A method and system for promoting load balancing of an industrial park to accommodate new energy

Abstract

The application relates to the technical field of power systems, and discloses an industrial park load balancing method and system for promoting new energy consumption, wherein the method comprises the following steps: inputting new energy power generation data and industrial park load data into a time sequence prediction model to obtain power consumption balance state prediction results of each subregion; calculating the load utilization rate of each subregion based on new energy power generation resource quantization data and industrial park load resource quantization data, and determining a high load subregion and a low load subregion in the subregion according to the load utilization rate; obtaining a to-be-migrated task list of the high load subregion and an idle resource list of the low load subregion, and performing optimization matching processing on to-be-migrated tasks in the to-be-migrated task list and idle resources in the idle resource list to obtain a resource allocation optimization result; and executing a cross-region task migration instruction generated from the resource allocation optimization result. The method improves the new energy consumption efficiency.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a load balancing method and system for industrial parks that promotes the consumption of new energy sources. Background Technology

[0002] As the global energy structure transitions towards a low-carbon model, the proportion of new energy sources such as wind and solar power in the power system continues to increase. However, the intermittent and highly volatile nature of new energy power generation poses a significant challenge to the real-time balance and efficient absorption of the power grid.

[0003] Existing load balancing methods rely on experience-based scheduling or static planning, which can easily lead to resource misallocation problems. For example, if a sudden snowfall causes a sharp drop in photovoltaic power output in city A, while the load demand of the local industrial park is strained due to a surge in computing tasks, and the neighboring city B has an excess of wind power output and a low load in its local industrial park, traditional methods rely on preset scheduling plans or manual decision-making. These methods cannot capture the power imbalance between the two regions in real time, nor can they quickly calculate the optimal resource allocation path. Ultimately, city A faces the risk of overload due to power shortages, while city B is forced to abandon power because it cannot transmit renewable energy to other regions. This severely restricts the efficiency of renewable energy absorption and the stability of grid operation. Summary of the Invention

[0004] This invention provides a load balancing method and system for industrial parks to promote the consumption of new energy sources, in order to solve the technical problem of how to improve existing load balancing methods, thereby improving the efficiency of new energy consumption and the stability of power grid operation.

[0005] To address the aforementioned technical problems, this invention provides, in one aspect, a method for load balancing in industrial parks to promote the absorption of new energy sources, comprising:

[0006] In response to load balance signals, acquire new energy power generation data and industrial park load data for each sub-region of the target area;

[0007] The new energy power generation data and the load data are input into a pre-trained time series prediction model for prediction, and the power balance state prediction results of each sub-region are obtained; wherein the power balance state prediction results include the quantitative data of new energy power generation resources and the quantitative data of load resources at multiple time points in each sub-region within a future time period;

[0008] The load utilization rate of each sub-region is calculated based on the quantitative data of new energy power generation resources and the quantitative data of load resources, and the high-load sub-regions and low-load sub-regions in the sub-regions are determined according to the load utilization rate.

[0009] Obtain the list of tasks to be migrated in the high-load sub-region and the list of idle resources in the low-load sub-region, and perform optimization matching on the tasks to be migrated in the list of tasks to be migrated and the idle resources in the list of idle resources based on the improved adaptive genetic algorithm to obtain the resource allocation optimization result;

[0010] Execute the cross-regional task migration instruction generated by the resource allocation optimization result, so that the tasks to be migrated in the high-load sub-region are migrated to the corresponding low-load sub-region.

[0011] As one preferred embodiment, the step of inputting the new energy power generation data and the load data into a pre-trained time-series prediction model for prediction, to obtain the predicted power balance state of each of the sub-regions, includes:

[0012] An initial time-series prediction model based on an LSTM neural network is constructed, and historical power generation data and historical load data of new energy sources in each sub-region of the target area are obtained.

[0013] The initial time-series prediction model is trained based on the historical power generation data of the new energy source and the historical load data to obtain a trained time-series prediction model.

[0014] Based on the time-series prediction model, the renewable energy generation and load demand in the sub-region are predicted hour by hour to obtain the quantitative sequence of renewable energy generation resources and the quantitative sequence of load resources in the future time period.

[0015] Based on the quantitative sequence of new energy power generation resources and the quantitative sequence of load resources, the dynamic change characteristics of the power generation and consumption balance status of each sub-region are extracted to obtain the power generation and consumption balance status prediction results that include resource fluctuation trends.

[0016] As one preferred embodiment, the step of calculating the load utilization rate of each sub-region based on the quantitative data of new energy power generation resources and the quantitative data of load resources, and determining the high-load sub-regions and low-load sub-regions within the sub-regions based on the load utilization rate, includes:

[0017] Based on the quantitative data of new energy power generation resources, the total number of processing units and storage units in each sub-region are determined, and based on the quantitative data of load resources, the total number of occupied processing units and storage units in each sub-region are determined.

[0018] The load utilization rate of each sub-region is calculated based on the total number of processing units, the total number of storage units, the total number of occupied processing units, and the total number of occupied storage units.

[0019] The average load utilization rate is calculated based on the load utilization rate of each sub-region. The load utilization rate The sub-regions are marked as high-load sub-regions, and the load utilization rate is... The sub-region is marked as a low-load sub-region, where, This is a floating threshold.

[0020] As one preferred embodiment, the improved adaptive genetic algorithm is used to perform optimized matching between the tasks to be migrated in the task-to-migrate list and the idle resources in the idle resource list to obtain resource allocation optimization results, including:

[0021] The list of tasks to be migrated and the list of free resources are sorted and encoded, and the pairing relationship between each task to be migrated and the target sub-region is mapped to a chromosome sequence based on the encoding results to generate an initial task allocation population.

[0022] The initial task allocation population is iteratively optimized to obtain an optimized task allocation population. The iterative optimization process is designed to perform linear expansion of the initial task allocation population during the exploration phase and linear contraction of the initial task allocation population during the convergence phase.

[0023] The fitness values ​​of chromosome sequences in the task allocation population are predicted based on a pre-trained fitness prediction model, and parent chromosomes are selected based on the fitness values.

[0024] After crossover and recombination are performed on the parent chromosomes to generate offspring, mutation operations are performed at a preset probability until the number of iterations reaches a preset threshold or the population size is smaller than the initial size. The resource allocation optimization results are then output, which include the optimal matching relationship between each task to be migrated and the target low-load sub-region.

[0025] As one preferred embodiment, executing the cross-regional task migration instruction generated by the resource allocation optimization result to migrate the tasks to be migrated from the high-load sub-region to the corresponding low-load sub-region includes:

[0026] Based on the resource optimization allocation results, extract the matching relationship between the task to be migrated and the target low-load sub-region, and obtain the task data volume of each migration task and the transmission rate between the corresponding target sub-regions;

[0027] The transmission delay of the task migration is calculated based on the task data volume and the transmission rate, and the arrival time of each migration task in the target sub-region is determined by combining the task migration start time.

