Multi-time scale energy scheduling method and system based on electricity price prediction

Abstract

The invention discloses a multi-time scale energy scheduling method and system based on electricity price prediction. The method comprises the steps of obtaining electricity price prediction values on multiple time scales in the future; generating an energy scheduling strategy on multiple time scales; executing an energy scheduling strategy on multiple time scales, monitoring an actual electricity price in real time, and comparing the actual electricity price with a corresponding electricity price prediction value to determine a prediction error; judging whether the prediction error is greater than an error threshold; in response to the fact that the prediction error is larger than an error threshold value, updating parameters of the electricity price prediction model according to the prediction error, and returning to the step of obtaining electricity price prediction values on multiple time scales in the future; and taking the current energy scheduling strategy on the multi-time scale as a final energy scheduling strategy on the multi-time scale in response to the fact that the prediction error is not greater than the error threshold value. By using the scheme of the invention, the electricity price trend in the short period, the middle period and even the long period in the future can be pre-judged, and a solid foundation is provided for formulating a comprehensive and fine energy scheduling strategy.

Description

Technical Field

[0001] This application generally relates to the field of energy dispatching technology. More specifically, this application relates to a multi-timescale energy dispatching method and system based on electricity price forecasting. Background Technology

[0002] Under the dual pressures of continuously growing global energy demand and increasingly severe climate change, improving energy efficiency and promoting a sustainable energy structure have become key global issues. Against this backdrop, Integrated Energy Systems (IES), as an advanced energy system capable of unified planning, scheduling, and management of multiple energy forms such as electricity, heat, cooling, and gas, have become an important technological means to achieve cascaded energy utilization, reduce overall energy consumption, and promote the absorption of renewable energy.

[0003] To optimize the economic operation of integrated energy systems, existing technologies typically employ dispatch strategies based on electricity price forecasts. By analyzing historical electricity price data and utilizing machine learning models, future electricity prices can be predicted. This guides the system to store or purchase electricity during off-peak hours and release energy or reduce electricity purchases during peak hours, thereby lowering operating costs.

[0004] However, existing electricity price forecasting and energy dispatching methods still have the following obvious technical shortcomings: Single scheduling scale: Most existing scheduling methods only optimize on a single time scale (e.g., day-ahead or real-time), lacking comprehensive consideration and coordinated planning of short-term, medium-term, and long-term electricity price trends. This single-scale scheduling strategy often leads to neglecting one aspect for another and failing to achieve global optimization. For example, charging and discharging behavior pursued for short-term arbitrage may conflict with seasonal long-term energy storage goals, thus limiting the overall economic benefits of the system.

[0005] Lack of error feedback and model adaptability: Electricity price forecasting models inevitably produce forecasting errors, which directly affect the actual execution effectiveness and economic efficiency of dispatching schemes. Existing methods are mostly open-loop systems, meaning that once a dispatching scheme is formulated, it is rarely or never adjusted based on real-time forecasting errors, and the forecasting model itself cannot learn from new error data and self-correct. This static, feedback-lacking mechanism makes the system unable to adapt to dynamically changing market environments, leading to a deterioration in dispatching effectiveness over time.

[0006] In view of this, there is an urgent need to provide a multi-timescale energy dispatching scheme based on electricity price forecasting to solve the problems of single dispatching scale and inability to perform closed-loop feedback optimization based on forecasting errors in existing technologies. Summary of the Invention

[0007] In order to at least address one or more of the technical problems mentioned above, this application proposes a multi-timescale energy dispatch scheme based on electricity price forecasting in several aspects.

[0008] In a first aspect, this application provides a multi-timescale energy dispatching method based on electricity price forecasting, comprising: obtaining electricity price forecasts for multiple future timescales using an electricity price forecasting model based on historical data; generating a multi-timescale energy dispatching strategy based on the electricity price forecasts for multiple future timescales; executing the multi-timescale energy dispatching strategy and monitoring the actual electricity price in real time, comparing it with the corresponding electricity price forecast to determine the forecast error; determining whether the forecast error is greater than an error threshold; in response to the forecast error being greater than the error threshold, updating the parameters of the electricity price forecasting model according to the forecast error, and returning to the step of obtaining electricity price forecasts for multiple future timescales using the electricity price forecasting model based on historical data; and in response to the forecast error not being greater than the error threshold, using the current multi-timescale energy dispatching strategy as the final multi-timescale energy dispatching strategy.

[0009] In some embodiments, the electricity price forecasts over multiple time scales include short-term, medium-term, and long-term electricity price forecasts.

