Distributed energy and load collaborative resource optimization scheduling method under machine learning

Abstract

The application discloses a distributed energy and load collaborative resource optimization scheduling method under machine learning, relates to the technical field of distributed energy scheduling, and comprises the following steps: obtaining an energy monitoring sequence by monitoring a distributed energy system in real time, and synchronously obtaining a load monitoring sequence on the user side; obtaining a first energy load scheduling group through transaction matching under trend prediction according to the two sequences; obtaining a second group according to predetermined balance optimization; screening a third group through a multi-level scheduling evaluation model; and determining an energy load collaborative management strategy through attention breeding optimization of the second group and the third group. The application solves the technical problems of traditional scheduling methods, such as insufficient prediction accuracy, limited balance and comprehensive evaluation capability, and difficulty in realizing real-time, accurate and low-energy collaborative optimization scheduling, achieves the technical effects of fast response to market changes, accurate matching of distributed energy and load demand, and intelligent and efficient energy load collaborative scheduling transaction.

Description

Technical Field

[0001] This invention relates to the field of distributed energy dispatching technology, and in particular to a resource optimization dispatching method for distributed energy and load coordination based on machine learning. Background Technology

[0002] Distributed generation is widely present in modern energy systems, including various forms such as solar, wind, and energy storage. Its coordinated scheduling with user-side loads has a significant impact on energy utilization, power supply reliability, and security. Current technologies primarily employ traditional mathematical optimization methods and single prediction models for resource allocation and scheduling decisions to address the problem of coordinated optimization scheduling between distributed generation systems and loads. These methods have played a crucial role in scenarios with relatively stable loads and energy supply. However, with the increasing proportion of renewable energy and intensified fluctuations in electricity load, traditional scheduling technologies exhibit poor adaptability, insufficient prediction accuracy, and limited coordination capabilities when applied to distributed generation systems, making it difficult to meet the refined and real-time scheduling requirements under conditions of high renewable energy integration. Summary of the Invention

[0003] This application provides a resource optimization scheduling method for distributed energy and load coordination based on machine learning, which solves the technical problems of insufficient prediction accuracy, limited balance and comprehensive evaluation capabilities of traditional scheduling methods, making it difficult to achieve real-time, accurate and low-energy-consumption coordinated optimization scheduling.

[0004] The first aspect of this application provides a resource optimization scheduling method for distributed energy and load coordination under machine learning. The method includes: real-time monitoring of a distributed energy system to obtain an energy monitoring sequence, and synchronously acquiring the load monitoring sequence corresponding to the user side of the distributed energy system; making transaction matching decisions based on trend prediction based on the energy monitoring sequence and the load monitoring sequence to obtain a first energy load scheduling group, and performing energy load balance analysis optimization on the first energy load scheduling group based on a predetermined energy load balance degree to obtain a second energy load scheduling group; performing scheduling evaluation optimization on the second energy load scheduling group according to a multi-level scheduling evaluation model to obtain a third energy load scheduling group, and performing attention propagation optimization based on the second energy load scheduling group and the third energy load scheduling group to determine an energy load coordination management strategy.

[0005] A second aspect of this application provides a resource optimization scheduling system for distributed energy and load coordination under machine learning. The system includes: an energy monitoring sequence acquisition module, used to monitor the distributed energy system in real time, obtain an energy monitoring sequence, and simultaneously acquire the load monitoring sequence corresponding to the user side of the distributed energy system; an energy load scheduling group acquisition module, used to make transaction matching decisions based on trend prediction of the energy monitoring sequence and the load monitoring sequence, obtain a first energy load scheduling group, and perform energy load balance analysis optimization on the first energy load scheduling group based on a predetermined energy load balance degree, to obtain a second energy load scheduling group; and an energy load coordination management strategy determination module, used to perform scheduling evaluation optimization on the second energy load scheduling group according to a multi-level scheduling evaluation model, obtain a third energy load scheduling group, and perform attention propagation optimization based on the second energy load scheduling group and the third energy load scheduling group, to determine an energy load coordination management strategy.

[0006] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the resource optimization scheduling method for distributed energy and load coordination under machine learning provided in this application.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] This application achieves real-time monitoring of distributed energy systems and user-side loads, obtaining energy monitoring sequences and load monitoring sequences. It utilizes multiple machine learning models for trend prediction and generates an energy load prediction map. An initial scheduling scheme group is obtained through transaction matching decisions. Further optimization is performed based on a predetermined energy load balance, combined with a multi-level scheduling evaluation model for further selection. Attention-based optimization and energy consumption minimization optimization are introduced to ultimately generate a precise and reliable energy load collaborative management strategy. This achieves efficient collaborative scheduling of distributed energy and loads, enabling rapid response to market changes, accurate matching of distributed energy and load demands, and intelligent and efficient energy load collaborative scheduling transactions. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the resource optimization scheduling method for distributed energy and load coordination under machine learning provided in this application embodiment.

[0011] Figure 2 This is a schematic diagram of the structure of the distributed energy and load coordination resource optimization scheduling system under machine learning provided in the embodiments of this application.

[0012] Figure labeling: Energy monitoring sequence acquisition module 1, energy load dispatch group acquisition module 2, energy load collaborative management strategy determination module 3. Detailed Implementation

[0013] This application provides a resource optimization scheduling method for distributed energy and load coordination based on machine learning, which solves the technical problems of insufficient prediction accuracy, limited balance and comprehensive evaluation capabilities of traditional scheduling methods, making it difficult to achieve real-time, accurate and low-energy-consumption coordinated optimization scheduling.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0016] Example 1, as Figure 1 As shown, a resource optimization scheduling method for distributed energy and load coordination under machine learning is described, wherein the method includes:

[0017] The distributed energy system is monitored in real time to obtain the energy monitoring sequence, and the load monitoring sequence corresponding to the user side of the distributed energy system is obtained simultaneously.

[0018] In this embodiment, a distributed energy system refers to a small-scale energy system that supplies power to nearby locations. The user side refers to the electricity-consuming side that corresponds to the distributed energy system, can obtain load monitoring sequences through synchronous monitoring, and needs to perform energy-coordinated dispatch with the distributed energy system.

[0019] Specifically, at the critical operating nodes of each energy device in the distributed energy system—including solar photovoltaic modules, small wind turbines, and energy storage battery packs—voltage sensors, current sensors, power transmitters, and high-precision energy metering instruments are deployed. These devices continuously collect real-time operating parameters of the energy devices according to a preset sampling frequency, which can be flexibly adjusted between once per second and once per minute based on actual scheduling needs. These parameters include the output power of photovoltaic modules, the speed and power generation of wind turbines, and the remaining charge and discharge power of energy storage battery packs. The collected raw data undergoes a smoothing process to remove outliers caused by sensor malfunctions, standardize data units and storage formats to reduce the impact of short-term random fluctuations, and then organize the processed data into a continuous dataset in chronological order. This dataset constitutes the energy monitoring sequence, which comprehensively records the energy supply capacity and equipment operating status of the distributed energy system at different points in time.

