New energy output prediction method and system based on Monte Carlo Dropout

Abstract

This application provides a method and system for predicting renewable energy output based on Monte Carlo Dropout. It uses Monte Carlo Dropout technology to probabilistically predict renewable energy output, generating dynamic prediction results with confidence intervals to quantify the uncertainties of meteorological changes and equipment status. Secondly, it constructs a multi-stage stochastic dynamic programming model, discretizing the prediction interval into multiple scenario inputs, and designs a nonlinear objective function based on discharge depth to simultaneously optimize electricity purchase costs and energy storage aging costs. Finally, it combines a model predictive control framework to achieve rolling optimization of the system, dynamically adjusting the weight of energy storage aging costs by updating prediction data and dispatch instructions online and embedding a real-time feedback mechanism for energy storage health status. This significantly improves the grid integration rate of photovoltaic and wind power, reduces the total lifecycle cost of energy storage systems, and ensures the robustness of dispatch strategies under extreme weather conditions.

Description

Technical Field

[0001] This application relates to the field of image segmentation, recognition and analysis technology, and in particular to a new energy output prediction method and device based on Monte Carlo Dropout. Background Technology

[0002] As the global energy structure accelerates its shift towards cleaner energy, the penetration rate of new energy sources such as wind power and photovoltaics in the power system continues to increase. However, their output is significantly affected by the coupling of meteorological factors and equipment status, exhibiting significant short-term fluctuations. Therefore, the power system needs to rely on the rapid charging and discharging of energy storage devices to achieve power balance. Simultaneously, the formulation of dispatch strategies highly depends on the ability to make high-precision short-term predictions and quantify risks. However, in new energy prediction-dispatch systems, traditional deterministic prediction models only output single-point estimates and cannot quantify the probability distribution of prediction errors, leading to overly conservative or risky dispatch strategies. Secondly, the optimization of energy storage system charging and discharging is mostly based on fixed aging cost coefficients, failing to dynamically correlate with actual battery operating conditions (such as SOC, temperature, and cycle depth), resulting in an imbalance between lifespan reduction and economic efficiency. Especially when participating in electricity spot market transactions, these shortcomings amplify price fluctuation risks, leading to a significant increase in the operator's revenue loss rate.

[0003] As the global energy structure accelerates its shift towards cleaner energy, the penetration rate of new energy sources such as wind power and photovoltaics in the power system continues to increase. However, their output is significantly affected by the coupling of meteorological factors and equipment status, exhibiting significant short-term fluctuations. Therefore, the power system needs to rely on the rapid charging and discharging of energy storage devices to achieve power balance. Simultaneously, the formulation of dispatch strategies highly depends on the ability to make high-precision short-term predictions and quantify risks. However, in new energy prediction-dispatch systems, traditional deterministic prediction models only output single-point estimates and cannot quantify the probability distribution of prediction errors, leading to overly conservative or risky dispatch strategies. Secondly, the optimization of energy storage system charging and discharging is mostly based on fixed aging cost coefficients, failing to dynamically correlate with actual battery operating conditions (such as SOC, temperature, and cycle depth), resulting in an imbalance between lifespan reduction and economic efficiency. Especially when participating in electricity spot market transactions, these shortcomings amplify price fluctuation risks, leading to a significant increase in the operator's revenue loss rate.

[0004] Early renewable energy dispatch systems adopted a decoupled "prediction-optimization" architecture. For example, deterministic predictions were generated based on algorithms such as ARIMA or SVM, and then energy storage plans were formulated through linear programming. These methods often led to a mismatch between dispatch instructions and actual output because they ignored the propagation of prediction uncertainty. In recent years, some studies have introduced stochastic optimization or robust optimization to handle uncertainty, but these rely on pre-set probability distribution assumptions and are difficult to characterize the true form of prediction errors under complex weather conditions. At the same time, deep learning models have improved prediction accuracy, but their black-box nature leads to insufficient quantification of uncertainty and a lack of coordination mechanism with physical constraints. On the energy storage dispatch side, existing dynamic programming methods mostly use simplified linear aging models, which cannot reflect the nonlinear characteristics of battery degradation. Although some studies have attempted to combine reinforcement learning to optimize charging and discharging strategies, offline training modes are difficult to adapt to real-time weather changes and do not form a closed-loop feedback with prediction uncertainty.

[0005] Therefore, designing renewable energy dispatching methods and systems that can significantly improve the absorption rate of photovoltaic and wind power, reduce the total life cycle cost of energy storage systems, and ensure the robustness of dispatching strategies under extreme weather conditions in a hybrid operation environment of the power market and renewable energy microgrids is an important research topic for engineers in this field. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a new energy dispatching method and system based on Monte Carlo Dropout and dynamic programming. It aims to significantly improve the absorption rate of photovoltaic and wind power, reduce the total life cycle cost of energy storage systems, and ensure the robustness of dispatching strategies under extreme weather conditions in a hybrid operation environment of the electricity market and new energy microgrids.

[0007] In the first aspect, the embodiments of this application provide a new energy output prediction method based on Monte Carlo Dropout, step 1: collect historical new energy output data, meteorological and environmental data, and real-time data of energy storage system and market electricity price;

[0008] Step 2: Preprocess the acquired data;

[0009] Step 3: Perform multiple forward propagations based on the Bi-LSTM model with Dropout layer to generate the probability distribution of new energy output prediction. Extract the quantile interval of the prediction results using the Monte Carlo method to construct multi-scenario input.

[0010] Step 4: With the optimization objectives of minimizing electricity purchase cost and energy storage aging cost, and with power balance and energy storage capacity as constraints, establish a nonlinear dynamic programming model for the new energy-energy storage dispatch system. Use dynamic programming to solve for the optimal charging and discharging strategy to achieve cost minimization under risk constraints.