[0010] In some embodiments, the energy dispatch strategy across multiple time scales includes a short-term energy dispatch strategy, a medium-term energy dispatch strategy, and a long-term energy dispatch strategy.

[0011] In some embodiments, a short-term energy dispatch strategy is obtained by minimizing the short-term electricity purchase cost; wherein, the expression for the short-term electricity purchase cost is: , For short-term electricity purchase costs, For short-term scheduling duration, The amount of electricity purchased at time step t. Let be the short-term electricity price forecast at time step t. To comprehensively consider the short-term energy storage dispatch cost of energy in time step t, To save energy storage dispatch costs during load demand response at time step t; , The amount of charge at time step t. Let be the amount of charge discharged at time step t. This represents the short-term electricity price forecast at time step t. , This represents the amount of electricity reduced during the load demand response period at time step t. Let be the short-term electricity price forecast at time step t.

[0012] In some embodiments, a medium-term energy dispatch strategy is obtained by minimizing the medium-term electricity purchase cost; wherein the expression for the medium-term electricity purchase cost is: , For medium-term electricity purchase costs, The duration of the intermediate scheduling. The amount of electricity purchased at time step t. This represents the amount of electricity sold at time step t. The medium-term electricity price forecast at time step t. To save on electricity purchase costs by generating renewable energy in time steps, For integrated energy in time step t Energy storage dispatch costs; , ; , For the amount of electricity generated by renewable energy at time step t, .

[0013] In some embodiments, a long-term energy dispatch strategy is obtained by minimizing the long-term electricity purchase cost; wherein the expression for the long-term electricity purchase cost is: , For long-term electricity purchase costs, The duration of long-term scheduling. The amount of electricity purchased at time step t. This represents the amount of electricity sold at time step t. Let be the long-term electricity price forecast at time step t. To comprehensively consider the long-term energy storage dispatch cost over time step t, To optimize the configuration of power generation equipment at time step t and save on electricity purchase costs; , The amount of charge at time step t. Let be the amount of charge discharged at time step t. This represents the long-term electricity price forecast at time step t. , This represents the amount of electricity generated by optimizing the power generation equipment configuration at time step t. Let be the long-term electricity price forecast at time step t.

[0014] In some embodiments, the following calculation formula is used when updating the parameters of the electricity price prediction model based on the prediction error: , For the updated electricity price forecasting model parameters, These are the parameters for the current electricity price prediction model. For learning rate, It is a constant. Let be the bias-corrected first-moment estimate of the prediction error gradient at time t. This is the bias-corrected second-moment estimate of the prediction error gradient at time t.

[0015] In some embodiments, the expression for the first-moment estimate of the prediction error gradient after bias correction is: , Let be the bias-corrected first-moment estimate of the prediction error gradient at time t. Let be the decay rate of the first moment estimate at time t. Let be the first moment estimate of the gradient of the prediction error at time t. , The first moment estimate of the prediction error gradient at time t-1. Let be the gradient of the loss function at time t.

[0016] In some embodiments, the expression for the bias-corrected second-moment estimate of the prediction error gradient is: , This is the bias-corrected second-moment estimate of the prediction error gradient at time t. Let be the second moment estimate of the prediction error gradient at time t. Let be the decay rate of the second moment estimate at time t. , The second moment estimate of the prediction error gradient at time t-1 is given. Let be the gradient of the loss function at time t.

[0017] In a second aspect, this application provides a multi-timescale energy dispatch system based on electricity price forecasting. The system employs the multi-timescale energy dispatching method based on electricity price forecasting as described in any embodiment of the first aspect. The system includes: an electricity price forecasting module, used to obtain predicted electricity prices for multiple future timescales based on historical data using an electricity price forecasting model; an energy dispatching strategy generation module, used to generate multi-timescale energy dispatching strategies based on the predicted electricity prices for multiple future timescales; a prediction error determination module, used to execute the multi-timescale energy dispatching strategies and monitor the actual electricity price in real time, comparing it with the corresponding predicted electricity price to determine the prediction error; a prediction error judgment module, used to determine whether the prediction error is greater than an error threshold; a loop execution module, used to update the parameters of the electricity price forecasting model according to the prediction error in response to the prediction error being greater than the error threshold, and return to the step of obtaining predicted electricity prices for multiple future timescales based on historical data using the electricity price forecasting model; and a final strategy acquisition module, used to use the current multi-timescale energy dispatching strategy as the final multi-timescale energy dispatching strategy in response to the prediction error not being greater than the error threshold.

