A method, device and storage medium for acquiring link delay using acquisition terminal

Abstract

The present invention discloses a method for acquiring link delay using a collection terminal, comprising the following steps: S1. Collecting data for data preprocessing; S2. Selecting the Adam model as the initial training model for data training; S3. Using the collected data set to train the selected model; S4. Using the trained model, predicting the communication delay in a specific scenario according to the input features, analyzing the influence of each feature on the model output delay data, and generating corresponding influence factors; S5. The training model outputs a basic predicted delay value, further optimizes the predicted delay value according to the influence factor, obtains the optimized data and stores it; S6. Regularly retrains and optimizes the model according to the stored data. The present invention can adjust the learning rate, effectively deal with the influence of different features on the model training speed and accuracy, realize the secondary optimization of the predicted delay by combining the feature influence factor, and improve the delay prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method, device and storage medium for acquiring link delay using a collection terminal. Background Art

[0002] The current application of carrier communication technology generally relies on power lines as a carrier for information transmission. However, power lines themselves are not ideal communication media. In particular, changes in power quality have become a major problem, mainly due to the operation of various devices in the power grid - their start-up, stop and operation process will inject instantaneous voltage and current shocks into the power lines, resulting in waveform distortion and continuous harmonic pollution. These interferences not only increase the difficulty of signal decoding, but also directly cause the instability and delay uncertainty of power line communication.

[0003] Therefore, when the carrier communication system processes communication messages of different lengths, there are often the following problems:

[0004] (1) Different communication message lengths are reflected in the carrier delay, which is also a discrete component. Therefore, even a small change in message size may cause significant differences in delay. For example, simply applying the average delay data for 200-byte messages to 50-byte messages is often not feasible, because the actual communication delay is closely related to multiple links such as message processing, encoding and decoding. The contribution of each link to the delay varies with the size of the message, which undoubtedly increases the complexity of delay management.

[0005] (2) The power line environment is susceptible to interference from certain devices: The start and stop of these devices and their operation will cause voltage and current shocks to the power line environment, waveform distortion, and the generation of continuous harmonics. In reality, for example, in a certain area, due to the simultaneous activation of multiple high-power electrical devices, the collective load changes can easily constitute an impact on the power line communication environment, resulting in a significant increase in communication delay, which may cause the carrier communication delay performance to fluctuate abnormally at a certain time of the day.

[0006] At the same time, in specific experiments, it was found that the traditional method of inferring carrier communication delay based on a single dimension is feasible in a laboratory environment, because the laboratory environment can use the control variable method to change the specific relationship between a single dimension and the carrier communication delay, and then infer the carrier delay. However, the accuracy is too poor in actual applications. The fundamental reason is that the field environment is the result of the combined effect of multiple complex factors, and the combined effect value cannot accurately calculate the proportion of factors affecting the carrier communication delay.

[0007] To this end, the present application specifically proposes a method for acquiring link delay using a collection terminal to effectively cope with the combined impact of power quality fluctuations, message length diversity, and load mutations in specific time periods, and to solve the above technical problems. Summary of the invention

[0008] The main purpose of the present invention is to provide a method for acquiring link delay using a collection terminal to solve the technical problems raised in the background technology.

[0009] The present invention adopts the following technical solutions to solve the above technical problems:

[0010] A method for acquiring link delay using a collection terminal includes the following specific steps:

[0011] S1. Collect multiple time delay data as training data sets for machine learning models for data preprocessing;

[0012] S2. Select the Adam model based on the adaptive learning rate algorithm as the initial training model for data training;

[0013] S3. Train the selected model using the collected dataset;

[0014] S4. After the training is completed, the trained model is used to predict the communication delay in a specific scenario based on the input power quality Q, data transmission and reception time point T, and message length L as features, and the influence of each feature on the model output delay data is analyzed, and the corresponding influence factor is generated;

[0015] S5. The training model outputs a basic predicted delay value, further optimizes the predicted delay value according to the influencing factors obtained by feature analysis, obtains optimized predicted delay data, and stores the data in the predicted delay database;

[0016] S6. Based on the stored data in the prediction delay database, the model is retrained and optimized regularly at a preset time to improve the accuracy and generalization ability of the prediction.

