Source load prediction method and system for zero-carbon digital park collaborative optimization

Abstract

The invention relates to a zero-carbon digital park collaborative optimization-oriented source load prediction method and system. The method comprises the following steps: collecting multi-source historical sample data of a zero-carbon digital park; jointly training a to-be-trained power supply prediction sub-model and a to-be-trained load prediction sub-model according to the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model comprises a pre-trained power supply prediction sub-model, a pre-trained load prediction sub-model and a collaborative optimization unit; the collaborative optimization unit is used for adjusting the output of the pre-trained power supply prediction sub-model and the pre-trained load prediction sub-model based on the physical constraint of energy supply and demand balance; wherein in the process of training the to-be-trained power supply prediction sub-model and the to-be-trained load prediction sub-model, through sharing the weight of the LSTM layer, the to-be-trained power supply prediction sub-model and the to-be-trained load prediction sub-model learn the source-load correlation characteristics in training. By adopting the method, the source load prediction precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a source-load prediction method and system for collaborative optimization of zero-carbon digital parks. Background Art

[0002] With the proposed "dual carbon" goals, the construction of zero-carbon digital parks has become a crucial path to achieving carbon neutrality. These parks incorporate diverse elements, including distributed energy, energy storage systems, and flexible loads. Their energy systems feature multi-energy complementarity, source-load interaction, and spatiotemporal coupling. Traditional source-load forecasting methods struggle to accurately capture the complex characteristics of zero-carbon digital parks, resulting in low forecast accuracy and difficulty supporting the coordinated and optimized operation of these parks. Summary of the Invention

[0003] Based on this, it is necessary to provide a source-load prediction method, system, computer equipment, computer-readable storage medium and computer program product for collaborative optimization of zero-carbon digital parks, which can improve the accuracy of source-load prediction and address the above technical problems.

[0004] In the first aspect, this application provides a source-load prediction method for collaborative optimization of zero-carbon digital parks, including:

[0005] Collect multi-source historical sample data of the zero-carbon digital park; the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data, and carbon emission data;

[0006] Jointly training the power prediction sub-model to be trained and the load prediction sub-model to be trained according to the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model includes a pre-trained power prediction sub-model, a pre-trained load prediction sub-model and a collaborative optimization unit;

[0007] The pre-trained power prediction sub-model is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park; the pre-trained load prediction sub-model is used to predict the park electrical load, thermal load and cooling load of the zero-carbon digital park; the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance;

[0008] In which, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. In the process of training the power prediction sub-model to be trained and the load prediction sub-model to be trained, the power prediction sub-model to be trained and the load prediction sub-model to be trained learn the source-load correlation characteristics during training by sharing the weights of the LSTM layer; and the physical constraints of energy supply and demand balance are introduced, and the prediction results that deviate from the constraints are punished by the loss function.

[0009] In one embodiment, the method of jointly training a power source prediction sub-model to be trained and a load prediction sub-model to be trained based on the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model includes:

[0010] Extracting sample target features related to source-load prediction from the multi-source historical sample data; the sample target features include sample time features, sample meteorological features, sample load features, and sample power generation features;

[0011] Inputting the sample time features, the sample meteorological features, and the sample power generation features into the power prediction sub-model to be trained, and outputting predicted power generation data;

[0012] Inputting the sample time characteristics, the sample meteorological characteristics, and the sample load characteristics into the load prediction sub-model to be trained, and outputting predicted load data;

[0013] According to the first difference between the predicted power generation data and the actual power generation data, and the second difference between the predicted load data and the actual load data, the power source prediction sub-model to be trained and the load prediction sub-model to be trained are jointly trained to obtain the pre-trained source-load collaborative prediction model.

[0014] In one embodiment, the method further comprises:

[0015] Inputting the predicted power generation data and the actual power generation data into a preset first loss function, and outputting a first loss function value; the first loss function value represents the first difference;

[0016] The predicted load data and the actual load data are input into a preset second loss function, and a second loss function value is output; the second loss function value represents the second difference.

[0017] In one embodiment, the jointly training the to-be-trained power prediction sub-model and the to-be-trained load prediction sub-model based on the first difference between the predicted power generation data and the actual power generation data and the second difference between the predicted load data and the actual load data comprises:

[0018] determining a third difference between the predicted load data and the predicted power generation data based on the physical constraints of the energy supply and demand balance;

[0019] The power supply prediction sub-model to be trained and the load prediction sub-model to be trained are jointly trained according to the first difference, the second difference and the third difference.