[0018] Through the multi-timescale energy dispatch scheme based on electricity price forecasting provided above, the embodiments of this application can predict future short-term, medium-term, and even long-term electricity price trends by forecasting electricity prices at multiple time scales. This provides a solid foundation for formulating comprehensive and refined energy dispatch strategies, avoiding the limitations that may arise from single-timescale dispatching. Through a real-time monitoring and prediction error feedback mechanism, the electricity price forecasting model can continuously learn and iteratively optimize based on deviations in actual operation until the final multi-timescale energy dispatch strategy is obtained. This means that the model is no longer static but can dynamically adapt to constantly changing market conditions and unforeseen events, thereby improving its prediction accuracy.

[0019] Furthermore, in some embodiments, a comprehensive and hierarchical energy management framework is constructed by explicitly defining short-term, medium-term, and long-term electricity price forecasts and corresponding energy dispatch strategies, ensuring optimal decision-making across different time dimensions. This allows the system to use the most suitable models and strategies at different time dimensions, ensuring that short-term operations are conducted in service of medium- and long-term objectives.

[0020] Furthermore, in some embodiments, the perspective of energy dispatch is elevated from short-term operational considerations to long-term strategic investment and planning. Cost savings from optimizing power generation equipment configuration are incorporated as a core optimization objective. This signifies a shift from merely optimizing existing equipment at the operational level to guiding more fundamental and forward-looking investment decisions. By unifying all long-term critical decisions—including equipment configuration, energy storage strategies, and power purchase and sale transactions—under the clear economic objective of minimizing total long-term costs, a scientific and actionable basis is provided for achieving global optimization and maximizing long-term profitability of the integrated energy system. This ensures overall economic efficiency from asset investment to long-term operation, significantly enhancing the system's long-term competitiveness and sustainable development capabilities.

[0021] Furthermore, in some embodiments, unlike traditional methods that use a single fixed learning rate, by utilizing second-moment estimation of the gradient, an independent, adaptive learning rate can be computed for each parameter of the model. This means that for parameters with large gradients, a smaller learning step size is used, and for parameters with small gradients, a larger step size is used to accelerate training. This significantly improves optimization efficiency and the final performance of the model. By introducing first-moment estimation of the gradient, past gradient directions are accumulated when updating parameters. This makes it easier to find the global optimum while also accelerating the convergence process. Attached Figure Description

[0022] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein: Figure 1 An exemplary flowchart of a multi-timescale energy dispatching method based on electricity price forecasting according to an embodiment of this application is shown; Figure 2 An exemplary structural block diagram of a multi-timescale energy dispatch system based on electricity price forecasting according to an embodiment of this application is shown. Detailed Implementation

[0023] 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 some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be understood that the terms "comprising" and "including" used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0025] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0026] Figure 1 An exemplary flowchart of a multi-timescale energy dispatching method 100 based on electricity price forecasting according to an embodiment of this application is shown.

[0027] like Figure 1 As shown, in step S110, the electricity price forecast values ​​for multiple future time scales are obtained based on historical data through the electricity price forecasting model.

[0028] In the embodiments of this application, the aforementioned electricity price prediction model is obtained by training an LSTM neural network using historical electricity price data from a comprehensive energy system.

[0029] Specifically, the historical electricity price data of the integrated energy system is recorded in time series by hour, day, and month, including: electricity value, timestamp, and geographical location. The timestamp is the time corresponding to each electricity price record, and the geographical location is the geographical location corresponding to each electricity price record.

[0030] In the embodiments of this application, during the training of the LSTM neural network using historical electricity price data from the integrated energy system, the historical electricity price data of the integrated energy system is first preprocessed. Then, the LSTM neural network is trained based on the preprocessed historical electricity price data to obtain an electricity price prediction model.

[0031] In the embodiments of this application, preprocessing is used to ensure the quality of historical electricity price data and to provide it for subsequent use in building electricity price prediction models. Preprocessing includes removing outlier data, filling in missing values, standardization, and time series formatting.

[0032] Specifically, historical electricity price data is converted into data with a mean of 0 and a standard deviation of 1 through standardization processing. The formula is as follows: , Historical electricity price data, It is the average of historical electricity price data. It is the standard deviation of historical electricity price data. This represents historical electricity price data after standardization.

[0033] Specifically, time series formatting facilitates processing by LSTM neural networks. During time series formatting, historical electricity price data needs to be sorted chronologically and formatted using a sliding window. Let the prediction time step be... It is divided according to a fixed-length time window.

