An automated scheduling method based on active-active switching control platform

Abstract

The present invention discloses an automated scheduling method based on an active-active switching control platform, the method comprising data collection, data preprocessing, establishing a scheduling decision model for the control platform, optimizing the hyperparameters of the scheduling decision model, and automated scheduling. The present invention relates to the field of automated scheduling technology, specifically to an automated scheduling method based on an active-active switching control platform, the present solution constructs a scheduling decision model by designing an activation function, a model structure, a loss function, a gradient change factor, and a smooth learning rate, thereby improving the performance of the model, enhancing the fitting ability and generalization ability of the model, and improving the training efficiency and stability of the model; optimizing the hyperparameters of the model by designing an adaptive threshold, an adaptive bidirectional search function, and a progressively adjusted search function, thereby avoiding falling into a local optimal solution, improving the search ability of the model optimization algorithm, and improving the search efficiency and stability.

Description

Technical Field

[0001] The present invention relates to the technical field of automated scheduling, and in particular to an automated scheduling method based on a dual-active switching management and control platform. Background Art

[0002] An automated scheduling method based on a dual-active switching control platform is to use information technology, artificial intelligence algorithms and data analysis to monitor the operating status of the control platform in real time, respond quickly and schedule automatically. However, the scheduling decision model of the traditional control platform has problems such as low performance, weak fitting and generalization capabilities, and low training efficiency; the traditional model optimization algorithm is prone to falling into local optimal solutions, weak search capabilities, low search efficiency and poor stability. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an automated scheduling method based on a dual-active switching control platform. In view of the problems of low performance, weak fitting and generalization capabilities, and low training efficiency in the scheduling decision model of the traditional control platform, this scheme constructs a scheduling decision model by designing activation functions, designing model structures, designing adaptive zeroing layers, designing loss functions, designing gradient change factors, designing smoothing learning rates, selecting learning rates and parameter learning, thereby improving the performance of the model, enhancing the fitting and generalization capabilities of the model, reducing the risk of overfitting, and improving the training efficiency and stability of the model; in view of the problems that the traditional model optimization algorithm is easy to fall into the local optimal solution, has weak search capabilities, low search efficiency, and poor stability, this scheme optimizes the model hyperparameters by optimizing initialization, designing adaptive thresholds, designing adaptive bidirectional search functions, and designing progressive adjustment search functions, thereby avoiding falling into the local optimal solution, improving the search capability of the model optimization algorithm, and improving the search efficiency and stability.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an automated scheduling method based on a dual-active switching control platform, the method comprising the following steps:

[0005] Step S1: data collection;

[0006] Step S2: data preprocessing;

[0007] Step S3: Establish a scheduling decision model for the management and control platform;

[0008] Step S4: Optimize the scheduling decision model hyperparameters;

[0009] Step S5: Automated scheduling.

[0010] Furthermore, in step S1, the data collection is to collect control platform load data, node status data, task execution data, fault information, environmental data and scheduling decision data; the control platform load data includes CPU usage, memory occupancy, and network load; the node status data includes the node's online and offline status, health status, and load status; the task execution data includes the task execution status, completion time, and resource consumption; the fault information includes fault events, abnormal conditions, and error logs; the environmental data includes temperature and humidity.

[0011] Furthermore, in step S2, the data preprocessing includes data cleaning, feature selection, data standardization and label setting, which specifically includes the following steps:

[0012] Step S21: Data cleaning, processing missing values, outliers and duplicate values;

[0013] Step S22: feature selection, evaluating the impact of data on the management and control platform, selecting data with high relevance, and removing data with low relevance;

[0014] Step S23: data standardization, unifying the data into the same range;

[0015] Step S24: Setting a label, and setting the scheduling decision data as label data.