[0020] In one embodiment, the physical constraint for energy supply and demand balance is that power output is greater than or equal to load demand, and determining the third difference between the predicted load data and the predicted power generation data based on the physical constraint for energy supply and demand balance includes:

[0021] Inputting the predicted load data and the predicted power generation data into a third loss function associated with the physical constraint, and outputting a third loss function value; the third loss function value represents the third difference;

[0022] The formula of the third loss function is:

[0023] ;

[0024] in, is used to characterize the third loss function value, It is the constraint weight coefficient, which is used to control the penalty intensity for supply and demand imbalance.

[0025] In one embodiment, the jointly training the to-be-trained power prediction sub-model and the to-be-trained load prediction sub-model based on the first difference, the second difference, and the third difference includes:

[0026] determining a target difference based on the first difference, the second difference, and the third difference;

[0027] According to the target difference, simultaneously updating the parameters of the shared LSTM layer, the model parameters of the power prediction sub-model to be trained, and the model parameters of the load prediction sub-model to be trained;

[0028] The power prediction sub-model and the load prediction sub-model after the model parameters are updated are retrained until the trained power prediction sub-model and the trained load prediction sub-model meet the preset training end condition.

[0029] Secondly, this application also provides a source-load prediction system for collaborative optimization of zero-carbon digital parks, including:

[0030] A data acquisition module is used to collect multi-source historical sample data of the zero-carbon digital park; the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data, and carbon emission data;

[0031] A model training and optimization module, configured to jointly train a power prediction sub-model to be trained and a load prediction sub-model to be trained based on the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model includes a pre-trained power prediction sub-model, a pre-trained load prediction sub-model, and a collaborative optimization unit;

[0032] The pre-trained power prediction sub-model is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park; the pre-trained load prediction sub-model is used to predict the park electrical load, thermal load and cooling load of the zero-carbon digital park; the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance;

[0033] Among them, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. In the process of training the power prediction sub-model to be trained and the load prediction sub-model to be trained, the model training and optimization module is specifically used to enable the power prediction sub-model to be trained and the load prediction sub-model to be trained to learn source-load correlation characteristics during training by sharing the weights of the LSTM layer; and, introduce the physical constraints of energy supply and demand balance, and punish the prediction results that deviate from the constraints through the loss function.

[0034] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the above method are implemented.

[0035] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0036] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0037] The above-mentioned source-load prediction method, system, computer device, computer-readable storage medium and computer program product for collaborative optimization of zero-carbon digital parks collect multi-source historical sample data of the zero-carbon digital park; the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data and carbon emission data; a power prediction sub-model to be trained and a load prediction sub-model to be trained are jointly trained based on the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model includes a pre-trained power prediction sub-model, a pre-trained load prediction sub-model and a collaborative optimization unit; the pre-trained power prediction sub-model is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park; The pre-trained load prediction sub-model is used to predict the park electricity load, heating load and cooling load of the zero-carbon digital park; the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance; wherein, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. In the process of training the power prediction sub-model to be trained and the load prediction sub-model to be trained, the power prediction sub-model to be trained and the load prediction sub-model to be trained learn the source-load correlation characteristics during training by sharing the weights of the LSTM layer; and, the physical constraints of energy supply and demand balance are introduced, and the prediction results that deviate from the constraints are penalized by the loss function.

[0038] In this way, through collaborative optimization, the spatiotemporal coupling relationship between source and load is learned, enabling dynamic matching of source and load forecasts across both time and space. This resolves the issue of independent predictions but operational conflicts in related technologies, improving source and load forecast accuracy. Furthermore, in related technologies, power and load forecast errors can compound (e.g., overestimating power generation and underestimating load demand), leading to operational risks. In this application, collaborative optimization reduces overall uncertainty through joint modeling, effectively improving source and load forecast accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 A schematic flow chart of a source-load prediction method for collaborative optimization of a zero-carbon digital park in one embodiment;

[0041] Figure 21. A flowchart illustrating steps for jointly training a power prediction sub-model to be trained and a load prediction sub-model to be trained based on multi-source historical sample data in one embodiment;

[0042] Figure 3 A schematic flow chart of a source-load prediction method for collaborative optimization of a zero-carbon digital park in another embodiment;

[0043] Figure 4 This is a structural framework diagram of a source-load prediction system for collaborative optimization of a zero-carbon digital park in one embodiment;

[0044] Figure 5 A schematic diagram of a data synchronization process in one embodiment;

[0045] Figure 6 A schematic diagram of the structure of a model building module in one embodiment;

[0046] Figure 7 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0048] It should be noted that the terms "first", "second", etc. used in this application may be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "including" and "having" used in this application and any variations thereof are intended to cover non-exclusive inclusions. The term "plurality" used in this application refers to two or more. The term "and / or" used in this application refers to one of the solutions or any combination of multiple solutions.

