New energy power generation capacity prediction application parallel strategy optimization method and system
By automatically generating and optimizing the optimization strategy for the prediction of new energy power generation capacity, combined with historical databases and real-time monitoring data, the problem of insufficient static and adaptability of optimization strategies in the existing technology is solved, and the accuracy and system of new energy power generation forecasts are improved. stability.
Patent Information
- Application Number
- CN202410921322.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-26
AI Technical Summary
The existing new energy power generation capacity prediction optimization system relies on manually formulated static optimization strategies, resulting in limited optimization effects and efficiency, and the inability to adaptive dynamic scheduling, which reduces the robustness and adaptive performance of the system.
Through the random generation tool combined with the historical database to generate matching optimization strategies, the fitness function filtering and cross-mutation operations are used to automatically find the global optimal optimization strategy, and adaptive scheduling is performed when real-time monitoring of device status and environment changes to realize computing resources dynamic allocation.
The optimization effect and efficiency of the prediction and calculation tasks of new energy power generation capacity are improved, the robustness and adaptive performance of the system are enhanced, manual intervention is reduced, and the performance of optimization strategies is ensured.
Smart Images

Figure CN119026827A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of renewable energy power generation, and in particular relates to a parallel strategy optimization method and system for predicting renewable energy power generation capacity. Background Art
[0002] Renewable energy generation capacity forecast refers to the process of predicting and evaluating the generation capacity of power generation systems using renewable energy, such as wind power, solar power, etc. This forecast is crucial for the operation and planning of power systems, because the generation of renewable energy has certain uncertainties and fluctuations, which is different from traditional energy sources such as coal and nuclear power. Therefore, accurately predicting the generation capacity of renewable energy can help power system managers to reasonably arrange power production and transmission and ensure the stable operation of the power system;
[0003] In recent years, the installed capacity of renewable energy generation has shown a rapid growth trend worldwide, with wind power, solar power and other energy sources becoming the main forms of renewable energy generation. This growth has brought new challenges and opportunities, requiring effective prediction and management of renewable energy generation capacity.
[0004] Renewable energy generation capacity prediction usually uses various mathematical models and algorithms, including models based on physical principles: such as meteorological models, wind energy models, photovoltaic models, etc., models based on statistical methods: such as time series analysis, regression analysis, etc., machine learning algorithms: such as neural networks, support vector machines, random forests, etc. These models and algorithms can use historical data, real-time monitoring data and other relevant information to predict and estimate future renewable energy generation capacity.
[0005] The prediction and calculation of renewable energy generation capacity needs to rely on computers and related strategies. Computers provide the necessary computing power and data processing capabilities, making it possible to process large-scale data and perform complex calculations and analyses. At the same time, the use of appropriate computing strategies and algorithms can improve the accuracy and efficiency of the prediction model.
[0006] The existing optimization system has the following defects: the existing optimization system usually optimizes the prediction and calculation tasks of renewable energy power generation capacity based on the optimization strategy formulated in advance manually. This optimization method has the following defects: first, the optimization strategy formulation is highly subjective, so the performance of the optimization strategy cannot be guaranteed, which may reduce the optimization effect or optimization efficiency of the optimization strategy; second, the manually formulated optimization strategy is a static optimization strategy, that is, after the optimization system adopts the manually formulated optimization strategy, it can only optimize the prediction and calculation tasks of renewable energy power generation capacity using the methods in the optimization strategy. In actual applications, when affected by factors such as equipment status or environment, the optimization system cannot perform adaptive dynamic scheduling, which reduces the robustness and adaptive performance of the optimization system. Summary of the invention
[0007] In view of the above-mentioned existing problems, the present invention can automatically find the global optimal optimization strategy based on the current renewable energy power generation capacity prediction calculation task combined with the historical big database, reduce manual intervention, ensure the performance of the optimization strategy, and perform global or local adaptive scheduling of the optimization strategy according to the real-time equipment status and environment, so as to improve the robustness and adaptive performance of the optimization system.
[0008] In order to solve the above technical problems, a parallel strategy optimization method for renewable energy power generation capacity prediction is proposed, including:
[0009] The optimization system generates matching optimization strategies based on the current prediction and calculation tasks of renewable energy power generation capacity through random generation tools combined with historical big databases; generates fitness functions according to application requirements and optimization goals and calculates the fitness value of each optimization scheme; selects optimization strategies based on fitness values through selection algorithms; after the screening, crossover and mutation operations are performed on the selected optimization strategies and a strategy set is established, and the fitness value is calculated again; repeats the selection, crossover and mutation operations for cyclic iterations, and generates a set of strategy sets for each iteration. When the stop condition is met, the cyclic iteration is exited and all strategy sets are output; all strategy sets are integrated into a strategy list, and the optimization system selects the strategy set in the strategy list The optimization strategy with the largest fitness value is used as the optimization strategy for the current renewable energy power generation capacity prediction and calculation task; in the process of the optimization system optimizing the current renewable energy power generation capacity prediction and calculation task based on the optimization strategy, when any subtask computing node fails, the computing resources of the failed node are evenly distributed to the non-faulty nodes; the data flow changes and data processing volume changes of all nodes are monitored in real time, and the data floating coefficient is generated according to the data flow changes and data processing volume changes of the nodes. The computing load status of the node is analyzed by comparing the data floating coefficient with the coefficient threshold, and all nodes are divided into sets according to the analysis results, and the node computing resource allocation is allocated and scheduled according to the set division results.
[0010] As a preferred scheme of a parallel strategy optimization method for predicting new energy power generation capacity described in the present invention, wherein: the matching optimization strategy includes generating a matching optimization strategy through a random generation tool combined with a historical big database, and the content of each optimization strategy includes the number of parallel subtasks divided, the optimization model selection and the node computing resource allocation, wherein the number of parallel subtasks divided, the optimization model selection and the node computing resource allocation are processed through binary coding, and each content is reflected in the optimization strategy in the form of binary coding, and the contents of the matching optimization strategies are different.
[0011] Get the constraints for computing task optimization, including the maximum and minimum number of subtask divisions, the type of prediction model, and the maximum allocation of computing resources:
[0012] maximize(orminimize) f(D,M′)
[0013] subject to W min ≤|D|≤W max
[0014]
[0015] D={D1,D2,...,D |D|}
[0016]
[0017] Where D is the total computing task, D d is the dth subtask, W min and W max is the minimum and maximum number of subtask divisions, M is the set of prediction models, M m is the mth prediction model, R max is the maximum allocation of computing resources, f(D,M') is the subtask D d Using Model M m The amount of computing resources required.
[0018] The optimization strategy includes dividing the number of parallel subtasks under the constraints, optimizing the model selection and allocating node computing resources, establishing four parent optimization strategies, and then establishing sub-optimization strategies in turn according to the parent optimization strategies.
[0019] The four parent optimization strategies include: optimization strategy 1 is to divide into three parallel subtasks, select a random forest optimization model, and evenly distribute computing resources to each subtask.
[0020] Optimization strategy 2 is to divide the task into five parallel subtasks, select the neural network optimization model, and allocate more computing resources to the intermediate nodes.