[0016] Furthermore, in step S3, the establishment of the management and control platform scheduling decision model specifically includes the following steps:

[0017] Step S31: Design an activation function and introduce smoothness, which is expressed as follows:

[0018] ;

[0019] Among them, Gaf(x) represents the gradient activation function, and x represents the input value of the gradient activation function;

[0020] Step S32: Design a basic model structure, the model has an input layer for receiving feature data, a hidden layer with five sub-layers and an output layer, and the output of each hidden sub-layer is adaptively set to zero;

[0021] Step S33: Design an input layer for receiving input feature data, as shown below:

[0022] ;

[0023] in, represents the output of the input layer, S represents the number of input features, , and Represents the first, second and Sth feature data of the input;

[0024] Step S34: Design the hidden layer, which is expressed as follows:

[0025] ;

[0026] Among them, c represents the layer index of the hidden sublayer, represents the output of the cth hidden sublayer, represents the weight of the cth hidden sublayer, represents the input of the cth hidden sublayer, represents the bias of the cth hidden sublayer;

[0027] Step S35: Design the output layer, which is expressed as follows:

[0028] ;

[0029] Among them, Yo represents the output of the output layer, softmax(·) represents the normalized exponential function, represents the input of the output layer, and Represent the weight and bias of the output layer respectively;

[0030] Step S36: Design a zero probability function to calculate the probability of dropout according to the activation value of the neuron, which is expressed as follows:

[0031] ;

[0032] in, represents the output of neurons in the hidden layer, represents the zeroing probability function, Maxdrp represents the maximum zeroing probability, sfactor represents the scaling factor, Represents the activation value obtained by normalizing the output value of the neurons in the hidden layer to the range of [0,1];

[0033] Step S37: Design an adaptive zeroing layer, which is expressed as follows:

[0034] ;

[0035] in, represents the input value of the adaptive zeroing layer, represents the output of the adaptive zeroing layer, Dropout(·) represents the zeroing function, It means that the input value is set to zero with a probability of p.

[0036] Step S38: Design the loss function, introduce the smoothness of the gradient activation function, and perceive the error amplification, as shown below:

[0037] ;

[0038] in, are the weight and bias parameters of the model, represents the loss function, k represents the index of the sample, Indicates the actual value, represents the predicted value;

[0039] Step S39: Design the gradient magnitude change factor, expressed as follows:

[0040] ;

[0041] Among them, j represents the number of parameter learning, Indicates the gradient change factor during the j-th parameter learning, represents the modulo symbol, represents the weight and bias parameters of the model when the number of parameter learning is j times, represents the weight and bias parameters of the model when the number of parameter learning is j-1 times, represents the gradient of the loss function;

[0042] Step S310: Design the gradient direction change factor, which is expressed as follows:

[0043] ;

[0044] in, represents the gradient direction change factor during the j-th parameter learning. sign(·) is a sign function used to determine the sign attribute of the input value.

[0045] Step S311: Design a smooth learning rate, which is expressed as follows:

[0046] ;

[0047] in, It represents the smooth learning rate when the number of parameter learning is j+1 times, It represents the smooth learning rate when the number of parameter learning is j times, represents the smoothing factor;

[0048] Step S312: Select a learning rate, which is expressed as follows:

[0049] ;

[0050] in, represents the learning rate selected when the number of parameter learning times is j+1, min(·) represents the minimum value, and max(·) represents the maximum value. and They represent the upper and lower bounds of the learning rate, respectively. α and β are adjustment factors used to adjust the amplitude of the learning rate change. Indicates that the gradient direction change factor is greater than 0 and the gradient magnitude change factor is greater than 1. Indicates that the gradient direction change factor is less than 0 and the gradient magnitude change factor is less than 1, and one or more of them are satisfied. Otherwise indicates other situations.

[0051] Step S313: parameter learning, expressed as follows:

[0052] ;

[0053] in, Represents the weight and bias parameters of the model when the number of parameter learning is j+1 times.