[0011] Step 5: Based on the synergistic cooperation of Monte Carlo prediction and dynamic programming, realize the rolling optimization framework for new energy output prediction and scheduling strategies.

[0012] Optionally, in one implementation of the first aspect of the present invention, step 1: collecting historical power output data of new energy sources, meteorological and environmental data, and real-time data of energy storage systems and market electricity prices, includes:

[0013] The data collection consists of three parts: historical output data of new energy sources, meteorological and environmental data, and parameters of energy storage systems and market electricity prices.

[0014] Among them, the historical power output data of new energy refers to the historical multi-scale power output data of photovoltaic and wind power with a time span from 15 minutes to 72 hours;

[0015] The meteorological environmental data includes the inherent meteorological numerical characteristics of the new energy source, such as wind speed, irradiance, temperature, and cloud cover.

[0016] The data in the energy storage and electricity price market mainly includes energy storage system parameters, load demand curves, and real-time electricity price data.

[0017] Optionally, in one implementation of the first aspect of the present invention, step 2: preprocessing the acquired data includes:

[0018] Linear interpolation was used to fill in short-term missing data and remove outliers.

[0019] The data is cleaned and processed, and the resulting feature data is used to construct temporal and spatial features;

[0020] Min-Max standardization was performed on historical power output data of new energy sources and dynamic characteristics of meteorological information.

[0021] ,

[0022] in, Represents the original data. Indicates the standard value. This represents the minimum value in the dataset. This indicates the maximum value of the dataset.

[0023] Optionally, in one implementation of the first aspect of the present invention, step 3: performing multiple forward propagations based on a Bi-LSTM model with a Dropout layer to generate a probability distribution for predicting new energy output, extracting the quantile intervals of the prediction results using the Monte Carlo method, and constructing a multi-scenario input, including:

[0024] A two-layer bidirectional Bi-LSTM deep neural network is used as the basic time series prediction model, which considers both forward and backward information of the sequence to capture the dependencies in the sequence.

[0025] The model inputs time-series features, meteorological features, and time codes. The input dimension is the number of features × the historical time step. The time codes are represented by a periodic cosine function encoding method.

[0026] A Dropout layer is added after each LSTM layer, maintaining the active state during both training and inference, and dropping the probability. During each training session, the Dropout layer randomly discards some neurons with a set probability, preventing the model from becoming overly reliant on certain neurons and forcing the model to learn more robust feature representations.

[0027] The model's output layer is a fully connected network responsible for mapping the features extracted from the previous layers to the final prediction result. The output layer is used to predict the output value of new energy sources in the next 24 hours. At the same time, a 90% confidence interval is generated. To reflect the uncertainty of the prediction results, among which Indicates in The lowest value of the predicted value at any given time. exist The time indicates the highest value of the predicted value.

[0028] Optionally, in one implementation of the first aspect of the present invention, the introduction of the Monte Carlo method to enhance the uncertainty estimation of the Bi-LSTM model includes:

[0029] During the inference phase of the Bi-LSTM model, the same input data is processed... In each forward propagation, the Dropout layer randomly discards neurons, generating different prediction results, which constitute the prediction sample set. ;

[0030] Different combinations of dropped neurons can lead to different predictions from the model. By sampling multiple times to obtain multiple different prediction results, the model can capture the uncertainty estimate of the input data.

[0031] The Bi-LSTM model processes the input time-series features, meteorological features, and time codes, and outputs the predicted value of renewable energy output for the next 24 hours through forward propagation. Based on the prediction results of Bi-LSTM and the Monte Carlo method, confidence intervals are calculated to provide probabilistic input for subsequent scheduling optimization, including: sorting the sampling results in ascending order.

[0032] set up :

[0033] Determine the quantiles based on the required confidence level. For a 90% confidence interval, calculate the 5% and 95% quantiles as the upper and lower limits of the quantiles.

[0034]

[0035]

[0036] in, , These are the upper and lower limits of the confidence interval. It is a quantile function;

[0037] If the prediction results follow a Gaussian distribution, then the mean of the sample set is further calculated. and standard deviation Generate interval ,in, The value is an adjustment factor, determined according to the required confidence level. For a 90% confidence interval, under a Gaussian distribution... The value is approximately 1.645;

[0038] If the prediction results do not follow a Gaussian distribution, the nonparametric fractional method is used to generate confidence intervals.

[0039] Optionally, in one implementation of the first aspect of the present invention, step 4: taking the minimization of electricity purchase cost and energy storage aging cost as the optimization objective, and power balance and energy storage capacity as constraints, establishing a nonlinear dynamic programming model of the new energy-energy storage dispatch system, and using dynamic programming to solve for the optimal charging and discharging strategy to achieve cost minimization under risk constraints, includes:

[0040] S31. Dynamic Programming Model Construction:

[0041] The variables in the model's state equations are defined as follows:

[0042]

[0043] in: for Energy storage capacity during a given period Forecast power output range for new energy sources , The lower limit of the predicted output range for new energy sources This represents the upper limit of the predicted power output range for new energy sources. For real-time electricity prices, For load demand;

[0044] The decision variable equation is defined as follows:

[0045] Energy storage charging and discharging power satisfy Charging is positive, discharging is negative. Indicates the maximum discharge power;

[0046] The state transition, i.e., the energy storage capacity update equation, is defined as follows:

[0047]

[0048] in, , For charge / discharge efficiency;

[0049] The model's objective function minimizes the system's electricity purchase cost and energy storage aging cost as the total cost. When the output of new energy sources and the discharge of energy storage are insufficient to meet the load, the purchased electricity is:

