Regulation and control method and system for power grid demand side resources

Abstract

The invention provides a regulation and control method and system for power grid demand side resources, and the method comprises the steps: collecting the production data of a power grid, and carrying out the preprocessing of the collected data; determining a corresponding calculation amount according to demand information of a demand side, processing the preprocessed production data through a pre-trained evaluation model, predicting the corresponding calculation amount and a distribution scheme of the calculation amount, and obtaining a corresponding regulation and control strategy; wherein the evaluation model determines a problem type needing to be optimized according to a calculation amount, and calculates input production data through a preset specific algorithm to determine a resource allocation scheme corresponding to the problem type needing to be optimized; and adjusting the power grid demand side resources according to the regulation and control strategy to realize regulation and control of the power grid demand side resources. According to actual demands and resource characteristics, the artificial intelligence algorithm is selected and optimized, high matching of the algorithm and the problem is ensured, and the prediction precision and decision-making efficiency of the model are further improved through algorithm optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automation, and particularly to a method and system for regulating and controlling grid demand-side resources. Background Art

[0002] Optimal regulation and control of demand-side resources refers to regulating and improving resource demand through various measures and strategies to achieve effective allocation of resources, improve energy efficiency, and reduce waste. In the energy industry, this involves encouraging consumers to use energy more rationally, reducing the overall demand for energy through technological innovation and policy guidance, while maintaining or enhancing the quality of life and production.

[0003] The optimal regulation and control of demand-side resources is mainly applied in the following aspects: Flexible regulation of the power system: During the construction of a new power system, demand-side resources such as distributed generation resources, load resources, and energy storage resources are integrated and aggregated through technologies such as demand response and virtual power plants, and resource clusters participate in multiple scenarios such as grid peak shaving, new energy power consumption, and market trading from three dimensions: energy, space, and time. Energy conservation, emission reduction, and environmental protection: The optimal regulation and control of demand-side resources encourages the adoption of efficient energy-using technologies and products, reduces the energy consumption of enterprises and residents, helps reduce greenhouse gas emissions, and combats climate change. At the same time, by providing energy-saving education and training, it enhances users' energy-saving awareness and promotes changes in energy-saving behaviors. Participation in the power market: In the power market environment, demand-side resources have participated in the ancillary service market, new energy trading, medium- and long-term power markets, and real-time spot markets. Through demand response, it participates in power balance, optimizes the operating cost of the power grid, and improves system stability. Dynamic balance between supply and demand: The power system is transforming from "generation follows load" to "generation and load interact", and the flexibility of demand-side resources has become the key to ensuring the safe, reliable, and economic operation of the power system. Through optimal regulation, the dynamic balance between power supply and demand is achieved, and system flexibility is improved.

[0004] In the prior art, there are some obvious defects in the method for optimizing the regulation and control of demand-side resources: The limitations of technology and equipment may lead to unsatisfactory regulation effects. Although there are many advanced technical means for optimizing the regulation and control of demand-side resources, the update speed of equipment and technology may not be able to keep up with the changes in energy demand, resulting in limited regulation effects in some cases. The incompleteness of data and information may also affect the accuracy of regulation. The optimal regulation and control of demand-side resources requires a large amount of data and information support, but if the data and information sources are inaccurate or incomplete, it may lead to deviations in regulation measures and affect the regulation effect. Summary of the Invention

[0005] The purpose of the present invention is to propose a method and system for regulating and controlling grid demand-side resources to solve the technical problem of how to improve the accuracy and efficiency of demand-side regulation.

[0006] On the one hand, a method for regulating resources on the demand side of the power grid is provided, including:

[0007] Collect production data of the power grid and preprocess the collected data; the production data at least includes production equipment status data, production task information data, and employee work condition data;

[0008] Determine the corresponding computing amount according to the demand information on the demand side, and process the preprocessed production data through a pre-trained evaluation model to predict the corresponding computing amount and the allocation scheme of the computing amount, and obtain the corresponding regulation strategy; wherein, the evaluation model determines the type of problem to be optimized according to the computing amount, and calculates the input production data through a preset specific algorithm to determine the resource allocation scheme corresponding to the type of problem to be optimized;

[0009] Adjust the resources on the demand side of the power grid according to the regulation strategy to achieve the regulation of the resources on the demand side of the power grid.

[0010] Preferably, it further includes collecting and displaying the corresponding monitoring data in real time according to preset monitoring indicators;

[0011] Evaluate the performance of the system according to the monitoring data and demand information, and adjust the resource scheduling strategy and load balancing strategy in the regulation strategy in real time according to the evaluation results.

[0012] Preferably, the preprocessing of the collected data includes cleaning, format conversion, and standardization of the production data according to preset standards; and integrating the production data from different data sources to obtain the corresponding data set.

[0013] Preferably, it further includes training the evaluation model according to the following process,

[0014] Divide the training data set into k mutually exclusive subsets with similar sizes, use the union of k - 1 subsets as the training set, and the remaining one subset as the test set;

[0015] Use the training set to train the preset algorithm model to obtain an initial model; and continuously adjust the internal parameters of the initial model on the training set through iterative training to minimize the loss function until the best fit of the data distribution is achieved, and output the final evaluation model.

[0016] Preferably, the prediction of the corresponding computing amount and the allocation scheme of the computing amount includes,

[0017] After the data set is input into the model, the evaluation model predicts the computing amount of the demand side task according to the historical data and the generated data.

[0018] Preferably, the calculation amount corresponding to the prediction and the calculation amount allocation scheme further include

[0019] dividing the tasks into multiple subsets according to the predicted calculation amount and allocating them to different computing nodes or processors;

[0020] determining the performance and load conditions of the nodes, and dynamically adjusting the task allocation until the performance and load conditions of each node reach the optimal overall performance, and outputting the allocation scheme at this time as the final calculation amount allocation scheme.