[0054] Furthermore, in step S4, the optimization of the scheduling decision model hyperparameters specifically includes the following steps:

[0055] Step S41: Optimize initialization, create a hyperparameter optimization search space, initialize the parameter search cluster, set the fitness value of the parameter individual with the opposite of the model's loss value, set the fitness threshold and the maximum number of iterations of the parameter search;

[0056] Step S42: Design an adaptive threshold, which is expressed as follows:

[0057] ;

[0058] Where z represents the adaptive threshold, tanh(·) is the hyperbolic tangent function, T represents the maximum number of iterations, and t represents the current number of iterations;

[0059] Step S43: Design an adaptive bidirectional search function, which is expressed as follows:

[0060] ;

[0061] in, Indicates the position searched by the adaptive bidirectional search function during the t+1th parameter search. represents the position searched by the adaptive bidirectional search function during the t-th parameter search, r1 represents a random number between 0 and 1, and Tas represents a random number that obeys a Gaussian distribution with a mean of 0 and a variance of 1. Indicates the position with the maximum current fitness value;

[0062] Step S44: Design a progressive adjustment search function, which is expressed as follows:

[0063] ;

[0064] in, Indicates the position found by gradually adjusting the search function during the t+1th parameter search. It indicates the position found by gradually adjusting the search function during the t-th parameter search. represents the control parameter that controls the search range;

[0065] Step S45: Optimize the model hyperparameters, calculate the fitness value of the position corresponding to the initial search cluster, obtain the maximum fitness value position, and then perform parameter optimization search of the adaptive bidirectional search function, calculate the fitness value of the searched point, update the maximum fitness value position, and then perform parameter optimization search of the progressive adjustment search function, calculate the fitness value of the searched point, update the maximum fitness value position again, count the number of positions whose fitness values ​​are greater than the fitness threshold, increase the number of iterations by one, if the fitness values ​​of more than 50% of the positions in a search are greater than the fitness threshold, stop the parameter optimization search, and output the parameters of the position with the highest fitness value; if the maximum number of iterations is reached, re-initialize and search the parameters; otherwise, continue the iterative search.

[0066] Furthermore, in step S5, the automated scheduling is to create a node cluster to be taken over on the control platform, and the model monitors the control platform in real time. When the model detects that the main operating node of the control platform is down, the corresponding node is automatically scheduled to take over according to the scheduling decision result output by the model to complete the active-active switching; at the same time, feedback information is collected to continuously improve the scheduling strategy and model.

[0067] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0068] (1) Aiming at the problems of low performance, weak fitting and generalization capabilities, and low training efficiency of the scheduling decision model of the traditional management and control platform, this solution constructs a scheduling decision model by designing activation functions, designing model infrastructure, designing zeroing probability functions, designing adaptive zeroing layers, designing loss functions, designing gradient change factors, designing smoothing learning rates, selecting learning rates and parameter learning, thereby improving the performance of the model, enhancing the model's fitting and generalization capabilities, reducing the risk of overfitting, and improving the training efficiency and stability of the model.

[0069] (2) In view of the problems that traditional model optimization algorithms are prone to falling into local optimal solutions, have weak search capabilities, low search efficiency, and poor stability, this scheme optimizes model hyperparameters by optimizing initialization, designing adaptive thresholds, designing adaptive bidirectional search functions, and designing progressive adjustment search functions, thereby avoiding falling into local optimal solutions, improving the search capabilities of the model optimization algorithm, and improving search efficiency and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A flowchart of an automated scheduling method based on a dual-active switching control platform provided by the present invention;

[0071] Figure 2 is a schematic flow chart of step S2;

[0072] Figure 3 is a schematic flow chart of step S3;

[0073] Figure 4 It is a schematic diagram of the process of step S4.

[0074] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0076] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying 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 should not be understood as limiting the present invention.

[0077] Example 1, see Figure 1 The present invention provides an automatic scheduling method based on a dual-active switching management and control platform, the method comprising the following steps:

[0078] Step S1: Data collection, collecting control platform load data, node status data, task execution data, fault information, environmental data and scheduling decision data;

[0079] Step S2: Data preprocessing, data cleaning, feature selection, data standardization and label setting;

[0080] Step S3: Establishing a scheduling decision model for a control platform, by designing an activation function, designing a model structure, designing a gradient change factor, selecting a learning rate and parameter learning;

[0081] Step S4: Optimizing the hyperparameters of the scheduling decision model by designing an adaptive threshold, an adaptive bidirectional search function, and a progressive adjustment search function;

[0082] Step S5: Automated scheduling: the model monitors the management and control platform in real time and automatically schedules nodes to take over based on the scheduling decision results output by the model.

