model performance evaluation and updating mechanism:

[0147]

[0148] wherein, MAE w denotes the average absolute error within the sliding window w, w denotes the size of the sliding window; denotes the final output prediction value after boundary constraints, i.e., the prediction result after "physical limitations". This formula is used to avoid the model output exceeding the actual capacity of the device (such as negative power generation or exceeding the rated capacity), to ensure that the prediction value can be used for actual scheduling, control and operation, and to prevent invalid prediction results from causing impact or misleading operation on the system; denotes the original prediction value of the model at time t; P max the maximum output limit value of the new energy device. MAE w>θ, where θ represents a preset error tolerance threshold. Exceeding this threshold triggers a model retraining mechanism. This application dynamically evaluates the overall performance of the entire prediction model system through a model performance evaluation and update mechanism, triggering updates to the overall or partial models. The prediction model system includes sub-models such as CNN, LSTM, and GRU, a dynamic ensemble mechanism, an error correction module, and a scheduling bias constraint module. These modules collectively constitute different layers of the prediction system, and performance evaluation assesses the overall output after their integration.

[0149] According to a second aspect of the embodiments of this application, a fast prediction system deployed at edge nodes is provided. Considering that the prediction model is deployed at an edge node, such as a local control system of a wind farm, referencing... Figure 3 As shown, the implementation steps include: using a compressed trained model to reduce computational load; the model input is the meteorological and power output data for the past 12 hours, and the output is the predicted value for the next hour; the power output prediction results are used to optimize wind turbine control strategies and grid connection plans; an integrated error feedback mechanism allows for independent operation without a network, improving system adaptability. Specifically, the combination of several key model modules and formulas provided in this application indirectly demonstrates the application of the power output prediction results. Mainly through power output boundary control, scheduling consistency modeling, error correction and evaluation, the controllability, stability, and executability of the prediction results are achieved, providing direct numerical input for wind turbine control and grid connection plans, and laying the foundation for the prediction system to operate independently at edge nodes.

[0150] Thirdly, this invention provides an online learning mechanism to enhance system stability and adapt to sudden output fluctuations caused by extreme weather or equipment failures. This includes: triggering adaptive adjustment when the prediction error exceeds a threshold for an extended period; introducing a label delay compensation mechanism, which uses past actual output data for fine-tuning; and online updating through a training-and-inference approach to iterate the prediction model without interrupting business processes.

[0151] The beneficial effects of this invention are compared with those of the prior art:

[0152] 1. This invention uses a combination of multiple neural network structures such as CNN, LSTM, and GRU for modeling. By extracting time-series features, nonlinear features, and local trends through the complementary characteristics of the models, it can more comprehensively characterize the multi-source influencing factors of wind power / solar power output, improve the model's ability to characterize and adapt to intermittent and fluctuating new energy power output, and thus significantly improve prediction accuracy.

[0153] 2. This invention constructs a dynamic weighted integration mechanism to adjust the weights of each sub-model in real time based on model error feedback, effectively suppressing the performance degradation of individual models caused by sudden environmental changes or training errors, and enhancing the robustness and generalization ability of the overall model.

[0154] 3. Introducing the ARMA residual prediction mechanism, the secondary correction of the initial prediction result, effectively identifying and compensating for systematic deviations and random errors in prediction, making the prediction model have certain "memory" and "learning" ability, and can adaptively adjust the strategy, significantly improving the prediction deviation.

[0155] 4. The application further fuses the scheduling plan information and the scheduling curve characteristics, and through the construction of the scheduling deviation constraint and the peak-valley ratio punishment mechanism, the prediction result is more in line with the scheduling executability requirement, and the system adaptability and practical value of the model are improved.

[0156] It can be understood that the scheduling curve in the application = the change curve of the scheduling target output over time;

[0157] The related scheduling information variables (such as the planned output, the change rate, and the peak-valley ratio) are key control information extracted and characterized according to the scheduling curve, which is used as the model input and the loss function constraint term, so as to realize the deep fusion and collaborative optimization of prediction and scheduling, and is derived from the scheduling curve;

[0158] Finally, the "scheduling information input" is formed for training the deep prediction model.