[0020] Next, for the user side of the distributed energy system, smart meters are deployed at key locations on the user's electricity circuit, based on the differences in user type, such as industrial users, commercial users, and residential users. To ensure accurate time synchronization between load data and energy data, these user-side monitoring devices maintain strict time synchronization with the energy-side monitoring devices through network time protocols or GPS clock synchronization technology. The monitoring devices first collect data such as the user's real-time power consumption, cumulative power consumption, time-period distribution characteristics of power load, and operating status of key electrical equipment. Then, these raw load data are processed, including removing abnormal power consumption data caused by electrical equipment failures, correcting transmission errors during data collection, and standardizing data statistical standards. Finally, a load monitoring sequence synchronized with the energy monitoring sequence is formed, which can accurately reflect the scale and changing patterns of the user's electricity demand at different times.

[0021] The energy supply capacity of distributed energy systems and the electricity demand on the user side are both dynamically changing. Obtaining only energy monitoring sequences or load monitoring sequences alone cannot accurately capture their temporal matching relationship. This results in a lack of complete and synchronized data support for subsequent trend forecasting and transaction matching decisions, thus affecting the accuracy of distributed energy and load coordinated scheduling. However, by obtaining energy monitoring sequences and load monitoring sequences separately through the aforementioned steps and ensuring their time synchronization, comprehensive perception and real-time tracking of both the distributed energy supply side and user demand side can be achieved, providing a core data foundation for subsequent scheduling-related work based on these two types of sequences.

[0022] By deploying monitoring equipment on the distributed energy system side and obtaining energy monitoring sequences through data processing, and deploying time-synchronized monitoring equipment on the user side and obtaining load monitoring sequences through data processing, the system achieves comprehensive, real-time, and time-synchronized data collection and status recording for both the distributed energy supply side and the user demand side.

[0023] Based on the energy monitoring sequence and the load monitoring sequence, a transaction matching decision is made under trend prediction to obtain a first energy load scheduling group. Then, based on a predetermined energy load balance degree, the first energy load scheduling group is analyzed and optimized to obtain a second energy load scheduling group.

[0024] In this embodiment of the application, the energy load balance is an indicator used to evaluate the degree of matching between energy supply and load demand in an energy load scheduling scheme.

[0025] Optionally, firstly, multiple machine learning models are used to perform trend prediction on the energy monitoring sequence and the load monitoring sequence, respectively, to obtain the corresponding energy feature prediction results and load feature prediction results; then, these two types of feature prediction results are organized to form an energy load prediction map; finally, transaction matching decisions are made based on the energy load prediction map, thereby generating the first energy load dispatch group.

[0026] Next, the first energy load scheduling scheme is extracted from the first energy load scheduling group, and then the energy load balance is retrieved historically based on the scheme to obtain the first energy load balance historical sequence. Subsequently, the first fitted energy load balance is obtained by lumped value calculation on the historical sequence. If the first fitted energy load balance is greater than or equal to the predetermined energy load balance, the first energy load scheduling scheme is added to the second energy load scheduling group. Finally, the energy load balance is analyzed and optimized in the first energy load scheduling group based on the predetermined energy load balance to update the second energy load scheduling group.

[0027] The second energy load scheduling group is evaluated and optimized using a multi-level scheduling evaluation model to obtain a third energy load scheduling group. Then, attention propagation optimization is performed based on the second and third energy load scheduling groups to determine the energy load collaborative management strategy.

[0028] In one embodiment of this application, a multi-level dispatch evaluation model that includes multi-level dispatch evaluation indicators such as power supply quality and power supply security is first activated. Then, the model is used to conduct dispatch evaluation on each energy load dispatch scheme in the second energy load dispatch group to obtain multiple dispatch evaluation sequences. Subsequently, constraints are configured according to the multi-level dispatch evaluation indicators to determine the dispatch evaluation constraints. Finally, the dispatch evaluation constraints and multiple dispatch evaluation sequences are combined to optimize and screen the second energy load dispatch group, thereby generating the third energy load dispatch group.

[0029] Next, attention is multiplied on the second energy load scheduling group based on the third energy load scheduling group to obtain the first scheduling multiplication space; then, the first scheduling multiplication space is evaluated and optimized through a multi-level scheduling evaluation model to obtain the second scheduling multiplication space; subsequently, the third energy load scheduling group is expanded with the second scheduling multiplication space to form the fourth energy load scheduling group; finally, the fourth energy load scheduling group is optimized for minimizing energy consumption, thereby generating an energy load collaborative management strategy.

[0030] Furthermore, the method provided in this application embodiment includes:

[0031] Load the historical energy monitoring data of the distributed energy system; perform supervised training on the historical energy monitoring data using the multiple machine learning models to obtain multiple energy trend prediction models; conduct adversarial tests on the multiple energy trend prediction models using the historical energy monitoring data to generate multiple energy prediction loss data; perform integrated training on the multiple energy trend prediction models using the multiple energy prediction loss data to obtain energy trend prediction nodes; input the energy monitoring sequence into the energy trend prediction nodes to generate the energy feature prediction results.

[0032] Optionally, firstly, load the energy monitoring history set of the distributed energy system. This set contains energy operation data of the distributed energy system at different times in the past, such as the output power of solar photovoltaic modules, the power generation of small wind turbines, the remaining power and charging and discharging power of energy storage battery packs, etc. Obtain this historical data to form an energy monitoring history data set that can be used by subsequent models.

[0033] Next, supervised training is conducted on several machine learning models familiar to those skilled in the art to obtain multiple energy trend prediction models. Specifically, taking linear regression models, Long Short-Term Memory (LSTM) network models, and Random Forest models as examples:

[0034] For the linear regression model, during supervised training, the time, environmental parameters, and corresponding energy output power of the historical energy monitoring data are used as input feature data, and the actual energy output power of that period is used as output label data. During training, the mean squared error is used as the loss function, and the linear coefficient is iteratively adjusted to gradually reduce the loss function value to below 0.01. At the same time, the R² coefficient of determination is used to evaluate the model's fit, ensuring that the R² coefficient of determination is not lower than 0.85 after training. If regularization is introduced, L2 regularization is used, and the regularization parameter λ ranges from 0.01 to 0.05 to avoid model overfitting. Finally, the training is completed to obtain an energy trend prediction model based on linear regression.

[0035] For the LSTM model, during supervised training, the input sequence length is first determined to be 12 to 24 hours, that is, the input is energy operation data of 12 to 24 consecutive hours in the historical energy monitoring set, and the output is the energy prediction value for the next period. During training, the hidden layer dimension is set to 64 to 128, the batch size is set to 32 to 64, the training epochs are set to 50 to 100, Adam is selected as the optimizer and the learning rate is controlled between 0.001 and 0.005, the mean squared error is used as the loss function, and the hidden layer parameters are continuously optimized based on the loss function value through the backpropagation algorithm to reduce the deviation between the predicted value and the historical actual value until the loss function value is stable within 0.02, and finally the energy trend prediction model based on LSTM is obtained.