[0050]

[0051] in, The aging cost of energy storage reflects the impact of depth of charge / discharge and state of charge (SOC) on battery life. The coefficient for dynamically updating the health status of energy storage based on the aging cycle of energy storage is used. This is the final time slot;

[0052]

[0053] in, This is the initial total capacity. For cumulative capacity loss, The coefficient representing the initial energy storage health status;

[0054] Establishing an energy storage aging cost function based on rainflow counting method:

[0055]

[0056] in, , ;

[0057] The new energy-storage dispatch model must satisfy power balance and consider energy storage charging / discharging and capacity constraints:

[0058] Power balance: , For purchased electricity, the randomness of the predicted range of renewable energy output is taken into account: ;

[0059] Energy storage safety capacity constraints: ;

[0060] Energy storage charging and discharging power limitations: ;

[0061] S32. Monte Carlo Multi-Scenario Input and Dynamic Programming Solution:

[0062] Obtaining system costs under different scenarios through Monte Carlo simulations helps evaluate the economics and stability of power dispatching systems under varying renewable energy output conditions, including:

[0063] The prediction interval generated by Monte Carlo Discretization into dynamic programming There are 10 input scenarios, each with a probability of 1. These scenarios represent the output state of new energy under different possibilities. By discretizing, the continuous distribution is transformed into multi-scenario input, which is simplified to dynamic programming for solution.

[0064] Each time period New energy output Follows a uniform distribution:

[0065]

[0066] In state Select the optimal charging and discharging power. The solution is to recursively solve the Bellman equation to minimize the sum of the current cost and the expected future cost.

[0067]

[0068] in: For immediate costs; For the next period The optimal value function under a given state;

[0069] By iterating the equation continuously, from the final time period Reverse recursion to the initial time period Finally, the optimal decision sequence for the entire scheduling cycle is obtained, and the optimal value function and strategy for each state are calculated, i.e., the charging and discharging power of energy storage at different times. The optimal value is to minimize the total cost.

[0070] Optionally, in one implementation of the first aspect of the present invention, step 5: realizing a rolling optimization framework for new energy output prediction and scheduling strategies based on the synergistic cooperation of Monte Carlo prediction and dynamic programming, includes:

[0071] Combining a model-based predictive control framework, confidence intervals are regenerated every 15 minutes based on a Bi-LSTM model and the Monte Carlo Dropout method. The scheduling strategy for the next few hours is dynamically updated using a dynamic programming algorithm, generating the latest energy storage scheduling instructions to respond to real-time data changes, including:

[0072] S41. Optimize the energy storage scheduling strategy within the window and execute the optimal action at the current moment. ;

[0073] S42. Collect actual power output data of new energy sources and energy storage status, and correct the confidence interval of new energy power output generated by the prediction model;

[0074] S43. Collect actual energy storage status and renewable energy output data, and update the energy storage aging cost coefficient. ;

[0075] The Bi-LSTM model, combined with the Monte Carlo Dropout method, generates confidence intervals for renewable energy output as inputs for dynamic programming prediction to optimization multi-scenario inputs. At the same time, it feeds back actual energy storage state data and renewable energy output data to the next round of optimization to prediction, thereby correcting the deviation between model prediction and scheduling.

[0076] Secondly, embodiments of this application provide a new energy output prediction system based on Monte Carlo Dropout, applied to the new energy output prediction method based on Monte Carlo Dropout as described in the first aspect, characterized in that it includes:

[0077] Data Acquisition Module: Collects historical power output data from new energy sources, meteorological and environmental data, and real-time data on energy storage systems and market electricity prices;

[0078] Preprocessing module: performs preprocessing on the acquired data;

[0079] Prediction module: Based on the Bi-LSTM model with Dropout layer, multiple forward propagations are performed to generate the probability distribution of new energy output prediction. The quantile interval of the prediction results is extracted by Monte Carlo method to construct multi-scenario input.

[0080] Solution module: With the optimization objectives of minimizing electricity purchase cost and energy storage aging cost, and with power balance and energy storage capacity as constraints, a nonlinear dynamic programming model of the new energy-energy storage dispatch system is established. Dynamic programming is used to solve for the optimal charging and discharging strategy to achieve cost minimization under risk constraints.

[0081] Execution module: Based on the collaborative cooperation of Monte Carlo prediction and dynamic programming, a rolling optimization framework for new energy output prediction and scheduling strategies is realized.

[0082] Thirdly, embodiments of this application provide an electronic device, including:

[0083] processor;

[0084] Memory used to store processor-executable instructions;

[0085] The processor is configured to implement the Monte Carlo Dropout-based new energy output prediction method as described in the first aspect when executing the instructions.

[0086] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program that instructs a device to execute the Monte Carlo Dropout-based renewable energy output prediction method as described in the first aspect.

[0087] The technical solution provided by this invention offers a method and device for predicting renewable energy output based on Monte Carlo Dropout. It uses Monte Carlo Dropout technology to probabilistically predict renewable energy output, generating dynamic prediction results with confidence intervals to quantify the uncertainties of meteorological changes and equipment status. Secondly, it constructs a multi-stage stochastic dynamic programming model, discretizing the prediction interval into multiple scenario inputs, and designs a nonlinear objective function based on discharge depth to simultaneously optimize electricity purchase costs and energy storage aging costs. Finally, it combines a model predictive control framework to achieve rolling optimization of the system, dynamically adjusting the weight of energy storage aging costs by updating prediction data and scheduling instructions online and embedding a real-time feedback mechanism for energy storage health status. This significantly improves the absorption rate of photovoltaic and wind power, reduces the total life-cycle cost of energy storage systems, and ensures the robustness of scheduling strategies under extreme weather conditions.