[0017] Preferably, the specific operation flow of data preprocessing in step S1 includes:

[0018] S11. Collect the terminal's delay data at different time points, under different power quality conditions, and under different message lengths. These data will be used as training data sets for machine learning models;

[0019] S12. Extract power quality parameters, data transmission and reception time points, and message length from each data point as feature X, and delay as target variable Y.

[0020] Preferably, in step S3, cross-validation can be used to evaluate the performance of the model during the training process to ensure that it is neither overfitting nor underfitting.

[0021] Preferably, the specific operation steps of training the selected model in step S3 include:

[0022] S31. For a given input vector X = [Q, T, L], use forward propagation and neural network transformation to obtain the primary neural network model and calculate the predicted delay

[0023] S32. Use the table to collect the actual value Y of the delay i , which is transformed by the neural network forward propagation to obtain the predicted value of the collected data point And use mean square error (MSE) as the loss function to measure the difference between the predicted value and the true value:

[0024]

[0025] Where k is the number of collected data points;

[0026] S33. Update the parameters of the Adam algorithm model according to the loss function L.

[0027] Preferably, in step S31, a primary neural network model is obtained and the predicted delay is calculated. The specific steps include:

[0028] (1) Input to the first hidden layer:

[0029] A [1] =σ(W [1] X+b [1] )

[0030] Among them, W [1] and b [1] represents the weight and bias of the first hidden layer, σ(·) represents the activation function;

[0031] (2) Iterate from the first hidden layer to the nth hidden layer, where the nth hidden layer can be the second hidden layer and any subsequent hidden layer:

[0032] A [n] =σ(W [n] A [n-1] +b [n] )

[0033] Among them, W [n] and b [n] Represents the weight and bias of the nth hidden layer, n is greater than or equal to 2;

[0034] (3) From the nth hidden layer to the output layer:

[0035]

[0036] Where W [n+1] and b [n+1]Represents the weight and bias of the output layer. ω represents the impact factor caused by power quality, time dimension and message length. When ω cannot be determined, the value of ω is 0.

[0037] (4) The output layer uses linear activation to directly provide prediction values

[0038] Preferably, the specific operation steps for updating the parameters in step S33 include:

[0039] (1) Calculate the gradient g of the loss function L with respect to the current parameters t , used to update the weights of the output layer,

[0040]

[0041] Among them, W [n+1] It represents the weight of the output layer in the primary neural network model, and t represents the number of gradient descent rounds;

[0042] (2) Define m0 = 0 and update the first-order moment estimate m of the gradient t (Gradient g t The expectation of ) and the second-order estimate v t (gradient Expectations):

[0043] m t =β1m t-1 +(1-β1)g t

[0044]

[0045] Among them, β1 and β2 are decay rates;

[0046] The calculation method of the decay rate β1 includes:

[0047]

[0048] The calculation method of the decay rate β2 includes:

[0049]

[0050] Where k is the number of collected data points;

[0051] (3) Bias correction is performed on the first-order and second-order moment estimates:

[0052]

[0053] in, Expressed as the corrected g t The first moment estimate of Expressed as the corrected gt Estimates of the second moment of

[0054] Bias correction helped correct the low estimate in the early stages;

[0055] (4) Update weight parameters:

[0056]

[0057] Here, α is the learning rate, ∈=1e-8 is used to avoid division by zero, and this is repeated for all parameters, including weights and biases of other hidden layers.

[0058] Preferably, the specific operation steps of analyzing the influence of each feature on the model output delay data and generating the corresponding influence factor in step S4 are:

[0059] S41. Under the conditions of the same data transmission and reception time point T, message length L and different power quality Q, the model outputs a set of predicted delay values ​​τ1;

[0060] S42. Under the conditions of the same message length L, power quality Q and different data transmission and reception time points T, the model outputs a set of predicted delay values ​​τ2;

[0061] S43. Under the conditions of the same data transmission and reception time point T, power quality Q and different message lengths L, the model outputs a set of predicted delay values ​​τ3

[0062] S44. The impact factor formula is defined as:

[0063]

[0064] q+t+l≤100;

[0065] Among them, Y represents the set of true values ​​under the same conditions, q represents the impact weight value of power quality, t represents the impact weight value of the data transmission and reception time point, l represents the impact weight value of the message length, and the values ​​of q, t, and l are always greater than 0;

[0066] S45. The value settings of q, t, and l can be adjusted at will until the influence factor ω approaches 0 infinitely. At this time, the final values ​​of q, t, and l can be obtained, and the influence factor values ​​in the neural network can be updated.