[0036] For the random forest model, the supervised training phase first uses multi-dimensional features from the historical energy monitoring data as input and the actual energy values ​​for the corresponding time period as output. During training, the number of trees is set to 100 to 200, the maximum tree depth is 10 to 15 layers, and the number of features selected at each split is one-half to two-thirds of the total number of features. The mean squared error is used as the basic evaluation index for the regression task. By constructing multiple decision trees and combining the prediction results of each tree, the number of trees and the branching rules are adjusted to ensure that the mean squared error of the model after training does not exceed 0.03, thus ensuring that the overall prediction accuracy meets the standard, and thus obtaining the energy trend prediction model based on random forest.

[0037] Subsequently, multiple energy trend prediction models were tested against disturbances based on historical energy monitoring data. Specifically, small, reasonable disturbances were added to a portion of the data in the historical energy monitoring set, such as adjusting the photovoltaic output power data for a certain period by ±5%. These disturbed historical data were then input into each energy trend prediction model, which output a corresponding energy prediction value. This prediction value was compared with the actual energy value for that period without disturbance, and the degree of difference between the two was calculated. This degree of difference represents the energy prediction loss data for a single model. This application uses a mean squared error calculation formula to calculate the specific prediction loss value, and the specific formula is as follows:

[0038] Where MSE represents the predicted loss value of a single energy trend forecasting model, and n represents the number of samples tested against disturbances. This represents the historical actual value of energy monitoring for the i-th test sample. This represents the model's energy prediction for the i-th test sample. This process is performed on all energy trend prediction models, ultimately generating multiple corresponding energy prediction loss data sets.

[0039] Subsequently, multiple energy trend prediction models were integrated and trained based on various energy prediction loss data. Specifically, the energy prediction loss data of each model was first analyzed. Models with smaller loss data were assigned higher prediction accuracy and thus higher weights; models with larger loss data were assigned lower weights. All energy trend prediction models were then combined according to their weight ratios, for example, by using a weighted average to integrate the prediction results. During the fusion process, the weight allocation of each model was continuously adjusted. By comparing the error between the predicted values ​​output by the fused model and the actual values ​​in the historical energy monitoring set, the weight ratios were gradually optimized until the overall prediction error of the fused model was minimized. At this point, the energy trend prediction node was obtained.

[0040] Finally, the energy monitoring sequence is input into the energy trend prediction node. This sequence contains real-time and recent energy operation data of the distributed energy system, such as the photovoltaic output power for the current hour and the current remaining power of the energy storage battery. The energy trend prediction node uses multiple internally integrated energy trend prediction models, combined with the weights of each model, to calculate the input monitoring sequence and output energy characteristic data for a specific future period, such as the photovoltaic output power change trend for the next 24 hours and the peak charging and discharging demand periods of the energy storage battery. This output data constitutes the energy characteristic prediction results.

[0041] By loading historical data, training multiple models separately, testing model loss, fusing and optimizing the model, and inputting real-time data for calculation, the advantages of multiple machine learning models are integrated to improve the accuracy of energy feature prediction.

[0042] Furthermore, the method provided in this application embodiment includes:

[0043] The energy monitoring sequence is trend-predicted using multiple machine learning models to obtain energy characteristic prediction results; the load monitoring sequence is trend-predicted using the same multiple machine learning models to obtain load characteristic prediction results; the energy characteristic prediction results and the load characteristic prediction results are combined to obtain an energy load prediction map; transaction matching decisions are made based on the energy load prediction map to generate the first energy load scheduling group.

[0044] Specifically, first, the historical energy monitoring data of the distributed energy system is loaded, and then supervised training is performed on the historical energy monitoring data using multiple machine learning models to obtain multiple energy trend prediction models. Subsequently, adversarial perturbation tests are performed on these energy trend prediction models based on the historical energy monitoring data to generate multiple energy prediction loss data. Next, the multiple energy trend prediction models are integrated and fused based on these energy prediction loss data to obtain an energy trend prediction node. Finally, the energy monitoring sequence is input into the energy trend prediction node to generate energy feature prediction results.

[0045] Next, the historical load monitoring data corresponding to the user side of the distributed energy system is loaded. This historical data includes load data such as power consumption and duration of electricity consumption at different times in the past. Similar to the previous steps, multiple machine learning models are selected and trained under supervised supervision using the historical load monitoring data. For example, coefficients are adjusted for the linear regression model to reduce prediction error, hidden layer parameters are optimized for the LSTM model to improve time-series prediction capability, and the number and depth of trees are adjusted for the random forest model to improve fitting accuracy, resulting in multiple load trend prediction models. Subsequently, each load trend prediction model is tested against disturbances using the historical load monitoring data. After fine-tuning some historical load data, the model is input to calculate the difference between the predicted and actual values, generating multiple load prediction loss data. Based on the loss data, multiple models are integrated and fused for training. Models with high prediction accuracy (i.e., smaller loss data) are assigned higher weights, forming a load trend prediction node. The real-time acquired load monitoring sequence is input into this node to obtain load characteristic prediction results containing the scale and trend of user load demand at different times in the future.

[0046] Next, the energy characteristic prediction results obtained from the aforementioned steps using multiple machine learning models to predict the trend of the energy monitoring sequence are acquired. These results include the energy supply capacity and changing trends of the distributed energy system in different future time periods. The energy characteristic prediction results and load characteristic prediction results are aligned along the time dimension, with each hourly energy supply prediction value and load demand prediction value mapped one-to-one. Following the structure of: time period - energy supply prediction value - load demand prediction value, the aligned results are organized into a structured dataset, forming an energy load prediction map that clearly shows the correspondence between energy supply and load demand in different future time periods.

[0047] Next, examine the energy supply and load demand matching in the energy load forecast map for each time period. For periods where the energy supply forecast is greater than the load demand forecast, formulate a preliminary dispatch plan for excess energy allocation; for periods where supply is less than demand, formulate a preliminary dispatch plan for supplementary energy; for periods where supply and demand are close to balance, formulate a preliminary dispatch plan to maintain the current supply status. Eliminate unreasonable preliminary plans, such as those with allocation volumes exceeding the system's carrying capacity, and collect and integrate the selected reasonable plans to generate the first energy load dispatch group.

[0048] By analogy with the steps of energy trend forecasting, load characteristic forecasting results are obtained, two types of forecasting results are organized into a graph, and scheduling schemes are formulated and screened based on the graph. This achieves the effects of accurately predicting load demand, clearly presenting the relationship between energy and load supply and demand, and generating a reasonable first energy load scheduling group.

[0049] Furthermore, the method provided in this application embodiment includes:

[0050] The first energy load scheduling scheme is extracted from the first energy load scheduling group; the energy load balance history is retrieved based on the first energy load scheduling scheme to obtain the first energy load balance history sequence; the central value is calculated on the first energy load balance history sequence to obtain the first fitted energy load balance; if the first fitted energy load balance is greater than or equal to the predetermined energy load balance, the first energy load scheduling scheme is added to the second energy load scheduling group; the energy load balance is further analyzed and optimized based on the predetermined energy load balance, and the second energy load scheduling group is updated.