[0049] In one embodiment, Figure 1 As shown, a source-load prediction method for collaborative optimization of a zero-carbon digital park is provided. This embodiment uses the method applied to a computer device as an example. It is understandable that the computer device can be a terminal, a server, or a system including a terminal and a server. In this embodiment, the method includes the following steps:

[0050] Step S110: Collect multi-source historical sample data of the zero-carbon digital park.

[0051] Among them, the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data, and carbon emission data.

[0052] The meteorological data includes at least one item of data such as temperature, humidity, wind speed, and irradiance.

[0053] Among them, energy production data includes at least one data such as photovoltaic power generation and wind power generation.

[0054] Among them, energy consumption data includes at least one item of data such as electricity load, heat load, and cooling load.

[0055] In specific implementation, computer equipment can collect multi-source historical sample data of the zero-carbon digital park, for example, collecting meteorological data, energy production data, energy consumption data, and carbon emission data of the zero-carbon digital park in the past year.

[0056] Step S120 , jointly training the power source prediction sub-model to be trained and the load prediction sub-model to be trained based on multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model.

[0057] The pre-trained source-load collaborative prediction model includes a pre-trained power source prediction sub-model, a pre-trained load prediction sub-model and a collaborative optimization unit.

[0058] Among them, the pre-trained power prediction sub-model (power prediction sub-model) is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park.

[0059] Among them, the pre-trained load forecasting sub-model is used to predict the park electricity load, heating load and cooling load of the zero-carbon digital park.

[0060] Among them, the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance.

[0061] In this way, load forecasting and power forecasting of the power system can be realized, providing data support for optimizing scheduling strategies, resource allocation, and equipment operation management, which will help the power system achieve more effective resource management and energy allocation, thereby improving its operating efficiency and stability.

[0062] Among them, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. During the training process of the power prediction sub-model to be trained and the load prediction sub-model to be trained, the power prediction sub-model to be trained and the load prediction sub-model to be trained learn the source-load correlation characteristics during training by sharing the weights of the LSTM layer; and the physical constraints of energy supply and demand balance are introduced, and the prediction results that deviate from the constraints are penalized by the loss function.

[0063] In this way, the power and load sub-models are not trained independently, but are instead optimized simultaneously through shared layers and a joint loss function to reach a global optimal solution. The power prediction sub-model is connected to an independent LSTM layer after the shared LSTM layer, outputting photovoltaic and wind power generation forecasts; the load prediction sub-model is connected to an independent LSTM layer after the shared LSTM layer, outputting electricity, heating, and cooling load demand forecasts.

[0064] The power forecast submodel and the load forecast submodel share the first few layers of LSTM units to extract common time series features (such as weather periodicity and park operation patterns). This shared LSTM layer allows power and load forecasts to share underlying time series features, avoiding separate modeling.

[0065] Among them, the source-load correlation characteristics can refer to the dynamic coupling relationship between the power supply side (such as photovoltaic and wind power) and the load side (such as electrical equipment). These relationships are formed through multi-dimensional interactions such as time, space, and physical mechanisms, and can include time dimension correlation, space dimension correlation, etc.

[0066] In a specific implementation, the power supply prediction sub-model to be trained and the load prediction sub-model to be trained can be jointly trained based on multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model.

[0067] Specifically, multi-source historical sample data can be input into the power prediction sub-model to be trained to output predicted power generation data, and multi-source historical sample data can be input into the load prediction sub-model to be trained to output predicted load data; based on the first difference between the predicted power generation data and the actual power generation data, and the second difference between the predicted load data and the actual load data, the power prediction sub-model to be trained and the load prediction sub-model to be trained are jointly trained to obtain a pre-trained source-load collaborative prediction model.