[0067] Preferably, the specific operation steps for further optimization in step S5 include:

[0068] S51. Input new vector into the trained and optimized model And perform forward propagation through the trained model to calculate the predicted delay Where C represents the basic value of the previous weight revision, and the value of C is obtained by multiple training of the model;

[0069] S52. Further optimize the predicted delay value based on the influencing factors obtained by feature analysis to obtain the final predicted delay result in:

[0070]

[0071] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0072] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0073] It can be seen from the above technical solution that the present invention provides a method for acquiring link delay using a collection terminal. Compared with the prior art, the present invention has the following advantages:

[0074] 1. The present invention selects the Adam model for data training and utilizes its adaptive learning rate mechanism to automatically adjust the learning rate, effectively addressing the impact of different features on the model training speed and accuracy, making the model training process more efficient and the results more stable. At the same time, by combining the Adam model with the feature influencing factor, the secondary optimization of the prediction delay is achieved, which significantly improves the accuracy of the delay prediction.

[0075] 2. By setting a regular retraining mechanism, the present invention can ensure that the model is continuously updated over time, adapt to changes in the network environment, maintain the timeliness and generalization ability of the prediction model, and ultimately achieve long-term high-precision delay prediction, thereby improving the intelligence level of the entire Adam model use and management.

[0076] 3. Since the impact of a single dimension on carrier communication delay cannot be quantified and evaluated, the single-dimensional reasoning results cannot be used for actual on-site use. Therefore, the present invention adopts multi-dimensional evaluation reasoning to suit on-site usage scenarios, and can obtain more reasoning data feedback over time to achieve more accurate data reasoning results, which is convenient as an important bearing basis for subsequent carrier precise peer operations.

[0077] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become easy to understand through the following description. Of course, it is not necessary to achieve all of the advantages described above simultaneously for any product implementing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0079] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0080] Figure 2 A schematic diagram of a flow chart of outputting prediction results of the training model of the present invention;

[0081] Figure 3 The figure is a flow chart of the steps of obtaining the impact factor of the present invention. DETAILED DESCRIPTION

[0082] 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 of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments in 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.

[0083] In the embodiment, see Figures 1 to 3 .

[0084] like Figure 1 As shown. Since the impact of a single dimension on carrier communication delay cannot be quantified and evaluated, the single-dimensional reasoning result cannot be used for actual on-site use. For this reason, the embodiment of the present invention proposes a method for acquiring link delay using a collection terminal, which adopts multi-dimensional evaluation reasoning to fit the on-site use scenario, and can obtain more reasoning data feedback over time, so as to achieve more accurate data reasoning results, which is convenient as an important bearing basis for subsequent carrier precise peer-to-peer operations.

[0085] In order to calculate the power quality of the power grid in the substation area, the relationship between the data transmission and reception time point, the message length and the average delay, and further infer the precise delay in a specific scenario, this method considers using a big data self-learning model for analysis and prediction. The method includes the following specific steps:

[0086] S1. Collect multiple time delay data as training data sets for machine learning models for data preprocessing;

[0087] The specific operation process of data preprocessing includes:

[0088] S11. Collect the terminal's delay data at different time points, under different power quality conditions, and under different message lengths. These data will be used as training data sets for machine learning models;

[0089] S12. Extract power quality parameters, data transmission and reception time points, and message length from each data point as feature X, and delay as target variable Y.

[0090] S2. Select the Adam model based on the adaptive learning rate algorithm as the initial training model for data training.

[0091] S3. Use the collected data set to train the selected model. It should be noted that cross-validation can also be used to evaluate the performance of the model during the training process to ensure that it is neither overfitting nor underfitting;

[0092] Specifically, refer to Figure 3 , the specific steps for training the selected model include:

[0093] S31. Setting: Q represents power quality, T represents the time dimension, such as the time point of data transmission and reception, and L represents the message length;

[0094] For a given input vector X = [Q, T, L], the primary neural network model is obtained by forward propagation and transformation through the neural network, and the predicted delay is calculated.