[0051] Specifically, firstly, the generated first energy load scheduling group is obtained, which contains multiple preliminary scheduling schemes for distributed energy and load coordination. Then, by systematically analyzing each scheme, each independent scheduling scheme is extracted from the first energy load scheduling group, clarifying key information such as the energy supply allocation method, the matching objects for load demand, and the scheduling execution period, ultimately resulting in multiple independent first energy load scheduling schemes.

[0052] Next, for each extracted first energy load dispatching scheme, its core characteristic parameters are determined, including the energy type involved (e.g., solar, wind power), user-side load scale, dispatching cycle, etc. Based on these core characteristic parameters, the historical dispatching database of the distributed energy system is used to retrieve previously executed dispatching schemes with the same or highly similar characteristic parameters as the current scheme, and the energy load balance data recorded during the execution of these historical schemes is extracted. The retrieved historical energy load balance data are arranged in chronological order to form a historical sequence of first energy load balance corresponding to the current first energy load dispatching scheme.

[0053] Furthermore, the aforementioned historical scheduling database can directly retrieve the set of scheduling records stored in the past operation of the distributed energy system. This set includes the text of scheduling schemes executed over the years and for each time period, the corresponding energy supply data (such as solar power generation and wind power generation), and load consumption data. The specific indicator of energy load balance is the energy load balance rate. This application uses a balance rate calculation formula to calculate the specific energy load balance, and the specific formula is as follows:

[0054] Where B represents the energy load balance degree for a certain period, in percentage; S represents the actual energy supply of the distributed energy system for that period, in kW·h; and L represents the actual load demand of the user side for that period, in kW·h. The calculated energy load balance degree only reflects the degree of imbalance, without distinguishing the direction of imbalance. This is based on the core objective of distributed energy and load coordinated scheduling, which is to prioritize avoiding any form of supply and demand imbalance from affecting the system.

[0055] When the supply of a distributed energy system exceeds demand, excess energy may overload energy storage devices, lead to energy waste, and even affect the stable output of the distributed energy system. Conversely, when supply falls short of demand, a power shortage will occur, failing to meet the basic electricity needs of users and potentially causing equipment outages and a decline in user experience. For distributed energy systems, both of these imbalances disrupt the coordinated stability of energy and load. Therefore, when evaluating the balance performance of historical scheduling schemes, it is necessary to first uniformly measure the severity of the imbalance. This indicator is obtained by extracting the actual energy supply and actual load demand recorded during the execution of the corresponding historical scheme from the historical scheduling database, substituting them into the formula for calculation, and then arranging this indicator of multiple historical schemes in chronological order to form the first historical sequence of energy load balance.

[0056] Then, the obtained historical sequence of the first energy load balance degree is processed. The arithmetic mean method is used to calculate the lumped value. All energy load balance degree values ​​in the historical sequence are added together and then divided by the total number of values. This calculation method yields a first fitted energy load balance degree that represents the overall level of the historical sequence, thus reflecting the average balance level of the current first energy load scheduling scheme in historical execution. The specific formula for calculating the arithmetic mean used in the above content is as follows:

[0057] ,in Let represent the first fitted energy load balance degree, and k represent the number of samples in the historical sequence of the first energy load balance degree. This represents the energy load balance of the j-th sample in the historical sequence.

[0058] Next, the calculated first fitted energy load balance degree is compared with the preset predetermined energy load balance degree. If the value of the first fitted energy load balance degree is greater than or equal to the predetermined energy load balance degree, it indicates that the first energy load scheduling scheme can meet the system's basic requirements for energy and load balance in historical execution. The scheme is then included in the candidate range of the second energy load scheduling group, completing the screening and addition of a single scheme.

[0059] The predetermined energy load balance can be set at 85%. This is achieved by referencing historical energy load balance data from the past three years of stable operation of the distributed energy system, taking the minimum balance value for over 80% of the time periods, and combining this with the industry's minimum balance standard for safe operation of distributed energy systems. This ensures stable power supply and avoids equipment failures or power outages due to energy and load imbalances, while also referencing historical balance ranges that meet user electricity needs with low system energy consumption. Next, the calculated first fitted energy load balance is compared with this 85% predetermined energy load balance. If the first fitted energy load balance is greater than or equal to 85%, it indicates that the first energy load dispatching scheme has met the system's basic requirements for energy and load balance in historical implementations. This scheme is then included in the candidate range of the second energy load dispatching group, completing the selection and addition of a single scheme.

[0060] Finally, following the steps described above, all remaining energy load dispatching schemes in the first energy load dispatching group are extracted, historically retrieved, lumped value calculated, and balance compared one by one. Each dispatching scheme that meets the predetermined energy load balance requirement is added to the second energy load dispatching group in sequence, while schemes that do not meet the predetermined energy load balance requirement are removed. The content of the second energy load dispatching group is continuously updated until all schemes in the first energy load dispatching group have been screened.

[0061] By extracting schemes from the scheduling group, retrieving historical balance data, calculating the fit balance, comparing and screening schemes, and continuously updating the scheduling group, the goal of selecting scheduling schemes that meet the predetermined requirements for energy load balance and forming a high-quality second energy load scheduling group is achieved.

[0062] Furthermore, the method provided in this application embodiment includes:

[0063] The multi-level scheduling evaluation model is activated, which includes multi-level scheduling evaluation indicators, including power supply quality and power supply security. The scheduling evaluation of each energy load scheduling scheme within the second energy load scheduling group is performed based on the multi-level scheduling evaluation model to obtain multiple scheduling evaluation sequences. Constraints are configured based on the multi-level scheduling evaluation indicators to obtain scheduling evaluation constraints. The second energy load scheduling group is then optimized and screened based on the scheduling evaluation constraints and the multiple scheduling evaluation sequences to generate the third energy load scheduling group.

[0064] Specifically, firstly, a neural network model is constructed and trained as a multi-level scheduling evaluation model. The model structure is defined as an input layer - three hidden layers - an output layer. Specifically, the input layer has 8 neurons, corresponding to the eight core features of the energy load scheduling scheme: energy supply type, supply stability, load matching rate, voltage fluctuation range, frequency fluctuation range, fault response time, reserve capacity ratio, and historical execution success rate. The first hidden layer has 32 neurons, the second hidden layer has 16 neurons, and the third hidden layer has 8 neurons. The output layer has 2 neurons, corresponding to the power supply quality evaluation coefficient and the power supply safety evaluation coefficient, respectively. The ReLU activation function is used in the hidden layers to improve the model's nonlinear fitting ability, and the Sigmoid function is used in the output layer to limit the evaluation coefficient range to 0-1 for easier quantitative comparison. The optimizer is Adam, with a learning rate of 0.003, 50 training epochs, a batch size of 32, and the mean squared error loss function, which minimizes the deviation between the model's predicted evaluation coefficients and the actual evaluation results. The training data consisted of 1,000 effective scheduling schemes of distributed energy systems over the past three years, along with corresponding actual power supply quality and security evaluation data. The model was trained until the loss function value stabilized within 0.01. The trained model was then activated, and its multi-level scheduling evaluation indicators were identified as power supply quality and power supply security.