[0159] Embodiment one

[0160] An adaptive new energy output prediction method based on an integrated deep learning algorithm, comprising:

[0161] S110: Data preprocessing and feature construction. The new energy output historical data and its influencing factors (such as wind speed, irradiance, temperature, humidity, etc.) are normalized to improve the model training efficiency and convergence speed, and a prediction input feature set is constructed. A sliding window method is used to fuse historical output, environmental characteristics, time labels, etc. to form a multi-dimensional input vector;

[0162] S120: Deep learning model construction, an integrated deep learning model containing CNN, LSTM and GRU is constructed in the embodiment, which is used for multi-feature extraction and dynamic time series modeling;

[0163] S130: Model integration and dynamic weighted prediction;

[0164] S140: Residual modeling and correction mechanism;

[0165] S150: Scheduling target constraint and tracking mechanism.

[0166] S160: Model performance evaluation and updating mechanism.

[0167] Please refer to Figure 4 The embodiment of the application also provides a new energy output prediction device, which can realize the above method, and the device comprises:

[0168] The acquisition module 410 is configured to acquire historical output data, meteorological data and a scheduling plan.

[0169] The preprocessing module 420 is configured to preprocess the historical output data, the meteorological data and the scheduling plan to obtain input data.

[0170] The prediction module 430 is configured to input the input data into a first model and a second model in a prediction model respectively to obtain a first prediction result and a second prediction result, wherein the prediction model at least includes the first model and the second model.

[0171] The determination module 440 is configured to determine a target prediction result according to the first prediction result, a first weight, the second prediction result and a second weight, wherein the first weight and the second weight are corrected by an error between the target prediction result and a real output.

[0172] It can be understood that the content in the above method embodiments is applicable to the device embodiments, the device embodiments specifically implement the functions of the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0173] The embodiment of the application further provides an electronic device, which includes a memory and a processor, the memory stores a computer program, and the processor implements the above method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer and the like.

[0174] It can be understood that the content in the above method embodiments is applicable to the device embodiments, the device embodiments specifically implement the functions of the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0175] Please refer to Figure 5 , Figure 5 The hardware structure of the electronic device of another embodiment is illustrated, and the electronic device includes:

[0176] The processor 901 can be implemented in a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, and is used to execute related programs to implement the technical solutions provided by the embodiments of the application.

[0177] The memory 902 can be implemented in the form of a Read Only Memory (ROM), a static storage device, a dynamic storage device, or a Random Access Memory (RAM), etc. The memory 902 can store an operating system and other application programs, and when the technical solutions provided by the embodiments of the present specification are implemented by software or firmware, the related program codes are stored in the memory 902 and are called and executed by the processor 901 to perform the above-mentioned method of the embodiments of the present application;

[0178] The input / output interface 903 is configured to realize information input and output.

[0179] The communication interface 904 is configured to realize the communication interaction between the device and other devices, and can realize the communication through a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0180] The bus 905 is configured to transmit information between various components (for example, the processor 901, the memory 902, the input / output interface 903, and the communication interface 904) of the device.

[0181] The processor 901, the memory 902, the input / output interface 903, and the communication interface 904 are connected to each other through the bus 905 to realize the communication connection between the device.

[0182] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the above-mentioned method.

[0183] It can be understood that the contents in the above-mentioned method embodiments are all applicable to the present storage medium embodiments. The functions specifically implemented by the present storage medium embodiments are the same as those of the above-mentioned method embodiments, and the beneficial effects achieved by the present storage medium embodiments are also the same as those achieved by the above-mentioned method embodiments.

[0184] The embodiments of the present application also provide a computer program product, which includes a computer program. The computer program is executed by a processor to implement the above-mentioned method.

[0185] It can be understood that the contents in the above-mentioned method embodiments are all applicable to the present program product embodiments. The functions specifically implemented by the present program product embodiments are the same as those of the above-mentioned method embodiments, and the beneficial effects achieved by the present program product embodiments are also the same as those achieved by the above-mentioned method embodiments.

[0186] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory can include a high-speed random access memory and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory disposed remotely from the processor, which can be connected to the processor through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0187] The embodiments described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0188] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and can include more or fewer steps than shown in the figures, or combine certain steps or different steps.