[0021] Optimization strategy 3 is to divide the task into four parallel subtasks, select support vector machine optimization, allocate most of the computing resources to subtasks with higher prediction accuracy, and allocate a small amount of resources to other subtasks.
[0022] Optimization strategy 4 is to divide the task into two parallel subtasks, select linear regression optimization model selection, and allocate all computing resources to a single subtask.
[0023] The number of parallel subtask divisions, optimization model selection, and node computing resource allocation are encoded as binary sequences respectively.
[0024] As a preferred solution of the parallel strategy optimization method for predicting the new energy power generation capacity described in the present invention, the fitness function includes performance indicators obtained according to application requirements and optimization goals, including parallel acceleration efficiency, throughput and response time. The fitness function requires maximizing the parallel acceleration efficiency and throughput and minimizing the response time. The expression is:
[0025]
[0026] Among them, Fitness_value is the fitness value, PAE is the parallel acceleration efficiency, T is the throughput, RT is the response time, α, β, and δ are the proportional coefficients of parallel acceleration efficiency, throughput, and response time, respectively, and α, β, and δ are all greater than 0.
[0027]
[0028] Among them, PAE min is the minimum acceptable value of parallel acceleration efficiency, γ is the growth rate of the adjustment scale factor, ε is the influence of control throughput on the scale factor, and RT opt is the optimal response time, k r is to increase the sensitivity of the response time to the proportionality factor.
[0029] Calculate the fitness value Fitness_value of each optimization strategy n , the optimization strategy is screened according to the fitness value by selecting the algorithm.
[0030] As a preferred solution of the parallel strategy optimization method for predicting the power generation capacity of new energy described in the present invention, the screening of the optimization strategies includes sorting all the obtained optimization strategies from large to small according to the fitness value, generating a strategy sorting table, selecting the same number of optimization strategies as 2α according to the positive order of the strategy sorting table, wherein 2α is twice the required number threshold, and calculating the area of the roulette sector region for the screened optimization strategies:
[0031]
[0032] Among them, Fitness_value n Represents the fitness value of each optimization strategy, N is the total number of all optimization strategies, and the area of the roulette sector of each optimization strategy is mapped to the virtual roulette, and the virtual roulette is started to rotate. When the virtual roulette stops rotating, the roulette pointer points to the optimization strategy corresponding to the sector area and is selected. When the optimization strategy is selected, it needs to be deleted from the virtual roulette, and the area of the roulette sector of the remaining optimization strategies is recalculated. When the number of selected optimization strategies is equal to α, the selection of optimization strategies is stopped.
[0033] As a preferred solution of the parallel strategy optimization method for predicting the power generation capacity of new energy described in the present invention, the strategy set includes stopping the selection of optimization strategies when the number of selected optimization strategies is equal to α, and performing crossover and mutation operations on the selected optimization strategies:
[0034] A random intersection is selected from two optimization strategies, the content before the intersection is retained, and the content after the intersection is swapped. A mutation operation is performed on the content of the optimization strategy after the intersection. A random position is selected for flipping and replacement. The optimization strategies that complete the selection, crossover, and mutation operations are used to establish a strategy set, and the fitness value of each optimization strategy in the strategy set is calculated again through the fitness function.
[0035] The output of all strategy sets includes repeated selection, crossover, and mutation operations for cyclic iteration. Each iteration generates a set of strategy sets. When stop condition 1 or stop condition 2 is met, the cyclic iteration is exited.
[0036] The stop condition 1 is to record the number of loop iterations when performing loop iterations, and to exit the loop iteration when the number of loop iterations is equal to the number threshold.
[0037] The stopping condition 2 is that when performing loop iteration, when there is at least one optimization strategy in the output strategy set whose fitness value is greater than the fitness threshold, the loop iteration is jumped out. At this time, it indicates that the optimization system has found an optimization strategy that meets the current new energy power generation capacity prediction calculation task and can converge in advance.
[0038] After jumping out of the loop iteration, all strategy sets are output, and all strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task.
[0039] As a preferred solution of the parallel strategy optimization method for predicting the power generation capacity of renewable energy described in the present invention, wherein: the average allocation of computing resources of the faulty nodes to the non-faulty nodes includes predicting the computing node state W(τ) during the optimization process of the optimization system optimizing the current power generation capacity prediction computing task based on the optimization strategy:
[0040]
[0041] Among them, τ is the prediction calculation time window, A is the total number of computing nodes, E a is the actual state value of the ath computing node, C a is the expected state value of the ath computing node, s is the sensitivity coefficient of node state prediction, U a is the timestamp of the ath computing node, Z(E a ,U a) is the node status and time complex information filtering function:
[0042]
[0043] B is the total number of failed nodes, is the normalization parameter, X b is the feature vector of the bth faulty node, Y b is the expected feature vector of the bth faulty node, Φ(X b ,Y b ) is the matching function between the fault node feature vector and the expected feature vector:
[0044]
[0045] Among them, θ is the sensitivity coefficient of state difference, is the average value of all computing node states, λ is the time decay coefficient, U current is the current timestamp, L is the dimension of the feature vector, X bl is the l-th dimension eigenvalue of the b-th faulty node, Y bl is the expected eigenvalue of the lth dimension of the bth faulty node.
[0046] As the W value increases, the system performance and the computing node status are better. As the W value decreases, the system performance and the computing node status are worse. At this time, according to the predicted status, the computing node status in the monitoring system can be detected in time.
[0047]
[0048] Among them, Q(O) is the comprehensive score of the computing node status, O is the total number of computing nodes, and t o is the detection time period, g o (t) is the performance index of the oth computing node at time t, is the average performance index of all computing nodes at time t, is the fault detection sensitivity coefficient.
[0049] According to the subtasks processed by the faulty node, determine the affected subtasks, count the number of all normal nodes except the faulty node, and The computing resources of the faulty nodes are evenly distributed to the normal nodes based on the result.
[0050] As a preferred scheme of a parallel strategy optimization method for predicting new energy power generation capacity described in the present invention, wherein: the data floating coefficient includes when the data flow increase rate of the node exceeds the data flow increase rate threshold, it indicates that the data flow of the node increases too fast. At this time, the computing load of the node increases, and the time period when the current data flow increase rate exceeds the data flow increase rate threshold is recorded as the data flow warning period.
[0051] When the data processing volume increase rate of a node exceeds the data processing volume increase rate threshold, it indicates that the data processing volume of the node increases too fast. At this time, the computing load of the node increases. The period when the current data processing volume increase rate exceeds the data processing volume increase rate threshold is recorded as the data processing volume warning period. The data flow warning period and the data processing volume warning period are integrated to obtain the data floating coefficient. The expression is:
[0052]
[0053] The larger the value of the data floating coefficient is, the longer the node's data flow, data processing volume, and data flow combined with data processing increase too fast. f is the data floating coefficient, F(t) is the calculated load change, [t x ,t y ] is the data traffic warning period, [t i ,t j ] is the data processing volume warning period.
[0054] The allocation scheduling includes coefficient thresholds including a first load threshold and a second load threshold, and the first load threshold is smaller than the second load threshold. The first load threshold is used to determine the idle state of the node, and the second load threshold is used to determine the load state of the node.