[0065] Next, for each energy load dispatching scheme within the second energy load dispatching group, eight core feature data points are extracted and standardized. This involves converting each feature value into a 0-1 range to eliminate the influence of dimensions. The standardized feature data is then input into an activated neural network model. The multi-level dispatching evaluation model, through forward propagation, combines indicators related to power quality, such as voltage fluctuation range, frequency fluctuation range, and supply stability, as well as indicators related to power supply safety, such as fault response time, reserve capacity ratio, and historical execution success rate. Ultimately, it outputs power quality evaluation coefficients and power supply safety evaluation coefficients. Each dispatching scheme generates a dispatching evaluation sequence containing these two coefficients, resulting in multiple dispatching evaluation sequences consistent with the number of schemes in the second energy load dispatching group.

[0066] Then, scheduling evaluation constraints are configured based on the power supply quality and power supply security indicators of the multi-level scheduling evaluation system. Referring to the basic requirements for ensuring normal power supply to users in distributed energy systems and the evaluation standards of high-quality scheduling schemes in the industry, the constraint threshold for the power supply quality evaluation coefficient is set at no less than 0.8 to ensure that voltage and frequency remain stable within a reasonable range. The constraint threshold for the power supply security evaluation coefficient is set at no less than 0.85 to ensure rapid fault response and sufficient backup capacity, thus forming a scheduling evaluation constraint condition with a power supply quality evaluation coefficient ≥ 0.8 and a power supply security evaluation coefficient ≥ 0.85.

[0067] Finally, each energy load dispatching scheme and its corresponding dispatching evaluation sequence within the second energy load dispatching group are verified one by one: if a scheme's power supply quality evaluation coefficient is ≥0.8 and its power supply safety evaluation coefficient is ≥0.85, it indicates that the scheme meets the system operation requirements in terms of power supply quality and safety, and is included in the candidate range of the third energy load dispatching group; if either coefficient is lower than the corresponding constraint threshold, it indicates that the scheme has power supply instability or safety risks, and is eliminated. After all schemes have been verified, the schemes in the candidate range are integrated to generate the third energy load dispatching group.

[0068] By constructing a neural network model with clearly defined training parameters, extracting features of scheduling schemes to calculate evaluation coefficients, configuring constraint thresholds, and screening schemes that meet the constraints, the system achieves the effect of accurately evaluating the power supply quality and security of scheduling schemes and selecting schemes that meet the system operation requirements to form a high-quality third energy load scheduling group.

[0069] Furthermore, the method provided in this application embodiment includes:

[0070] Based on the third energy load scheduling group, the second energy load scheduling group is identified to obtain a scheduling difference feature group; the scheduling difference feature group is mutated and adjusted to obtain a reproductive attention feature group; the second energy load scheduling group is adjusted based on the reproductive attention feature group to obtain a scheduling adjustment space; the scheduling adjustment space is analyzed and optimized based on the predetermined energy load balance degree to generate the first scheduling reproductive space.

[0071] Specifically, firstly, for the third and second energy load dispatch groups obtained in the aforementioned steps, the core features of each energy load dispatch scheme within each dispatch group are extracted. These core features include energy supply type, load matching rate, power supply quality evaluation coefficient, power supply security evaluation coefficient, and dispatch cycle. Each scheme in the third energy load dispatch group is paired one-to-one with similar schemes in the second energy load dispatch group, and the core feature values ​​of each pair are compared. For example, if a scheme in the third energy load dispatch group has a load matching rate of 92%, while a similar scheme in the second energy load dispatch group has a load matching rate of 83%, then a 9% difference in load matching rate is considered a difference feature point. Similarly, if a scheme in the third energy load dispatch group has a power supply quality evaluation coefficient of 0.91, while a similar scheme in the second energy load dispatch group has a coefficient of 0.82, then a 0.09 difference in power supply quality evaluation coefficient is considered another difference feature point. All difference feature points are collected according to the above steps to form a dispatch difference feature group.

[0072] Next, the variation of each difference feature point in the scheduling difference feature group is adjusted: those skilled in the art first set a reasonable variation range according to actual needs to ensure that the adjusted features still conform to the operating logic of the distributed energy system. For example, the variation range of load matching rate difference is set to ±4%, and the variation range of power quality evaluation coefficient difference is set to ±0.04. For each difference feature point, its value is adjusted within the set variation range. For example, the load matching rate difference of 9% in the above example is adjusted to 7%, and the power quality evaluation coefficient difference of 0.09 is adjusted to 0.07 to avoid the subsequent adjustment scheduling scheme exceeding the system's carrying capacity due to excessive variation. All adjusted difference feature points are integrated to form a reproductive attention feature group. Each feature point in this feature group is a key basis for optimizing the second energy load scheduling group.

[0073] Then, the second energy load dispatching group is adjusted based on the reproductive attention feature group. For each energy load dispatching scheme within the second energy load dispatching group, the corresponding difference feature point in the reproductive attention feature group is identified, and the core parameters of the scheme are adjusted according to the adjustment direction and value of the feature point. For example, if the reproductive attention feature point corresponding to a certain scheme is that the load matching rate needs to be improved by 7%, then the load matching strategy of the scheme is adjusted to increase the duration of distributed energy supply, so that the load matching rate is increased from the original 83% to 90%; if the corresponding feature point is that the power quality evaluation coefficient needs to be improved by 0.07, then the voltage stability control measures of the scheme are optimized, so that the power quality evaluation coefficient is increased from 0.82 to 0.89. After all schemes within the second energy load dispatching group have been adjusted, these adjusted schemes are integrated to form a dispatching adjustment space.

[0074] Finally, similar to the steps described above, the energy load balance of the scheduling adjustment space is analyzed and optimized based on the predetermined energy load balance. Each adjusted energy load scheduling scheme is extracted from the scheduling adjustment space. Based on the core characteristics of each scheme, such as energy supply scale and total load demand, energy load balance data from past executions of the scheme or similar schemes are retrieved from the historical scheduling database of the distributed energy system, forming a historical energy load balance sequence for each scheme. A lumped value calculation is performed on each historical sequence. The arithmetic mean method is used to add all balance values ​​in the sequence and divide by the number of values ​​to obtain the fitted energy load balance for each scheme. The fitted energy load balance is compared with the predetermined energy load balance. If the fitted value is greater than or equal to the predetermined value, the adjusted scheme is included in the candidate range. After all schemes have been calculated and compared, the schemes in the candidate range are integrated to generate the first scheduling breeding space.

[0075] By extracting features, comparing and identifying differences, setting ranges to adjust the difference features, optimizing the scheduling scheme based on the features, and calculating the balance degree to select the scheme, the first scheduling breeding space that meets the predetermined energy load balance degree requirements is generated, thus providing a high-quality scheme foundation for subsequent energy load collaborative scheduling optimization.