[0068] The above-mentioned source-load prediction method, system, computer device, computer-readable storage medium and computer program product for collaborative optimization of zero-carbon digital parks collect multi-source historical sample data of the zero-carbon digital park; the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data and carbon emission data; a power prediction sub-model to be trained and a load prediction sub-model to be trained are jointly trained based on the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model includes a pre-trained power prediction sub-model, a pre-trained load prediction sub-model and a collaborative optimization unit; the pre-trained power prediction sub-model is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park; The pre-trained load prediction sub-model is used to predict the park electricity load, heating load and cooling load of the zero-carbon digital park; the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance; wherein, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. In the process of training the power prediction sub-model to be trained and the load prediction sub-model to be trained, the power prediction sub-model to be trained and the load prediction sub-model to be trained learn the source-load correlation characteristics during training by sharing the weights of the LSTM layer; and, the physical constraints of energy supply and demand balance are introduced, and the prediction results that deviate from the constraints are penalized by the loss function.

[0069] In this way, through collaborative optimization, the spatiotemporal coupling relationship between source and load is learned, enabling dynamic matching of source and load forecasts across both time and space. This resolves the issue of independent predictions but operational conflicts in related technologies, improving source and load forecast accuracy. Furthermore, in related technologies, power and load forecast errors can compound (e.g., overestimating power generation and underestimating load demand), leading to operational risks. In this application, collaborative optimization reduces overall uncertainty through joint modeling, effectively improving source and load forecast accuracy.

[0070] In one embodiment, Figure 2 As shown, step S120, jointly training the power prediction sub-model to be trained and the load prediction sub-model to be trained based on multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model, includes the following steps:

[0071] Step S210: extracting sample target features related to source load prediction from multi-source historical sample data.

[0072] Among them, the sample target characteristics include sample time characteristics, sample meteorological characteristics, sample load characteristics and sample power generation characteristics.

[0073] The sample time features may include year, month, day, hour, holiday, etc.

[0074] Among them, the sample meteorological characteristics may include temperature, humidity, wind speed, irradiance, etc.

[0075] The sample load characteristics may include the electric load, heating load, cooling load, etc. in the past 24 hours.

[0076] The sample power generation characteristics may include photovoltaic power generation, wind power generation, etc. in the past 24 hours.

[0077] In a specific implementation, the computer device can extract sample target features related to source-load prediction from multi-source historical sample data; the sample target features include sample time features, sample meteorological features, sample load features and sample power generation features.

[0078] Step S220 , inputting the sample time features, sample meteorological features, and sample power generation features into the power prediction sub-model to be trained, and outputting predicted power generation data.

[0079] In a specific implementation, the computer device can input sample time features, sample meteorological features, and sample power generation features into the power prediction sub-model to be trained, and output predicted power generation data.

[0080] For example, the sample time features, sample meteorological features, and sample power generation features corresponding to the multi-source historical sample data of the previous day can be obtained and input into the power prediction sub-model to be trained to predict the power generation data of the next day and output the predicted power generation data.

[0081] Step S230: input the sample time characteristics, sample meteorological characteristics, and sample load characteristics into the load prediction sub-model to be trained, and output the predicted load data.

[0082] In a specific implementation, the computer device can input sample time characteristics, sample meteorological characteristics, and sample load characteristics into the load prediction sub-model to be trained, and output predicted load data.

[0083] For example, the sample time features, sample meteorological features, and sample load features corresponding to the multi-source historical sample data of the previous day can be obtained and input into the load forecasting sub-model to be trained to predict the load data of the next day and output the predicted load data.

[0084] Step S240, based on the first difference between the predicted power generation data and the actual power generation data, and the second difference between the predicted load data and the actual load data, jointly train the power prediction sub-model to be trained and the load prediction sub-model to be trained to obtain a pre-trained source-load collaborative prediction model.

[0085] In a specific implementation, the computer device can jointly train the power source prediction sub-model to be trained and the load prediction sub-model to be trained based on the first difference between the predicted power generation data and the actual power generation data, and the second difference between the predicted load data and the actual load data to obtain a pre-trained source-load collaborative prediction model.

[0086] For example, continuing with the previous example, after outputting the predicted power generation data for the next day and the predicted load data for the next day, the power source prediction sub-model to be trained and the load prediction sub-model to be trained can be jointly trained based on the first difference between the predicted power generation data for the next day and the actual power generation data for the next day, as well as the second difference between the predicted load data for the next day and the actual load data for the next day to obtain a pre-trained source-load collaborative prediction model.

[0087] In some embodiments, the method also includes: inputting the predicted power generation data and the actual power generation data into a preset first loss function, and outputting a first loss function value; the first loss function value represents the first difference; inputting the predicted load data and the actual load data into a preset second loss function, and outputting a second loss function value; the second loss function value represents the second difference.