[0095] The primary neural network model is obtained and the predicted delay is calculated The specific steps include:

[0096] (1) Input to the first hidden layer:

[0097] A [1] =σ(W [1] X+b [1] )

[0098] Among them, W [1] and b [1] represents the weight and bias of the first hidden layer, σ(·) represents the activation function;

[0099] (2) Iterate from the first hidden layer to the nth hidden layer, where the nth hidden layer can be the second hidden layer and any subsequent hidden layer:

[0100] A [n] =σ(W [n] A [n-1] +b [n] )

[0101] Among them, W [n] and b [n]Represents the weight and bias of the nth hidden layer, n is greater than or equal to 2;

[0102] (3) From the nth hidden layer to the output layer:

[0103]

[0104] Where W [n+1] and b [n+1] Represents the weight and bias of the output layer, ω represents the impact factor caused by power quality, time dimension and message length. When ω cannot be determined, the value of ω is 0;

[0105] (4) The output layer uses linear activation to directly provide prediction values

[0106] S32. Use the table to collect the actual value Y of the delay i , which is transformed by the neural network forward propagation to obtain the predicted value of the collected data point And use mean square error (MSE) as the loss function to measure the difference between the predicted value and the true value:

[0107]

[0108] Where k is the number of collected data points;

[0109] S33. Update the parameters of the Adam algorithm model according to the loss function L. The specific operation steps include:

[0110] (1) Calculate the gradient g of the loss function L with respect to the current parameters t , used to update the weights of the output layer,

[0111]

[0112] Among them, W [n+1] It represents the weight of the output layer in the primary neural network model, and t represents the number of gradient descent rounds;

[0113] (2) Define m0 = 0 and update the first-order moment estimate m of the gradient t (Gradient g t The expectation of ) and the second-order estimate v t (gradient Expectations):

[0114] m t =β1m t-1 +(1-β1)g t

[0115]

[0116] Among them, β1 and β2 are decay rates;

[0117] The calculation method of the decay rate β1 includes:

[0118]

[0119] The calculation method of the decay rate β2 includes:

[0120]

[0121] Where k is the number of collected data points;

[0122] (3) Bias correction is performed on the first-order and second-order moment estimates:

[0123]

[0124] in, Expressed as the corrected g t The first moment estimate of Expressed as the corrected g t Estimates of the second moment of

[0125] Bias correction helped correct the low estimate in the early stages;

[0126] (4) Update weight parameters:

[0127]

[0128] Where α is the learning rate, ∈ = 1e-8 is used to avoid division by zero, and is repeated for all parameters, including weights and biases of other hidden layers;

[0129] It can be explained here that by selecting the Adam model for data training and utilizing its adaptive learning rate mechanism, the learning rate can be automatically adjusted to effectively deal with the impact of different features on the model training speed and accuracy, making the model training process more efficient and the results more stable.

[0130] S4. After the training is completed, the trained model is used to predict the communication delay in a specific scenario based on the input power quality Q, data transmission and reception time point T, and message length L as features, and the influence of each feature on the model output delay data is analyzed, and the corresponding influence factor ω is generated;

[0131] refer to Figure 3 , where the specific steps for analyzing the impact of each feature on the model output delay data and generating the corresponding impact factor are:

[0132] S41. Under the conditions of the same data transmission and reception time point T, message length L and different power quality Q, the model outputs a set of predicted delay values ​​τ1;

[0133] S42. Under the conditions of the same message length L, power quality Q and different data transmission and reception time points T, the model outputs a set of predicted delay values ​​τ2;

[0134] S43. Under the conditions of the same data transmission and reception time point T, power quality Q and different message lengths L, the model outputs a set of predicted delay values ​​τ3

[0135] S44. The impact factor formula is defined as:

[0136]

[0137] q+t+l≤100;

[0138] Among them, Y represents the set of true values ​​under the same conditions, q represents the impact weight value of power quality, t represents the impact weight value of the data transmission and reception time point, l represents the impact weight value of the message length, and the values ​​of q, t, and l are always greater than 0;

[0139] S45. The value settings of q, t, and l can be adjusted at will until the influence factor ω approaches 0 infinitely. At this time, the final values ​​of q, t, and l can be obtained, and the influence factor values ​​in the neural network can be updated;

[0140] It should be noted at this point that when the impact factor ω approaches 0 infinitely, it means that the predicted value and the true value are close in value and shape, proving that the predicted value can reflect the characteristics of the real data to a large extent. The q, t, l, and ω values ​​here can be effectively used as weights to reduce the error between the predicted value and the true value.