[0076] Furthermore, the method provided in this application embodiment includes:

[0077] Attention is multiplied in the second energy load scheduling group based on the third energy load scheduling group to obtain a first scheduling multiplication space; scheduling evaluation and optimization are performed on the first scheduling multiplication space based on the multi-level scheduling evaluation model to obtain a second scheduling multiplication space; the third energy load scheduling group is expanded based on the second scheduling multiplication space to obtain a fourth energy load scheduling group; energy consumption minimization is performed on the fourth energy load scheduling group to generate the energy load collaborative management strategy.

[0078] In one embodiment, firstly, the second energy load scheduling group is identified by the third energy load scheduling group to obtain a scheduling difference feature group; then, the scheduling difference feature group is mutated and adjusted to form a reproductive attention feature group; next, the second energy load scheduling group is adjusted according to the reproductive attention feature group to obtain a scheduling adjustment space; finally, the scheduling adjustment space is optimized by energy load balance analysis according to the predetermined energy load balance degree to generate the first scheduling reproductive space.

[0079] Next, similar to the steps described above, the activation of the multi-level dispatch evaluation model is performed. This model includes multi-level dispatch evaluation indicators for power quality and power security. The core features of each energy load dispatch scheme within the first dispatch breeding space are extracted. These core features include voltage fluctuation range, frequency fluctuation range, fault response time, and reserve capacity ratio. This feature data is then input into the activated multi-level dispatch evaluation model. The model evaluates the power quality and power security of each scheme, outputting a dispatch evaluation sequence containing power quality evaluation coefficients and power security evaluation coefficients. Subsequently, dispatch evaluation constraints are configured according to the multi-level dispatch evaluation indicators, setting the power quality evaluation coefficient to be no less than 0.8 and the power security evaluation coefficient to be no less than 0.85. The dispatch evaluation sequence of each scheme is verified to ensure it meets these constraints. Schemes that meet the constraints are selected and integrated to form the second dispatch breeding space.

[0080] Next, the existing energy load dispatching schemes within the third energy load dispatching group are reviewed, and the core characteristics of each scheme are recorded to avoid duplication, such as energy supply type, load matching rate, and dispatching cycle. All energy load dispatching schemes within the second dispatching propagation space are extracted, and their core characteristics are compared one by one with those of the schemes in the third energy load dispatching group. If the characteristics of a scheme in the second dispatching propagation space are completely identical to those in the third energy load dispatching group, the duplicate scheme is removed; if the characteristics differ or there are optimization differences, it is included in the third energy load dispatching group. The updated third energy load dispatching group is then integrated to obtain the fourth energy load dispatching group.

[0081] Finally, energy consumption-related data for each energy load dispatching scheme within the fourth energy load dispatching group are extracted, including the total energy consumption of solar photovoltaic modules, small wind turbines, and energy storage battery packs in the distributed energy system, as well as the energy consumption ratio of each equipment unit within the dispatching cycle. The energy consumption data for each scheme is quantified to obtain the total energy consumption value of a single scheme; for example, one scheme has a total energy consumption value of 500 kWh, while another scheme has 480 kWh. All schemes are sorted in ascending order of total energy consumption value, and the top 30% of schemes with the lowest total energy consumption value are selected. The specific percentage can be adjusted by those skilled in the art based on the actual operational needs of the distributed energy system. These low-energy-consumption schemes are further integrated, clarifying the execution period, energy allocation method, and load matching strategy for each scheme, generating the final energy load collaborative management strategy.

[0082] By employing the steps of screening solutions that meet constraints through scheduling evaluation, expanding the scheduling group to integrate high-quality resources, and selecting low-energy-consumption solutions through energy consumption ranking, the goal of determining an energy load collaborative management strategy that balances power supply quality, power supply security, and low energy consumption needs was achieved.

[0083] In summary, the resource optimization scheduling method for distributed energy and load coordination based on machine learning provided in this application has the following technical effects:

[0084] This application obtains energy monitoring sequences and user-side load monitoring sequences through real-time monitoring of distributed energy systems. Based on multiple machine learning models, it performs trend prediction on these two types of sequences and organizes them into an energy load prediction map. Combining this map, a first energy load scheduling group is generated. Then, based on a predetermined energy load balance, the first energy load scheduling group is analytically optimized to obtain a second energy load scheduling group. A third energy load scheduling group is obtained by evaluating and screening the second energy load scheduling group through a multi-level scheduling evaluation model. Finally, attention-based optimization is performed on the second and third energy load scheduling groups. Through scheduling evaluation, group expansion, and energy consumption minimization optimization, an energy load collaborative management strategy is determined. This achieves collaborative resource optimization scheduling of distributed energy and load, enabling the scheduling scheme to balance power supply, power quality, and low energy consumption, improving scheduling accuracy and reliability. It achieves the technical effects of rapid response to market changes, accurate matching of distributed energy and load demand, and intelligent and efficient collaborative energy load scheduling transactions.

[0085] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a resource optimization scheduling system for distributed energy and load coordination under machine learning, the system comprising:

[0086] Energy monitoring sequence acquisition module 1 is used to monitor the distributed energy system in real time, obtain the energy monitoring sequence, and simultaneously acquire the load monitoring sequence corresponding to the user side of the distributed energy system.

[0087] The energy load scheduling group acquisition module 2 performs transaction matching decisions based on trend prediction of the energy monitoring sequence and the load monitoring sequence to obtain a first energy load scheduling group, and performs energy load balance analysis optimization on the first energy load scheduling group based on a predetermined energy load balance degree to obtain a second energy load scheduling group.

[0088] The energy load collaborative management strategy determination module 3 is used to perform scheduling evaluation and optimization on the second energy load scheduling group according to the multi-level scheduling evaluation model to obtain the third energy load scheduling group, and to perform attention reproduction optimization based on the second energy load scheduling group and the third energy load scheduling group to determine the energy load collaborative management strategy.

[0089] Furthermore, the energy load dispatch group acquisition module 2 includes:

[0090] An energy feature prediction result acquisition unit is used to perform trend prediction on the energy monitoring sequence based on multiple machine learning models to obtain energy feature prediction results; a load feature prediction result acquisition unit is used to perform trend prediction on the load monitoring sequence based on the multiple machine learning models to obtain load feature prediction results; an energy load prediction map acquisition unit is used to organize the energy feature prediction results and the load feature prediction results to obtain an energy load prediction map; and a first energy load scheduling group generation unit is used to perform transaction matching decisions based on the energy load prediction map to generate the first energy load scheduling group.

[0091] Furthermore, the energy load dispatch group acquisition module 2 includes:

[0092] The system includes: an energy monitoring history set loading subunit, used to load the energy monitoring history set of the distributed energy system; an energy trend prediction model acquisition subunit, used to perform supervised training on the energy monitoring history set according to the multiple machine learning models to obtain multiple energy trend prediction models; an energy prediction loss data generation subunit, used to perform adversarial perturbation tests on the multiple energy trend prediction models according to the energy monitoring history set to generate multiple energy prediction loss data; an energy trend prediction node acquisition subunit, used to perform integrated fusion training on the multiple energy trend prediction models according to the multiple energy prediction loss data to obtain energy trend prediction nodes; and an energy feature prediction result generation subunit, used to input the energy monitoring sequence into the energy trend prediction node to generate the energy feature prediction result.