[0021] On the other hand, a regulation system for power grid demand-side resources is also provided, which is used to implement the regulation method for power grid demand-side resources described above, including

[0022] a data acquisition module, which is used to acquire the production data of the power grid and preprocess the acquired data; the production data at least includes production equipment status data, production task information data, and employee work condition data;

[0023] a data evaluation module, which is used to determine the corresponding calculation amount according to the demand information on the demand side, and process the preprocessed production data through a pre-trained evaluation model to predict the corresponding calculation amount and the calculation amount allocation scheme, and obtain the corresponding regulation strategy; wherein, the evaluation model determines the problem type to be optimized according to the calculation amount, and calculates the input production data through a preset specific algorithm to determine the resource allocation scheme corresponding to the problem type to be optimized;

[0024] a regulation module, which is used to adjust the power grid demand-side resources according to the regulation strategy to realize the regulation of the power grid demand-side resources.

[0025] Preferably, the regulation module is further used to

[0026] acquire and display the corresponding monitoring data in real time according to the preset monitoring indicators;

[0027] evaluate the performance of the system according to the monitoring data and demand information, and adjust the resource scheduling strategy and load balancing strategy in the regulation strategy in real time according to the evaluation results.

[0028] Preferably, the data acquisition module is specifically used to

[0029] clean, format convert, and standardize the production data according to the preset standards; and integrate the production data from different data sources to obtain the corresponding data set.

[0030] Determine the performance and load conditions of the nodes, and dynamically adjust the task allocation until the performance and load conditions of each node reach the optimal overall performance. Output the allocation scheme at this time as the final calculation amount allocation scheme.

[0031] In summary, implementing the embodiments of the present invention has the following beneficial effects:

[0032] The regulation method and system for power grid demand-side resources provided by the present invention select and optimize artificial intelligence algorithms based on actual demands and resource characteristics, not only ensuring a high degree of matching between the algorithms and the problems, but also further improving the prediction accuracy and decision-making efficiency of the model through algorithm optimization, providing a scientific basis for resource allocation; by real-time monitoring the task execution situation and dynamically evaluating and allocating the calculation amount, it helps to timely adjust the resource usage strategy, avoid resource waste and bottleneck problems, and achieve flexible scheduling and efficient utilization of computing resources; automatically adjusting the number of computing resources according to the real-time changes in business demands can ensure a high degree of matching between resource supply and business demands, reduce operating costs, and improve the overall operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, obtaining other drawings without creative efforts still belongs to the scope of the present invention.

[0034] Figure 1 It is a main process schematic diagram of a regulation method for power grid demand-side resources in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings.

[0036] As Figure 1 shown, it is a schematic diagram of an embodiment of a regulation method for power grid demand-side resources provided by the present invention. In this embodiment, the method includes the following steps:

[0037] Step S1, collect the production data of the power grid and preprocess the collected data; the production data at least includes production equipment status data, production task information data, and employee work situation data; it is understandable that the data sources to be collected are determined, including production equipment status, production task information, employee work situation, etc.; when collecting production data, use sensors, RFID, log records, etc. to collect production data in real time; and store the collected production data in a database or data warehouse for subsequent processing and analysis.

[0038] The production equipment status data can be monitored in real time through the Internet of Things. Install sensors on the production equipment to collect operating parameters (such as temperature, pressure, vibration frequency, etc.) and fault warning signals. The data sources cover the real-time data directly transmitted by the sensors, as well as the built-in diagnostic systems and maintenance records of the equipment, which jointly provide decision-making support for equipment preventive maintenance. The collection of production task information can be carried out through the ERP (Enterprise Resource Planning) system and the enterprise internal database. The ERP system integrates information such as production plans, material requirements, and order status. Through the user interface or API interface, the progress of production tasks, priority changes, delivery dates, etc. can be obtained and tracked in real time. In addition, the historical production task data stored in the database helps to analyze production efficiency and trends. The collection of employee work situation information can be carried out using the HRIS (Human Resources Information System) and the intelligent working hours management system. The HRIS records employees' basic information, attendance, skill training records, etc., providing a basis for human resources allocation and optimization. The intelligent working hours management system accurately calculates individual and team working hours, and evaluates work efficiency and saturation through functions such as employee clock-in, task assignment, and completion confirmation. The data sources include reports generated by the system, self-evaluation and peer-evaluation data of employees, and instant work feedback submitted through mobile devices.

[0039] Sensors are the key components for realizing the collection of real-time data of equipment. According to the characteristics of the equipment and monitoring requirements, different types of sensors can be selected, such as temperature sensors, pressure sensors, vibration sensors, and flow sensors, etc. These sensors are installed at the designated parts of the equipment. For example, temperature sensors are usually installed in the hot spots of the equipment to monitor temperature changes; vibration sensors are installed at the bearings of the equipment to monitor vibration conditions. The location selection of sensors is crucial because it directly affects the accuracy and integrity of the data. The data collected by the sensors is transmitted to the data acquisition system, such as a programmable logic controller (PLC) or a distributed control system (DCS), through wired or wireless means. Wired connections usually use serial communication protocols, such as RS-232, RS-485, etc.; wireless connections use wireless communication technologies such as Wi-Fi and Bluetooth.

[0040] RFID realizes the automatic identification and data collection of objects through RFID tags, readers, and a backend management system. In a production environment, RFID tags are attached to items that need to be tracked, and unique identification information such as numbers, categories, production dates, etc. is written into the tags. When an item enters the reading range of the reader, the reader activates the tag through radio waves, and the tag returns the stored information to complete data collection. The RFID system supports contactless identification and can identify multiple tags simultaneously in an extremely short time, making it suitable for scenarios that require high-efficiency processing of large amounts of data, such as production line tracking and warehouse management. The reader transmits the collected data to the backend management system through a wired or wireless network, and the backend management system parses, validates, and stores the data to achieve dynamic management and real-time tracking of the data. RFID not only improves the efficiency and accuracy of data collection but also protects the security of data transmission through encryption technology to prevent information from being illegally read or tampered with. This makes RFID of great significance in achieving the transparency and accurate traceability of the production chain and supply chain.