[0034] In some embodiments of this application, historical electricity price data from the past 24 hours is used to predict the electricity price for the next hour, historical electricity price data from the past 7 days is used to predict the electricity price for the next day, and historical electricity price data from the past 6 months is used to predict the electricity price for the next month. Specifically, the standardized historical electricity price data from the past 24 hours is converted into a time window format: The expression for the electricity price in the next hour is: .

[0035] In the embodiments of this application, the preprocessed historical electricity price data is split into a training set and a test set, and the order of the time series is preserved. 70% of the preprocessed historical electricity price data is divided into the training set and 30% of the preprocessed historical electricity price data is divided into the test set.

[0036] In the embodiments of this application, the architecture of the LSTM neural network includes an input layer, an LSTM layer, and an output layer. The input layer inputs the training set into the LSTM layer. The LSTM layer processes the training set through the same number of LSTM units as the time step, capturing the trend and periodicity in the time series. The LSTM layer includes a forget gate, an input gate, and an output gate. The output layer is a fully connected layer that passes the feature vector output by the LSTM layer to the fully connected layer, further converting it into an electricity price prediction value.

[0037] In the embodiments of this application, the goal of model training is to adjust the parameters in the electricity price prediction model by minimizing the loss function, thereby reducing the error between the predicted electricity price and the actual electricity price, so that the electricity price prediction model can accurately derive the predicted electricity price; the training process updates the parameters of the LSTM neural network through the backpropagation algorithm and gradient descent.

[0038] In the embodiments of this application, a loss function based on mean squared error is used. This loss function measures the error between the electricity price prediction value generated by the LSTM neural network and the actual electricity price value. The calculation formula is as follows: ,in, Indicates the first The true value of electricity price for each training sample. Indicates the first Electricity price predictions for each training sample The number of samples in the training set. This represents the loss function.

[0039] In the embodiments of this application, the gradient of the loss function with respect to each parameter is calculated using the backpropagation algorithm. In an LSTM network, backpropagation calculates the derivative of the loss function's gradient with respect to each parameter. The gradient represents the rate of change of the loss function with respect to the parameters. Through backpropagation, the network weights can be updated layer by layer, thereby reducing prediction errors.

[0040] In the embodiments of this application, the loss function is minimized by calculating the gradient and adjusting the weights along the negative direction of the gradient. The formula for updating the parameters of the LSTM network is as follows: ,in, This represents the parameters of the electricity price prediction model at time step t+1. The learning rate controls the step size for each update. This represents the derivative of the loss function with respect to the parameter gradient of the electricity price prediction model.

[0041] In the embodiments of this application, after training, an electricity price prediction model is obtained. Corresponding historical data is input into the model to obtain short-term, medium-term, and long-term electricity price predictions. The short-term electricity price prediction is obtained by using historical electricity price data preprocessed over the past 24 hours; the medium-term electricity price prediction is obtained by using historical electricity price data preprocessed over the past 7 days; and the long-term electricity price prediction is obtained by using historical electricity price data preprocessed over the past 6 months.

[0042] After completing step S110, in step S120, an energy dispatch strategy for multiple time scales is generated based on the electricity price forecasts for multiple future time scales.

[0043] In the embodiments of this application, the energy dispatch strategy across multiple time scales includes a short-term energy dispatch strategy, a medium-term energy dispatch strategy, and a long-term energy dispatch strategy.

[0044] In the embodiments of this application, a short-term energy dispatch strategy is obtained by minimizing short-term electricity purchase costs. Short-term electricity purchase costs include short-term base electricity purchase costs, integrated energy system energy storage dispatch, and load demand response. Based on short-term electricity price forecasts, the electricity purchase cost at each time point is calculated. Electricity is purchased when electricity prices are low, and external electricity purchases are reduced by discharging energy storage when electricity prices are high.

[0045] In the embodiments of this application, the expression for short-term electricity purchase cost is: , For short-term electricity purchase costs, For short-term scheduling duration, The amount of electricity purchased at time step t. Let be the short-term electricity price forecast at time step t. To comprehensively consider the short-term energy storage dispatch cost of energy in time step t, This is to save energy storage dispatch costs during the load demand response period at time step t.

[0046] Specifically, , The amount of charge at time step t. Let be the amount of charge discharged at time step t. Let be the short-term electricity price forecast at time step t.

[0047] Specifically, , This represents the amount of electricity reduced during the load demand response period at time step t. Let be the short-term electricity price forecast at time step t.