[0083] Example 2, see Figure 1 This embodiment is based on the above embodiment, and the data collection is to collect the control platform load data, node status data, task execution data, fault information, environmental data and scheduling decision data; the control platform load data includes CPU usage, memory occupancy, and network load; the node status data includes the node's online and offline status, health status, and load status; the task execution data includes the task execution status, completion time, and resource consumption; the fault information includes fault events, abnormal conditions, and error logs; the environmental data includes temperature and humidity.

[0084] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The data preprocessing performs data cleaning, feature selection, data standardization and label setting, and specifically includes the following steps:

[0085] Step S21: Data cleaning, processing missing values, outliers and duplicate values;

[0086] Step S22: feature selection, evaluating the impact of data on the management and control platform, selecting data with high relevance, and removing data with low relevance;

[0087] Step S23: data standardization, unifying the data into the same range;

[0088] Step S24: Setting a label, and setting the scheduling decision data as label data.

[0089] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment, and the establishment of the management and control platform scheduling decision model specifically includes the following steps:

[0090] Step S31: Design an activation function and introduce smoothness, which is expressed as follows:

[0091] ;

[0092] Among them, Gaf(x) represents the gradient activation function, and x represents the input value of the gradient activation function;

[0093] Step S32: Design a basic model structure, the model has an input layer for receiving feature data, a hidden layer with five sub-layers and an output layer, and the output of each hidden sub-layer is adaptively set to zero;

[0094] Step S33: Design an input layer for receiving input feature data, as shown below:

[0095] ;

[0096] in, represents the output of the input layer, S represents the number of input features, , and Represents the first, second and Sth feature data of the input;

[0097] Step S34: Design the hidden layer, which is expressed as follows:

[0098] ;

[0099] Among them, c represents the layer index of the hidden sublayer, represents the output of the cth hidden sublayer, represents the weight of the cth hidden sublayer, represents the input of the cth hidden sublayer, represents the bias of the cth hidden sublayer;

[0100] Step S35: Design the output layer, which is expressed as follows:

[0101] ;

[0102] Among them, Yo represents the output of the output layer, softmax(·) represents the normalized exponential function, represents the input of the output layer, and Represent the weight and bias of the output layer respectively;

[0103] Step S36: Design a zero probability function to calculate the probability of dropout according to the activation value of the neuron, which is expressed as follows:

[0104] ;

[0105] in, represents the output of neurons in the hidden layer, represents the zeroing probability function, Maxdrp represents the maximum zeroing probability, sfactor represents the scaling factor, Represents the activation value obtained by normalizing the output value of the neurons in the hidden layer to the range of [0,1];

[0106] Step S37: Design an adaptive zeroing layer, which is expressed as follows:

[0107] ;

[0108] in, represents the input value of the adaptive zeroing layer, represents the output of the adaptive zeroing layer, Dropout(·) represents the zeroing function, It means that the input value is set to zero with a probability of p.

[0109] Step S38: Design the loss function, introduce the smoothness of the gradient activation function, and perceive the error amplification, as shown below:

[0110] ;

[0111] in, are the weight and bias parameters of the model, represents the loss function, k represents the index of the sample, Indicates the actual value, represents the predicted value;

[0112] Step S39: Design the gradient magnitude change factor, expressed as follows:

[0113] ;

[0114] Among them, j represents the number of parameter learning, Indicates the gradient change factor during the j-th parameter learning, represents the modulo symbol, represents the weight and bias parameters of the model when the number of parameter learning is j times, represents the weight and bias parameters of the model when the number of parameter learning is j-1 times, represents the gradient of the loss function;

[0115] Step S310: Design the gradient direction change factor, which is expressed as follows:

[0116] ;

[0117] in, represents the gradient direction change factor during the j-th parameter learning. sign(·) is a sign function used to determine the sign attribute of the input value.