[0041] Logging is another important way of data collection, especially when recording production task information and employees' work conditions. Web servers such as Nginx, Tomcat, etc. have built-in logging functions and can record system events and operation logs. In addition, enterprises can also use specialized log collection tools such as Chukwa of Hadoop, Flume of Cloudera, etc. These tools adopt a distributed architecture and can meet the needs of high-speed log data collection and transmission.

[0042] In a production environment, the work conditions of employees can be recorded through custom collection methods, such as using JavaScript code to monitor user behavior and making AJAX asynchronous requests to the backend to record logs. In terms of data synchronization, methods such as direct data source synchronization, generating data file synchronization, or database log synchronization can be adopted to ensure the real-time nature and integrity of the data.

[0043] In one embodiment, the preprocessing of the collected data includes cleaning, format conversion, and standardization of production data according to preset criteria; and integrating the production data from different data sources to obtain the corresponding data set. It can be understood that cleaning the production data means removing invalid data such as duplicate data and abnormal data to ensure data quality. Converting the production data means converting the data into a format suitable for algorithm processing, such as converting text data into numerical data. Standardizing the production data means performing normalization processing on the production data to eliminate the influence of different dimensions. Integrating the production data means integrating the production data from different data sources to form a complete data set. Checking the consistency of the production data means ensuring that the data from different data sources is semantically consistent.

[0044] Specifically, when cleaning production data, first identify and remove duplicate data. Use the deduplication function of the database management system or data structures in programming languages (such as sets in Python) to screen and eliminate duplicate records. During this process, carefully define the criteria for duplicate data, such as exact matches based on device ID, task number, or employee ID number, to ensure that no potential valid information is missed. Then, for the detection and handling of abnormal data, a combination of statistical analysis and machine learning algorithms can be used. By setting reasonable threshold ranges, such as the device running time should not exceed the normal working period, the production task completion rate should be within a reasonable range, and the employee work efficiency should conform to the historical average level, etc., mark the data that exceeds these ranges. At the same time, using machine learning techniques such as clustering analysis can automatically identify and isolate the outliers that deviate significantly from most data points, improving the accuracy and efficiency of data cleaning. In addition, a data quality monitoring mechanism can be established to regularly audit production data to ensure the continuous optimization and improvement of the data cleaning process. This includes reviewing the effectiveness of data cleaning rules, monitoring the frequency and reasons for the occurrence of abnormal data, and adjusting the data cleaning strategy in a timely manner according to business changes.

[0045] When converting production data, for production equipment status data, such as the equipment operation status (running, stopped, under repair, etc.), these text statuses can be converted into numerical codes through a mapping strategy. For example, "running" can be mapped to 1, "stopped" to 0, "under repair" to 2, and so on. This numerical representation not only simplifies data presentation but also facilitates subsequent data analysis and algorithm processing. Production task information often contains multiple attributes, such as task priority, estimated completion time, etc. For text descriptions of priority (such as "high", "medium", "low"), numerical coding can also be used for conversion. The estimated completion time can be directly converted into a timestamp or a unified time unit (such as hours, minutes) for time series analysis or the application of scheduling algorithms. Employee work situation data, such as attendance rate, work efficiency, etc., usually already appear in numerical form, but attention should be paid to the consistency and integrity of the data. For the attendance rate, ensure that the data format is in percentage or decimal form for direct calculation by algorithms. Work efficiency may need to be combined with specific task completion time and quality indicators to obtain a numerical result through formula calculation. All data conversion processes can record the conversion rules for data traceability and verification. At the same time, conduct quality checks on the converted data and store it in a unified data structure, such as a database or a data warehouse, after ensuring it is correct, providing a solid data foundation for subsequent algorithm model training, predictive analysis, and decision support. Through this series of fine data conversion processes, the utilization efficiency and value of production data can be significantly improved.

[0046] When standardizing production data, normalization processing is performed on the production data to eliminate the influence of different dimensions. The purpose of normalization processing is to convert various types of data to the same scale for subsequent data analysis and mining. For production equipment status data, such as temperature, pressure, vibration frequency, etc., the min-max normalization method can be used, that is, the data is scaled proportionally between 0 and 1, and the formula is: (X - X_min) / (X_max - X_min). This processing can effectively eliminate the problem of inconsistent dimensions caused by differences in equipment models and performance. For production task information, such as production quantity, completion time, etc., considering that these data may have different orders of magnitude, the Z-score standardization method can be used, that is, the data is divided by the standard deviation after subtracting the mean, so that the processed data conforms to the standard normal distribution. This method helps to identify the relative advantages and disadvantages of production task completion and facilitates the optimization and adjustment of production plans. The data on employees' work conditions, such as work efficiency, attendance rate, etc., also apply the min-max normalization or Z-score standardization method. The specific choice should be based on the distribution characteristics of the data and the subsequent analysis requirements. Through normalization processing, the performance differences between different positions and different employees can be intuitively compared, providing a basis for human resource allocation and incentive mechanism design.

[0047] More specifically, when integrating production data, integrating production data from different data sources is a key step in improving operational efficiency and optimizing resource allocation. This step aims to fuse diverse data such as production equipment status, production task information, and employees' work conditions into a complete dataset. First, establish unified data standards and formats to ensure seamless docking of the information provided by each data source. This includes defining the naming specifications, data types, and data precisions of data fields, laying a solid foundation for subsequent data integration. Then, use data integration tools or platforms, such as ETL (Extract, Transform, Load) systems, to achieve data extraction, transformation, and loading. In this process, corresponding data extraction strategies, such as scheduled extraction, event-triggered extraction, etc., need to be designed according to the characteristics of different data sources to ensure the timeliness and accuracy of the data. At the same time, through the data transformation step, the data in different formats is unified into the standard format of the target system for subsequent processing and analysis. Then build a data warehouse or data lake as the carrier for storing the integrated dataset. These storage systems not only need to have high availability and scalability but also support complex data query and analysis operations to meet the diverse needs in production management.