[0055] When the node data floating coefficient sj f When it is greater than the second load threshold, it is determined that the node calculation load is large, and the current node is classified into the first-level load set.
[0056] When the node data floating coefficient sj f When it is greater than the first load threshold and the data floating coefficient is less than or equal to the second load threshold, the node is judged to calculate the load balance and the current node is classified into the secondary load set.
[0057] When the node data floating coefficient sj f When it is less than or equal to the first load threshold, the node computing load is judged to be small, and the current node is classified into a three-level load set, among which the first-level load set is larger than the second-level load set, which is larger than the third-level load set. The nodes in the second-level load set do not need to reallocate computing resources, and the computing resources of the nodes in the third-level load set are allocated to the first-level load set.
[0058] The optimization system first obtains the maximum amount of allocated computing resources in the three-level load set, and then obtains the minimum amount of computing resources required in the first-level load set. If the maximum amount of allocated computing resources is greater than or equal to the minimum amount of computing resources required, the optimization system directly allocates the computing resources. If the maximum amount of allocated computing resources is less than the minimum amount of computing resources required, the optimization system sends a warning signal to the administrator, indicating that the computing resources are insufficient.
[0059] Another object of the present invention is to provide a parallel strategy optimization system for predicting renewable energy power generation capacity. The present invention automatically adjusts the parallel strategy of computing tasks through an intelligent optimization algorithm, thereby improving the efficiency and accuracy of renewable energy power generation capacity prediction, while achieving reasonable allocation and scheduling of computing resources, thereby improving the operating efficiency and stability of the entire renewable energy power generation system.
[0060] As a preferred solution of the parallel strategy optimization system for predicting the new energy power generation capacity described in the present invention, it is characterized by including a strategy generation module, an operation module, an iterative output module and a scheduling module.
[0061] The strategy generation module generates a matching optimization strategy based on the current new energy power generation capacity prediction calculation task through a random generation tool combined with a historical big database. The content of each optimization strategy includes the number of parallel subtasks, the optimization model selection and the node computing resource allocation. Among them, the number of parallel subtasks, the optimization model selection and the node computing resource allocation are processed through binary coding. Each content is reflected in the optimization strategy in the form of binary coding, and the content of the optimization strategy is different.
[0062] The operation module generates a fitness function according to application requirements and optimization objectives, calculates the fitness value of each optimization scheme according to the fitness function, screens the optimization strategy according to the fitness value through a selection algorithm, and when the selected optimization strategy meets the requirements, performs crossover and mutation operations on the selected optimization strategy, thereby increasing the randomness and diversity of the optimization strategy and avoiding the local optimal solution of the optimization system line. The optimization strategy that completes the selection, crossover and mutation operations is established into a strategy set, and the fitness value of each optimization strategy in the strategy set is calculated again through the fitness function.
[0063] The iterative output module repeats the selection, crossover and mutation operations to perform cyclic iterations, and generates a set of strategy sets in each iteration. When the stopping condition is met, the cyclic iteration is jumped out and all strategy sets are output, and all strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task. The optimization system ensures the performance of the optimization strategy by automatically searching for the global optimal optimization strategy.
[0064] The scheduling module, during the process of optimizing the current new energy power generation capacity prediction calculation task by the optimization system based on the optimization strategy, when any subtask computing node fails, distributes the computing resources of the failed node evenly to the non-faulty nodes, and monitors the data flow changes and data processing volume changes of all nodes in real time, generates a data floating coefficient based on the data flow changes and data processing volume changes of the nodes, analyzes the computing load status of the nodes through the comparison results of the data floating coefficient and the coefficient threshold, divides all nodes into sets according to the analysis results, and allocates and schedules the node computing resource allocation according to the set division results.
[0065] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of a method described in the parallel strategy optimization of new energy power generation capacity prediction are implemented.
[0066] A computer-readable storage medium stores a computer program thereon, characterized in that when the computer program is executed by a processor, the steps of a method described in the parallel strategy optimization for predicting new energy power generation capacity are implemented.
[0067] Beneficial effects of the present invention: The present invention calculates the fitness value of each optimization scheme according to the fitness function, selects several optimization strategies according to the fitness value through the selection algorithm, and after selecting a sufficient number of optimization strategies, performs crossover and mutation operations on the selected optimization strategies, thereby increasing the randomness and diversity of the optimization strategies, avoiding the local optimal solution of the optimization system line, and establishing a strategy set for the optimization strategies that complete the selection, crossover and mutation operations, and again calculates the fitness value of each optimization strategy in the strategy set through the fitness function, repeats the selection, crossover and mutation operations for cyclic iteration, and generates a set of strategy sets for each iteration. When the stop condition is met, the cyclic iteration is jumped out and all strategy sets are output, and all strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task; the optimization system automatically searches for the global optimal optimization strategy, which can effectively guarantee the performance of the optimization strategy. It can automatically find the global optimal optimization strategy based on the current new energy power generation capacity prediction calculation task in combination with the historical big database, reduce manual intervention, and guarantee the performance of the optimization strategy.
[0068] In the process of optimizing the current new energy power generation capacity prediction calculation task based on the optimization strategy of the optimization system of the present invention, when a certain subtask calculation node fails, the calculation resources of the failed node are evenly distributed to other nodes, and the data flow changes and data processing volume changes of all nodes are monitored in real time, and the data floating coefficient is generated according to the data flow changes and data processing volume changes of the nodes, and the calculation load status of the nodes is analyzed by comparing the data floating coefficient with the coefficient threshold, and all nodes are divided into sets according to the analysis results, and the node calculation resource allocation is allocated and scheduled according to the set division results. The optimization strategy is globally or locally adaptively scheduled according to the real-time equipment status and environment, so as to improve the robustness and adaptive performance of the optimization system. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:
[0070] Figure 1 An overall flow chart of a parallel strategy optimization method for predicting renewable energy power generation capacity provided by an embodiment of the present invention.
[0071] Figure 2 A system solution flow chart of a parallel strategy optimization system for predicting new energy power generation capacity provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0072] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.
[0073] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0074] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is mutually exclusive with other embodiments, either individually or selectively.
[0075] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.
[0076] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.
[0077] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0078] Example 1
[0079] Reference Figure 1 , which is the first embodiment of the present invention, and provides a parallel strategy optimization method for predicting the power generation capacity of new energy, including:
[0080] S1: The optimization system generates several optimization strategies based on the current renewable energy power generation capacity prediction calculation task through random generation tools combined with historical big data database. The content of each optimization strategy includes the number of parallel subtasks, optimization model selection and node computing resource allocation. The number of parallel subtasks, optimization model selection and node computing resource allocation are processed through binary coding. Each content is reflected in the optimization strategy in the form of binary coding. The contents of several optimization strategies are different, which can be completely different or partially different, and the number of generated optimization strategies needs to be sufficient to cover the diversity of the solution space.
[0081] Get the constraints for computing task optimization, including the maximum and minimum number of subtask divisions, the type of prediction model, and the maximum allocation of computing resources:
[0082] maximize(orminimize) f(D,M′)
[0083] subject to W min ≤|D|≤W max
[0084]
[0085] D={D1,D2,...,D |D|}
[0086]
[0087] Where D is the total computing task, D d is the dth subtask, W min and W max is the minimum and maximum number of subtask divisions, M is the set of prediction models, M m is the mth prediction model, R max is the maximum allocation of computing resources, f(D,M') is the subtask D d Using Model M m The amount of computing resources required.