[0118] Step S311: Design a smooth learning rate, which is expressed as follows:

[0119] ;

[0120] in, It represents the smooth learning rate when the number of parameter learning is j+1 times, It represents the smooth learning rate when the number of parameter learning is j times, represents the smoothing factor;

[0121] Step S312: Select a learning rate, which is expressed as follows:

[0122] ;

[0123] in, represents the learning rate selected when the number of parameter learning times is j+1, min(·) represents the minimum value, and max(·) represents the maximum value. and They represent the upper and lower bounds of the learning rate, respectively. α and β are adjustment factors used to adjust the amplitude of the learning rate change. Indicates that the gradient direction change factor is greater than 0 and the gradient magnitude change factor is greater than 1. Indicates that the gradient direction change factor is less than 0 and the gradient magnitude change factor is less than 1, and one or more of them are satisfied. Otherwise indicates other situations.

[0124] Step S313: parameter learning, expressed as follows:

[0125] ;

[0126] in, Represents the weight and bias parameters of the model when the number of parameter learning is j+1 times.

[0127] By performing the above operations, in order to address the problems of low performance, weak fitting and generalization capabilities, and low training efficiency in the scheduling decision model of the traditional management and control platform, this solution constructs a scheduling decision model by designing activation functions, designing model infrastructure, designing zeroing probability functions, designing adaptive zeroing layers, designing loss functions, designing gradient change factors, designing smoothing learning rates, selecting learning rates and parameter learning, thereby improving the performance of the model, enhancing the model's fitting and generalization capabilities, reducing the risk of overfitting, and improving the training efficiency and stability of the model.

[0128] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment, and the optimization of the scheduling decision model hyperparameters specifically includes the following steps:

[0129] Step S41: Optimize initialization, create a hyperparameter optimization search space, initialize the parameter search cluster, set the fitness value of the parameter individual with the opposite of the model's loss value, set the fitness threshold and the maximum number of iterations of the parameter search;

[0130] Step S42: Design an adaptive threshold, which is expressed as follows:

[0131] ;

[0132] Where z represents the adaptive threshold, tanh(·) is the hyperbolic tangent function, T represents the maximum number of iterations, and t represents the current number of iterations;

[0133] Step S43: Design an adaptive bidirectional search function, which is expressed as follows:

[0134] ;

[0135] in, Indicates the position searched by the adaptive bidirectional search function during the t+1th parameter search. represents the position searched by the adaptive bidirectional search function during the t-th parameter search, r1 represents a random number between 0 and 1, and Tas represents a random number that obeys a Gaussian distribution with a mean of 0 and a variance of 1. Indicates the position with the maximum current fitness value;

[0136] Step S44: Design a progressive adjustment search function, which is expressed as follows:

[0137] ;

[0138] in, Indicates the position found by gradually adjusting the search function during the t+1th parameter search. It indicates the position found by gradually adjusting the search function during the t-th parameter search. represents the control parameter that controls the search range;

[0139] Step S45: Optimize the model hyperparameters, calculate the fitness value of the position corresponding to the initial search cluster, obtain the maximum fitness value position, and then perform parameter optimization search of the adaptive bidirectional search function, calculate the fitness value of the searched point, update the maximum fitness value position, and then perform parameter optimization search of the progressive adjustment search function, calculate the fitness value of the searched point, update the maximum fitness value position again, count the number of positions whose fitness values ​​are greater than the fitness threshold, increase the number of iterations by one, if the fitness values ​​of more than 50% of the positions in a search are greater than the fitness threshold, stop the parameter optimization search, and output the parameters of the position with the highest fitness value; if the maximum number of iterations is reached, re-initialize and search the parameters; otherwise, continue the iterative search.

[0140] By performing the above operations, the traditional model optimization algorithm has the problems of easy falling into local optimal solutions, weak search capabilities, low search efficiency and poor stability. This solution optimizes the model hyperparameters by optimizing initialization, designing adaptive thresholds, designing adaptive bidirectional search functions, and designing progressive adjustment search functions, avoiding falling into local optimal solutions, improving the search capabilities of the model optimization algorithm, and improving search efficiency and stability.