[0028] Based on the task arrival time and the idle resource data of the target low-load sub-region, the sum of the load utilization variation coefficients at each time within the preset time period is evaluated to ensure that the load balancing constraints are met.

[0029] The task region matching relationship that meets the load balancing constraint is converted into a standardized cross-region task migration instruction, which is then sent to the task scheduling node of the high-load sub-region and the resource receiving node of the low-load sub-region to execute the task migration operation.

[0030] Another aspect of the present invention provides a load balancing system for industrial parks that promotes the consumption of new energy sources, comprising:

[0031] The acquisition module is used to acquire new energy power generation data and load data of each sub-region in the target area in response to the load balance signal;

[0032] The prediction module is used to input the new energy power generation data and the load data into a pre-trained time series prediction model for prediction, and obtain the power balance state prediction results of each sub-region; wherein the power balance state prediction results include the quantitative data of new energy power generation resources and the quantitative data of load resources at multiple time points in the future time period for each sub-region.

[0033] The calculation module is used to calculate the load utilization rate of each sub-region based on the quantitative data of new energy power generation resources and the quantitative data of load resources, and to determine the high-load sub-regions and low-load sub-regions in the sub-regions according to the load utilization rate;

[0034] The optimization module is used to obtain the list of tasks to be migrated in the high-load sub-region and the list of idle resources in the low-load sub-region, and to perform optimization matching processing on the tasks to be migrated in the list of tasks to be migrated and the idle resources in the list of idle resources based on an improved adaptive genetic algorithm to obtain the resource allocation optimization result.

[0035] The execution module is used to execute cross-regional task migration instructions generated by the resource allocation optimization results, so as to migrate the tasks to be migrated in the high-load sub-region to the corresponding low-load sub-region.

[0036] As one preferred embodiment, the prediction module is specifically used for:

[0037] An initial time-series prediction model based on an LSTM neural network is constructed, and historical power generation data and historical load data of new energy sources in each sub-region of the target area are obtained.

[0038] The initial time-series prediction model is trained based on the historical power generation data of the new energy source and the historical load data to obtain a trained time-series prediction model.

[0039] Based on the time-series prediction model, the renewable energy generation and load demand in the sub-region are predicted hour by hour to obtain the quantitative sequence of renewable energy generation resources and the quantitative sequence of load resources in the future time period.

[0040] Based on the quantitative sequence of new energy power generation resources and the quantitative sequence of load resources, the dynamic change characteristics of the power generation and consumption balance status of each sub-region are extracted to obtain the power generation and consumption balance status prediction results that include resource fluctuation trends.

[0041] As one preferred embodiment, the computing module is specifically used for:

[0042] Based on the quantitative data of new energy power generation resources, the total number of processing units and storage units in each sub-region are determined, and based on the quantitative data of load resources, the total number of occupied processing units and storage units in each sub-region are determined.

[0043] The load utilization rate of each sub-region is calculated based on the total number of processing units, the total number of storage units, the total number of occupied processing units, and the total number of occupied storage units.

[0044] The average load utilization rate is calculated based on the load utilization rate of each sub-region. The load utilization rate The sub-regions are marked as high-load sub-regions, and the load utilization rate is... The sub-region is marked as a low-load sub-region, where, This is a floating threshold.

[0045] As one preferred embodiment, the optimization module is specifically used for:

[0046] The list of tasks to be migrated and the list of free resources are sorted and encoded, and the pairing relationship between each task to be migrated and the target sub-region is mapped to a chromosome sequence based on the encoding results to generate an initial task allocation population.

[0047] The initial task allocation population is iteratively optimized to obtain an optimized task allocation population. The iterative optimization process is designed to perform linear expansion of the initial task allocation population during the exploration phase and linear contraction of the initial task allocation population during the convergence phase.

[0048] The fitness values ​​of chromosome sequences in the task allocation population are predicted based on a pre-trained fitness prediction model, and parent chromosomes are selected based on the fitness values.

[0049] After crossover and recombination are performed on the parent chromosomes to generate offspring, mutation operations are performed at a preset probability until the number of iterations reaches a preset threshold or the population size is smaller than the initial size. The resource allocation optimization results are then output, which include the optimal matching relationship between each task to be migrated and the target low-load sub-region.

[0050] As one preferred embodiment, the execution module is specifically used for:

[0051] Based on the resource optimization allocation results, extract the matching relationship between the task to be migrated and the target low-load sub-region, and obtain the task data volume of each migration task and the transmission rate between the corresponding target sub-regions;

[0052] The transmission delay of the task migration is calculated based on the task data volume and the transmission rate, and the arrival time of each migration task in the target sub-region is determined by combining the task migration start time.

[0053] Based on the task arrival time and the idle resource data of the target low-load sub-region, the sum of the load utilization variation coefficients at each time within the preset time period is evaluated to ensure that the load balancing constraints are met.

[0054] The task region matching relationship that meets the load balancing constraint is converted into a standardized cross-region task migration instruction, which is then sent to the task scheduling node of the high-load sub-region and the resource receiving node of the low-load sub-region to execute the task migration operation.

[0055] Compared with the prior art, the beneficial effects of the present invention are at least one of the following:

[0056] 1) This invention effectively enhances the dynamic coordination capability and resource utilization efficiency of green power consumption by integrating LSTM time-series prediction with an improved adaptive genetic algorithm. The LSTM model, through its gating mechanism and cell state design, accurately captures the long-term temporal dependency between industrial park load and clean energy generation, enabling quantitative prediction of power generation and consumption status at multiple future moments. This provides a forward-looking decision-making basis for resource allocation, fundamentally alleviating the problems of uneven resource allocation and idleness caused by fluctuations in new energy output and dynamic changes in tasks. The improved genetic algorithm reduces redundant information through permutation encoding, adjusts the balance exploration and convergence efficiency through adaptive population size adjustment, and accelerates the computation process through a fitness prediction neural network. This significantly reduces the computational overhead and decision lag of cross-regional task migration, enabling scheduling decisions to break through the limitations of local optima and quickly find globally optimal migration strategies to alleviate power pressure in high-load areas.

[0057] 2) This invention achieves dynamic balance and efficient coordination of power generation and consumption across multiple regions through quantitative load management and optimized target design. By setting load utilization thresholds to accurately distinguish between high and low load areas, and combining this with a target of minimizing the coefficient of variation, it ensures balanced load distribution across regions after task migration, avoids load state reversal, and significantly improves the proportion of clean energy consumption. Simultaneously, the algorithm's adaptive mechanism and elite retention strategy enhance the system's robustness and scalability, enabling it to adapt to large-scale computing nodes and heterogeneous task scenarios. This provides technical support for multi-level collaborative modes of cross-regional task migration instructions, contributes to the construction of green and low-carbon industrial parks, and promotes the transformation and upgrading of the power system towards an efficient and low-carbon dynamic collaborative mode. Attached Figure Description

[0058] Figure 1 This is a schematic flowchart of a method for promoting the consumption of new energy in industrial parks to achieve load balancing, according to one embodiment of the present invention.