[0048] Check the consistency of production data: Ensure that the data from different data sources is semantically consistent. First, establish a unified set of data standards and definitions, clarify the indicators and meanings of production equipment status, production tasks, and employees' work conditions, and ensure that each data source follows the same specifications when collecting and recording data. This helps reduce misunderstandings and errors caused by definition differences. Then, implement a regular data comparison and verification mechanism. Through automated tools, cross-verify the data from different data sources to check for contradictions or inconsistencies. Focus on the logical relationships and business rules between the data to ensure the accuracy of the data at the semantic level. Next, establish a data feedback and correction process. Once a data inconsistency problem is discovered, immediately initiate the feedback mechanism to notify the person in charge of the relevant data source for verification and correction. At the same time, record the reasons for the problem occurrence and the solution process to provide reference for subsequent data management and optimization.

[0049] In addition, adopting a data governance platform is also an important means to ensure data consistency. Through the data governance platform, it is possible to monitor and manage the entire life cycle of data, including data collection, storage, processing, and usage. This helps to promptly discover and solve data quality problems and improve the overall credibility of the data.

[0050] Step S2, determine the corresponding amount of computation according to the demand information on the demand side, and process the preprocessed production data through a pre-trained evaluation model to predict the corresponding amount of computation and the allocation scheme of the amount of computation, and obtain the corresponding regulation strategy; wherein, the evaluation model determines the type of problem to be optimized according to the amount of computation, and calculates the input production data through a preset specific algorithm to determine the resource allocation scheme corresponding to the type of problem to be optimized; it is understandable that according to specific requirements and resource characteristics, select an artificial intelligence algorithm and optimize the artificial intelligence algorithm. According to business requirements, determine the type of optimization problem to be solved, such as classification, regression, clustering, etc. According to the characteristics of the problem and the data, select a suitable algorithm, such as neural network, support vector machine, decision tree, etc. Algorithm framework selection: Select a suitable algorithm framework, such as TensorFlow, PyTorch, etc., for implementing the algorithm. Algorithm parameter configuration: Configure the corresponding parameters according to the requirements of the algorithm, such as learning rate, number of iterations, etc. Algorithm optimization: Cross-validation: Use the cross-validation method to evaluate the performance of the algorithm and select the optimal parameter combination. Feature selection: Through the feature selection method, remove redundant features to improve the running efficiency and accuracy of the algorithm.

[0051] It should be noted that business requirements specifically refer to the specific requirements from users, the market, or within the system for aspects such as resource allocation, efficiency improvement, and cost control, including but not limited to improving energy utilization efficiency, optimizing the customer service experience, and reducing operating costs. Analyze the business requirements to clarify their goals, constraints, and key performance indicators (KPIs). Based on the results of the requirements analysis, determine which category of optimization problems need to be solved. For example, if it is necessary to predict future energy demand to reasonably arrange supply, it may involve a regression problem; if it is necessary to divide user groups according to specific attributes to implement a differential service strategy, it may involve a clustering problem; if it is necessary to determine whether a certain service meets specific criteria to decide whether to provide it, it may involve a classification problem.

[0052] In one embodiment, the corresponding calculation amount and the allocation scheme of the calculation amount include that after the data set is input into the model, the evaluation model predicts the calculation amount of the demand-side task according to historical data and generated data. It also includes dividing the task into multiple subsets according to the predicted calculation amount and allocating them to different computing nodes or processors; determining the performance and load conditions of the nodes, and dynamically adjusting the task allocation until the performance and load conditions of each node reach the optimal overall performance, and outputting the allocation scheme at this time as the final allocation scheme of the calculation amount. It can be understood that by monitoring the task execution situation in real time, evaluating and allocating the calculation amounts of different tasks, it is possible to ensure the full utilization of resources while avoiding congestion. Dynamic load balancing can significantly improve the utilization rate of resources, reduce the response time of the system, and improve the overall performance of the system.

[0053] Specifically, use machine learning algorithms to predict the calculation amount of tasks according to historical data and real-time data. Apply the trained model to real-time data prediction. After the real-time data is input into the model, the model can quickly output the predicted calculation amount, providing decision support for task scheduling and resource allocation. According to the predicted calculation amount, allocate the tasks to different computing nodes or processors.

[0054] Analyze and predict the entire computing task to understand the computing resources required for each part. This involves a comprehensive evaluation of factors such as the data volume of the task, algorithm complexity, and computing efficiency. Based on these predictions, divide the task into multiple subsets. These subsets can be uniform or non-uniform, depending on the data distribution and the complexity of the computing task. These subsets are then allocated to different computing nodes or processors.

[0055] During the allocation process, the performance and load conditions of the computing nodes are considered to ensure the reasonable utilization of resources and load balancing. This is achieved through a scheduling system that can monitor the status of each node and dynamically adjust task allocation to optimize the overall performance. After receiving tasks, each computing node or processor executes the computing tasks in parallel, processing the data subsets assigned to them. During this process, data exchange and synchronization may be required between nodes to ensure the correctness of the computing results. After all computing tasks are completed, the system combines the computing results on each node to obtain the final computing result.

[0056] Load balancing strategy: Dynamic scheduling: Use dynamic scheduling algorithms, such as container scheduling frameworks like Kubernetes, to dynamically adjust task allocation based on the execution status of tasks and the load condition of the system. Dynamic scheduling dynamically adjusts task allocation by monitoring the execution status of tasks and system load in real time to optimize resource utilization and performance.