[0088] Analyze historical data to understand the characteristics, data distribution, computing load, etc. of the renewable energy power generation capacity prediction task, and use random generation tools to generate multiple optimization strategies. The optimization strategies include the number of parallel subtasks, optimization model selection, and node computing resource allocation.
[0089] The optimization strategy includes dividing the number of parallel subtasks under the constraints, optimizing the model selection and allocating node computing resources, establishing four parent optimization strategies, and then establishing sub-optimization strategies in turn according to the parent optimization strategies.
[0090] The four parent optimization strategies include: optimization strategy 1 is to divide into three parallel subtasks, select a random forest optimization model, and evenly distribute computing resources to each subtask.
[0091] Optimization strategy 2 is to divide the task into five parallel subtasks, select the neural network optimization model, and allocate more computing resources to the intermediate nodes.
[0092] Optimization strategy 3 is to divide the task into four parallel subtasks, select support vector machine optimization, allocate most of the computing resources to subtasks with higher prediction accuracy, and allocate a small amount of resources to other subtasks.
[0093] Optimization strategy 4 is to divide the task into two parallel subtasks, select linear regression optimization model selection, and allocate all computing resources to a single subtask.
[0094] The number of parallel subtasks, the optimization model selection, and the node computing resource allocation are encoded as binary sequences respectively; for example, for the number of subtasks, a fixed-length binary number can be used to represent it, such as 3 subtasks can be encoded as "011"; for the optimization model selection, binary encoding can be used to represent different model selections, such as the random forest model can be encoded as "001", the neural network model can be encoded as "010", etc.; for the node computing resource allocation, binary encoding can be used to represent different allocation schemes, such as evenly allocating computing resources to each subtask can be encoded as "001", allocating more resources to intermediate nodes can be encoded as "010", etc.
[0095] The relevant Python code for binary encoding is as follows:
[0096] defencode_parallel_tasks(num_tasks,total_tasks):#Encode the number of subtasks into a fixed-length binary number
[0097] binary_encoding=bin(num_tasks)[2:].zfill(total_tasks)
[0098] return binary_encoding
[0099] defencode_model_selection(model):#Encode the optimization model selection as a binary sequence
[0100] model_encoding = {'random_forest':'001','neural_network':'010','svm':'011'} # Example model encoding mapping
[0101] return model_encoding.get(model,'000')#If the model is unknown, encode it as a sequence of all zeros def encode_resource_allocation(resource_allocation):#Encode node computing resource allocation as a binary sequence
[0102] resource_encoding = {'even':'001','more_to_middle':'010'}# Example resource allocation encoding mapping
[0103] return resource_encoding.get(resource_allocation,'000')#If the resource allocation is unknown, the encoding is a sequence of all zeros
[0104] #Example: Encode the number of 3 subtask divisions into a binary sequence
[0105] num_tasks=3
[0106] total_tasks=3
[0107] encoded_tasks=encode_parallel_tasks(num_tasks,total_tasks)
[0108] print("Encoded parallel tasks:",encoded_tasks)
[0109] #Example: Encoding a Random Forest Model as a Binary Sequence
[0110] model = 'random_forest'
[0111] encoded_model=encode_model_selection(model)
[0112] print("Encoded model selection:",encoded_model)
[0113] #Example: Allocate more resources to intermediate nodes encoded as binary sequence
[0114] resource_allocation='more_to_middle'
[0115] encoded_resource=encode_resource_allocation(resource_allocation)
[0116] print("Encoded resource allocation:",encoded_resource).
[0117] Each optimization strategy is represented by a binary sequence and includes the encoding of the number of subtask divisions, optimization model selection, and node computing resource allocation.
[0118] S2: Generate a fitness function according to application requirements and optimization goals, calculate the fitness value of each optimization scheme based on the fitness function, and obtain performance indicators including parallel acceleration efficiency, throughput, and response time according to application requirements and optimization goals.
[0119] When optimizing the prediction and calculation tasks of renewable energy power generation capacity, these indicators are often used to evaluate the effect and performance of the optimization strategy. Improving parallel acceleration efficiency can speed up the prediction calculation, improving throughput can increase the system processing capacity, and shortening the response time can improve the real-time performance and user experience of the system.
[0120] The performance indicators obtained according to application requirements and optimization goals include parallel acceleration efficiency, throughput, and response time. The fitness function requires maximizing parallel acceleration efficiency and throughput and minimizing response time. The expression is:
[0121]
[0122] Among them, Fitness_value is the fitness value, PAE is the parallel acceleration efficiency, T is the throughput, RT is the response time, α, β, and δ are the proportional coefficients of parallel acceleration efficiency, throughput, and response time, respectively, and α, β, and δ are all greater than 0.
[0123]
[0124] Among them, PAE min is the minimum acceptable value of parallel acceleration efficiency, γ is the growth rate of the adjustment scale factor, ε is the influence of control throughput on the scale factor, and RT opt is the optimal response time, k r is to increase the sensitivity of the response time to the proportionality factor.
[0125] Calculate the fitness value Fitness_value of each optimization strategy n , the optimization strategy is screened according to the fitness value by selecting the algorithm.
[0126] S3: Screening several optimization strategies according to the fitness value through the selection algorithm, including the following steps:
[0127] In this application, the selection algorithm includes a roulette wheel selection algorithm and a sorting selection algorithm.
[0128] Sort all the optimization strategies obtained by the fitness value from large to small, generate a strategy sorting table, select the same number of optimization strategies as 2α according to the positive order of the strategy sorting table, where 2α is twice the required number threshold, and calculate the sector area of the roulette wheel for the selected optimization strategies:
[0129]
[0130] Among them, Fitness_value n Represents the fitness value of each optimization strategy, N is the total number of all optimization strategies, and the area of the roulette sector of each optimization strategy is mapped to the virtual roulette, and the virtual roulette is started to rotate. When the virtual roulette stops rotating, the roulette pointer points to the optimization strategy corresponding to the sector area and is selected. When the optimization strategy is selected, it needs to be deleted from the virtual roulette, and the area of the roulette sector of the remaining optimization strategies is recalculated. When the number of selected optimization strategies is equal to α, the selection of optimization strategies is stopped.