[0059] Figure 2 This is a schematic diagram of an industrial park load balancing system for promoting the consumption of new energy in one embodiment of the present invention;

[0060] Figure label:

[0061] The module consists of: 11. Acquisition module; 12. Prediction module; 13. Calculation module; 14. Optimization module; and 15. Execution module. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0063] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0064] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0065] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0066] One embodiment of the present invention provides a load balancing method for industrial parks to promote the consumption of new energy sources. For details, please refer to [link to specific documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart illustrating a method for load balancing in industrial parks to promote the consumption of new energy sources, according to one embodiment of the present invention, which includes steps S1-S5:

[0067] S1: In response to the load balance signal, acquire new energy power generation data and industrial park load data for each sub-region in the target area;

[0068] In this embodiment, the load balancing signal typically originates from the system's real-time monitoring of regional power generation and consumption status. When it detects that some sub-regions have idle renewable energy power generation resources, excessive power load in industrial parks leading to power consumption imbalance, or when the overall system deviates from the source-grid-load-storage coordinated balance state, the system automatically triggers this signal and initiates the data acquisition process.

[0069] The target area refers to a geographical area that has the conditions for cross-sub-region load migration. It can be defined by administrative division (such as city or district) or power grid coverage. It must be pre-divided into at least two independent sub-regions. Each sub-region is equipped with independent new energy power generation facilities, industrial park clusters and data acquisition nodes. The sub-region boundary is based on the power grid partition boundary to ensure power grid compatibility for load migration.

[0070] New energy power generation data refers to quantitative data reflecting the amount of available clean energy resources in a sub-region, covering three core indicators. The definitions, collection methods, and uses of each indicator are as follows: First, real-time power generation, referring to the current power output of clean energy sources such as wind power and solar energy, measured in kilowatts (kW), collected every minute by smart meters (accuracy level ≥ 0.5) attached to the power generation facilities; Second, historical power generation curves, referring to the time-series data of daily power generation over the past 30 days (recorded at 15-minute granularity), reflecting the fluctuation pattern of new energy output, extracted from the historical database of the sub-region's edge nodes; Third, installed capacity, referring to the total installed capacity of wind turbines and solar photovoltaic panels in the sub-region, measured in megawatts (MW), entered into the system after confirmation by the new energy power generation project filing documents, serving as the benchmark for calculating the maximum power generation. All data is collected through power Internet of Things terminals to ensure data authenticity and traceability.

[0071] Industrial park load data refers to data that quantitatively characterizes the electricity load pressure of industrial parks, focusing on the demand and occupancy status of computing resources. Specifically, it includes two core indicators: First, the demand for computing resources, which refers to the theoretical demand for CPU and memory from the electricity consumption tasks (including information management systems, production control systems, industrial IoT terminal control tasks, etc.) in the sub-region of the industrial park. The unit of CPU demand is cores, and the unit of memory demand is gigabytes (GB), which is extracted from the task application parameters of the task scheduling system. Second, the amount of computing resources already occupied, which refers to the number of CPU cores and memory capacity currently used by the servers in the industrial park. This is collected once every minute through server operating system monitoring tools (such as the top command in Linux systems and performance monitors in Windows systems) to ensure that the collection frequency is consistent with the real-time power generation data, which is convenient for subsequent time series matching.

[0072] Data acquisition and transmission rely on a wide-area transmission architecture of "Power Internet of Things + 5G". The specific process is as follows: Sub-region edge nodes (deployed in industrial park substations or new energy power plants) act as data acquisition terminals. After collecting new energy power generation data and load data in real time, they are transmitted to the global node (system core control node) through a 5G private network (transmission rate ≥1Gbps, latency ≤20ms). During transmission, encryption protocols (such as SSL / TLS) are used to ensure data security, and data verification mechanisms (such as CRC32 check) are used to remove erroneous data to ensure data integrity. After receiving the data, the global node stores it in a structured database (such as PostgreSQL) in the format of "sub-region identifier-data type-collection timestamp". The data storage retention period is 90 days, providing historical data support for the subsequent S2 time series prediction model.

[0073] S2: Input the new energy power generation data and the load data into the pre-trained time series prediction model for prediction, and obtain the power balance prediction results of each sub-region; wherein the power balance prediction results include the quantitative data of new energy power generation resources and the quantitative data of load resources at multiple time points in the future time period for each sub-region.

[0074] Preferably, in one embodiment of the present invention, the step of inputting the new energy power generation data and the load data into a pre-trained time series prediction model for prediction to obtain the power balance prediction results for each of the sub-regions includes:

[0075] An initial time-series prediction model based on an LSTM neural network is constructed, and historical power generation data and historical load data of new energy sources in each sub-region of the target area are obtained.

[0076] The initial time-series prediction model is trained based on the historical power generation data of the new energy source and the historical load data to obtain a trained time-series prediction model.

[0077] Based on the time-series prediction model, the renewable energy generation and load demand in the sub-region are predicted hour by hour to obtain the quantitative sequence of renewable energy generation resources and the quantitative sequence of load resources in the future time period.

[0078] Based on the quantitative sequence of new energy power generation resources and the quantitative sequence of load resources, the dynamic change characteristics of the power generation and consumption balance status of each sub-region are extracted to obtain the power generation and consumption balance status prediction results that include resource fluctuation trends.

[0079] The time series prediction model in this embodiment refers to a time series prediction model built based on LSTM (Long Short-Term Memory Network). Its core advantage is that it solves the "gradient vanishing" problem of traditional recurrent neural networks (RNN) through a gating mechanism (meaning that when training long sequence data, the gradient of backpropagation gradually approaches 0 with the increase of time steps, resulting in the inability to learn long-term dependencies). The power generation and consumption balance prediction result is a structured result containing quantitative data and fluctuation trends for future time periods (specifically the next 24 hours, divided into 15-minute granularities, totaling 96 time points). Among them, the quantitative data of new energy power generation resources is in "kilowatts (kW)" and the quantitative data of load resources is in "CPU cores (core)" and "memory capacity (GB)". The two need to be correlated through energy efficiency coefficients (e.g., 1kW of new energy power generation can support 1.176 core CPUs running at full load, and the energy efficiency coefficient is calculated based on the server power efficiency of 0.85).

[0080] In this embodiment, the initial time-series prediction model based on the LSTM neural network uses a three-gate control mechanism (forget gate, input gate, output gate) and a cell state design to solve the gradient vanishing problem of traditional recurrent neural networks, effectively capturing the long-term dependency between renewable energy generation and load changes. The computational flow for each time step is as follows:

[0081] ForgetGate: Determines which information to discard from the cell state.

[0082]

[0083] in, It is the output vector of the forget gate. The Sigmoid function compresses values ​​to... interval, and These are the weight matrix and the bias terms. It is the hidden state from the previous moment. This is the current input.

[0084] Input Gate: Determines which new information to store in the cell state.

[0085]

[0086]

[0087] in, It controls the percentage of new information retained. This is the candidate cell state (activated using tanh).

[0088] Cell state update: The cell state is updated by combining the results of the forget gate and the input gate.