[0088] Beneficial effects:

[0089] 1. By coupling Monte Carlo Dropout with dynamic programming, a confidence interval for new energy output is generated and discretized into multi-scenario inputs. The impact of uncertainty on scheduling is quantified, the robustness of decision-making under extreme weather conditions is improved, and the loss of revenue caused by prediction bias is reduced.

[0090] 2. Based on the dynamic embedding of charge / discharge depth and nonlinear aging cost, the degradation characteristics of energy storage are accurately reflected, extending the life of energy storage while reducing operation and maintenance costs;

[0091] 3. Design a "prediction-optimization" closed-loop system that dynamically updates the interval prediction and scheduling strategy every 15 minutes. Combined with hardware resource support, it achieves second-level response to ensure that the power system meets real-time power balance and improves arbitrage profits in the context of power market fluctuations. Attached Figure Description

[0092] Figure 1 This is a schematic diagram of a new energy output prediction method based on Monte Carlo Dropout provided in an embodiment of this application.

[0093] Figure 2 This is a schematic diagram of a new energy dispatching method provided in an embodiment of this application.

[0094] Figure 3 This is a schematic diagram of a new energy output prediction system module based on Monte Carlo Dropout, provided as an embodiment of this application.

[0095] Figure 4 A schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0096] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0097] It should be noted that, in the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. 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 in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0098] It should be noted that in the embodiments of this application, the terms "first," "second," etc., are used only for descriptive purposes and should not be construed as indicating or implying relative importance, nor as indicating or implying order. Features specified as "first" or "second" can explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0099] Based on the embodiments described in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0100] Example 1

[0101] Figure 1 This is a schematic diagram of a new energy output prediction method based on Monte Carlo Dropout provided in an embodiment of this application.

[0102] like Figure 1 As shown, this application provides a new energy output prediction method based on Monte Carlo Dropout, which includes the following steps.

[0103] Step 1: Collect historical power output data of new energy sources, meteorological and environmental data, and real-time data of energy storage systems and market electricity prices.

[0104] Specifically, the data collection mainly consists of three parts: historical output data of new energy sources, meteorological and environmental data, and energy storage system and market electricity price parameters. Among them, historical output data of new energy sources refers to historical multi-scale output data of photovoltaic and wind power with a time span from 15 minutes to 72 hours; meteorological and environmental data includes the inherent meteorological numerical characteristics of the new energy sources, such as wind speed, irradiance, temperature, and cloud cover; while the data of energy storage and electricity price market mainly includes energy storage system parameters (capacity, efficiency, SOC limit), load demand curves, and real-time electricity price data.

[0105] Step 2: Preprocess the acquired data.

[0106] Specifically, step 2 involves preprocessing the acquired data, including:

[0107] Linear interpolation was used to fill in short-term missing data and remove outliers.

[0108] The data is cleaned and processed, and the resulting feature data is used to construct temporal and spatial features;

[0109] Min-Max standardization was performed on historical power output data of new energy sources and dynamic characteristics of meteorological information.

[0110] ,

[0111] in, Represents the original data. Indicates the standard value. This represents the minimum value in the dataset. This indicates the maximum value of the dataset.

[0112] Specifically, the acquired data undergoes preprocessing, such as using linear interpolation to fill in short-term missing data (≤3 time points) and removing outliers. The cleaned feature data is then used to construct temporal features (lag variables, sliding window statistics (24-hour moving mean, variance)) and spatial features (constructing a power output correlation matrix of neighboring stations based on Pearson correlation coefficient (>0.8).

[0113] Step 3: Perform multiple forward propagations based on the Bi-LSTM model with Dropout layer to generate the probability distribution of new energy output prediction. Extract the quantile interval of the prediction results using the Monte Carlo method to construct multi-scenario input.

[0114] Specifically, Figure 2 This is a schematic diagram of a new energy dispatching method provided in an embodiment of this application. Figure 2As shown, step 2 involves performing multiple forward propagations based on a Bi-LSTM model with a Dropout layer to generate a probability distribution for predicting new energy output. The Monte Carlo method is then used to extract the quantile intervals of the prediction results, and a multi-scenario input is constructed, including:

[0115] A two-layer bidirectional Bi-LSTM deep neural network is used as the basic time series prediction model, which considers both forward and backward information of the sequence to capture the dependencies in the sequence.

[0116] The model inputs time-series features, meteorological features, and time codes. The input dimension is the number of features × the historical time step. The time codes are represented by a periodic cosine function encoding method.

[0117] A Dropout layer is added after each LSTM layer, maintaining the active state during both training and inference, and dropping the probability. During each training session, the Dropout layer randomly discards some neurons with a set probability, preventing the model from becoming overly reliant on certain neurons and forcing the model to learn more robust feature representations.

[0118] The model's output layer is a fully connected network responsible for mapping the features extracted from the previous layers to the final prediction result. The output layer is used to predict the output value of new energy sources in the next 24 hours. At the same time, a 90% confidence interval is generated. To reflect the uncertainty of the prediction results, among which Indicates in The lowest value of the predicted value at any given time. exist The time indicates the highest value of the predicted value.

[0119] Specifically, a two-layer bidirectional Bi-LSTM deep neural network is used as the basic time series prediction model, which considers both forward and backward information of the sequence to more comprehensively capture the dependencies in the sequence. The model inputs time series features, meteorological features, and time codes (time is represented by a periodic cosine function encoding method), and the input dimension is the number of features × the historical time step.