[0131] The process of creating a virtual roulette wheel based on the fitness value and selecting the optimization strategy based on the roulette wheel is implemented through Python. In this example, it is assumed that the fitness value of each optimization strategy is stored in the list fitness_values, and the threshold of the number of optimization strategies is num_selected_policies;
[0132] import random
[0133] def select_policy_with_roulette(fitness_values,num_selected_policies):#Calculate the sum of fitness values
[0134] total_fitness = sum(fitness_values) # Calculate the roulette sector area for each optimization strategy
[0135] sectors=[fitness_value / total_fitness for fitness_value in fitness_values]
[0136] selected_policies = [] # Repeatedly select optimization strategies until the quantity threshold is reached
[0137] while len(selected_policies) <num_selected_policies:
[0138] #Create a virtual roulette wheel
[0139] virtual_roulette = []
[0140] for idx,sector in enumerate(sectors):
[0141] virtual_roulette.extend([idx]*int(sector*100))#The sector area is mapped to the virtual roulette
[0142] #Virtual roulette wheel spinning
[0143] selected_index=random.choice(virtual_roulette)
[0144] selected_policy = selected_index #Assume that the index of the optimization policy is consistent with the list index
[0145] selected_policies.append(selected_policy)
[0146] #Delete the selected optimization strategy from the virtual roulette wheel
[0147] remaining_sectors=sectors[:selected_index]+sectors[selected_index+1:]
[0148] total_remaining_fitness=sum(remaining_sectors)
[0149] sectors=[sector / total_remaining_fitness for sectorin remaining_sectors]
[0150] return selected_policies
[0151] #Example usage
[0152] fitness_values = [0.5, 0.3, 0.7, 0.2] # Example fitness values
[0153] num_selected_policies = 3 # The threshold for the number of selected optimization policies
[0154] selected_policies=select_policy_with_roulette(fitness_values,num_selected_policies)
[0155] print("Selected policies:",selected_policies).
[0156] All optimization strategies are sorted according to the fitness value through Python, and the optimization strategy with the same quantity threshold is selected in the sorted order. The code is as follows:
[0157] def select_policies_by_rank(fitness_values,num_selected_policies):#Create a strategy ranking table
[0158] sorted_policies = sorted(range(len(fitness_values)), key = lambda k:fitness_values[k], reverse = True) # Select the same optimization strategy as the quantity threshold
[0159] selected_policies=sorted_policies[:num_selected_policies]
[0160] return selected_policies
[0161] #Example usage
[0162] fitness_values = [0.5, 0.3, 0.7, 0.2] # Example fitness values
[0163] num_selected_policies = 3 # The threshold for the number of selected optimization policies
[0164] selected_policies=select_policies_by_rank(fitness_values,num_selected_policies)
[0165] print("Selected policies:",selected_policies).
[0166] S4: After a sufficient number of optimization strategies are selected, crossover and mutation operations are performed on the selected optimization strategies to increase the randomness and diversity of the optimization strategies and avoid the local optimal solution of the optimization system lines. The optimization strategies that have completed the selection, crossover and mutation operations are used to establish a strategy set, and the fitness value of each optimization strategy in the strategy set is calculated again through the fitness function.
[0167] A crossover point is randomly selected from two optimization strategies. The content before the crossover point is retained, and the content after the crossover point is swapped. For example, assuming that the optimization strategy of parent 1 is "011-001-010" and the optimization strategy of parent 2 is "101-010-001", after randomly selecting the crossover point, you can get child 1 "011-010-001" and child 2 "101-001-010";
[0168] Perform mutation operations on the optimization strategy content completed by crossover, such as randomly selecting a position for flipping or replacement. For example, assuming that the optimization strategy of the parent generation is "011-001-010", after randomly selecting a position for mutation, the offspring "011-001-011" or "011-000-010" can be obtained;
[0169] The code for crossover and mutation operations on the selected optimization strategy through Python is as follows:
[0170] import random
[0171] defcrossover(parent1,parent2):#Randomly select crossover points
[0172] crossover_point = random.randint(1,len(parent1)-1) # Perform crossover operation
[0173] child1=parent1[:crossover_point]+parent2[crossover_point:]
[0174] child2=parent2[:crossover_point]+parent1[crossover_point:]
[0175] return child1,child2
[0176] defmutation(parent):#Randomly select a position for mutation operation
[0177] mutation_point = random.randint(0, len(parent) - 1) # Perform mutation operation
[0178] mutated_gene = '0' if parent[mutation_point] == '1' else '1'
[0179] child = parent[:mutation_point] + mutated_gene + parent[mutation_point + 1:]
[0180] return child
[0181] # Example
[0182] parent1 = "011-001-010"
[0183] parent2 = "101-010-001"
[0184] mutation_parent = "011-001-010"
[0185] child1, child2 = crossover(parent1, parent2)
[0186] mutated_child = mutation(mutation_parent)
[0187] print("Crossover result:")
[0188] print("Child 1:", child1)
[0189] print("Child 2:", child2)
[0190] print("\nMutation result:")
[0191] print("Mutated child:", mutated_child)。
[0192] S5: Repeat the selection, crossover, and mutation operations to perform loop iterations. Each iteration generates a set of strategy sets. When the stopping condition is met, the loop iteration is exited and all strategy sets are output. All strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task; the optimization system automatically searches for the global optimal optimization strategy, which can effectively ensure the performance of the optimization strategy.
[0193] Meeting the stopping conditions includes reaching the target number of iterations or reaching the target fitness value in the continuously generated optimization strategies;
[0194] Stop condition 1: When performing loop iterations, record the number of loop iterations. When the number of loop iterations is equal to the number threshold, jump out of the loop iteration. The number of loop iterations is used as a benchmark to determine whether to stop the loop iteration. This can effectively prevent the optimization system from generating too many optimization strategies, which will reduce the calculation efficiency of the optimization system and reduce the operation burden of the optimization system.
[0195] Stop condition 2: During the loop iteration, when there is at least one optimization strategy in the output strategy set whose fitness value is greater than the fitness threshold, the loop iteration is exited. This indicates that the optimization system has found an optimization strategy that meets the current renewable energy power generation capacity prediction calculation task and can converge in advance.
[0196] After jumping out of the loop iteration, all strategy sets are output, and all strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task.
[0197] S6: When the optimization system optimizes the current renewable energy power generation capacity prediction calculation task based on the optimization strategy, when a subtask computing node fails, the computing resources of the failed node are evenly distributed to other nodes.
[0198] In the process of optimizing the current new energy power generation capacity prediction calculation task based on the optimization strategy, the prediction calculation node state W(τ) is:
[0199]
[0200] Among them, τ is the prediction calculation time window, A is the total number of computing nodes, E a is the actual state value of the ath computing node, C a is the expected state value of the ath computing node, s is the sensitivity coefficient of node state prediction, U a is the timestamp of the ath computing node, Z(E a ,U a) is the node status and time complex information filtering function:
[0201]
[0202] B is the total number of faulty nodes, ρ is the normalization parameter, X b is the feature vector of the bth faulty node, Y b is the expected feature vector of the bth faulty node, Φ(X b ,Y b ) is the matching function between the fault node feature vector and the expected feature vector:
[0203]
[0204] Among them, θ is the sensitivity coefficient of state difference, is the average value of all computing node states, λ is the time decay coefficient, U current is the current timestamp, L is the dimension of the feature vector, X bl is the l-th dimension eigenvalue of the b-th faulty node, Y bl is the expected eigenvalue of the lth dimension of the bth faulty node.
[0205] As the W value increases, the system performance and the computing node status are better. As the W value decreases, the system performance and the computing node status are worse. At this time, according to the predicted status, the computing node status in the monitoring system can be detected in time.
[0206]
[0207] Among them, Q(O) is the comprehensive score of the computing node status, O is the total number of computing nodes, and t o is the detection time period, g o (t) is the performance index of the oth computing node at time t, is the average performance index of all computing nodes at time t, is the fault detection sensitivity coefficient.