[0093] Furthermore, the energy load dispatch group acquisition module 2 includes:

[0094] A first energy load scheduling scheme acquisition unit is used to extract a first energy load scheduling scheme based on the first energy load scheduling group; a first energy load balance degree historical sequence acquisition unit is used to perform energy load balance degree historical retrieval based on the first energy load scheduling scheme to obtain a first energy load balance degree historical sequence; a first fitted energy load balance degree acquisition unit is used to perform lumped value calculation on the first energy load balance degree historical sequence to obtain a first fitted energy load balance degree; a second energy load scheduling group receiving unit is used to add the first energy load scheduling scheme to the second energy load scheduling group if the first fitted energy load balance degree is greater than or equal to the predetermined energy load balance degree; a second energy load scheduling group updating unit is used to continue to perform energy load balance degree analysis and optimization on the first energy load scheduling group based on the predetermined energy load balance degree, and update the second energy load scheduling group.

[0095] Furthermore, the energy load collaborative management strategy determination module 3 includes:

[0096] A multi-level dispatch evaluation model activation unit is used to activate the multi-level dispatch evaluation model, which includes multi-level dispatch evaluation indicators, including power supply quality and power supply security; a dispatch evaluation sequence acquisition unit is used to perform dispatch evaluation on each energy load dispatch scheme in the second energy load dispatch group according to the multi-level dispatch evaluation model to obtain multiple dispatch evaluation sequences; a dispatch evaluation constraint acquisition unit is used to configure constraints according to the multi-level dispatch evaluation indicators to obtain dispatch evaluation constraints; and a third energy load dispatch group generation unit is used to perform optimization screening on the second energy load dispatch group according to the dispatch evaluation constraints and the multiple dispatch evaluation sequences to generate the third energy load dispatch group.

[0097] Furthermore, the energy load collaborative management strategy determination module 3 includes:

[0098] The first scheduling propagation space acquisition unit is used to perform attention propagation on the second energy load scheduling group based on the third energy load scheduling group to obtain a first scheduling propagation space; the second scheduling propagation space acquisition unit is used to perform scheduling evaluation optimization on the first scheduling propagation space based on the multi-level scheduling evaluation model to obtain a second scheduling propagation space; the fourth energy load scheduling group acquisition unit is used to expand the third energy load scheduling group based on the second scheduling propagation space to obtain a fourth energy load scheduling group; the energy load collaborative management strategy generation unit is used to perform energy consumption minimization optimization on the fourth energy load scheduling group to generate the energy load collaborative management strategy.

[0099] Furthermore, the energy load collaborative management strategy determination module 3 includes:

[0100] The system includes a scheduling difference feature group acquisition subunit, used to identify differences in the second energy load scheduling group based on the third energy load scheduling group, and obtain a scheduling difference feature group; a reproductive attention feature group acquisition subunit, used to perform mutation adjustment on the scheduling difference feature group, and obtain a reproductive attention feature group; a scheduling adjustment space acquisition subunit, used to adjust the second energy load scheduling group based on the reproductive attention feature group, and obtain a scheduling adjustment space; and a first scheduling reproductive space generation subunit, used to perform energy load balance analysis and optimization on the scheduling adjustment space based on the predetermined energy load balance, and generate the first scheduling reproductive space.

[0101] In embodiment three, this application also provides a computer-readable storage medium that can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the resource optimization scheduling method for distributed energy and load coordination under machine learning in the embodiments of this application, thereby realizing the above-mentioned resource optimization scheduling method for distributed energy and load coordination under machine learning.

[0102] It should be understood that the embodiments disclosed in this application and the above description enable those skilled in the art to implement this application. However, this application is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this application.

Claims

1. A resource optimization scheduling method for distributed energy and load coordination based on machine learning, characterized in that, include: Real-time monitoring of the distributed energy system is performed to obtain the energy monitoring sequence, and the load monitoring sequence corresponding to the user side of the distributed energy system is obtained simultaneously. Based on the energy monitoring sequence and the load monitoring sequence, a transaction matching decision is made under trend prediction to obtain a first energy load scheduling group. Then, based on a predetermined energy load balance degree, the first energy load scheduling group is analyzed and optimized to obtain a second energy load scheduling group. The second energy load scheduling group is evaluated and optimized according to the multi-level scheduling evaluation model to obtain the third energy load scheduling group. Then, attention reproduction optimization is performed based on the second energy load scheduling group and the third energy load scheduling group to determine the energy load collaborative management strategy. Based on the energy monitoring sequence and the load monitoring sequence, a transaction matching decision is made under trend prediction to obtain a first energy load dispatch group, including: The energy monitoring sequence is trend-predicted using multiple machine learning models to obtain energy characteristic prediction results. The load monitoring sequence is trend-predicted based on the multiple machine learning models to obtain load characteristic prediction results. By combining the energy characteristic prediction results and the load characteristic prediction results, an energy load prediction map is obtained. Based on the energy load forecast map, a transaction matching decision is made to generate the first energy load scheduling group; The energy monitoring sequence is trend-predicted using multiple machine learning models to obtain energy characteristic prediction results, including: Load the energy monitoring history set of the distributed energy system; The energy monitoring historical set is trained under supervision using the multiple machine learning models to obtain multiple energy trend prediction models. Based on the energy monitoring history set, the multiple energy trend prediction models are subjected to anti-disturbance tests to generate multiple energy prediction loss data. The multiple energy trend prediction models are integrated and trained based on the multiple energy prediction loss data to obtain energy trend prediction nodes. The energy monitoring sequence is input into the energy trend prediction node to generate the energy characteristic prediction result; Based on the second and third energy load scheduling groups, attention-based optimization is performed to determine energy load collaborative management strategies, including: The second energy load scheduling group is subjected to attention propagation based on the third energy load scheduling group to obtain a first scheduling propagation space; The first scheduling breeding space is evaluated and optimized according to the multi-level scheduling evaluation model to obtain the second scheduling breeding space. The third energy load scheduling group is expanded according to the second scheduling breeding space to obtain a fourth energy load scheduling group; The fourth energy load scheduling group is optimized for minimizing energy consumption, and the energy load collaborative management strategy is generated. Based on the third energy load scheduling group, attention is multiplied on the second energy load scheduling group to obtain a first scheduling multiplication space, including: Based on the third energy load scheduling group, the second energy load scheduling group is identified to obtain a scheduling difference feature group; The scheduling difference feature group is mutated and adjusted to obtain the reproductive attention feature group; The second energy load scheduling group is adjusted according to the reproductive attention feature group to obtain the scheduling adjustment space; Based on the predetermined energy load balance, the scheduling adjustment space is analyzed and optimized to generate the first scheduling breeding space.