[0120] To enhance the model's generalization ability, a Dropout layer is added after each LSTM layer, maintaining the active state and dropping probabilities during both training and inference. In other words, during each training session, the Dropout layer randomly discards some neurons with a set probability, preventing the model from becoming overly reliant on certain specific neurons and thus forcing the model to learn more robust feature representations.

[0121] The model's output layer is a fully connected network responsible for mapping the features extracted from the previous layers to the final prediction result. The output layer not only predicts the output value of new energy sources for the next 24 hours, but also... It will also generate a 90% confidence interval. This is to reflect the uncertainty of the prediction results.

[0122] Specifically, the Monte Carlo method is introduced to enhance the uncertainty estimation of the Bi-LSTM model, including:

[0123] During the inference phase of the Bi-LSTM model, the same input data is processed... In each forward propagation, the Dropout layer randomly discards neurons, generating different prediction results, which constitute the prediction sample set. ;

[0124] Different combinations of dropped neurons can lead to different predictions from the model. By sampling multiple times to obtain multiple different prediction results, the model can capture the uncertainty estimate of the input data.

[0125] The Bi-LSTM model processes the input time-series features, meteorological features, and time codes, and outputs the predicted value of renewable energy output for the next 24 hours through forward propagation. Based on the prediction results of Bi-LSTM and the Monte Carlo method, confidence intervals are calculated to provide probabilistic input for subsequent scheduling optimization, including: sorting the sampling results in ascending order.

[0126] set up :

[0127] Determine the quantiles based on the required confidence level. For a 90% confidence interval, calculate the 5% and 95% quantiles as the upper and lower limits of the quantiles.

[0128]

[0129]

[0130] in, , These are the upper and lower limits of the confidence interval. It is a quantile function;

[0131] If the prediction results follow a Gaussian distribution, then the mean of the sample set is further calculated. and standard deviation Generate interval ,in, The value is an adjustment factor, determined according to the required confidence level. For a 90% confidence interval, under a Gaussian distribution... The value is approximately 1.645;

[0132] If the prediction results do not follow a Gaussian distribution, the nonparametric fractional method is used to generate confidence intervals.

[0133] Specifically, this invention introduces the Monte Carlo method to enhance the uncertainty estimation of the Bi-LSTM model, approximating the solution to the problem through extensive random sampling. The inference phase of the Bi-LSTM model is similar to Monte Carlo sampling, performing sampling on the same input data. In each forward pass, the Dropout layer randomly discards neurons, generating different prediction results. These results constitute the prediction sample set. Because different combinations of discarded neurons can lead to different predictions from the model, multiple samplings can be used to obtain multiple different prediction results, thereby capturing the model's estimation of uncertainty in the input data.

[0134] The Bi-LSTM model processes the input time-series features, meteorological features, and time codes, and outputs a predicted value for the renewable energy output for the next 24 hours through forward propagation. Based on the Bi-LSTM prediction results and the Monte Carlo method, confidence intervals are calculated to provide probabilistic input for subsequent scheduling optimization. Specifically, the sampling results are sorted in ascending order:

[0135] set up :

[0136] Determine the quantiles based on the required confidence level. For a 90% confidence interval, calculate the 5% and 95% quantiles as the upper and lower limits of the quantiles:

[0137]

[0138]

[0139] in, , These are the upper and lower limits of the confidence interval. This is a quantile function.

[0140] If the prediction results follow a Gaussian distribution, then the mean of the sample set can be further calculated. and standard deviation Generate interval ,in, The value is determined based on the required confidence level; for a 90% confidence interval, under a Gaussian distribution... The value is approximately 1.645; if the prediction result does not follow a Gaussian distribution, then the above nonparametric score method is used to generate confidence intervals.

[0141] Step 4: With the optimization objectives of minimizing electricity purchase cost and energy storage aging cost, and with power balance and energy storage capacity as constraints, establish a nonlinear dynamic programming model for the new energy-energy storage dispatch system. Use dynamic programming to solve for the optimal charging and discharging strategy to achieve cost minimization under risk constraints.

[0142] Specifically, step 4: With minimizing electricity purchase cost and energy storage aging cost as optimization objectives, and power balance and energy storage capacity as constraints, a nonlinear dynamic programming model of the new energy-energy storage dispatch system is established. Dynamic programming is used to solve for the optimal charging and discharging strategy to achieve cost minimization under risk constraints, including:

[0143] S31. Dynamic Programming Model Construction:

[0144] The variables in the model's state equations are defined as follows:

[0145]

[0146] in: for Energy storage capacity during a given period Forecast power output range for new energy sources , The lower limit of the predicted output range for new energy sources This represents the upper limit of the predicted power output range for new energy sources. For real-time electricity prices, For load demand;

[0147] The decision variable equation is defined as follows:

[0148] Energy storage charging and discharging power satisfy Charging is positive, discharging is negative. Indicates the maximum discharge power;

[0149] The state transition, i.e., the energy storage capacity update equation, is defined as follows:

[0150]

[0151] in, , For charge / discharge efficiency;

[0152] The model's objective function minimizes the system's electricity purchase cost and energy storage aging cost as the total cost. When the output of new energy sources and the discharge of energy storage are insufficient to meet the load, the purchased electricity is:

[0153]

[0154] in, The aging cost of energy storage reflects the impact of depth of charge / discharge and state of charge (SOC) on battery life. The coefficient for dynamically updating the health status of energy storage based on the aging cycle of energy storage is used. This is the final time slot;

[0155]

[0156] in, This is the initial total capacity. For cumulative capacity loss, The coefficient representing the initial energy storage health status;

[0157] Establishing an energy storage aging cost function based on rainflow counting method:

[0158]

[0159] in, , ;

[0160] The new energy-storage dispatch model must satisfy power balance and consider energy storage charging / discharging and capacity constraints:

[0161] Power balance: , For purchased electricity, the randomness of the predicted range of renewable energy output is taken into account: ;

[0162] Energy storage safety capacity constraints: ;

[0163] Energy storage charging and discharging power limitations: .