[0057] In Kubernetes, the Scheduler component is responsible for the scheduling and allocation of Pods. It discovers unscheduled Pods in the cluster through the watch mechanism and filters them according to a series of scheduling principles (such as resource requests, computing resources, memory, storage, network, port numbers, etc.) to select nodes that meet the Pod requirements. Subsequently, the Scheduler scores each candidate node according to the scoring rules (such as the resource utilization rate and performance of the node) and selects the node with the highest score to run the Pod. This process ensures that Pods are assigned to the most suitable nodes to improve the overall performance and reliability of the system.

[0058] When the system load or task execution status changes, the dynamic scheduling algorithm can adjust task allocation in a timely manner. For example, if a certain node has a high load, the Scheduler can migrate some tasks to nodes with lower loads to maintain system balance. In addition, Kubernetes also supports label-based scheduling strategies. By adding labels to nodes and Pods, more flexible and fine-grained scheduling control can be achieved. Dynamic scheduling improves resource utilization and system performance through real-time monitoring and intelligent adjustment of task allocation. In container scheduling frameworks such as Kubernetes, this process is efficiently and automatically implemented, providing strong support for complex and changing business scenarios.

[0059] Reserve a certain amount of computing resources for critical tasks to ensure their normal execution under high load conditions. Set critical tasks and non-critical tasks according to the actual operation of power calculation. Resource reservation is an important mechanism to ensure the normal execution of critical tasks under high load conditions. Analyze the resource requirements of critical tasks, which include computing, storage, and network resources, etc. Through historical data and task characteristics, predict their resource consumption under high load. According to the analysis results, pre-allocate and lock a certain amount of resources in the system. These resources will be specifically used to support the execution of critical tasks and avoid resource contention with other tasks. It can be achieved through the resource management tools of the cloud computing platform, such as setting static reservation or policy-based dynamic reservation.

[0060] After implementing resource reservation, continuously monitor the system load and resource usage. Through monitoring tools, keep track of the resource utilization rate and the performance of critical tasks in real time. Once resource tension or task performance degradation is detected, immediately trigger the warning mechanism to take optimization measures in a timely manner.

[0061] For regression problems, models such as linear regression and decision tree regression can be selected; for clustering problems, methods such as K-means and hierarchical clustering can be considered; for classification problems, classifiers such as logistic regression and support vector machines can be used. Determine whether to use TensorFlow or PyTorch according to the actual requirements and the characteristics of production data. The advantages of TensorFlow are its strong distributed computing ability and optimization, which are suitable for processing large-scale datasets and complex models; while the advantages of PyTorch are its dynamic computational graph and intuitive and easy-to-use API, which are more suitable for rapid prototyping and experiments. After selecting the algorithm framework, perform environment configuration. This includes installing the corresponding version of TensorFlow or PyTorch, as well as configuring the necessary dependent libraries and Python environment. For TensorFlow, it is also necessary to configure the Conda source to accelerate package management and updates.

[0062] It is necessary to clarify the specific requirements of the algorithm used for parameters. For example, for the gradient descent algorithm, the learning rate and the number of iterations are key hyperparameters, which directly affect the training effect and convergence speed of the model. The learning rate determines the step size of each parameter update, while the number of iterations determines the total number of forward and backward propagations of the model on the training data. According to the characteristics of the algorithm and the data distribution, initially select the values of the learning rate and the number of iterations, and set a reasonable initial range for these parameters. Then, use the validation set to evaluate the performance of the model, and find the optimal configuration by continuously adjusting the parameter values.

[0063] An embodiment further includes training an evaluation model according to the following process. Divide the training dataset into k mutually exclusive subsets of similar sizes, use the union of k - 1 subsets as the training set, and the remaining one subset as the test set. Use the training set to train a preset algorithm model to obtain an initial model. And through iterative training, continuously adjust the internal parameters of the initial model on the training set to minimize the loss function until the best fit to the data distribution is achieved, and output the final evaluation model. Use the processed data to train the selected algorithm and improve the performance of the model through optimization methods. Through model training and optimization, a model with excellent performance can be obtained, providing strong support for subsequent resource optimization and regulation.

[0064] It can be understood that model training specifically includes:

[0065] Training data preparation: Divide the processed data into a training set and a test set. Adopt the hold-out method: directly divide the dataset into two mutually exclusive parts, where one part is used as the training set and the other part is used as the test set. The key to this method is to ensure that the data distributions of the training set and the test set are as consistent as possible to avoid adverse effects on the model training results due to data distribution differences. Usually, this goal is achieved through stratified sampling to ensure that the training set and the test set are consistent in class distribution. Adopt the cross-validation method: divide the dataset into k mutually exclusive subsets of similar sizes, and then each time use the union of k - 1 subsets as the training set, and the remaining one subset as the test set. In this way, k times of training and testing can be carried out, and finally the mean of the k test results is taken as the evaluation result. The stability and fidelity of the cross-validation method largely depend on the value of k. A common value of k is usually 10, that is, 10-fold cross-validation.

[0066] Model training process: Use the training set data to train the algorithm to obtain an initial model. Through the iterative training process, the algorithm continuously adjusts its internal parameters (such as weights) on the training set to minimize the loss function (such as mean squared error, cross-entropy, etc.), so as to achieve the best fit to the data distribution. This process involves optimization algorithms such as gradient descent and stochastic gradient descent.

[0067] Model optimization: Regularization: Prevent overfitting by adding a regularization term. Overfitting means that the model performs well on the training data but has poor generalization ability on new data. Regularization is a technique that helps prevent overfitting by adding an additional penalty term to the loss function of the model to limit the magnitude of the model parameters. The regularization term is used to control the complexity of the model, making the model more inclined to choose simple solutions during training. This simplicity is achieved by restricting the magnitude or sparsity of the model parameters. There are various specific forms of the regularization term, including L0 regularization, L1 regularization, L2 regularization, and stochastic regularization, etc.