[0189] The device embodiments described above are only schematic, and the units described as separate components can or can not be physically separate, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0190] Those skilled in the art can understand that all or some of the steps in the above disclosed method, the functional modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0191] The terms "first", "second", "third", "fourth" and the like used in the specification of the present application and the above-described drawings (if any) are used to distinguish similar objects, and do not necessarily have to describe a particular order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0192] It should be understood that, in the application, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B and A and B existing at the same time, wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the front and rear associated objects. "At least one of the following" or the like means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b and c can be single or multiple.

[0193] In several embodiments provided in the application, it should be understood that the disclosed 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 above units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed mutual units can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0194] The units described above as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0195] In addition, the functional units in each embodiment of the application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0196] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0197] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, and are not limited to the scope of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A new energy output prediction method, characterized in that, The method comprises the following steps: Obtaining historical output data, meteorological data and scheduling plan; Preprocessing the historical output data, the meteorological data and the scheduling plan to obtain input data; Inputting the input data into a first model and a second model in a prediction model respectively to obtain a first prediction result and a second prediction result; wherein the prediction model comprises at least the first model and the second model; Determining a target prediction result according to the first prediction result, a first weight, the second prediction result and a second weight; wherein the first weight and the second weight are corrected by an error between the target prediction result and a real output.

2. The method of claim 1, wherein, The target prediction result is corrected by the following steps: Determining a prediction residual sequence by an autoregressive moving average model; Determining a correction error value according to the prediction residual sequence, an autoregressive coefficient, a moving average coefficient and white noise interference; Correcting the target prediction result according to the correction error value to update the target prediction result.

3. The method of claim 2, wherein, The first weight is corrected by the following steps: Determining a first prediction error according to the first prediction result and the real output; Determining a second prediction error according to the second prediction result and the real output; Correcting the first weight according to the first prediction error, the second prediction error and a dynamic weight adjustment factor.

4. The method of claim 1, wherein, The prediction model is obtained by the following steps: Obtaining historical output data samples, meteorological data samples and scheduling plan samples; Preprocessing the historical output data samples, the meteorological data samples and the scheduling plan samples to obtain input data samples; Inputting the input data samples into a prediction model respectively to obtain target prediction samples; Constructing a loss function according to the target prediction samples and real results, and training the prediction model according to the loss function to obtain a trained prediction model; wherein the loss function comprises a scheduling bias loss and a peak-valley ratio bias loss.

5. The method of claim 4, wherein, The loss function is determined by the following steps: Determining a scheduling bias loss according to the target prediction samples and a scheduling target output; the scheduling target output is related to the scheduling plan samples; Determining a predicted peak output and a predicted valley output according to the target prediction samples; Determining a predicted peak-valley ratio according to the predicted peak output, the predicted valley output and a predetermined constant; Determining a peak-valley ratio bias loss according to the predicted peak-valley ratio and a scheduling peak-valley ratio; wherein the scheduling peak-valley ratio is related to the scheduling plan samples; Determining a loss function according to the scheduling bias loss, the peak-valley ratio bias loss and a regular loss.

6. The method of claim 1, wherein, The method further comprises: Determining a constraint prediction value according to the target prediction result and a maximum output value of a new energy device; Determining a mean absolute error according to the constraint prediction value, the real output and a sliding time window; If the mean absolute error is greater than an error tolerance threshold, retraining the prediction model.

7. The method of claim 1, wherein, The method further comprises: Obtaining a scheduling curve; Feature encoding the scheduling curve to determine a scheduling peak-valley ratio and a scheduling target output.

8. A new energy output prediction device characterized by comprising: The device comprises: An acquisition module is configured to acquire historical output data, meteorological data and a scheduling plan; A preprocessing module is configured to preprocess the historical output data, the meteorological data and the scheduling plan to obtain input data; A prediction module is configured to input the input data into a first model and a second model in a prediction model respectively to obtain a first prediction result and a second prediction result; wherein the prediction model at least includes the first model and the second model; A determination module is configured to determine a target prediction result according to the first prediction result, a first weight, the second prediction result and a second weight; wherein the first weight and the second weight are corrected by an error between the target prediction result and a real output.

9. An electronic device, comprising: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method of any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the method of any one of claims 1 to 7.

Images