[0088] In a specific implementation, the computer device can input the predicted power generation data and the actual power generation data into a preset first loss function, and output a first loss function value representing the first difference; input the predicted load data and the actual load data into a preset second loss function, and output a second loss function value representing the second difference.

[0089] The first loss function and the second loss function may adopt loss functions such as Mean Square Error (MSE) and Mean Absolute Error (MAE), which are not specifically limited here.

[0090] In some embodiments, based on the first difference between the predicted power generation data and the actual power generation data, and the second difference between the predicted load data and the actual load data, the power supply prediction sub-model to be trained and the load prediction sub-model to be trained are jointly trained, including: determining the third difference between the predicted load data and the predicted power generation data based on the physical constraints of the energy supply and demand balance; and jointly training the power supply prediction sub-model to be trained and the load prediction sub-model to be trained based on the first difference, the second difference and the third difference.

[0091] In a specific implementation, the computer device can determine the third difference between the predicted load data and the predicted power generation data based on the physical constraints of energy supply and demand balance during the process of jointly training the power prediction sub-model to be trained and the load prediction sub-model to be trained based on the first difference between the predicted power generation data and the actual power generation data and the second difference between the predicted load data and the actual load data; and jointly train the power prediction sub-model to be trained and the load prediction sub-model to be trained based on the first difference, the second difference and the third difference.

[0092] In some embodiments, the physical constraint for energy supply and demand balance is that the power output is greater than or equal to the load demand. Determining a third difference between the predicted load data and the predicted power generation data based on the physical constraint for energy supply and demand balance includes: inputting the predicted load data and the predicted power generation data into a third loss function associated with the physical constraint, and outputting a third loss function value; the third loss function value represents the third difference; the formula for the third loss function is: .in, is used to characterize the value of the third loss function, It is the constraint weight coefficient, which is used to control the penalty intensity for supply and demand imbalance.

[0093] In a specific implementation, when determining the third difference between the predicted load data and the predicted power generation data based on the physical constraint of energy supply and demand balance, the computer device may input the predicted load data and the predicted power generation data into a third loss function associated with the physical constraint. The formula of the third loss function is: , is the constraint weight coefficient, which is used to control the penalty intensity of supply-demand imbalance, thereby outputting the third loss function value representing the third difference.

[0094] In this way, the physical rules of energy balance are converted into mathematical loss functions to ensure that the prediction results are consistent with the actual operating logic. By dynamically penalizing the supply and demand gap (load > power supply), the power supply and load forecast values ​​are forced to meet the actual operating constraints, thereby avoiding the risk of energy shortage caused by the supply and demand split in traditional independent forecasts, significantly improving the feasibility of the prediction results and the stability of the system, and providing high-reliability decision-making support for zero-carbon parks.

[0095] In some embodiments, the power prediction sub-model to be trained and the load prediction sub-model to be trained are jointly trained based on the first difference, the second difference and the third difference, including: determining the target difference based on the first difference, the second difference and the third difference; simultaneously updating the parameters of the shared LSTM layer, the model parameters of the power prediction sub-model to be trained and the model parameters of the load prediction sub-model to be trained based on the target difference; and retraining the power prediction sub-model and the load prediction sub-model after the model parameters are updated until the trained power prediction sub-model and the trained load prediction sub-model meet the preset training end conditions.

[0096] In a specific implementation, the computer device can determine the target difference based on the first difference, the second difference, and the third difference during the process of jointly training the power prediction sub-model to be trained and the load prediction sub-model to be trained according to the first difference, the second difference, and the third difference. Specifically, the target loss function value representing the target difference can be determined based on the sum of the first loss function value, the second loss function value, and the third loss function value. For example, .in, Represents the target loss function value, Characterize the first loss function value, Represents the value of the second loss function.

[0097] In this way, according to the target difference, the parameters of the shared LSTM layer, the model parameters of the power prediction sub-model to be trained, and the model parameters of the load prediction sub-model to be trained can be updated at the same time, and the power prediction sub-model and the load prediction sub-model after the model parameters are updated can be retrained until the trained power prediction sub-model and the trained load prediction sub-model meet the preset training end conditions.

[0098] During training, the gradient simultaneously updates the parameters of the shared LSTM layer and sub-models, enabling the model to proactively reduce the supply-demand imbalance while optimizing prediction accuracy.

[0099] During the training process, the optimizer configuration can use the Adam optimizer, with an initial learning rate of 0.001 and a learning rate decay strategy.

[0100] Cross-validation can also be used: divide the data from the past year into 5 folds by week to avoid seasonal and holiday effects. After training each fold, verify the MAPE (mean absolute percentage error) and select the optimal model parameters.