[0068] L1 regularization adds the sum of the absolute values of the model parameters as a penalty term to the loss function, which encourages some parameters to become zero, thus achieving feature selection and sparsity. This method helps to exclude irrelevant or redundant features, reduce the complexity of the model, and thus reduce the risk of overfitting.

[0069] L2 regularization reduces the magnitude of the parameters by taking the sum of the squares of the model parameters as a penalty term, thereby reducing the complexity of the model. Different from L1 regularization, L2 regularization tends to shrink all parameters to varying degrees rather than making some parameters zero. This method can make the model parameters distributed within a smaller range, reduce the collinearity problem between parameters, and improve the stability and generalization ability of the model.

[0070] Random regularization simplifies the network structure by randomly setting some function parameters to 0, so as to avoid overfitting of the model. In a neural network, this usually manifests as inactivating some neurons, such as using a dropout network structure or using activation functions such as Relu or Gelu to set some parameters to 0.

[0071] Ensemble Learning: Use ensemble learning methods such as Bagging, Boosting, etc. to improve the generalization ability of the model. Using ensemble learning methods such as Bagging and Boosting can enhance the generalization ability of the model. These methods combine multiple different learners together for joint learning and prediction, which can significantly improve the overall learning and prediction performance. The Bagging algorithm realizes ensemble learning by simultaneously constructing multiple weak classifiers. It first randomly samples the training data to form multiple datasets, and each dataset is used to train a weak classifier. This random sampling method enables each weak classifier to learn different data features. In the prediction stage, the Bagging algorithm uses voting (for classification problems) or calculating the mean (for regression problems) to synthesize the prediction results of all weak classifiers, thereby obtaining the final prediction result. This method reduces the variance of the model and enhances the generalization ability of the model. A typical implementation of the Bagging algorithm is random forest, which uses decision trees as weak classifiers and further enhances the diversity of the model by randomly selecting some features at the nodes for splitting. The Boosting algorithm gradually constructs weak learners and continuously adjusts the weights of examples according to the error rate, thereby improving the performance of the overall model. In each round of training, the Boosting algorithm focuses on the data misclassified in the previous round and assigns higher weights to these data, enabling the new weak learner to learn more about these difficult-to-classify data. Finally, the Boosting algorithm performs a weighted sum of the prediction results of all weak learners to obtain the final prediction result. Since each weak learner is optimized for difficult-to-classify data, the Boosting algorithm can significantly reduce the bias of the model and improve the generalization ability of the model. Common implementations of the Boosting algorithm include Adaboost, GBDT, and XGBoost, etc.

[0072] Hyperparameter Tuning: Use methods such as grid search and random search to tune the hyperparameters of the model. Hyperparameter tuning directly affects the performance and prediction accuracy of the model.

[0073] Grid search is a systematic tuning method that finds the best combination by exhaustively searching the specified hyperparameter combinations. First, determine the possible value ranges of each hyperparameter, and then permute and combine these hyperparameters to form all possible parameter combinations. Next, for each parameter combination, evaluate the model performance using cross-validation or a validation set, and select the parameter combination with the best performance as the hyperparameters of the final model. The advantage of grid search is that it is simple and easy to understand and can find the global optimal solution.

[0074] Random search randomly samples parameter combinations in the parameter space instead of systematically traversing all combinations. The theoretical basis of this method is that in many cases, the optimal parameter combination may not be at the edge of the parameter space but at some intermediate position. By randomly sampling, the optimal solution can be found more quickly. In addition, random search can be easily parallelized, thus further improving the tuning efficiency.

[0075] Model evaluation: Selection of evaluation metrics: According to the type of problem, select appropriate evaluation metrics such as accuracy, recall, F1-score, etc. When choosing appropriate evaluation metrics to measure the performance of a model, it is necessary to consider the type of problem and the aspects emphasized by various evaluation metrics. First, clarify whether the problem is a classification, regression, or other type of task. Taking the classification task as an example, common evaluation metrics include accuracy, recall, and F1-score, etc. For binary or multi-class classification problems, accuracy is an important metric to measure the proportion of correctly predicted samples in the total samples. However, when the class distribution is imbalanced, accuracy may not be comprehensive enough because it may ignore the prediction of minority class samples. Recall (also known as sensitivity) focuses on measuring the ability of the model to identify samples of a certain class (usually the positive class), that is, the proportion of samples that are actually positive and are predicted as positive by the model. A high recall means that the model can identify as many positive class samples as possible, but it may come at the cost of some correct predictions of negative class samples. The F1-score is the harmonic mean of accuracy and recall, which provides a comprehensive metric to measure the performance of the model. The higher the F1-score, the better the balance the model achieves between accuracy and recall.

[0076] Evaluation using the test set: Use the test set data to evaluate the trained model and verify its performance. Divide the dataset into a training set, a validation set, and a test set. The training set is used to build the model, the validation set is used to adjust the model parameters and select the optimal model, while the test set is used to evaluate the model's prediction ability for future data.

[0077] After the model training is completed, use the test set data for prediction. In this process, it is necessary to ensure that the test set data has the same format and preprocessing steps as the training set data to ensure the fairness of the evaluation. Compare the model's prediction results with the true labels of the test set, and evaluate the model's performance on the test set by calculating metrics such as accuracy, precision, recall, F1-score, etc. These metrics can comprehensively reflect the model's performance on different classes and its ability to distinguish between positive and negative samples. Finally, optimize and improve the model according to the evaluation results. If the model performs poorly on the test set, adjust the model structure, increase the training data, or adopt other optimization strategies. Through continuous iteration and optimization, the generalization ability and prediction accuracy of the model can be improved.