[0048] By integrating basic electricity purchase, energy storage dispatch, and load demand response, and using real-time predicted short-term electricity prices as a unified value measure, the economic logic of charging and discharging is defined, treating charging as an investment during low electricity prices and discharging as a return during high electricity prices. This provides a precise mathematical basis for the system to execute energy arbitrage by buying low and selling high. Instead of simply reducing load, the formula for saving costs through demand response directly transforms users' load reduction behavior into calculable and tangible economic benefits, providing a clear economic signal for incentivizing and dispatching user-side resources.

[0049] In the embodiments of this application, a medium-term energy dispatch strategy is obtained by minimizing medium-term electricity purchase costs. Medium-term electricity purchase costs include medium-term base electricity purchase costs, energy storage dispatch costs of the integrated energy system, and renewable energy generation dispatch costs. When electricity prices are high, the integrated energy system tends to prioritize the discharge of energy storage batteries or the generation of local renewable energy, while when electricity prices are low, it chooses to purchase external electricity.

[0050] In the embodiments of this application, the expression for the medium-term electricity purchase cost is: , For medium-term electricity purchase costs, The duration of the intermediate scheduling. The amount of electricity purchased at time step t. This represents the amount of electricity sold at time step t. The medium-term electricity price forecast at time step t. To save on electricity purchase costs by generating renewable energy in time steps, For integrated energy in time step t Energy storage dispatch costs.

[0051] Specifically, , .

[0052] Specifically, , For the amount of electricity generated by renewable energy at time step t, .

[0053] By clearly defining the goal of minimizing total medium-term costs, this approach closely links three key elements: medium-term power purchase and sales, energy storage dispatch, and local renewable energy consumption. Quantifying and maximizing the economic value of renewable energy enables the most informed economic decisions regarding the use of green electricity and the retention of energy storage to cope with future higher electricity prices, thereby maximizing the true value of renewable energy.

[0054] In the embodiments of this application, a long-term energy dispatch strategy is obtained by minimizing long-term electricity purchase costs. Long-term electricity purchase costs include long-term base electricity purchase costs, electricity purchase costs for power generation equipment configuration, and energy storage dispatch costs of the integrated energy system. At this stage, it is necessary to decide whether to increase investment in renewable energy and optimize the energy storage capacity configuration of the integrated energy system based on long-term electricity price trends.

[0055] In the embodiments of this application, the expression for long-term electricity purchase cost is: , For long-term electricity purchase costs, The duration of long-term scheduling. The amount of electricity purchased at time step t. This represents the amount of electricity sold at time step t. Let be the long-term electricity price forecast at time step t. To comprehensively consider the long-term energy storage dispatch cost over time step t, To optimize the configuration of power generation equipment at time step t and save on electricity purchase costs.

[0056] Specifically, , The amount of charge at time step t. Let be the amount of charge discharged at time step t. This represents the long-term electricity price forecast at time step t. Specifically, , This represents the amount of electricity generated by optimizing the power generation equipment configuration at time step t. Let be the long-term electricity price forecast at time step t.

[0057] By minimizing a cost function that spans a longer time horizon, the long-term trend of future electricity prices is directly linked to the capacity configuration and return on investment of renewable energy and energy storage systems. It no longer relies on rough estimates or market intuition, but instead uses formulas to precisely quantify the potential economic savings of new generation or energy storage assets over their entire lifecycle. This allows policymakers to clearly calculate the potential rate of return on each investment based on long-term market forecasts, thus making the most informed and lowest-risk capital expenditure decisions. By optimizing the configuration of generation equipment and energy storage, the entire energy system is ensured to have the most economical and efficient asset portfolio from its inception, enabling it to confidently cope with market conditions and energy challenges in the coming years and achieve long-term cost optimization.

[0058] After completing step S120, in step S130, an energy dispatch strategy on multiple time scales is executed, and the actual electricity price is monitored in real time and compared with the corresponding electricity price forecast to determine the forecast error.

[0059] In the embodiments of this application, after obtaining short-term, medium-term, and long-term energy dispatch strategies, the energy dispatch strategies are executed in actual operation. The execution process includes electricity purchase, charging and discharging of energy storage devices, renewable energy generation dispatch, and load demand response. Specifically, the required electricity is purchased according to the electricity purchase decision in the energy dispatch strategy. The energy storage devices are charged or discharged according to the energy storage charging and discharging strategy in the energy dispatch strategy. The power generation of renewable energy devices is adjusted at different time steps according to the energy dispatch strategy. The load is adjusted according to the demand response strategy to reduce electricity demand, especially during peak electricity price periods.