[0089]

[0090] in, This represents element-wise multiplication (Hadamard product).

[0091] Output Gate: Determines the output of the hidden state at the current moment.

[0092]

[0093]

[0094] LSTM precisely controls gradient flow through gating mechanisms: cell state The update uses an additive operation, which ensures that the gradient is propagated at a constant rate during backpropagation, mitigating gradient vanishing. The gradients of the sigmoid functions of each gate are between [0, 0.25], but through the parameter sharing mechanism, the network can learn long-term dependencies.

[0095] LSTM backpropagation calculates parameter gradients using the chain rule:

[0096]

[0097] Based on the LSTM prediction model, this study analyzes historical clean energy power generation data for each region and the historical load pressure of industrial parks in each region (the CPU and memory resource requirements of tasks generated by local users). The historical data is used as input to train the LSTM model to optimize its parameters. The LSTM model then predicts the regional power generation and consumption balance at various points in time within the future period and reports this prediction to global nodes as input to a genetic algorithm.

[0098] The model training data comes from the historical power generation data of new energy sources (such as historical power generation and power curves of wind power and solar power) and historical load data (such as CPU usage, memory requirements and historical load fluctuation records of industrial park tasks) of each sub-region of the target area. After data cleaning and feature engineering (such as extracting time period features and seasonal features), the data is input into the model. The weight matrix and bias terms are optimized through the backpropagation algorithm so that the model can fit the time series characteristics of power generation and consumption data, and finally obtain the trained time series prediction model.

[0099] Furthermore, the trained time-series prediction model is used to extrapolate the renewable energy generation and load demand of each sub-region on a time-by-time basis, outputting quantitative sequences of renewable energy generation resources and load resources for future time periods. The quantitative data of renewable energy generation resources is represented by the total CPU and memory usage of the sub-region, directly reflecting the available scale of clean energy resources; the quantitative data of load resources uses occupied CPU and memory as quantitative indicators, reflecting the load pressure of the industrial park's tasks. By extracting the dynamic change characteristics of these two types of sequences, the model can capture the trends of renewable energy output fluctuations and load dynamic changes (such as resource fluctuation patterns during peak and off-peak periods). The final generated power balance prediction results not only include quantitative values ​​at each time point but also incorporate resource fluctuation trend information, providing data support for global nodes to perceive the risk of power imbalance at multiple future time points, ensuring that subsequent task migration decisions are forward-looking rather than reactive.

[0100] S3: Calculate the load utilization rate of each sub-region based on the quantitative data of the new energy power generation resources and the quantitative data of the load resources, and determine the high-load sub-region and low-load sub-region in the sub-region according to the load utilization rate.

[0101] Preferably, in one embodiment of the present invention, the step of calculating the load utilization rate of each of the sub-regions based on the quantitative data of new energy power generation resources and the quantitative data of load resources, and determining the high-load sub-regions and low-load sub-regions in the sub-regions according to the load utilization rate, includes:

[0102] Based on the quantitative data of new energy power generation resources, the total number of processing units and storage units in each sub-region are determined, and based on the quantitative data of load resources, the total number of occupied processing units and storage units in each sub-region are determined.

[0103] The load utilization rate of each sub-region is calculated based on the total number of processing units, the total number of storage units, the total number of occupied processing units, and the total number of occupied storage units.

[0104] The average load utilization rate is calculated based on the load utilization rate of each sub-region. The load utilization rate The sub-regions are marked as high-load sub-regions, and the load utilization rate is... The sub-region is marked as a low-load sub-region.

[0105] In this embodiment, the total number of processing units corresponds to the total number of CPUs, a quantitative indicator of the clean energy power generation capacity of a sub-region, and the total number of storage units corresponds to the total number of memory units. Both together characterize the scale of clean energy resources available in the region. The total number of occupied processing units and the total number of occupied storage units correspond to the number of CPUs and memory units already occupied by tasks in the industrial park, respectively, directly reflecting the current load pressure of the region. Based on this, the sub-region... Load utilization rate Calculated using the weighted formula:

[0106]

[0107] in, and They are regions The total CPU and total memory are used to represent the clean energy power generation situation; and They are regions The number of CPUs and memory already in use are used to indicate the load status of the industrial park. and It is a weighting factor for CPU and memory.

[0108] The logic behind setting the average load utilization rate and the floating threshold is to ensure that the criteria for determining high and low loads are objective and quantifiable. Specifically, the average load utilization rate of all sub-regions is first calculated. For example, if the total number of sub-regions n=5, the average load utilization rate of each sub-region is calculated. The percentages are 68%, 75%, 52%, 48%, and 62%, respectively. =61%; Secondly, set a floating threshold. The value is determined based on the industrial park's load fluctuation tolerance (10% in this example, i.e.) =0.1), this value is verified through historical data (when When the threshold is 10%, it can effectively distinguish between sub-regions of "tasks that need to be migrated" and "tasks that can be received", and avoid frequent migration triggers.

[0109] Furthermore, the criteria for identifying high-load sub-regions, low-load sub-regions, and balanced sub-regions, as well as their subsequent processing, are clarified.

[0110] The average load utilization rate of all regions was obtained by statistical analysis. And set a floating threshold If the regional load utilization rate If the area is considered to be under high load, then the area load utilization rate is considered to be high. This indicates a low load; if the regional load utilization rate... If the region is in a balanced state, it will not participate in this round of mission migration.

[0111] S4: Obtain the list of tasks to be migrated in the high-load sub-region and the list of idle resources in the low-load sub-region, and perform optimization matching on the tasks to be migrated in the list of tasks to be migrated and the idle resources in the list of idle resources based on the improved adaptive genetic algorithm to obtain the resource allocation optimization result;

[0112] Among them, the list of tasks to be migrated is from the high-load sub-region ( The set of transferable tasks selected from the list must include a unique task identifier, the number of CPU cores required for the task, the required memory capacity (mneed), and the latest completion time of the task. The selection criteria are that the delayed execution of the task will not affect core business (such as non-real-time industrial control tasks and batch data calculation tasks). The data comes from the task scheduling system of the high-load sub-region; the list of idle resources is selected from the low-load sub-region. The set of acceptable resources counted in the statistics must include the unique identifier of the sub-region, the number of idle CPU cores, the capacity of idle memory, and the duration of resource availability. The data comes from the resource monitoring system of the low-load sub-region. The fields of the two types of lists must correspond one-to-one to lay the foundation for subsequent matching.

[0113] Preferably, in one embodiment of the present invention, the optimized matching process of the tasks to be migrated in the task-to-migrate list and the idle resources in the idle resource list based on the improved adaptive genetic algorithm to obtain the resource allocation optimization result includes:

[0114] The list of tasks to be migrated and the list of free resources are sorted and encoded, and the pairing relationship between each task to be migrated and the target sub-region is mapped to a chromosome sequence based on the encoding results to generate an initial task allocation population.