[0078] Step S3: Adjust the power grid demand-side resources according to the above-mentioned regulation strategy to achieve the regulation of the power grid demand-side resources. Understandably, according to the predicted demand, the number of computing resources is automatically adjusted, such as increasing or decreasing virtual machines, containers, etc. Set monitoring metrics, which are predicted based on historical data and business trends to determine when and how to adjust resources. For example, during an e-commerce promotion, a surge in traffic is expected. At this time, thresholds can be set based on key metrics such as CPU utilization, memory usage, or network traffic. When the monitoring tool detects that these metrics reach or exceed the predefined thresholds, the automated system will trigger corresponding adjustment operations. This includes starting new virtual machines or containers, or shutting down instances that are no longer needed to match the current demand level.

[0079] In a cloud environment, such as the AWS EC2 Auto Scaling service, this process is particularly efficient. This service dynamically adjusts the number of Amazon EC2 instances according to predefined policies and monitoring metrics, ensuring sufficient computing power during traffic peaks and reducing resource waste during traffic lows. In addition, by adopting containerization technologies such as Docker and Kubernetes, the scalability and utilization of resources can be further improved. These technologies allow containers to be quickly deployed or destroyed according to demand, thus achieving more refined resource management.

[0080] The specific embodiment further includes: collecting and displaying corresponding monitoring data in real time according to preset monitoring metrics; evaluating the performance of the system based on the monitoring data and demand information, and making real-time adjustments to the resource scheduling strategy and load balancing strategy in the regulation strategy. Understandably, monitor and evaluate the operating status of the system and resource usage, and discover and solve problems in a timely manner. Monitoring and evaluation can ensure the stable operation of the system and the effective utilization of resources, discover and solve problems in a timely manner, and improve the overall performance of the system and user satisfaction.

[0081] Selection of monitoring metrics: Select key monitoring metrics, such as CPU utilization, memory occupancy, network bandwidth, etc. CPU utilization can intuitively reflect the system's ability to process tasks and whether there is a risk of overload. Memory occupancy helps to discover problems such as memory leaks or shortages, ensuring reasonable allocation of system resources. Monitoring network bandwidth by continuously tracking inbound and outbound traffic can timely detect network congestion or abnormal traffic conditions.

[0082] Monitoring Tool Selection: Select appropriate monitoring tools such as Prometheus, Grafana, etc. for real-time collection and display of monitoring data. Tools like Prometheus and Grafana provide a rich library of monitoring metrics and flexible configuration options, facilitating the rapid setup of an efficient monitoring system. Additionally, in combination with business characteristics and historical data, monitoring metrics can be customized and optimized to ensure they accurately reflect the system status and provide strong support for operation and maintenance decisions.

[0083] Monitoring Data Analysis: Data Visualization: Use visualization tools to present monitoring data in the form of charts for easy observation and analysis. Map the processed data to visual elements such as lines, bars, pies, etc. For example, line charts are suitable for showing the changing trends of time-series data, while bar charts are suitable for comparing data of different categories. Organize and filter the visualization charts through human-computer interaction to make them both aesthetically pleasing and easy to understand. Avoid page bloat and chaos to ensure that users can focus and quickly obtain information.

[0084] Anomaly Detection: Use machine learning algorithms such as clustering and classification to detect anomalies in monitoring data and promptly identify abnormal situations in the system. Collect and preprocess the monitoring data, including steps such as data cleaning and normalization, to ensure data quality and consistency. Subsequently, select appropriate machine learning algorithms for model training. For example, clustering algorithms can group normal data points together to form distinct clusters, while abnormal data lies outside the clusters or forms small clusters. Classification algorithms learn the characteristics of normal data to build classifiers for distinguishing normal and abnormal data. After the model training is completed, apply it to real-time monitoring data to assign an anomaly score or classification to each data point. When the anomaly score of a data point exceeds a preset threshold or is classified as abnormal, the system will trigger an alarm to notify relevant personnel for handling.

[0085] In a specific embodiment, the collected production data is stored in a database or a data warehouse for subsequent processing and analysis. Databases mainly include two categories: relational databases and non-relational databases. Relational databases, such as MySQL, Oracle, and SQL Server, use two-dimensional tables and their relationships to organize data, featuring clear data structures, easy to understand and operate. They support SQL language, facilitating complex queries and modifications. The advantages of relational databases lie in easy maintenance, convenient use, and support for complex operations. Non-relational databases, also known as NoSQL databases, such as MongoDB, Redis, and Cassandra, do not rely on the relational model, and data storage does not require a fixed table schema. Non-relational databases have fast processing speeds and strong scalability, especially suitable for processing big data and high-concurrency access. Key-value storage databases and document-oriented databases are the main types of non-relational databases. The former has high performance and is easy to deploy, while the latter has strong flexibility and is suitable for website back-end databases and small file systems, etc.

[0086] An embodiment of the present invention further provides a regulation system for grid demand-side resources, used to implement the regulation method for grid demand-side resources described above, including,

[0087] A data collection module, used to collect the production data of the power grid and preprocess the collected data; the production data at least includes production equipment status data, production task information data, and employee work condition data;

[0088] A data evaluation module, used to determine the corresponding calculation amount according to the demand information on the demand side, and process the preprocessed production data through a pre-trained evaluation model, predict the corresponding calculation amount and the allocation scheme of the calculation amount, and obtain the corresponding regulation strategy; wherein, the evaluation model determines the problem type to be optimized according to the calculation amount, and calculates the input production data through a preset specific algorithm to determine the resource allocation scheme corresponding to the problem type to be optimized;

[0089] A regulation module, used to adjust the grid demand-side resources according to the regulation strategy to achieve the regulation of grid demand-side resources.