[0060] During implementation, the operating status of the integrated energy system is monitored and fed back in real time. If the equipment does not operate as expected, the integrated energy system can be quickly adjusted to ensure the stability and economy of the integrated energy system.

[0061] In the embodiments of this application, the following calculation formula is used in the process of calculating the error between the actual real-time electricity price and the predicted electricity price at each time scale: ,in, This represents the electricity price forecast. This indicates the actual value of the electricity price. This represents the difference between the actual electricity price and the predicted electricity price, i.e., the prediction error.

[0062] After step S130 is completed, in step S140, it is determined whether the prediction error is greater than the error threshold.

[0063] Electricity price volatility is a key factor in setting the error threshold. Market electricity prices themselves fluctuate, especially during periods of high demand or when the weather is abnormal. In such cases, the error threshold should not be too low, otherwise the system will frequently trigger adjustments, leading to instability in the dispatching scheme.

[0064] In the embodiments of this application, the error threshold is set to 4% of the predicted electricity price.

[0065] In response to the prediction error being greater than the error threshold, in step S150, the parameters of the electricity price prediction model are updated according to the prediction error, and the process returns to step S110, which is the step of obtaining the electricity price prediction values ​​for multiple future time scales based on historical data through the electricity price prediction model.

[0066] In the embodiments of this application, the following calculation formula is used when updating the parameters of the electricity price prediction model based on the prediction error: , For the updated electricity price forecasting model parameters, These are the parameters for the current electricity price prediction model. For learning rate, It is a constant. Let be the bias-corrected first-moment estimate of the prediction error gradient at time t. This is the bias-corrected second-moment estimate of the prediction error gradient at time t.

[0067] In the embodiments of this application, This represents a very small constant set to prevent the denominator from being zero. .

[0068] In the embodiments of this application, the expression for the first-order moment estimate after bias correction of the prediction error gradient is: , Let be the bias-corrected first-moment estimate of the prediction error gradient at time t. Let be the decay rate of the first moment estimate at time t. Let be the first moment estimate of the gradient of the prediction error at time t. , The first moment estimate of the prediction error gradient at time t-1. Let be the gradient of the loss function at time t.

[0069] In the embodiments of this application, the expression for the second-order moment estimate after bias correction of the prediction error gradient is: , This is the bias-corrected second-moment estimate of the prediction error gradient at time t. Let be the second moment estimate of the prediction error gradient at time t. Let be the decay rate of the second moment estimate at time t. , The second moment estimate of the prediction error gradient at time t-1 is given. Let be the gradient of the loss function at time t.

[0070] In response to the prediction error being greater than the error threshold, in step S160, the current energy dispatch strategy on multiple time scales is adopted as the final energy dispatch strategy on multiple time scales.

[0071] In summary, through the multi-timescale energy dispatch scheme based on electricity price forecasting provided above, the embodiments of this application can predict future short-term, medium-term, and even long-term electricity price trends by forecasting electricity prices at multiple time scales. This provides a solid foundation for formulating comprehensive and refined energy dispatch strategies, avoiding the limitations that may arise from single-timescale dispatching. Through real-time monitoring and a feedback mechanism for prediction errors, the electricity price forecasting model can continuously learn and iteratively optimize based on deviations in actual operation until the final multi-timescale energy dispatch strategy is obtained. This means that the model is no longer static but can dynamically adapt to constantly changing market conditions and unforeseen events, thereby improving its prediction accuracy.

[0072] Furthermore, in some embodiments, a comprehensive and hierarchical energy management framework is constructed by explicitly defining short-term, medium-term, and long-term electricity price forecasts and corresponding energy dispatch strategies, ensuring optimal decision-making across different time dimensions. This allows the system to use the most suitable models and strategies at different time dimensions, ensuring that short-term operations are conducted in service of medium- and long-term objectives.

[0073] Furthermore, in some embodiments, the perspective of energy dispatch is elevated from short-term operational considerations to long-term strategic investment and planning. Cost savings from optimizing power generation equipment configuration are incorporated as a core optimization objective. This signifies a shift from merely optimizing existing equipment at the operational level to guiding more fundamental and forward-looking investment decisions. By unifying all long-term critical decisions—including equipment configuration, energy storage strategies, and power purchase and sale transactions—under the clear economic objective of minimizing total long-term costs, a scientific and actionable basis is provided for achieving global optimization and maximizing long-term profitability of the integrated energy system. This ensures overall economic efficiency from asset investment to long-term operation, significantly enhancing the system's long-term competitiveness and sustainable development capabilities.