[0141] Example 6, see Figure 1 This embodiment is based on the above embodiment. The automated scheduling is to create a node cluster to be taken over on the management and control platform. The model monitors the management and control platform in real time. When the model detects that the main running node of the management and control platform is down, the corresponding node is automatically scheduled to take over according to the scheduling decision result output by the model to complete the active-active switching. At the same time, feedback information is collected to continuously improve the scheduling strategy and model.

[0142] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0143] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0144] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An automated scheduling method based on a dual-active switching control platform, characterized in that: The method comprises the following steps: Step S1: Data collection, collecting control platform load data, node status data, task execution data, fault information, environmental data and scheduling decision data; Step S2: Data preprocessing, data cleaning, feature selection, data standardization and label setting; Step S3: Establish a scheduling decision model for the control platform, and construct the scheduling decision model by designing an activation function, designing a model infrastructure, designing a zeroing probability function, designing an adaptive zeroing layer, designing a loss function, designing a gradient change factor, designing a smooth learning rate, selecting a learning rate, and parameter learning; Design the activation function, expressed as follows: ； Among them, Gaf(x) represents the gradient activation function, and x represents the input value of the gradient activation function; Design the basic structure of the model. The model has an input layer for receiving feature data, a hidden layer with five sub-layers, and an output layer. The output of each hidden sub-layer is adaptively set to zero. Design a zero probability function and calculate the probability of dropout according to the activation value of the neuron, which is expressed as follows: ； in, represents the output of neurons in the hidden layer, represents the zeroing probability function, Maxdrp represents the maximum zeroing probability, sfactor represents the scaling factor, Represents the activation value obtained by normalizing the output value of the neurons in the hidden layer to the range of [0,1]; Design an adaptive zeroing layer, expressed as follows: ； in, represents the input value of the adaptive zeroing layer, represents the output of the adaptive zeroing layer, Dropout(·) represents the zeroing function, It means that the input value is set to zero with a probability of p. Step S4: Optimizing the hyperparameters of the scheduling decision model by designing an adaptive threshold, an adaptive bidirectional search function, and a progressive adjustment search function; Step S5: Automated scheduling: the model monitors the management and control platform in real time and automatically schedules nodes to take over based on the scheduling decision results output by the model.

2. According to claim 1, an automated scheduling method based on a dual-active switching control platform is characterized in that: In step S3, the establishment of the control platform scheduling decision model specifically includes the following steps: Design the input layer to receive input feature data, as follows: ； in, represents the output of the input layer, S represents the number of input features, , and Represents the first, second and Sth feature data of the input; Design the hidden layer, expressed as follows: ； Among them, c represents the layer index of the hidden sublayer, represents the output of the cth hidden sublayer, represents the weight of the cth hidden sublayer, represents the input of the cth hidden sublayer, represents the bias of the cth hidden sublayer; Design the output layer, expressed as follows: ； Among them, Yo represents the output of the output layer, softmax(·) represents the normalized exponential function, represents the input of the output layer, and Represent the weight and bias of the output layer respectively; The loss function is designed, the smoothness of the gradient activation function is introduced, and the error amplification perception is expressed as follows: ； in, are the weight and bias parameters of the model, represents the loss function, k represents the index of the sample, Indicates the actual value, represents the predicted value; Design the gradient size change factor, expressed as follows: ； Among them, j represents the number of parameter learning, Indicates the gradient change factor during the j-th parameter learning, represents the modulo symbol, represents the weight and bias parameters of the model when the number of parameter learning is j times, represents the weight and bias parameters of the model when the number of parameter learning is j-1 times, represents the gradient of the loss function; Design the gradient direction change factor, expressed as follows: ； in, represents the gradient direction change factor during the j-th parameter learning. sign(·) is a sign function used to determine the sign attribute of the input value. Design a smooth learning rate, expressed as follows: ； in, It represents the smooth learning rate when the number of parameter learning is j+1 times, It represents the smooth learning rate when the number of parameter learning is j times, represents the smoothing factor; Choose a learning rate, expressed as follows: ； in, represents the learning rate selected when the number of parameter learning times is j+1, min(·) represents the minimum value, and max(·) represents the maximum value. and They represent the upper and lower bounds of the learning rate, respectively. α and β are adjustment factors used to adjust the amplitude of the learning rate change. Indicates that the gradient direction change factor is greater than 0 and the gradient magnitude change factor is greater than 1. Indicates that the gradient direction change factor is less than 0 and the gradient magnitude change factor is less than 1, and one or more of them are satisfied. Otherwise indicates other situations. Parameter learning is expressed as follows: ； in, Represents the weight and bias parameters of the model when the number of parameter learning is j+1 times.