[0090] In one embodiment, the regulation module is further used to, collect and display the corresponding monitoring data in real time according to preset monitoring indicators; evaluate the performance of the system according to the monitoring data and demand information, and adjust the resource scheduling strategy and load balancing strategy in the regulation strategy in real time according to the evaluation results.

[0091] In one embodiment, the data collection module is specifically used to clean, format convert, and standardize the production data according to preset standards; and integrate the production data from different data sources to obtain the corresponding data set.

[0092] In one embodiment, the data evaluation module is specifically configured to, after the data set is input into the model, the evaluation model predicts the computing amount of the demand-side task according to historical data and generated data; according to the predicted computing amount, divide the task into multiple subsets and allocate them to different computing nodes or processors; determine the performance and load conditions of the nodes, and dynamically adjust the task allocation until the performance and load conditions of each node reach the optimal overall performance, and output the allocation scheme at this time as the final allocation scheme of the computing amount.

[0093] It should be noted that the system described in the above embodiment corresponds to the method described in the above embodiment. Therefore, the parts not detailed in the system described in the above embodiment can be obtained by referring to the content of the method described in the above embodiment, and will not be elaborated here.

[0094] In summary, implementing the embodiments of the present invention has the following beneficial effects:

[0095] The method and system for regulating power grid demand-side resources provided by the present invention select and optimize artificial intelligence algorithms based on actual demands and resource characteristics, not only ensuring a high degree of matching between the algorithms and the problems, but also further improving the prediction accuracy and decision-making efficiency of the model through algorithm optimization, providing a scientific basis for resource allocation; by monitoring the task execution situation in real time and dynamically evaluating and allocating the computing amount, it helps to timely adjust the resource usage strategy, avoid resource waste and bottleneck problems, and achieve flexible scheduling and efficient utilization of computing resources; automatically adjusting the number of computing resources according to the real-time changes in business demands can ensure a high degree of matching between resource supply and business demands, reduce operating costs, and improve overall operating efficiency.

[0096] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for regulating power grid demand side resources, characterized in that: include: Collecting production data of the power grid and preprocessing the collected data; the production data at least includes production equipment status data, production task information data and employee work status data; Determine the corresponding computing amount according to the demand information on the demand side, and process the preprocessed production data through the pre-trained evaluation model to predict the corresponding computing amount and the allocation plan of the computing amount, and obtain the corresponding control strategy; wherein, the evaluation model determines the type of problem to be optimized according to the computing amount, and calculates the input production data through a preset specific algorithm to determine the resource allocation plan corresponding to the type of problem to be optimized; According to the control strategy, the resources on the demand side of the power grid are adjusted to achieve the control of the resources on the demand side of the power grid.

2. The method according to claim 1, characterized in that It also includes collecting and displaying corresponding monitoring data in real time according to preset monitoring indicators; The system performance is evaluated based on monitoring data and demand information, and the resource scheduling strategy and load balancing strategy in the control strategy are adjusted in real time based on the evaluation results.

3. The method according to claim 2, characterized in that The preprocessing of the collected data includes cleaning, format conversion and standardization of the production data according to preset standards; and integrating the production data from different data sources to obtain a corresponding data set.

4. The method according to claim 3, characterized in that It also includes training and evaluating the model according to the following process: Divide the training data set into k mutually exclusive subsets of similar size, use the union of k-1 subsets as the training set, and the remaining subset as the test set; The preset algorithm model is trained using the training set to obtain the initial model; and the internal parameters of the initial model are continuously adjusted on the training set through iterative training to minimize the loss function until the best fit to the data distribution is achieved, and the final evaluation model is output.

5. The method according to claim 4, characterized in that The calculation amount corresponding to the prediction and the allocation scheme of the calculation amount include: After the data set is input into the model, the evaluation model predicts the computational workload of the demand-side task based on historical data and generated data.

6. The method according to claim 5, characterized in that The calculation amount corresponding to the prediction and the allocation scheme of the calculation amount also include: According to the predicted computing amount, the tasks are divided into multiple subsets and assigned to different computing nodes or processors; Determine the performance and load of the nodes, and dynamically adjust the task allocation until the performance and load of each node reach the optimal overall performance, and output the allocation plan at this time as the final computing allocation plan.

7. A control system for power grid demand side resources, used to implement the method according to any one of claims 1 to 6, characterized in that: include, A data acquisition module is used to collect production data of the power grid and pre-process the collected data; the production data at least includes production equipment status data, production task information data and employee work status data; The data evaluation module is used to determine the corresponding computing amount according to the demand information on the demand side, and to process the preprocessed production data through the pre-trained evaluation model, predict the corresponding computing amount and the allocation scheme of the computing amount, and obtain the corresponding control strategy; wherein, the evaluation model determines the type of problem to be optimized according to the computing amount, and calculates the input production data through a preset specific algorithm to determine the resource allocation scheme corresponding to the type of problem to be optimized; The control module is used to adjust the power grid demand side resources according to the control strategy to achieve the control of the power grid demand side resources.

8. The system according to claim 7, characterized in that The control module is also used to collect and display corresponding monitoring data in real time according to preset monitoring indicators; The system performance is evaluated based on monitoring data and demand information, and the resource scheduling strategy and load balancing strategy in the control strategy are adjusted in real time based on the evaluation results.

9. The system according to claim 8, characterized in that The data acquisition module is specifically used to clean, convert and standardize the production data according to preset standards; and integrate the production data from different data sources to obtain a corresponding data set.

10. The system according to claim 9, characterized in that The data evaluation module is specifically used to, after the data set is input into the model, evaluate the model to predict the computational amount of the demand-side task based on the historical data and the generated data; According to the predicted computing amount, the tasks are divided into multiple subsets and assigned to different computing nodes or processors; Determine the performance and load of the nodes, and dynamically adjust the task allocation until the performance and load of each node reach the optimal overall performance, and output the allocation plan at this time as the final computing allocation plan.

Images