2. The resource optimization scheduling method for distributed energy and load coordination under machine learning as described in claim 1, characterized in that, Based on a predetermined energy load balance, the first energy load scheduling group is analyzed and optimized to obtain a second energy load scheduling group, including: The first energy load scheduling scheme is extracted based on the first energy load scheduling group; Based on the first energy load scheduling scheme, a historical energy load balance retrieval is performed to obtain the first energy load balance historical sequence; The first fitted energy load balance is obtained by performing lumped value calculation on the first energy load balance historical sequence. If the first fitted energy load balance degree is greater than or equal to the predetermined energy load balance degree, the first energy load scheduling scheme is added to the second energy load scheduling group; Based on the predetermined energy load balance, the first energy load scheduling group is further analyzed and optimized to update the second energy load scheduling group.

3. The resource optimization scheduling method for distributed energy and load coordination under machine learning as described in claim 1, characterized in that, The second energy load scheduling group is evaluated and optimized based on a multi-level scheduling evaluation model to obtain a third energy load scheduling group, including: Activate the multi-level scheduling evaluation model, which includes multi-level scheduling evaluation indicators, including power supply quality and power supply security. Based on the multi-level scheduling evaluation model, each energy load scheduling scheme in the second energy load scheduling group is evaluated to obtain multiple scheduling evaluation sequences. Constraints are configured based on the multi-level scheduling evaluation indicators to obtain scheduling evaluation constraints; The second energy load scheduling group is optimized and screened based on the scheduling evaluation constraints and the multiple scheduling evaluation sequences to generate the third energy load scheduling group.

4. A resource optimization scheduling system for distributed energy and load coordination based on machine learning, characterized in that, The system is used to implement the machine learning-based distributed energy and load coordination resource optimization scheduling method according to any one of claims 1-3, the system comprising: The energy monitoring sequence acquisition module is used to monitor the distributed energy system in real time, obtain the energy monitoring sequence, and simultaneously acquire the load monitoring sequence corresponding to the user side of the distributed energy system. The energy load scheduling group acquisition module makes transaction matching decisions based on trend prediction of the energy monitoring sequence and the load monitoring sequence to obtain a first energy load scheduling group, and performs energy load balance analysis and optimization on the first energy load scheduling group based on a predetermined energy load balance degree to obtain a second energy load scheduling group. The energy load collaborative management strategy determination module is used to perform scheduling evaluation and optimization on the second energy load scheduling group according to the multi-level scheduling evaluation model to obtain the third energy load scheduling group, and to perform attention reproduction optimization based on the second energy load scheduling group and the third energy load scheduling group to determine the energy load collaborative management strategy. The energy load dispatch group acquisition module includes: An energy feature prediction result acquisition unit is used to perform trend prediction on the energy monitoring sequence based on multiple machine learning models to obtain energy feature prediction results. The load characteristic prediction result acquisition unit is used to perform trend prediction on the load monitoring sequence based on the multiple machine learning models to obtain load characteristic prediction results. An energy load forecast map acquisition unit is used to organize the energy characteristic forecast results and the load characteristic forecast results to obtain an energy load forecast map. The first energy load scheduling group generation unit is used to make transaction matching decisions based on the energy load prediction map and generate the first energy load scheduling group. The energy characteristic prediction result acquisition unit includes: An energy monitoring history set loading subunit is used to load the energy monitoring history set of the distributed energy system. The energy trend prediction model acquisition sub-unit is used to perform supervised training on the energy monitoring history set according to the multiple machine learning models to obtain multiple energy trend prediction models. The energy prediction loss data generation subunit is used to perform anti-disturbance tests on the multiple energy trend prediction models based on the energy monitoring history set, and generate multiple energy prediction loss data. The energy trend prediction node acquisition subunit is used to integrate and train the multiple energy trend prediction models based on the multiple energy prediction loss data to obtain energy trend prediction nodes. An energy characteristic prediction result generation subunit is used to input the energy monitoring sequence into the energy trend prediction node and generate the energy characteristic prediction result. The energy load collaborative management strategy determination module includes: The first scheduling breeding space acquisition unit is used to perform attention breeding on the second energy load scheduling group according to the third energy load scheduling group to obtain the first scheduling breeding space. The second scheduling breeding space acquisition unit is used to perform scheduling evaluation and optimization on the first scheduling breeding space according to the multi-level scheduling evaluation model to obtain the second scheduling breeding space. The fourth energy load scheduling group acquisition unit is used to expand the third energy load scheduling group according to the second scheduling propagation space to obtain the fourth energy load scheduling group. An energy load collaborative management strategy generation unit is used to perform energy consumption minimization optimization on the fourth energy load scheduling group and generate the energy load collaborative management strategy. The first scheduling and breeding space acquisition unit includes: The scheduling difference feature group acquisition subunit is used to identify the differences between the second energy load scheduling group and the third energy load scheduling group to obtain the scheduling difference feature group. A reproductive attention feature group acquisition subunit is used to perform mutation adjustment on the scheduling difference feature group to obtain the reproductive attention feature group; The scheduling adjustment space acquisition subunit is used to adjust the second energy load scheduling group according to the reproductive attention feature group to obtain the scheduling adjustment space; The first scheduling breeding space generation subunit is used to perform energy load balance analysis and optimization on the scheduling adjustment space according to the predetermined energy load balance, and generate the first scheduling breeding space.

5. The resource optimization and scheduling system for distributed energy and load coordination under machine learning as described in claim 4, characterized in that, The energy load dispatch group acquisition module includes: The first energy load scheduling scheme acquisition unit is used to extract the first energy load scheduling scheme based on the first energy load scheduling group. The first energy load balance history sequence acquisition unit is used to perform energy load balance history retrieval according to the first energy load scheduling scheme to obtain the first energy load balance history sequence. The first fitted energy load balance degree acquisition unit is used to perform lumped value calculation on the historical sequence of the first energy load balance degree to obtain the first fitted energy load balance degree. The second energy load scheduling group receiving unit is used to add the first energy load scheduling scheme to the second energy load scheduling group if the first fitted energy load balance degree is greater than or equal to the predetermined energy load balance degree. The second energy load scheduling group update unit is used to continue to perform energy load balance analysis and optimization on the first energy load scheduling group according to the predetermined energy load balance, and update the second energy load scheduling group.

6. The resource optimization scheduling system for distributed energy and load coordination under machine learning as described in claim 4, characterized in that, The energy load collaborative management strategy determination module includes: A multi-level scheduling evaluation model activation unit is used to activate the multi-level scheduling evaluation model, which includes multi-level scheduling evaluation indicators, including power supply quality and power supply security. The scheduling evaluation sequence acquisition unit is used to perform scheduling evaluation on each energy load scheduling scheme in the second energy load scheduling group according to the multi-level scheduling evaluation model, and obtain multiple scheduling evaluation sequences. The scheduling evaluation constraint acquisition unit is used to configure constraints according to the multi-level scheduling evaluation indicators to obtain scheduling evaluation constraints. The third energy load scheduling group generation unit is used to optimize and screen the second energy load scheduling group according to the scheduling evaluation constraints and the multiple scheduling evaluation sequences, and generate the third energy load scheduling group.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the resource optimization scheduling method for distributed energy and load coordination under machine learning as described in any one of claims 1 to 3.

Images