[0115] The initial task allocation population is iteratively optimized to obtain an optimized task allocation population. The iterative optimization process is designed to perform linear expansion of the initial task allocation population during the exploration phase and linear contraction of the initial task allocation population during the convergence phase.

[0116] The fitness values ​​of chromosome sequences in the task allocation population are predicted based on a pre-trained fitness prediction model, and parent chromosomes are selected based on the fitness values.

[0117] After crossover and recombination are performed on the parent chromosomes to generate offspring, mutation operations are performed at a preset probability until the number of iterations reaches a preset threshold or the population size is smaller than the initial size. The resource allocation optimization results are then output, which include the optimal matching relationship between each task to be migrated and the target low-load sub-region.

[0118] Among them, permutation encoding is a coding method that transforms the matching relationship between "tasks to be migrated - low-load sub-regions" into an algorithm-recognizable chromosome sequence. The specific mapping rule is as follows: Let the number of tasks to be migrated be m, and the number of low-load sub-regions be n. Each chromosome sequence is represented as chrom=[ , ,..., ],in ∈[1,n] (j=1,2,...,m), =k represents the j-th task to be migrated being matched to the k-th low-load sub-region (e.g., m=3, n=2, chromosome [1,2,1] represents task 1→region 1, task 2→region 2, task 3→region 1); the "initial task allocation population" is a set containing multiple chromosome sequences, the size of which is determined by an empirical function. Determined: When m×n≥100, =m×n×0.3 (covering more potential solutions); when m×n<100, =50 (fixed value, to avoid insufficient solution space coverage due to excessively small size). The initial population is constructed by randomly generating chromosome sequences (it is necessary to ensure that each task matches at least one region and that the region's free resources meet the task requirements).

[0119] In this embodiment, a permutation encoding method is used to map the matching relationship between the task to be migrated and the low-load sub-region to a chromosome sequence, where each chromosome specifically represents a set of task-region pairs. This encoding method eliminates redundant information, allowing the solution space search to focus more on effective matching relationships and improving algorithm efficiency. The initial task allocation population size is determined by an empirical function. The population size is determined, where m is the number of tasks to be migrated and n is the number of low-load sub-regions: when the number of tasks and regions is large, the population size is appropriately increased to cover more potential solutions; when the number is small, a fixed value is used. This ensures the rationality and computational feasibility of the initial population.

[0120] In this embodiment, the iterative optimization process balances the breadth and depth of the solution space search through a two-stage "exploration-convergence" adaptive population adjustment mechanism. The exploration stage corresponds to the first T1 iterations (or when the optimal solution is still being updated after N consecutive iterations), during which the population size linearly increases to twice the initial size. The goal is to broaden the potential optimal solution by expanding the population coverage, thus avoiding getting trapped in local optima; the convergence phase is triggered after N consecutive generations of no update of the optimal solution, and the population size is reduced linearly. The algorithm continues until the value is less than the initial size or the maximum number of iterations is reached. At this point, the algorithm concentrates its computing power to cultivate high-quality solution regions and accelerates convergence to a globally better solution.

[0121] The iterative optimization process is divided into an exploration phase and a convergence phase. The core is to balance the breadth of the solution space search and the convergence efficiency by dynamically adjusting the population size.

[0122] Exploration phase: Triggering condition is the number of iterations ≤ ( =20 generations, determined based on historical tests, covering most potential solutions) or the optimal solution for N consecutive generations (N=5 generations) is still updated, the operation is linear group expansion: ,in Given the current population size, ΔS=10 (expanding the population by 10 chromosomes each time to avoid computational overload due to a sudden increase in size), the aim is to expand the solution space coverage and avoid getting trapped in local optima;

[0123] Convergence Phase: The triggering condition is that the optimal solution has not been updated for N consecutive generations (N=5 generations), and the operation is linear shrinking of the group. until (When the population size is reduced to less than 50% of the initial size, focus on high-quality solutions.) At this point, the algorithm concentrates its computing power to optimize existing high-quality chromosomes, accelerating convergence to a globally better solution.

[0124] In this embodiment, the fitness prediction model is a pre-trained BP neural network (a multi-layer feedforward neural network that optimizes weights through backpropagation error). The training data consists of 100 sets of known matching relationships (chromosomes + corresponding precise fitness values). The input is a chromosome sequence (converted into a numerical vector), and the output is the predicted fitness value. During iteration, this model is used to predict fitness first. If the prediction error (the difference between the prediction error and a small number of precise calculated values) is >5%, the precise value is recalculated and the training data is updated to ensure prediction accuracy. The fitness function is designed based on the optimization objective, with the core being "the reciprocal of the sum of the coefficients of variation of load utilization rates of each low-load sub-region within a future time period." A higher fitness value indicates that the task allocation scheme can achieve better load balancing.

[0125] Furthermore, a roulette wheel selection mechanism is employed to select parent chromosomes based on fitness values, while retaining the two elite individuals with the highest fitness in each generation to directly enter the next generation, ensuring that optimal solutions are not lost. Crossover operations generate offspring by recombining parent task-region pairs, expanding the diversity of solutions. Mutation operations randomly adjust the migration target regions of tasks in the offspring with a preset probability d, helping the algorithm escape local optima traps. When the maximum number of iterations is reached or the population size is less than the initial size, the algorithm terminates and outputs the resource allocation optimization results, clarifying the optimal matching relationship between each task to be migrated and the target low-load sub-region, providing precise instructions for cross-region task migration.

[0126] The improved adaptive genetic algorithm LSTM-LGA is described below:

[0127] Input: List of high-load area migration tasks Report a list of predicted idle resources for low-load areas. Maximum number of iterations , mutation probability d.

[0128] Output: Optimal task migration strategy .

[0129] 1: Initialization:

[0130] 2: Calculate the initial population size .

[0131] 3: Randomly generate the initial population: Each chromosome This represents a migration strategy.

[0132] 4. Iterative optimization:

[0133] 5: for

[0134] 6: if (Exploration phase), meaning the optimal solution is updated within N generations:

[0135] 7: Linearly increase population size: .

[0136] 8: else (convergence phase) then:

[0137] 9: Linearly reduce population size: .

[0138] 10: endif

[0139] 11: for each chromosome :

[0140] 12: If neural network prediction mode is enabled, then:

[0141] 13: Input The pre-trained BP network outputs the predicted fitness. .

[0142] 14: else:

[0143] 15: Calculate the precise fitness value .

[0144] 16: endif

[0145] 17: endfor

[0146] 18: Use roulette wheel to select parent chromosomes based on fitness values ​​and retain the top-2 elite individuals.

[0147] 19: Perform crossover and recombination on the selected parent generation to generate the offspring population.

[0148] 20: The migration target region in the offspring chromosome is randomly mutated with probability d.

[0149] 21: Merge parent generation and child generation.

[0150] 22: endfor

[0151] 26: Return the optimal task migration strategy .

[0152] S5: Execute the cross-regional task migration instruction generated by the resource allocation optimization result, so that the tasks to be migrated in the high-load sub-region are migrated to the corresponding low-load sub-region.