[0101] During the forecasting phase, the model not only outputs results but also dynamically adjusts them based on real-time data (such as unexpected weather changes) to maintain a balance between supply and demand. These forecasts directly inform the campus energy management system's scheduling strategies, for example by enabling the energy storage system to activate in advance when supply and demand tensions are predicted.

[0102] In summary, through shared feature extraction and physical constraint embedding, we achieve deep coupling between power and load forecasting. The core of this approach lies in: extracting common time series features through a shared LSTM layer, and using a constrained loss function to enforce supply-demand balance, ensuring that forecasts are both accurate and meet actual operational requirements. This provides highly reliable and available source-load collaborative forecasting for zero-carbon parks, supporting refined energy management towards carbon neutrality goals.

[0103] In another embodiment, Figure 3 As shown, a flow chart of a source-load prediction method for collaborative optimization of a zero-carbon digital park is provided, which includes the following steps:

[0104] Step S302: Collect multi-source historical sample data of the zero-carbon digital park.

[0105] Step S304: extracting sample target features related to source load prediction from the multi-source historical sample data.

[0106] Step S306: input the sample time features, sample meteorological features, and sample power generation features into the power prediction sub-model to be trained, and output the predicted power generation data.

[0107] Step S308: Input the sample time characteristics, sample meteorological characteristics, and sample load characteristics into the load prediction sub-model to be trained, and output the predicted load data.

[0108] In step S310 , the predicted power generation data and the actual power generation data are input into a preset first loss function, and a first loss function value representing a first difference is output.

[0109] In step S312 , the predicted load data and the actual load data are input into a preset second loss function, and a second loss function value representing the second difference is output.

[0110] Step S314: input the predicted load data and the predicted power generation data into a third loss function associated with the physical constraint, and output a third loss function value representing the third difference.

[0111] Step S316: determining a target difference based on the first difference, the second difference, and the third difference.

[0112] Step S318: Based on the target difference, the parameters of the shared LSTM layer, the model parameters of the power prediction sub-model to be trained, and the model parameters of the load prediction sub-model to be trained are updated simultaneously.

[0113] Step S320 , retraining the power prediction sub-model and the load prediction sub-model after the model parameters are updated, until the trained power prediction sub-model and the trained load prediction sub-model meet a preset training end condition.

[0114] It should be noted that the specific limitations of the above steps can be found in the specific limitations of a source-load prediction method for collaborative optimization of a zero-carbon digital park.

[0115] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily performed in sequence in the order indicated by the arrows. Unless clearly stated herein, the execution of these steps is not strictly limited in order, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of the steps or stages in other steps or other steps. It is understandable that the various steps in different embodiments can be freely combined as needed, and the various non-contradictory schemes formed by the combination all fall within the scope of protection of this application.

[0116] Based on the same inventive concept, the embodiment of the present application also provides a source-load prediction system for collaborative optimization of zero-carbon digital parks for implementing the source-load prediction method for collaborative optimization of zero-carbon digital parks involved above. The implementation solution provided by this system is similar to the implementation solution described in the above method, so the specific limitations in one or more embodiments of the source-load prediction system for collaborative optimization of zero-carbon digital parks provided below can be found in the above limitations on the source-load prediction method for collaborative optimization of zero-carbon digital parks, and will not be repeated here.

[0117] A source-load prediction system for collaborative optimization of a zero-carbon digital park, comprising:

[0118] A data acquisition module is used to collect multi-source historical sample data of the zero-carbon digital park; the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data, and carbon emission data;

[0119] A model training and optimization module, configured to jointly train a power prediction sub-model to be trained and a load prediction sub-model to be trained based on the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model includes a pre-trained power prediction sub-model, a pre-trained load prediction sub-model, and a collaborative optimization unit;

[0120] The pre-trained power prediction sub-model is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park; the pre-trained load prediction sub-model is used to predict the park electrical load, thermal load and cooling load of the zero-carbon digital park; the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance;

[0121] Among them, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. In the process of training the power prediction sub-model to be trained and the load prediction sub-model to be trained, the model training and optimization module is specifically used to enable the power prediction sub-model to be trained and the load prediction sub-model to be trained to learn source-load correlation characteristics during training by sharing the weights of the LSTM layer; and, introduce the physical constraints of energy supply and demand balance, and punish the prediction results that deviate from the constraints through the loss function.