[0208] According to the subtasks processed by the faulty node, determine the affected subtasks, count the number of all normal nodes except the faulty node, and The computing resources of the faulty nodes are evenly distributed to the normal nodes based on the result.
[0209] For example, suppose there are three computing nodes in the system, namely Node A, Node B and Node C, which are responsible for processing three subtasks. The optimization strategy is "3-010-001", which means that the system is divided into three subtasks, the computing node state W(τ) is predicted, and the node computing resource allocation is to evenly distribute resources to each subtask;
[0210] Suddenly, a computing node failure occurs on Node B, causing Node B to be unable to continue processing the subtasks it is responsible for. In this case, the system needs to perform the following steps;
[0211] The monitoring system discovers that a computing node failure has occurred on node B and determines that the affected subtask is the subtask that node B is responsible for. The system counts the number of all normal nodes except the faulty node as 2, and evenly distributes the computing resources of the faulty node B to other normal nodes, that is, evenly distributes the resources of node B to node A and node C, and reschedules the affected subtask from the faulty node B to other nodes, that is, reallocates the subtask that node B is responsible for to node A and node C for processing.
[0212] S7: Monitor the data flow changes and data processing volume changes of all nodes in real time, and generate data floating coefficients according to the data flow changes and data processing volume changes of the nodes.
[0213] When the data traffic increase rate of a node exceeds the data traffic increase rate threshold, it indicates that the data traffic of the node increases too fast. At this time, the computing load of the node increases. Therefore, the period when the data traffic increase rate exceeds the data traffic increase rate threshold is recorded as the data traffic warning period.
[0214] When the data processing volume increase rate of a node exceeds the data processing volume increase rate threshold, it indicates that the data processing volume of the node increases too fast. At this time, the computing load of the node increases. Therefore, the period when the data processing volume increase rate exceeds the data processing volume increase rate threshold is recorded as the data processing volume warning period.
[0215] The increase in data sources and the access to new sensors, devices or systems will increase the amount of data received by the node, causing the data traffic to increase too quickly. The increase in data sampling rate: if the data is sampled more frequently, the data update rate will increase, causing the data traffic received by the node to increase too quickly. The complexity of the data processing algorithm increases: as the complexity of the data processing algorithm increases, such as the use of more feature extraction, data cleaning, model training and other operations, the amount of data processing at the node will increase too quickly. Data quality issues: when the data quality is poor, more data processing work may be required to clean, repair or filter the data, which will increase the amount of data processing at the node. System load increases: if the load of other nodes or services in the system increases, such as due to the expansion of the system scale or the increase in users, the amount of data processed by the node will also increase. Network topology changes: when the network topology changes, such as the connection relationship between network nodes, changes in the transmission path, etc., the data transmission path and traffic distribution may be affected, causing the data traffic at the node to increase too quickly. Data burst events: burst events, such as anomaly detection, fault events, etc., may cause a sudden large amount of data to be generated or processed, causing the data traffic and data processing volume of the node to increase too quickly.
[0216] Perform integral calculation on the data flow warning period and the data processing volume warning period to obtain the data floating coefficient. The expression is:
[0217]
[0218] The larger the value of the data floating coefficient is, the longer the node's data flow, data processing volume, and data flow combined with data processing increase too fast. f is the data floating coefficient, F(t) is the calculated load change, [t x ,t y ] is the data traffic warning period, [t i ,t j ] is the data processing volume warning period.
[0219] The allocation scheduling includes coefficient thresholds including a first load threshold and a second load threshold, and the first load threshold is smaller than the second load threshold. The first load threshold is used to determine the idle state of the node, and the second load threshold is used to determine the load state of the node.
[0220] When the node data floating coefficient sj f When it is greater than the second load threshold, it is determined that the node calculation load is large, and the current node is classified into the first-level load set.
[0221] When the node data floating coefficient sj fWhen it is greater than the first load threshold and the data floating coefficient is less than or equal to the second load threshold, the node is judged to calculate the load balance and the current node is classified into the secondary load set.
[0222] When the node data floating coefficient sj f When it is less than or equal to the first load threshold, the node computing load is judged to be small, and the current node is classified into a three-level load set, among which the first-level load set is larger than the second-level load set, which is larger than the third-level load set. The nodes in the second-level load set do not need to reallocate computing resources, and the computing resources of the nodes in the third-level load set are allocated to the first-level load set.
[0223] The optimization system first obtains the maximum amount of allocated computing resources in the three-level load set, and then obtains the minimum amount of computing resources required in the first-level load set. If the maximum amount of allocated computing resources is greater than or equal to the minimum amount of computing resources required, the optimization system directly allocates the computing resources. If the maximum amount of allocated computing resources is less than the minimum amount of computing resources required, the optimization system sends a warning signal to the administrator, indicating that the computing resources are insufficient. At this time, the administrator can manage it through the following methods:
[0224] Administrators increase the computing resources of the system by adding physical servers, virtual machine instances or cloud computing resources, which may include purchasing new hardware equipment, expanding the scale or configuration of cloud computing instances, etc. Administrators improve the utilization efficiency of computing resources by optimizing the resource utilization of the system, which may include adjusting the system's load balancing strategy, optimizing the task scheduling algorithm, and reasonably allocating task priorities. Administrators adjust the priority of tasks based on the importance and urgency of the tasks to ensure that critical tasks can obtain sufficient computing resources in a timely manner and improve the responsiveness and stability of the system. Administrators temporarily or permanently restrict non-essential tasks or services in the system to release computing resources for critical tasks, which can be achieved by pausing or stopping certain services, reducing the system load, etc. Administrators perform performance tuning on the system, including adjusting system parameters, optimizing software configuration, updating software versions, etc., to improve the computing performance and efficiency of the system, thereby reducing the occurrence of resource shortages. Administrators dynamically adjust the allocation of computing resources based on the real-time situation of the system, such as automatically expanding or shrinking computing resources based on load conditions to meet the current needs of the system.
[0225] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
[0226] Example 2
[0227] The third embodiment of the present invention is different from the first two embodiments in that:
[0228] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.
[0229] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.
[0230] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.
[0231] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0232] Example 3
[0233] Reference Figure 2 , which is the fourth embodiment of the present invention, and this embodiment provides a parallel strategy optimization system for new energy power generation capacity prediction application, including a strategy generation module, an operation module, an iterative output module, and a scheduling module.
[0234] The strategy generation module generates several optimization strategies based on the current renewable energy power generation capacity prediction calculation task through random generation tools combined with the historical big data database. The content of each optimization strategy includes the number of parallel subtasks, optimization model selection and node computing resource allocation. Among them, the number of parallel subtasks, optimization model selection and node computing resource allocation are processed through binary coding. Each content is reflected in the optimization strategy in the form of binary coding. The contents of several optimization strategies are different, which can be completely different or partially different, and the number of generated optimization strategies needs to be sufficient to cover the diversity of the solution space.