[0153] Preferably, in one embodiment of the present invention, executing the cross-regional task migration instruction generated by the resource allocation optimization result to migrate the tasks to be migrated in the high-load sub-region to the corresponding low-load sub-region includes:

[0154] Based on the resource optimization allocation results, extract the matching relationship between the task to be migrated and the target low-load sub-region, and obtain the task data volume of each migration task and the transmission rate between the corresponding target sub-regions;

[0155] The transmission delay of the task migration is calculated based on the task data volume and the transmission rate, and the arrival time of each migration task in the target sub-region is determined by combining the task migration start time.

[0156] Based on the task arrival time and the idle resource data of the target low-load sub-region, the sum of the load utilization variation coefficients at each time within the preset time period is evaluated to ensure that the load balancing constraints are met.

[0157] The task region matching relationship that meets the load balancing constraint is converted into a standardized cross-region task migration instruction, which is then sent to the task scheduling node of the high-load sub-region and the resource receiving node of the low-load sub-region to execute the task migration operation.

[0158] The input for this step is the resource allocation optimization result output by S4. Two key types of information need to be extracted from it: first, the task-region matching relationship; and second, the quantitative parameters of tasks and resources. The task data volume refers to the complete data volume of the task to be migrated, which is read from the task storage system of the high-load sub-region. The inter-sub-region transmission rate refers to the data transmission rate between the high-load sub-region i and the low-load sub-region j, which is determined by calculating the channel capacity using the Shannon formula. These two types of parameters directly determine the subsequent migration delay and execution timing.

[0159] In this embodiment, the matching relationship between the task to be migrated and the target low-load sub-region is extracted based on the resource allocation optimization results, i.e., the optimal task-region pair. For each task to be migrated... The amount of task data needs to be obtained. (Characterizing the data transfer scale of the task migration), and the transmission rate between the high-load sub-region and the target low-load sub-region. The channel capacity can be determined using Shannon's formula. Based on the quantification method for transmission delay, the transmission delay for task migration is:

[0160]

[0161] This delay directly affects the time it takes for the task to reach the target area and is a key parameter for subsequent load assessment. It is also considered in conjunction with the task migration start time. It can determine the time when the task arrives at the target sub-region. This ensures that the migration time point falls within the future time period predicted by LSTM. Within this framework, the time dimension of the forecast data should be kept consistent to avoid decision-making delays caused by time misalignment.

[0162] Furthermore, based on the task arrival time and idle resource data of the target low-load sub-region, the load distribution balance after migration is evaluated. Specifically, the load utilization rate of the target low-load sub-region at each time point within a preset time period is calculated, expressed as:

[0163]

[0164]

[0165] in, and It is a predicted low-load area At any moment The amount of CPU and memory already used at that time. and It is a low-load area In the time interval All received migrations The total number of CPUs and memory required.

[0166] The expression will... The predicted CPU and memory usage at each time step is compared with the data received at all time steps prior to step t. The actual CPU and memory usage at time t is obtained by adding the required CPU and memory usage.

[0167] Simultaneously, it is necessary to verify whether the core constraints are met: first, the task requirements do not exceed the idle resources of the low-load sub-region; second, the utilization rate of the low-load sub-region remains lower than that of the high-load region after migration, avoiding load state reversal; and third, the start and arrival times of migration are both within the prediction interval. Finally, by evaluating the sum of the coefficients of variation of load utilization at each moment within the preset time period, it is ensured that the load distribution of each region is more balanced after migration.

[0168] The optimization objective is:

[0169]

[0170]

[0171] Where constraints C1 and C2 represent the migration to All tasks need to require less CPU and memory. Idle resource quantity. Constraint C3 represents the load utilization rate of the previously low-load area after the migration is completed. It is still less than the original high-load area load utilization rate. To avoid load state reversal, constraint C4 indicates that the start and completion times of task migration must fall within the time range of resource forecasting.

[0172] The objective of this optimization is to minimize... The sum of the coefficients of variation of load utilization rates in all low-load areas at each point in time during the time period makes The load distribution across different regions of the system remains relatively balanced throughout the time period.

[0173] Furthermore, the task-region matching relationship verified by constraints is converted into standardized cross-region task migration instructions. Relying on a multi-level collaborative architecture of "cloud-edge-device" integrating the power IoT and 5G, global nodes precisely distribute instructions to the task scheduling nodes in high-load sub-regions (responsible for initiating the task migration process) and the resource receiving nodes in low-load sub-regions (responsible for preparing resource receiving tasks). The instructions clearly specify key information such as task identifier, target region identifier, migration start time, and resource reservation requirements, ensuring coordinated execution between the two nodes. That is, high-load regions migrate tasks according to the instructions, while low-load regions reserve corresponding CPU and memory resources in advance to receive tasks. Ultimately, this achieves the migration of tasks from high-load regions to low-load regions rich in clean energy, alleviating electricity pressure in high-load regions and improving the utilization rate of clean energy.

[0174] Another embodiment of the present invention provides a load balancing system for industrial parks that promotes the consumption of new energy sources. For details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram shown illustrates a load balancing system for industrial parks that promotes the consumption of new energy sources, as described in one embodiment of the present invention. The system includes:

[0175] The acquisition module 11 is used to acquire new energy power generation data and load data of each sub-region in the target area in response to the load balance signal;

[0176] Prediction module 12 is used to input the new energy power generation data and the load data into a pre-trained time series prediction model for prediction, and obtain the power balance state prediction results of each sub-region; wherein the power balance state prediction results include the quantitative data of new energy power generation resources and the quantitative data of load resources at multiple time points in the future time period for each sub-region.

[0177] Calculation module 13 is used to calculate the load utilization rate of each sub-region based on the quantitative data of new energy power generation resources and the quantitative data of load resources, and to determine the high-load sub-region and low-load sub-region in the sub-region according to the load utilization rate;

[0178] The optimization module 14 is used to obtain the list of tasks to be migrated in the high-load sub-region and the list of idle resources in the low-load sub-region, and to perform optimization matching processing on the tasks to be migrated in the list of tasks to be migrated and the idle resources in the list of idle resources based on the improved adaptive genetic algorithm to obtain the resource allocation optimization result.

[0179] Execution module 15 is used to execute cross-regional task migration instructions generated by the resource allocation optimization results, so as to migrate the tasks to be migrated in the high-load sub-region to the corresponding low-load sub-region.

[0180] Preferably, in one embodiment of the present invention, the prediction module is specifically used for:

[0181] An initial time-series prediction model based on an LSTM neural network is constructed, and historical power generation data and historical load data of new energy sources in each sub-region of the target area are obtained.

[0182] The initial time-series prediction model is trained based on the historical power generation data of the new energy source and the historical load data to obtain a trained time-series prediction model.

[0183] Based on the time-series prediction model, the renewable energy generation and load demand in the sub-region are predicted hour by hour to obtain the quantitative sequence of renewable energy generation resources and the quantitative sequence of load resources in the future time period.