[0122] In an exemplary embodiment, Figure 4 As shown in the figure, a structural framework diagram of a source-load prediction system for collaborative optimization of zero-carbon digital parks is provided, which includes seven functional modules: weather data synchronization module, data acquisition module, data preprocessing module, feature engineering module, model training and optimization module, prediction module, and visualization and output module. Specifically:

[0123] Weather data synchronization module: Synchronize weather data for the next three days with the cloud, including temperature, humidity, air pressure, wind direction and speed, precipitation, etc.; perform data cleaning; store data in the weather database to ensure that the data can be synchronized to the local server quickly and stably. Figure 5 As shown in FIG, a flowchart of data synchronization is provided, such as Figure 5 As shown, weather data can be read every 15 minutes, cleaned, and stored in the database.

[0124] Data Collection Module: This module is used to collect multi-source historical data from the Zero Carbon Digital Park, including meteorological data, energy production data, energy consumption data, and carbon emission data. The module collects historical data from the park over the past year, including meteorological data (temperature, humidity, wind speed, irradiance, etc.), energy production data (photovoltaic power generation, wind power generation, etc.), energy consumption data (electricity load, heating load, cooling load, etc.), and carbon emission data.

[0125] Data preprocessing module: performs preprocessing operations such as cleaning, denoising, and normalization on the collected data.

[0126] Feature Engineering Module: Extracts features related to source load prediction from preprocessed data, including:

[0127] (1) Time characteristics: year, month, day, hour, holiday, etc.

[0128] (2) Meteorological characteristics: temperature, humidity, wind speed, irradiance, etc.

[0129] (3) Historical load characteristics: electrical load, heat load, cooling load, etc. in the past 24 hours.

[0130] (4) Historical power generation characteristics: photovoltaic power generation, wind power generation, etc. in the past 24 hours.

[0131] Model building module: Based on LSTM neural network, build source-load coordination prediction model, such as Figure 6 As shown in FIG, a schematic diagram of the structure of a model building module is provided. The source-load collaborative prediction model includes:

[0132] (1) Power prediction sub-model: used to predict the photovoltaic power generation and wind power generation of the park. The input features include time characteristics, meteorological characteristics, historical power generation characteristics, etc.

[0133] (2) Load forecasting sub-model: used to predict the park's electrical load, heating load, and cooling load. Input features include time characteristics, meteorological characteristics, historical load characteristics, etc.

[0134] (3) Collaborative optimization module: used to coordinate the power prediction sub-model and the load prediction sub-model to achieve source-load collaborative optimization prediction.

[0135] Model training and optimization module: This module uses historical data from the past year to train the source-load coordination prediction model. It then optimizes the model using methods such as cross-validation and hyperparameter optimization to improve its prediction accuracy. For example, hyperparameter optimization methods include Bayesian optimization to search for the optimal parameter combination and early stopping to prevent overfitting (with a patience value of 10 rounds).

[0136] Prediction module: used to use the trained source-load collaborative prediction model to predict the park power output and load demand in the future.

[0137] Visualization and output module: The prediction results are visualized in the form of charts, etc., and the prediction data is output to provide decision support for the optimized operation of the energy system and carbon emission management of the zero-carbon digital park.

[0138] Each module in the aforementioned source-load prediction system for collaborative optimization of a zero-carbon digital park can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0139] In an exemplary embodiment, a computer device is provided. The computer device may be a source-load prediction system for collaborative optimization of a zero-carbon digital park. The internal structure diagram thereof may be as follows: Figure 7 As shown. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means can be implemented via WIFI, a mobile cellular network, near field communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a source-load prediction method for collaborative optimization of zero-carbon digital parks. The display unit of the computer device is used to form a visually visible image, and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0140] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0141] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0142] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0143] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0144] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0145] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0146] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0147] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A source-load prediction method for collaborative optimization of zero-carbon digital parks, characterized by: The method comprises: Collect multi-source historical sample data of the zero-carbon digital park; the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data, and carbon emission data; Jointly training the power prediction sub-model to be trained and the load prediction sub-model to be trained according to the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model includes a pre-trained power prediction sub-model, a pre-trained load prediction sub-model and a collaborative optimization unit; The pre-trained power prediction sub-model is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park; the pre-trained load prediction sub-model is used to predict the park electrical load, thermal load and cooling load of the zero-carbon digital park; the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance; In which, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. In the process of training the power prediction sub-model to be trained and the load prediction sub-model to be trained, the power prediction sub-model to be trained and the load prediction sub-model to be trained learn the source-load correlation characteristics during training by sharing the weights of the LSTM layer; and the physical constraints of energy supply and demand balance are introduced, and the prediction results that deviate from the constraints are punished by the loss function.