[0164] S32. Monte Carlo Multi-Scenario Input and Dynamic Programming Solution:

[0165] Obtaining system costs under different scenarios through Monte Carlo simulations helps evaluate the economics and stability of power dispatching systems under varying renewable energy output conditions, including:

[0166] The prediction interval generated by Monte Carlo Discretization into dynamic programming There are 10 input scenarios, each with a probability of 1. These scenarios represent the output state of new energy under different possibilities. By discretizing, the continuous distribution is transformed into multi-scenario input, which is simplified to dynamic programming for solution.

[0167] Each time period New energy output Follows a uniform distribution:

[0168]

[0169] In state Select the optimal charging and discharging power. The solution is to recursively solve the Bellman equation to minimize the sum of the current cost and the expected future cost.

[0170]

[0171] in: For immediate costs; For the next period The optimal value function under a given state;

[0172] By iterating the equation continuously, from the final time period Reverse recursion to the initial time period Finally, the optimal decision sequence for the entire scheduling cycle is obtained, and the optimal value function and strategy for each state are calculated, i.e., the charging and discharging power of energy storage at different times. The optimal value is to minimize the total cost.

[0173] Step 5: Based on the synergistic cooperation of Monte Carlo prediction and dynamic programming, realize the rolling optimization framework for new energy output prediction and scheduling strategies.

[0174] Specifically, step 5: Based on the synergistic cooperation of Monte Carlo prediction and dynamic programming, a rolling optimization framework for new energy output prediction and scheduling strategies is realized, including:

[0175] Combining a model-based predictive control framework, confidence intervals are regenerated every 15 minutes based on a Bi-LSTM model and the Monte Carlo Dropout method. The scheduling strategy for the next few hours is dynamically updated using a dynamic programming algorithm, generating the latest energy storage scheduling instructions to respond to real-time data changes, including:

[0176] S41. Optimize the energy storage scheduling strategy within the window and execute the optimal action at the current moment. ;

[0177] S42. Collect actual power output data of new energy sources and energy storage status, and correct the confidence interval of new energy power output generated by the prediction model;

[0178] S43. Collect actual energy storage status and renewable energy output data, and update the energy storage aging cost coefficient. ;

[0179] The Bi-LSTM model, combined with the Monte Carlo Dropout method, generates confidence intervals for renewable energy output as inputs for dynamic programming prediction to optimization multi-scenario inputs. At the same time, it feeds back actual energy storage state data and renewable energy output data to the next round of optimization to prediction, thereby correcting the deviation between model prediction and scheduling.

[0180] Specifically, the Bi-LSTM model, combined with the Monte Carlo Dropout method, generates confidence intervals for renewable energy output as multi-scenario inputs for dynamic programming (from prediction to optimization). At the same time, the actual energy storage status data and renewable energy output data are fed back to the next round of prediction (from optimization to prediction), thereby correcting the deviation between model prediction and scheduling.

[0181] Example 2

[0182] like Figure 3 As shown, this application provides a new energy output prediction system based on Monte Carlo Dropout, which is applied to the new energy output prediction method based on Monte Carlo Dropout as described in Embodiment 1, including: acquisition module 11, preprocessing module 12, prediction module 13, solution module 14, and execution module 15.

[0183] It is understood that in this embodiment, the acquisition module 11 acquires historical power output data of new energy sources, meteorological and environmental data, and real-time data of energy storage systems and market electricity prices.

[0184] It is understood that in this embodiment, the preprocessing module 12 preprocesses the acquired data;

[0185] It is understood that in this embodiment, the prediction module 13 performs multiple forward propagations based on the Bi-LSTM model with Dropout layer to generate the probability distribution of new energy output prediction, extracts the quantile interval of the prediction results through the Monte Carlo method, and constructs multi-scenario input.

[0186] It is understood that in this embodiment, the solution module 14: takes minimizing the cost of electricity purchase and the cost of energy storage aging as the optimization objective, and takes power balance and energy storage capacity as constraints, establishes a nonlinear dynamic programming model of the new energy-energy storage dispatch system, and uses dynamic programming to solve the optimal charging and discharging strategy to achieve cost minimization under risk constraints;

[0187] It is understood that in this embodiment, the execution module 15 is a rolling optimization framework for new energy output prediction and scheduling strategies based on the collaborative cooperation of Monte Carlo prediction and dynamic programming.

[0188] Figure 4 This is an electronic device provided in one embodiment of this application. For example... Figure 4 As shown, the electronic device includes at least the following components: processor 101 and memory 100, communication interface 103, and bus 102.

[0189] In this embodiment, the memory 100 is used to store executable instructions of the processor 101, which, when configured to execute instructions, implements... Figure 3 The device module shown.

[0190] In embodiments of this application, a computer-readable storage medium includes instructions that instruct a device to perform the method as described in the first aspect. For example, the instructions instruct the device to perform... Figure 1 The process steps in the process.

[0191] In one embodiment of this application, the program operating in the electronic device is a program that controls a central processing unit (CPU) or similar device to achieve the functions of the above-described embodiments of the present invention (a program that enables the computer to function). Information processed by these devices is then temporarily stored in random access memory (RAM) during processing, and subsequently stored in various ROMs such as read-only memory (FlashROM) and hard disk drives (HDDs), and read, corrected, and written by the CPU as needed.