[0235] The operation module generates a fitness function according to application requirements and optimization objectives, calculates the fitness value of each optimization scheme based on the fitness function, and screens several optimization strategies based on the fitness value through the selection algorithm. After a sufficient number of optimization strategies are selected, the selected optimization strategies are crossover and mutation operations are performed to increase the randomness and diversity of the optimization strategies, avoid the local optimal solution of the optimization system line, and establish a strategy set for the optimization strategies that have completed the selection, crossover and mutation operations, and calculate the fitness value of each optimization strategy in the strategy set again through the fitness function.
[0236] The iterative output module repeats the selection, crossover and mutation operations for cyclic iteration. Each iteration generates a set of strategy sets. When the stopping condition is met, it jumps out of the cyclic iteration and outputs all strategy sets. All strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task; the optimization system automatically searches for the global optimal optimization strategy, which can effectively ensure the performance of the optimization strategy.
[0237] In the process of optimizing the current renewable energy power generation capacity prediction calculation task based on the optimization strategy of the optimization system, when a subtask computing node fails, the scheduling module will evenly distribute the computing resources of the failed node to other nodes, and monitor the data flow changes and data processing volume changes of all nodes in real time. According to the data flow changes and data processing volume changes of the nodes, the data floating coefficient is generated, and the computing load status of the node is analyzed by comparing the data floating coefficient with the coefficient threshold. According to the analysis results, all nodes are divided into sets, and the node computing resource allocation is allocated and scheduled according to the set division results.
[0238] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A parallel strategy optimization method for predicting renewable energy power generation capacity, characterized by: include, The optimization system generates matching optimization strategies based on the current renewable energy power generation capacity prediction calculation task through random generation tools combined with the historical big database; Generate a fitness function based on application requirements and optimization goals and calculate the fitness value of each optimization solution. Use the selection algorithm to screen the optimization strategies based on the fitness value. After the screening, perform crossover and mutation operations on the selected optimization strategies and establish a strategy set, and calculate the fitness value again. Repeat the selection, crossover, and mutation operations to perform loop iterations. Each iteration generates a set of strategy sets. When the stop condition is met, the loop iteration is exited and all strategy sets are output. All strategy sets are integrated into a strategy list, and the optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task; In the process of optimizing the current renewable energy power generation capacity prediction calculation task based on the optimization strategy, when any subtask computing node fails, the computing resources of the failed node are evenly allocated to the non-faulty nodes; Monitor the changes in data flow and data processing volume of all nodes in real time, generate data floating coefficients based on the changes in data flow and data processing volume of the nodes, analyze the computing load status of the nodes by comparing the data floating coefficients with the coefficient thresholds, divide all nodes into sets based on the analysis results, and allocate and schedule node computing resources based on the set division results.
2. A parallel strategy optimization method for predicting renewable energy power generation capacity according to claim 1, characterized in that: The matching optimization strategy includes generating matching optimization strategies by combining a random generation tool with a historical big database, wherein the content of each optimization strategy includes the number of parallel subtasks, the selection of an optimization model, and the allocation of node computing resources, wherein the number of parallel subtasks, the selection of an optimization model, and the allocation of node computing resources are processed by binary coding, and each content is embodied in the optimization strategy in the form of binary coding, and the content of the matching optimization strategies is different; Get the constraints for computing task optimization, including the maximum and minimum number of subtask divisions, the type of prediction model, and the maximum allocation of computing resources: maximize(orminimize) f(D,M) subject to W min ≤|D|≤W max D={D1,D2,...,D |D| } Where D is the total computing task, D d is the dth subtask, W min and W max is the minimum and maximum number of subtask divisions, M is the set of prediction models, M m is the mth prediction model, R max is the maximum allocation of computing resources, f(D,M') is the subtask D d Using Model M m The amount of computing resources required; The optimization strategy includes the division of the number of parallel subtasks, the selection of optimization models and the allocation of node computing resources under the constraints, the establishment of four parent optimization strategies, and the establishment of sub-optimization strategies in turn according to the parent optimization strategies; The four parent optimization strategies include: optimization strategy 1 is to divide into three parallel subtasks, select a random forest optimization model, and evenly allocate computing resources to each subtask; Optimization strategy 2 is to divide the task into five parallel subtasks, select the neural network optimization model, and allocate more computing resources to the intermediate nodes; Optimization strategy 3 is to divide the task into four parallel subtasks, select support vector machine optimization, allocate most of the computing resources to the subtasks with higher prediction accuracy, and allocate a small amount of resources to other subtasks; Optimization strategy 4 is to divide the task into two parallel subtasks, select the linear regression optimization model selection, and allocate all computing resources to a single subtask; The number of parallel subtask divisions, optimization model selection, and node computing resource allocation are encoded as binary sequences respectively.
3. A parallel strategy optimization method for predicting renewable energy power generation capacity according to claim 2, characterized in that: The fitness function includes performance indicators obtained according to application requirements and optimization goals, including parallel acceleration efficiency, throughput and response time. The fitness function requires maximizing parallel acceleration efficiency and throughput and minimizing response time. The expression is: Among them, Fitness_value is the fitness value, PAE is the parallel acceleration efficiency, T is the throughput, RT is the response time, α, β, δ are the proportional coefficients of parallel acceleration efficiency, throughput, and response time, respectively, and α, β, δ are all greater than 0; Among them, PAE min is the minimum acceptable value of parallel acceleration efficiency, γ is the growth rate of the adjustment scale factor, ε is the influence of control throughput on the scale factor, and RT opt is the optimal response time, k r is to increase the sensitivity of the response time to the proportionality factor; Calculate the fitness value Fitness_value of each optimization strategy n , the optimization strategy is screened according to the fitness value by selecting the algorithm.
4. A parallel strategy optimization method for predicting new energy power generation capacity according to claim 3, characterized in that: The screening of the optimization strategies includes sorting all the obtained optimization strategies from large to small according to the fitness value, generating a strategy sorting table, selecting the same number of optimization strategies as 2α according to the positive order of the strategy sorting table, wherein 2α is twice the required number threshold, and calculating the area of the roulette sector region for the screened optimization strategies: Among them, Fitness_value n Represents the fitness value of each optimization strategy, N is the total number of all optimization strategies, and the area of the roulette sector of each optimization strategy is mapped to the virtual roulette, and the virtual roulette is started to rotate. When the virtual roulette stops rotating, the roulette pointer points to the optimization strategy corresponding to the sector area and is selected. When the optimization strategy is selected, it needs to be deleted from the virtual roulette, and the area of the roulette sector of the remaining optimization strategies is recalculated. When the number of selected optimization strategies is equal to α, the selection of optimization strategies is stopped.
5. A parallel strategy optimization method for predicting renewable energy power generation capacity according to claim 4, characterized in that: The strategy set includes stopping the selection of optimization strategies when the number of selected optimization strategies is equal to α, and performing crossover and mutation operations on the selected optimization strategies: Randomly select an intersection point from two optimization strategies, keep the content before the intersection point, swap the content after the intersection point, perform mutation operation on the optimization strategy content after the intersection, randomly select a position to flip and replace, establish a strategy set for the optimization strategies that have completed the selection, crossover, and mutation operations, and calculate the fitness value of each optimization strategy in the strategy set again through the fitness function; The output of all strategy sets includes repeated selection, crossover, and mutation operations for cyclic iteration. Each iteration generates a set of strategy sets. When stop condition 1 or stop condition 2 is met, the cyclic iteration is exited. The stop condition 1 is to record the number of loop iterations when performing loop iterations, and to exit the loop iteration when the number of loop iterations is equal to the number threshold; The stop condition 2 is that when performing loop iteration, when there is at least one optimization strategy in the output strategy set whose fitness value is greater than the fitness threshold, the loop iteration is jumped out, which indicates that the optimization system has found an optimization strategy that meets the current new energy power generation capacity prediction calculation task and can converge in advance; After jumping out of the loop iteration, all strategy sets are output, and all strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task.