2. The method according to claim 1, characterized in that The method of jointly training a power source prediction sub-model to be trained and a load prediction sub-model to be trained based on the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model includes: Extracting sample target features related to source-load prediction from the multi-source historical sample data; the sample target features include sample time features, sample meteorological features, sample load features, and sample power generation features; Inputting the sample time features, the sample meteorological features, and the sample power generation features into the power prediction sub-model to be trained, and outputting predicted power generation data; Inputting the sample time characteristics, the sample meteorological characteristics, and the sample load characteristics into the load prediction sub-model to be trained, and outputting predicted load data; According to the first difference between the predicted power generation data and the actual power generation data, and the second difference between the predicted load data and the actual load data, the power source prediction sub-model to be trained and the load prediction sub-model to be trained are jointly trained to obtain the pre-trained source-load collaborative prediction model.

3. The method according to claim 2, characterized in that The method further comprises: Inputting the predicted power generation data and the actual power generation data into a preset first loss function, and outputting a first loss function value; the first loss function value represents the first difference; The predicted load data and the actual load data are input into a preset second loss function, and a second loss function value is output; the second loss function value represents the second difference.

4. The method according to claim 2, characterized in that The method of jointly training the power prediction sub-model to be trained and the load prediction sub-model to be trained based on the first difference between the predicted power generation data and the actual power generation data and the second difference between the predicted load data and the actual load data includes: determining a third difference between the predicted load data and the predicted power generation data based on the physical constraints of the energy supply and demand balance; The power supply prediction sub-model to be trained and the load prediction sub-model to be trained are jointly trained according to the first difference, the second difference and the third difference.

5. The method according to claim 4, characterized in that The physical constraint of energy supply and demand balance is that power output is greater than or equal to load demand, and determining the third difference between the predicted load data and the predicted power generation data based on the physical constraint of energy supply and demand balance includes: Inputting the predicted load data and the predicted power generation data into a third loss function associated with the physical constraint, and outputting a third loss function value; the third loss function value represents the third difference; The formula of the third loss function is: ； in, is used to characterize the third loss function value, It is the constraint weight coefficient, which is used to control the penalty intensity for supply and demand imbalance.

6. The method according to claim 4, characterized in that The jointly training the to-be-trained power prediction sub-model and the to-be-trained load prediction sub-model according to the first difference, the second difference, and the third difference includes: determining a target difference based on the first difference, the second difference, and the third difference; According to the target difference, simultaneously updating the parameters of the shared LSTM layer, the model parameters of the power prediction sub-model to be trained, and the model parameters of the load prediction sub-model to be trained; The power prediction sub-model and the load prediction sub-model after the model parameters are updated are retrained until the trained power prediction sub-model and the trained load prediction sub-model meet the preset training end condition.

7. A source-load prediction system for collaborative optimization of zero-carbon digital parks, characterized by: The system comprises: A data acquisition module is used to collect multi-source historical sample data of the zero-carbon digital park; the multi-source historical sample data includes at least two of meteorological data, energy production data, energy consumption data, and carbon emission data; A model training and optimization module, configured to jointly train a power prediction sub-model to be trained and a load prediction sub-model to be trained based on the multi-source historical sample data to obtain a pre-trained source-load collaborative prediction model; the pre-trained source-load collaborative prediction model includes a pre-trained power prediction sub-model, a pre-trained load prediction sub-model, and a collaborative optimization unit; The pre-trained power prediction sub-model is used to predict the photovoltaic power generation and wind power generation of the zero-carbon digital park; the pre-trained load prediction sub-model is used to predict the park electrical load, thermal load and cooling load of the zero-carbon digital park; the collaborative optimization unit is used to adjust the outputs of the pre-trained power prediction sub-model and the pre-trained load prediction sub-model based on the physical constraints of energy supply and demand balance; Among them, the power prediction sub-model to be trained and the load prediction sub-model to be trained are constructed based on the LSTM neural network. In the process of training the power prediction sub-model to be trained and the load prediction sub-model to be trained, the model training and optimization module is specifically used to enable the power prediction sub-model to be trained and the load prediction sub-model to be trained to learn source-load correlation characteristics during training by sharing the weights of the LSTM layer; and, introduce the physical constraints of energy supply and demand balance, and punish the prediction results that deviate from the constraints through the loss function.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images