[0192] It should be noted that a portion of the electronic device described above can also be implemented using a computer. In this case, the program for implementing the control function can be recorded on a computer-readable recording medium, and the program recorded on the recording medium can be read into a computer and executed.

[0193] It should be noted that the term "computer" as used here refers to a computer built into an electronic device, employing hardware including an operating system and peripheral devices. Furthermore, "computer-readable recording media" refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard drives built into a computer.

[0194] Furthermore, a "computer-readable recording medium" can include: a medium that dynamically stores a program for a short period of time, such as a communication line used when transmitting a program via a network such as the Internet or a communication line such as a telephone line; and a medium that stores a program for a fixed period of time, such as volatile memory inside a computer that serves as a server or client in this case. In addition, the aforementioned program can be a program used to implement the above-mentioned functions, or it can be a program that can implement the above-mentioned functions by combining with a program already recorded in the computer.

[0195] Furthermore, the electronic device in the above embodiments can also be implemented as an assembly (device group) composed of multiple devices. Each device constituting the device group can possess some or all of the functions or functional blocks of the electronic device in the above embodiments. As a device group, it is sufficient to have all the functions or functional blocks of the electronic device.

[0196] Those skilled in the art should recognize that the above embodiments are only used to illustrate this application and are not intended to limit this application. Any appropriate changes and variations made to the above embodiments within the essential spirit and scope of this application fall within the scope of protection claimed by this application.

Claims

1. A new energy output prediction method based on Monte Carlo Dropout, characterized in that, The method includes: Step 1: Collect historical power output data of new energy sources, meteorological and environmental data, and real-time data on energy storage systems and market electricity prices; Step 2: Preprocess the acquired data; Step 3: Perform multiple forward propagations based on the Bi-LSTM model with Dropout layer to generate the probability distribution of new energy output prediction. Extract the quantile interval of the prediction results using the Monte Carlo method to construct multi-scenario input. Step 4: With the optimization objectives of minimizing electricity purchase cost and energy storage aging cost, and constrained by power balance and energy storage capacity, establish a nonlinear dynamic programming model for the new energy-energy storage dispatch system. Use dynamic programming to solve for the optimal charging and discharging strategy to minimize costs under risk constraints; including: S31. Dynamic Programming Model Construction: The variables in the model's state equations are defined as follows: ； in: for Energy storage capacity during a given period Forecast power output range for new energy sources , The lower limit of the predicted output range for new energy sources This represents the upper limit of the predicted power output range for new energy sources. For real-time electricity prices, For load demand; The decision variable equation is defined as follows: Energy storage charging and discharging power satisfy Charging is positive, discharging is negative. Indicates the maximum discharge power; The state transition, i.e., the energy storage capacity update equation, is defined as follows: ； in, , For charge / discharge efficiency; The model's objective function minimizes the system's electricity purchase cost and energy storage aging cost as the total cost. When the output of new energy sources and the discharge of energy storage are insufficient to meet the load, the purchased electricity is: ； in, The aging cost of energy storage reflects the impact of depth of charge / discharge and state of charge (SOC) on battery life. The coefficient for dynamically updating the health status of energy storage based on the aging cycle of energy storage is used. This is the final time slot; ； in, This is the initial total capacity. For cumulative capacity loss, The coefficient representing the initial energy storage health status; Establishing an energy storage aging cost function based on rainflow counting method: ； in, , ; The new energy-storage dispatch model must satisfy power balance and consider energy storage charging / discharging and capacity constraints: Power balance: , For purchased electricity, the randomness of the predicted range of renewable energy output is taken into account: ; Energy storage safety capacity constraints: ; Energy storage charging and discharging power limitations: ; S32. Monte Carlo Multi-Scenario Input and Dynamic Programming Solution: Obtaining system costs under different scenarios through Monte Carlo simulations helps evaluate the economics and stability of power dispatching systems under varying renewable energy output conditions, including: The prediction interval generated by Monte Carlo Discretization into dynamic programming There are 10 input scenarios, each with a probability of 1. These scenarios represent the output state of new energy under different possibilities. By discretizing, the continuous distribution is transformed into multi-scenario input, which is simplified to dynamic programming for solution. Each time period New energy output Follows a uniform distribution: ； In state Select the optimal charging and discharging power. The solution is to recursively solve the Bellman equation to minimize the sum of the current cost and the expected future cost. ； in: For immediate costs; For the next period The optimal value function under a given state; By iterating the equation continuously, from the final time period Reverse recursion to the initial time period Finally, the optimal decision sequence for the entire scheduling cycle is obtained, and the optimal value function and strategy for each state are calculated, i.e., the charging and discharging power of energy storage at different times. The optimal value is found to minimize the total cost; Step 5: Based on the synergistic cooperation of Monte Carlo prediction and dynamic programming, realize the rolling optimization framework for new energy output prediction and scheduling strategies.

2. The renewable energy output prediction method based on Monte Carlo Dropout according to claim 1, characterized in that, Step 1: Collect historical power output data of new energy sources, meteorological and environmental data, and real-time data on energy storage systems and market electricity prices, including: The data collection consists of three parts: historical output data of new energy sources, meteorological and environmental data, and parameters of energy storage systems and market electricity prices. Among them, the historical power output data of new energy refers to the historical multi-scale power output data of photovoltaic and wind power with a time span from 15 minutes to 72 hours; The meteorological environmental data includes the inherent meteorological numerical characteristics of the new energy source, such as wind speed, irradiance, temperature, and cloud cover. The data in the energy storage and electricity price market mainly includes energy storage system parameters, load demand curves, and real-time electricity price data.