[0184] Based on the quantitative sequence of new energy power generation resources and the quantitative sequence of load resources, the dynamic change characteristics of the power generation and consumption balance status of each sub-region are extracted to obtain the power generation and consumption balance status prediction results that include resource fluctuation trends.

[0185] Preferably, in one embodiment of the present invention, the computing module is specifically used for:

[0186] Based on the quantitative data of new energy power generation resources, the total number of processing units and storage units in each sub-region are determined, and based on the quantitative data of load resources, the total number of occupied processing units and storage units in each sub-region are determined.

[0187] The load utilization rate of each sub-region is calculated based on the total number of processing units, the total number of storage units, the total number of occupied processing units, and the total number of occupied storage units.

[0188] The average load utilization rate is calculated based on the load utilization rate of each sub-region. The load utilization rate The sub-regions are marked as high-load sub-regions, and the load utilization rate is... The sub-region is marked as a low-load sub-region.

[0189] Preferably, in one embodiment of the present invention, the optimization module is specifically used for:

[0190] The list of tasks to be migrated and the list of free resources are sorted and encoded, and the pairing relationship between each task to be migrated and the target sub-region is mapped to a chromosome sequence based on the encoding results to generate an initial task allocation population.

[0191] The initial task allocation population is iteratively optimized to obtain an optimized task allocation population. The iterative optimization process is designed to perform linear expansion of the initial task allocation population during the exploration phase and linear contraction of the initial task allocation population during the convergence phase.

[0192] The fitness values ​​of chromosome sequences in the task allocation population are predicted based on a pre-trained fitness prediction model, and parent chromosomes are selected based on the fitness values.

[0193] After crossover and recombination are performed on the parent chromosomes to generate offspring, mutation operations are performed at a preset probability until the number of iterations reaches a preset threshold or the population size is smaller than the initial size. The resource allocation optimization results are then output, which include the optimal matching relationship between each task to be migrated and the target low-load sub-region.

[0194] Preferably, in one embodiment of the present invention, the execution module is specifically used for:

[0195] Based on the resource optimization allocation results, extract the matching relationship between the task to be migrated and the target low-load sub-region, and obtain the task data volume of each migration task and the transmission rate between the corresponding target sub-regions;

[0196] The transmission delay of the task migration is calculated based on the task data volume and the transmission rate, and the arrival time of each migration task in the target sub-region is determined by combining the task migration start time.

[0197] Based on the task arrival time and the idle resource data of the target low-load sub-region, the sum of the load utilization variation coefficients at each time within the preset time period is evaluated to ensure that the load balancing constraints are met.

[0198] The task region matching relationship that meets the load balancing constraint is converted into a standardized cross-region task migration instruction, which is then sent to the task scheduling node of the high-load sub-region and the resource receiving node of the low-load sub-region to execute the task migration operation.

[0199] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. An industrial park load balancing method for promoting new energy consumption, characterized in that, The method comprises: in response to a load balancing signal, obtaining new energy power generation data and industrial park load data of each sub-region in a target region; wherein the new energy power generation data refers to quantitative data reflecting the available resource amount of clean energy in the sub-region, including real-time power generation, historical power generation curve and installed capacity; the industrial park load data refers to data quantitatively representing the power load pressure of the industrial park, including the calculated resource demand and the occupied resource amount, wherein the calculated resource demand refers to the theoretical demand value of CPU and memory of the industrial park power task in the sub-region, and the occupied resource amount refers to the number of CPU cores and the memory capacity used by the industrial park server at the current time; inputting the new energy power generation data and the load data into a pre-trained time series prediction model for prediction to obtain a power generation and consumption balance state prediction result of each sub-region; wherein the power generation and consumption balance state prediction result includes new energy power generation resource quantitative data and load resource quantitative data of each sub-region at multiple time points in a future time period; wherein the new energy power generation resource quantitative data is characterized by the total CPU and memory of the sub-region, directly reflecting the available resource scale of clean energy, and the load resource quantitative data is quantified by the occupied CPU and memory, reflecting the load pressure of the industrial park task; calculating the load utilization rate of each sub-region based on the new energy power generation resource quantitative data and the load resource quantitative data, and determining the high-load sub-region and the low-load sub-region in the sub-region according to the load utilization rate; obtaining a to-be-migrated task list of the high-load sub-region and an idle resource list of the low-load sub-region, and performing optimized matching processing on the to-be-migrated tasks in the to-be-migrated task list and the idle resources in the idle resource list based on an improved adaptive genetic algorithm to obtain a resource allocation optimization result; wherein the to-be-migrated task list is a set of migratable tasks selected from the high-load sub-region, including task unique identifier, required CPU core number, required memory capacity, and latest task completion time; the idle resource list is a set of receivable resources counted from the low-load sub-region, including sub-region unique identifier, idle CPU core number, idle memory capacity, and resource available duration; executing the cross-region task migration instruction generated by the resource allocation optimization result to migrate the to-be-migrated tasks of the high-load sub-region to the corresponding low-load sub-region; wherein the optimized matching processing of the to-be-migrated tasks in the to-be-migrated task list and the idle resources in the idle resource list based on the improved adaptive genetic algorithm to obtain the resource allocation optimization result comprises: performing permutation and coding on the to-be-migrated task list and the idle resource list, and mapping the pairing relationship between each to-be-migrated task and the target sub-region to a chromosome sequence based on the coding result to generate an initial task allocation population; The initial task allocation population is iteratively optimized to obtain an optimized task allocation population, wherein the iterative optimization process is designed to perform linear expansion population processing on the initial task allocation population in an exploration stage, and perform linear contraction population processing on the initial task allocation population in a convergence stage; The fitness values of the chromosome sequences in the task allocation population are predicted based on the pre-trained fitness prediction model, and parent chromosomes are screened based on the fitness values; According to the parent chromosomes, a mutation operation is performed on the offspring after cross-recombination with a preset probability until the number of iterations reaches a preset threshold or the population size is less than the initial size, and a resource allocation optimization result is output, which includes the optimal matching relationship between each task to be migrated and the target low-load sub-region.

2. The method of claim 1, wherein the method further comprises: The new energy power generation data and the load data are input into a pre-trained time series prediction model for prediction to obtain a power and load balance state prediction result of each sub-region, including: An initial time series prediction model based on an LSTM neural network is constructed, and historical new energy power generation data and historical load data of each sub-region in a target area are obtained; The initial time series prediction model is trained based on the historical new energy power generation data and the historical load data to obtain a trained time series prediction model; The new energy power generation and load demand in the sub-region are predicted by time according to the time series prediction model to obtain new energy power generation resource quantization sequences and load resource quantization sequences in a future time period; The dynamic change characteristics of the power and load balance state of each sub-region are extracted based on the new energy power generation resource quantization sequences and the load resource quantization sequences to obtain a power and load balance state prediction result containing resource fluctuation trends.