6. A parallel strategy optimization method for predicting renewable energy power generation capacity according to claim 5, characterized in that: The evenly allocating the computing resources of the faulty nodes to the non-faulty nodes includes predicting the computing node state W(τ) during the optimization process of the optimization system optimizing the current new energy power generation capacity prediction computing task based on the optimization strategy: Among them, τ is the prediction calculation time window, A is the total number of computing nodes, E a is the actual state value of the ath computing node, C a is the expected state value of the ath computing node, s is the sensitivity coefficient of node state prediction, U a is the timestamp of the ath computing node, Z(E a ,U a ) is the node status and time complex information filtering function: B is the total number of faulty nodes, ρ is the normalization parameter, X b is the feature vector of the bth faulty node, Y b is the expected feature vector of the bth faulty node, Φ(X b ,Y b ) is the matching function between the fault node feature vector and the expected feature vector: Among them, θ is the sensitivity coefficient of state difference, is the average value of all computing node states, λ is the time decay coefficient, U current is the current timestamp, L is the dimension of the feature vector, X bl is the l-th dimension eigenvalue of the b-th faulty node, Y bl is the expected eigenvalue of the lth dimension of the bth faulty node; As the W value increases, the system performance and the computing node status are better. As the W value decreases, the system performance and the computing node status are worse. At this time, according to the predicted status, the computing node status in the monitoring system can be detected in time. Among them, Q(O) is the comprehensive score of the computing node status, O is the total number of computing nodes, and t o is the detection time period, g o (t) is the performance index of the oth computing node at time t, is the average performance index of all computing nodes at time t, is the fault detection sensitivity coefficient; According to the subtasks processed by the faulty node, determine the affected subtasks, count the number of all normal nodes except the faulty node, and The computing resources of the faulty nodes are evenly distributed to the normal nodes based on the result.
7. A parallel strategy optimization method for predicting renewable energy power generation capacity according to claim 6, characterized in that: The data floating coefficient includes when the data flow increase rate of the node exceeds the data flow increase rate threshold, it indicates that the data flow of the node increases too fast, at this time the computing load of the node increases, and the period when the current data flow increase rate exceeds the data flow increase rate threshold is recorded as the data flow warning period; When the data processing volume increase rate of a node exceeds the data processing volume increase rate threshold, it indicates that the data processing volume of the node increases too fast. At this time, the computing load of the node increases. The period when the current data processing volume increase rate exceeds the data processing volume increase rate threshold is recorded as the data processing volume warning period. The data flow warning period and the data processing volume warning period are integrated to obtain the data floating coefficient. The expression is: The larger the value of the data floating coefficient is, the longer the node's data flow, data processing volume, and data flow combined with data processing increase too fast. f is the data floating coefficient, F(t) is the calculated load change, [t x ,t y ] is the data traffic warning period, [t i ,t j ] is the data processing volume warning period; The allocation scheduling includes a coefficient threshold including a first load threshold and a second load threshold, and the first load threshold is less than the second load threshold, the first load threshold is used to judge the idle state of the node, and the second load threshold is used to judge the load state of the node; When the node data floating coefficient sj f When the load is greater than the second load threshold, the node calculation load is judged to be large, and the current node is classified into the first-level load set; When the node data floating coefficient sj f When the load is greater than the first load threshold and the data floating coefficient is less than or equal to the second load threshold, the node is judged to calculate the load balance and the current node is classified into the secondary load set; When the node data floating coefficient sj f When the load is less than or equal to the first load threshold, the node computing load is judged to be small, and the current node is classified into a three-level load set, wherein the first-level load set is larger than the second-level load set, which is larger than the third-level load set. The nodes in the second-level load set do not need to re-allocate computing resources, and the computing resources of the nodes in the third-level load set are allocated to the first-level load set; The optimization system first obtains the maximum amount of allocated computing resources in the three-level load set, and then obtains the minimum amount of computing resources required in the first-level load set. If the maximum amount of allocated computing resources is greater than or equal to the minimum amount of computing resources required, the optimization system directly allocates the computing resources. If the maximum amount of allocated computing resources is less than the minimum amount of computing resources required, the optimization system sends a warning signal to the administrator, indicating that the computing resources are insufficient.
8. A system using a parallel strategy optimization method for predicting new energy power generation capacity as claimed in any one of claims 1 to 7, characterized in that: It includes strategy generation module, operation module, iteration output module and scheduling module; The strategy generation module generates matching optimization strategies based on the current new energy power generation capacity prediction calculation task through a random generation tool combined with a historical big database. The content of each optimization strategy includes the number of parallel subtasks, the optimization model selection, and the node computing resource allocation. The number of parallel subtasks, the optimization model selection, and the node computing resource allocation are processed by binary coding. Each content is embodied in the optimization strategy in the form of binary coding, and the content of the optimization strategy is different. The operation module generates a fitness function according to application requirements and optimization objectives, calculates the fitness value of each optimization scheme according to the fitness function, screens the optimization strategy according to the fitness value through a selection algorithm, and when the selected optimization strategy meets the requirements, performs crossover and mutation operations on the selected optimization strategy, thereby increasing the randomness and diversity of the optimization strategy, avoiding the local optimal solution of the optimization system line, establishing a strategy set with the optimization strategy that completes the selection, crossover and mutation operations, and again calculates the fitness value of each optimization strategy in the strategy set through the fitness function; The iterative output module repeats the selection, crossover, and mutation operations to perform cyclic iterations, and generates a set of strategy sets in each iteration. When the stop condition is met, the cyclic iteration is jumped out and all strategy sets are output, and all strategy sets are integrated into a strategy list. The optimization system selects the optimization strategy with the largest fitness value in the strategy list as the optimization strategy for the current new energy power generation capacity prediction calculation task. The optimization system automatically searches for the global optimal optimization strategy to ensure the performance of the optimization strategy; The scheduling module, during the process of optimizing the current new energy power generation capacity prediction calculation task by the optimization system based on the optimization strategy, when any subtask computing node fails, distributes the computing resources of the failed node evenly to the non-faulty nodes, and monitors the data flow changes and data processing volume changes of all nodes in real time, generates a data floating coefficient based on the data flow changes and data processing volume changes of the nodes, analyzes the computing load status of the nodes through the comparison results of the data floating coefficient and the coefficient threshold, divides all nodes into sets according to the analysis results, and allocates and schedules the node computing resource allocation according to the set division results.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.
Citation Information
Cited By
Resource scheduling system of prefabricated data center
CN119376901A
A resource scheduling system for a prefabricated data center
CN119376901B