3. The renewable energy output prediction method based on Monte Carlo Dropout according to claim 2, characterized in that, Step 2: Preprocessing the acquired data, including: Linear interpolation was used to fill in short-term missing data and remove outliers. The data is cleaned and processed, and the resulting feature data is used to construct temporal and spatial features; Min-Max standardization was performed on historical power output data of new energy sources and dynamic characteristics of meteorological information. ， in, Represents the original data. Indicates the standard value. This represents the minimum value in the dataset. This indicates the maximum value of the dataset.

4. The renewable energy output prediction method based on Monte Carlo Dropout as described in claim 1, characterized in that, Step 3: Based on the Bi-LSTM model with Dropout layer, multiple forward propagations are performed to generate the probability distribution of new energy output prediction. The quantile intervals of the prediction results are extracted using the Monte Carlo method to construct multi-scenario inputs, including: A two-layer bidirectional Bi-LSTM deep neural network is used as the basic time series prediction model, which considers both forward and backward information of the sequence to capture the dependencies in the sequence. The model inputs time-series features, meteorological features, and time codes. The input dimension is the number of features × the historical time step. The time codes are represented by a periodic cosine function encoding method. A Dropout layer is added after each LSTM layer, maintaining the active state during both training and inference, and dropping the probability. During each training session, the Dropout layer randomly discards some neurons with a set probability, preventing the model from becoming overly reliant on certain neurons and forcing the model to learn more robust feature representations. The model's output layer is a fully connected network responsible for mapping the features extracted from the previous layers to the final prediction result. The output layer is used to predict the output value of new energy sources in the next 24 hours. At the same time, a 90% confidence interval is generated. To reflect the uncertainty of the prediction results, among which Indicates in The lowest value of the predicted value at any given time. exist The time indicates the highest value of the predicted value.

5. The renewable energy output prediction method based on Monte Carlo Dropout according to claim 4, characterized in that, The Monte Carlo method is introduced to enhance the uncertainty estimation of the Bi-LSTM model, including: During the inference phase of the Bi-LSTM model, the same input data is processed... In each forward propagation, the Dropout layer randomly discards neurons, generating different prediction results, which constitute the prediction sample set. ; Different combinations of dropped neurons can lead to different predictions from the model. By sampling multiple times to obtain multiple different prediction results, the model can capture the uncertainty estimate of the input data. The Bi-LSTM model processes the input time-series features, meteorological features, and time codes, and outputs the predicted value of renewable energy output for the next 24 hours through forward propagation. Based on the prediction results of Bi-LSTM and the Monte Carlo method, confidence intervals are calculated to provide probabilistic input for subsequent scheduling optimization, including: sorting the sampling results in ascending order. set up : Determine the quantiles based on the required confidence level. For a 90% confidence interval, calculate the 5% and 95% quantiles as the upper and lower limits of the quantiles. ； ； in, , These are the upper and lower limits of the confidence interval. It is a quantile function; If the prediction results follow a Gaussian distribution, then the mean of the sample set is further calculated. and standard deviation Generate interval ,in, The value is an adjustment factor, determined according to the required confidence level. For a 90% confidence interval, under a Gaussian distribution... The value is approximately 1.645; If the prediction results do not follow a Gaussian distribution, the nonparametric fractional method is used to generate confidence intervals.

6. The renewable energy output prediction method based on Monte Carlo Dropout according to claim 1, characterized in that, Step 5: Based on the synergistic cooperation of Monte Carlo prediction and dynamic programming, a rolling optimization framework for new energy output prediction and scheduling strategies is realized, including: Combining a model-based predictive control framework, confidence intervals are regenerated every 15 minutes based on a Bi-LSTM model and the Monte Carlo Dropout method. The scheduling strategy for the next few hours is dynamically updated using a dynamic programming algorithm, generating the latest energy storage scheduling instructions to respond to real-time data changes, including: S41. Optimize the energy storage scheduling strategy within the window and execute the optimal action at the current moment. ; S42. Collect actual power output data of new energy sources and energy storage status, and correct the confidence interval of new energy power output generated by the prediction model; S43. Collect actual energy storage status and renewable energy output data, and update the energy storage aging cost coefficient. ; The Bi-LSTM model, combined with the Monte Carlo Dropout method, generates confidence intervals for renewable energy output as inputs for dynamic programming prediction to optimization multi-scenario inputs. At the same time, it feeds back actual energy storage state data and renewable energy output data to the next round of optimization to prediction, thereby correcting the deviation between model prediction and scheduling.

7. A new energy output prediction system based on Monte Carlo Dropout, applied to the new energy output prediction method based on Monte Carlo Dropout as described in any one of claims 1 to 6, characterized in that, include: Data Acquisition Module: Collects historical power output data from new energy sources, meteorological and environmental data, and real-time data on energy storage systems and market electricity prices; Preprocessing module: performs preprocessing on the acquired data; Prediction module: Based on the Bi-LSTM model with Dropout layer, multiple forward propagations are performed to generate the probability distribution of new energy output prediction. The quantile interval of the prediction results is extracted by Monte Carlo method to construct multi-scenario input. Solution module: With the optimization objectives of minimizing electricity purchase cost and energy storage aging cost, and with power balance and energy storage capacity as constraints, a nonlinear dynamic programming model of the new energy-energy storage dispatch system is established. Dynamic programming is used to solve for the optimal charging and discharging strategy to achieve cost minimization under risk constraints. Execution module: Based on the collaborative cooperation of Monte Carlo prediction and dynamic programming, a rolling optimization framework for new energy output prediction and scheduling strategies is realized.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the Monte Carlo Dropout-based new energy output prediction method as described in any one of claims 1 to 6 when executing the instructions.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that instructs the device to perform the Monte Carlo Dropout-based renewable energy output prediction method as described in any one of claims 1 to 6.

Images