3. The method of claim 1, wherein the method further comprises: The load utilization rate of each sub-region is calculated based on the new energy power generation resource quantization data and the load resource quantization data, and the high-load sub-region and the low-load sub-region in the sub-region are determined according to the load utilization rate, including: The total amount of processing units and the total amount of storage units of each sub-region are determined based on the new energy power generation resource quantization data, and the total amount of occupied processing units and the total amount of occupied storage units of each sub-region are determined based on the load resource quantization data; The load utilization rate of each sub-region is calculated according to the total amount of processing units, the total amount of storage units, the total amount of occupied processing units, and the total amount of occupied storage units; calculating an average load utilization based on the load utilizations of the sub-areas marking sub-areas with a load utilization below the average load utilization as low load sub-areas, wherein is a floating threshold value. is a floating threshold value.

4. The method of claim 1, wherein the method further comprises: The cross-region task migration instruction generated from the resource allocation optimization result is executed to migrate the tasks to be migrated in the high-load sub-region to the corresponding low-load sub-region, including: The matching relationship between the tasks to be migrated and the target low-load sub-region is extracted based on the resource optimization allocation result, the task data amount of each migration task and the transmission rate between the corresponding target sub-regions are obtained; The transmission delay of task migration is calculated according to the task data amount and the transmission rate, and the time when each migration task arrives at the target sub-region is determined in combination with the task migration start time. Based on the task arrival time and the idle resource data of the target low-load sub-region, the sum of the load utilization rate variation coefficients of each time point in a preset time period is evaluated to ensure that the load balancing constraint is met. The task region matching relationship that meets the load balancing constraint condition is converted into a standardized cross-region task migration instruction, which is sent to the task scheduling node of the high-load sub-region and the resource receiving node of the low-load sub-region to perform a task migration operation.

5. An industrial park load balancing system for promoting new energy consumption, characterized in that, Comprise: The acquisition module is configured to acquire new energy power generation data and load data of each sub-region in the target region in response to a load balancing signal; wherein the new energy power generation data refers to quantitative data reflecting the available resource amount of clean energy in the sub-region, including real-time power generation, historical power generation curve and installed capacity; the load data of the industrial park refers to data quantitatively representing the power load pressure of the industrial park, including the calculation resource demand and the occupied calculation resource amount; wherein the calculation resource demand refers to the theoretical demand value of CPU and memory of the industrial park power task in the sub-region; the occupied calculation resource amount refers to the number of CPU cores and the memory capacity used by the industrial park server at the current time; The prediction module is configured to input the new energy power generation data and the load data into a pre-trained time series prediction model for prediction to obtain a power balance state prediction result of each sub-region; wherein the power balance state prediction result includes new energy power generation resource quantitative data and load resource quantitative data of each sub-region at multiple time points in a future time period; wherein the new energy power generation resource quantitative data is characterized by the total CPU and memory of the sub-region, directly reflecting the available resource scale of clean energy; the load resource quantitative data takes the occupied CPU and occupied memory as the quantitative indicators, reflecting the load pressure of the industrial park task; The calculation module is configured to calculate the load utilization rate of each sub-region based on the new energy power generation resource quantitative data and the load resource quantitative data, and determine the high-load sub-region and the low-load sub-region in the sub-region according to the load utilization rate; The optimization module is configured to obtain a to-be-migrated task list of the high-load sub-region and a free resource list of the low-load sub-region, and perform optimization matching processing on the to-be-migrated tasks in the to-be-migrated task list and the free resources in the free resource list based on an improved adaptive genetic algorithm to obtain a resource allocation optimization result; wherein the to-be-migrated task list is a set of migratable tasks selected from the high-load sub-region, including task unique identifier, required CPU core number, required memory capacity, and latest task completion time; the free resource list is a set of receivable resources counted from the low-load sub-region, including sub-region unique identifier, free CPU core number, free memory capacity, and resource available duration; The optimization module is configured to: arrange and encode the to-be-migrated task list and the free resource list, and map the pairing relationship between each to-be-migrated task and the target sub-region to a chromosome sequence based on the encoding result to generate an initial task allocation population; iteratively optimizing the initial task allocation population to obtain an optimized task allocation population, wherein the iterative optimization process is designed to perform linear expansion population processing on the initial task allocation population in an exploration stage, and perform linear contraction population processing on the initial task allocation population in a convergence stage; predicting, based on the pre-trained fitness prediction model, fitness values of chromosome sequences in the task allocation population, and screening parent chromosomes based on the fitness values; performing cross-recombination based on the parent chromosomes to generate offspring, and performing mutation operation on the offspring with a preset probability until the number of iterations reaches a preset threshold or the population size is smaller than the initial size, and outputting a resource allocation optimization result, the resource allocation optimization result including optimal matching relationships between each task to be migrated and a target low-load sub-region; an execution module configured to execute cross-region task migration instructions generated based on the resource allocation optimization result, so that the tasks to be migrated in the high-load sub-region are migrated to the corresponding low-load sub-region. 6.The system of claim 5, wherein, The prediction module is specifically configured to: construct an initial time series prediction model based on an LSTM neural network, and obtain historical new energy generation data and historical load data of each sub-region in a target region; train the initial time series prediction model based on the historical new energy generation data and the historical load data to obtain a trained time series prediction model; predict the new energy generation and load demand in each time period in the sub-region based on the time series prediction model to obtain new energy generation resource quantization sequences and load resource quantization sequences in a future time period; extract dynamic change characteristics of power generation and consumption balance states of each sub-region based on the new energy generation resource quantization sequences and the load resource quantization sequences to obtain power generation and consumption balance state prediction results containing resource fluctuation trends.

7. The system of claim 5, wherein the system further comprises a power grid, wherein the power grid is connected to the power plant, the power storage device, and the power consumption device. The calculation module is specifically configured to: determine the total amount of processing units and the total amount of storage units of each sub-region based on the new energy generation resource quantization data, and determine the total amount of occupied processing units and the total amount of occupied storage units of each sub-region based on the load resource quantization data; calculate the load utilization rate of each sub-region based on the total amount of processing units, the total amount of storage units, the total amount of occupied processing units, and the total amount of occupied storage units; The average load utilization rate is calculated based on the load utilization rate of each sub-region. The load utilization rate The sub-regions are marked as high-load sub-regions, and the load utilization rate is... The sub-regions are marked as low-load sub-regions, where, This is a floating threshold. 8.The system of claim 5, wherein, The execution module is specifically configured to: extract matching relationships between tasks to be migrated and target low-load sub-regions based on the resource optimization allocation result, obtain task data amounts of each migration task and transmission rates between corresponding target sub-regions; calculate transmission delays of task migration based on the task data amounts and the transmission rates, and determine time points at which each migration task arrives at the target sub-region in combination with a task migration start time; evaluate total sums of load utilization rate variation coefficients at each time point in a preset time period based on the task arrival time points and idle resource data of the target low-load sub-regions, to ensure that a load balancing constraint is met; convert the task region matching relationship that meets the load balancing constraint condition into standardized cross-region task migration instructions, and deliver the standardized cross-region task migration instructions to task scheduling nodes of the high-load sub-region and resource receiving nodes of the low-load sub-region to perform task migration operations.

Images