[0074] Furthermore, in some embodiments, unlike traditional methods that use a single fixed learning rate, by utilizing second-moment estimation of the gradient, an independent, adaptive learning rate can be computed for each parameter of the model. This means that for parameters with large gradients, a smaller learning step size is used, and for parameters with small gradients, a larger step size is used to accelerate training. This significantly improves optimization efficiency and the final performance of the model. By introducing first-moment estimation of the gradient, past gradient directions are accumulated when updating parameters. This makes it easier to find the global optimum while also accelerating the convergence process.

[0075] This application also provides a multi-timescale energy dispatch system based on electricity price forecasting. It can use the aforementioned multi-timescale energy dispatching method 100 based on electricity price forecasting to perform multi-timescale energy dispatching based on electricity price forecasting, or it can use other methods to perform multi-timescale energy dispatching based on electricity price forecasting. This application does not limit it here.

[0076] Figure 2 An exemplary structural block diagram of a multi-timescale energy dispatch system based on electricity price forecasting according to an embodiment of this application is shown.

[0077] like Figure 2 As shown, the system 200 includes an electricity price prediction module 210, an energy dispatch strategy generation module 220, a prediction error determination module 230, a prediction error judgment module 240, a loop execution module 250, and a final strategy acquisition module 260.

[0078] Specifically, the electricity price forecasting module 210 is used to obtain electricity price forecasts for multiple future time scales based on historical data through an electricity price forecasting model.

[0079] Specifically, the energy dispatch strategy generation module 220 is used to generate energy dispatch strategies at multiple time scales based on electricity price forecasts at multiple future time scales.

[0080] Specifically, the prediction error determination module 230 is used to execute energy dispatch strategies on multiple time scales and monitor the actual electricity price in real time, comparing it with the corresponding electricity price prediction value to determine the prediction error.

[0081] Specifically, the prediction error judgment module 240 is used to determine whether the prediction error is greater than the error threshold.

[0082] Specifically, the loop execution module 250 is used to update the parameters of the electricity price prediction model according to the prediction error in response to the prediction error being greater than the error threshold, and return to the step of obtaining the electricity price prediction values ​​for multiple future time scales based on historical data through the electricity price prediction model.

[0083] Specifically, the final strategy acquisition module 260 is used to take the current energy dispatch strategy on multiple time scales as the final energy dispatch strategy on multiple time scales in response to the prediction error not being greater than the error threshold.

[0084] When system 200 performs multi-timescale energy dispatching based on electricity price prediction using the aforementioned multi-timescale energy dispatching method 100, the electricity price prediction module 210 executes the aforementioned step S110, the energy dispatching strategy generation module 220 executes the aforementioned step S120, the prediction error determination module 230 executes the aforementioned step S130, the prediction error judgment module 240 executes the aforementioned step S140, the loop execution module 250 executes the aforementioned step S150, and the final strategy acquisition module 260 executes the aforementioned step S160. The specific execution process can be found above and will not be repeated here.

[0085] While numerous embodiments of this application have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will arise for those skilled in the art without departing from the spirit and intent of this application. It should be understood that various alternatives to the embodiments of this application described herein may be employed in the practice of this application. The appended claims are intended to define the scope of protection of this application and therefore cover equivalents or alternatives within the scope of these claims.

Claims

1. A multi-timescale energy dispatch method based on electricity price forecasting, characterized in that, include: Electricity price forecasting models are used to obtain electricity price forecasts for multiple future time scales based on historical data. Energy dispatch strategies are generated based on electricity price forecasts over multiple future time scales. Implement energy dispatch strategies across multiple time scales and monitor actual electricity prices in real time, comparing them with corresponding electricity price forecasts to determine forecast errors; Determine whether the prediction error is greater than the error threshold; In response to a prediction error exceeding an error threshold, the parameters of the electricity price prediction model are updated based on the prediction error, and the process returns to the step of obtaining electricity price predictions for multiple future time scales based on historical data using the electricity price prediction model. In response to the prediction error not exceeding the error threshold, the current energy dispatch strategy on multiple time scales is adopted as the final energy dispatch strategy on multiple time scales.

2. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 1, characterized in that, The electricity price forecasts for the future include short-term, medium-term, and long-term forecasts.

3. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 2, characterized in that, The energy dispatch strategies across multiple time scales include short-term energy dispatch strategies, medium-term energy dispatch strategies, and long-term energy dispatch strategies.

4. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 3, characterized in that, A short-term energy dispatch strategy is derived by minimizing short-term electricity purchase costs; The expression for short-term electricity purchase cost is as follows: , For short-term electricity purchase costs, For short-term scheduling duration, The amount of electricity purchased at time step t. Let be the short-term electricity price forecast at time step t. To comprehensively consider the short-term energy storage dispatch cost of energy in time step t, To save energy storage dispatch costs during load demand response at time step t; , The amount of charge at time step t. Let be the amount of charge discharged at time step t. This represents the short-term electricity price forecast at time step t. , This represents the amount of electricity reduced during the load demand response period at time step t. Let be the short-term electricity price forecast at time step t.

5. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 3, characterized in that, A medium-term energy dispatch strategy is derived by minimizing medium-term electricity purchase costs; The expression for the medium-term electricity purchase cost is as follows: , For medium-term electricity purchase costs, The duration of the intermediate scheduling. The amount of electricity purchased at time step t. This represents the amount of electricity sold at time step t. The medium-term electricity price forecast at time step t. To save on electricity purchase costs by generating renewable energy in time steps, For integrated energy in time step t Energy storage dispatch costs; ， ； , For the amount of electricity generated by renewable energy at time step t, .

6. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 3, characterized in that, A long-term energy dispatch strategy is derived by minimizing long-term electricity purchase costs. The expression for long-term electricity purchase cost is as follows: , For long-term electricity purchase costs, The duration of long-term scheduling. The amount of electricity purchased at time step t. This represents the amount of electricity sold at time step t. Let be the long-term electricity price forecast at time step t. To comprehensively consider the long-term energy storage dispatch cost over time step t, To optimize the configuration of power generation equipment at time step t and save on electricity purchase costs; , The amount of charge at time step t. Let be the amount of charge discharged at time step t. This represents the long-term electricity price forecast at time step t. , This represents the amount of electricity generated by optimizing the power generation equipment configuration at time step t. Let be the long-term electricity price forecast at time step t.

7. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 1, characterized in that, The following calculation formula is used when updating the parameters of the electricity price prediction model based on the prediction error: , For the updated electricity price forecasting model parameters, These are the parameters for the current electricity price prediction model. For learning rate, It is a constant. Let be the bias-corrected first-moment estimate of the prediction error gradient at time t. This is the bias-corrected second-moment estimate of the prediction error gradient at time t.

8. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 7, characterized in that, The expression for the first-moment estimate of the prediction error gradient after bias correction is as follows: , Let be the bias-corrected first-moment estimate of the prediction error gradient at time t. Let be the decay rate of the first moment estimate at time t. Let be the first moment estimate of the gradient of the prediction error at time t. , The first moment estimate of the prediction error gradient at time t-1. Let be the gradient of the loss function at time t.

9. The multi-timescale energy dispatch method based on electricity price forecasting according to claim 7 or 8, characterized in that, The expression for the second-order moment estimate after bias correction of the prediction error gradient is: , This is the bias-corrected second-moment estimate of the prediction error gradient at time t. Let be the second moment estimate of the prediction error gradient at time t. Let be the decay rate of the second moment estimate at time t. , The second moment estimate of the prediction error gradient at time t-1 is given. Let be the gradient of the loss function at time t.

10. A multi-timescale energy dispatch system based on electricity price forecasting, characterized in that, The system employs the multi-timescale energy dispatching method based on electricity price forecasting as described in any one of claims 1-9, wherein the system comprises: The electricity price forecasting module is used to obtain electricity price forecasts for multiple future time scales based on historical data using an electricity price forecasting model. The energy dispatch strategy generation module is used to generate energy dispatch strategies for multiple time scales based on electricity price forecasts for multiple future time scales. The prediction error determination module is used to execute energy dispatch strategies on multiple time scales and monitor the actual electricity price in real time, comparing it with the corresponding electricity price forecast to determine the prediction error. The prediction error judgment module is used to determine whether the prediction error is greater than the error threshold. The loop execution module is used to update the parameters of the electricity price prediction model according to the prediction error when the prediction error is greater than the error threshold, and return to the step of obtaining the electricity price prediction values ​​for multiple future time scales based on historical data through the electricity price prediction model. The final strategy acquisition module is used to select the current multi-timescale energy dispatch strategy as the final multi-timescale energy dispatch strategy in response to the prediction error not being greater than the error threshold.

Images