3. According to the automatic scheduling method based on the active-active switching control platform of claim 1, it is characterized by: In step S4, the optimization of the scheduling decision model hyperparameters specifically includes the following steps: Step S41: Optimize initialization, create a hyperparameter optimization search space, initialize the parameter search cluster, set the fitness value of the parameter individual with the opposite of the model's loss value, set the fitness threshold and the maximum number of iterations of the parameter search; Step S42: Design an adaptive threshold, which is expressed as follows: ； Where z represents the adaptive threshold, tanh(·) is the hyperbolic tangent function, T represents the maximum number of iterations, and t represents the current number of iterations; Step S43: Design an adaptive bidirectional search function, which is expressed as follows: ； in, Indicates the position searched by the adaptive bidirectional search function during the t+1th parameter search. represents the position searched by the adaptive bidirectional search function during the t-th parameter search, r1 represents a random number between 0 and 1, and Tas represents a random number that obeys a Gaussian distribution with a mean of 0 and a variance of 1. Indicates the position with the maximum current fitness value; Step S44: Design a progressive adjustment search function, which is expressed as follows: ； in, Indicates the position found by gradually adjusting the search function during the t+1th parameter search. It indicates the position found by gradually adjusting the search function during the t-th parameter search. represents the control parameter that controls the search range; Step S45: Optimize the model hyperparameters, calculate the fitness value of the position corresponding to the initial search cluster, obtain the maximum fitness value position, and then perform parameter optimization search of the adaptive bidirectional search function, calculate the fitness value of the searched point, update the maximum fitness value position, and then perform parameter optimization search of the progressive adjustment search function, calculate the fitness value of the searched point, update the maximum fitness value position again, count the number of positions whose fitness values ​​are greater than the fitness threshold, increase the number of iterations by one, if the fitness values ​​of more than 50% of the positions in a search are greater than the fitness threshold, stop the parameter optimization search, and output the parameters of the position with the highest fitness value; if the maximum number of iterations is reached, re-initialize and search the parameters; otherwise, continue the iterative search.

4. The automatic scheduling method based on the active-active switching control platform according to claim 1 is characterized in that: In step S2, the data preprocessing includes data cleaning, feature selection, data standardization and label setting, which specifically includes the following steps: Step S21: Data cleaning, processing missing values, outliers and duplicate values; Step S22: feature selection, evaluating the impact of data on the management and control platform, selecting data with high relevance, and removing data with low relevance; Step S23: data standardization, unifying the data into the same range; Step S24: Setting a label, and setting the scheduling decision data as label data.

5. The automatic scheduling method based on the active-active switching control platform according to claim 1 is characterized in that: In step S1, the data collection is to collect the control platform load data, node status data, task execution data, fault information, environmental data and scheduling decision data; the control platform load data includes CPU usage, memory occupancy, and network load; the node status data includes the node's online and offline status, health status, and load status; the task execution data includes the task execution status, completion time, and resource consumption; the fault information includes fault events, abnormal conditions, and error logs; the environmental data includes temperature and humidity.

6. The automatic scheduling method based on the active-active switching control platform according to claim 1 is characterized in that: In step S5, the automated scheduling is to create a node cluster to be taken over by the management and control platform. The model monitors the management and control platform in real time. When the model detects that the main operating node of the management and control platform is down, the corresponding node is automatically scheduled to take over according to the scheduling decision result output by the model to complete the active-active switching. At the same time, feedback information is collected to continuously improve the scheduling strategy and model.

Images