[0141] S5. Figure 3 As shown, the training model outputs a basic predicted delay value, and the predicted delay value is further optimized according to the influencing factors obtained by feature analysis to obtain optimized predicted delay data, and the data is stored in the predicted delay database;

[0142] The specific steps for further optimization include:

[0143] S51. Input new vector into the trained and optimized model And perform forward propagation through the trained model to calculate the predicted delay Where C represents the basic value of the previous weight revision, and the value of C is obtained by multiple training of the model:

[0144] The initial C value is set to 100, and the model training is iterated upward from 100. If the predicted delay calculated in this round of training is If the error between the true value and the real value is smaller than that obtained in the previous round of training, the iterative training will continue, otherwise the training will be stopped. The C value at this time is the weight revision base value applicable to the vector;

[0145] S52. Further optimize the predicted delay value based on the influencing factors obtained by feature analysis to obtain the final predicted delay result in:

[0146]

[0147] At this time, the values ​​of q, t, and l can be used to optimize the prediction results in the early stage of data input. After the predicted delay results are calculated, the results are further optimized through ω, which can further improve the accuracy of the predicted delay results and is more suitable for more rigorous delay calculations.

[0148] In summary, this technical solution can achieve secondary optimization of the predicted delay by combining the Adam model with the characteristic influencing factor, significantly improving the accuracy of the delay prediction. In addition, this method can update the parameters in the neural network in an efficient and adaptive manner, thereby obtaining a more accurate delay prediction.

[0149] S6. Based on the stored data in the prediction delay database, the model is retrained and optimized regularly at preset times to ensure that the model is continuously updated over time and adapt to changes in the network environment. This improves and maintains the accuracy, timeliness and generalization ability of the prediction model, ultimately achieving long-term high-precision delay prediction and improving the intelligence level of the entire Adam model use and management.

[0150] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0151] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0152] In another embodiment provided by the present application, a computer program product including instructions is also provided. When the computer is run on the computer, the computer executes any of the methods for acquiring link delay using a collection terminal in the above embodiments.

[0153] It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above method.

[0154] The embodiment of the present application also provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus.

[0155] Memory, used to store computer programs;

[0156] The processor is used to implement the above-mentioned method for acquiring the link delay using the acquisition terminal when executing the program stored in the memory.

[0157] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0158] The communication interface is used for communication between the above electronic device and other devices.

[0159] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0160] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components.

[0161] It should also be noted that electronic devices also include terminal devices, which can also be called terminals, user equipment (UE), mobile stations (MS), mobile terminals (MT), etc. Terminal devices can be mobile phones, smart TVs, wearable devices, tablet computers (Pad), computers with wireless transceiver functions, virtual reality (VR) terminal devices, augmented reality (AR) terminal devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, etc. The embodiments of the present application do not limit the specific technology and specific device form adopted by the terminal devices.

[0162] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0163] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0164] In addition, it should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0165] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing in the full text includes three parallel schemes. Taking "A and / or B" as an example, it includes scheme A, or scheme B, or schemes that A and B meet at the same time. In addition, in the embodiments of the present invention, "multiple" refers to more than two. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A method for acquiring link delay using a collection terminal, characterized in that: The specific steps are as follows: S1. Collect multiple time delay data as training data sets for machine learning models for data preprocessing; S2. Select the Adam model based on the adaptive learning rate algorithm as the initial training model for data training; S3. Train the selected model using the collected dataset; S4. After the training is completed, the trained model is used to predict the communication delay in a specific scenario based on the input power quality Q, data transmission and reception time point T, and message length L as features, and the influence of each feature on the model output delay data is analyzed, and the corresponding influence factor is generated; The impact factor formula is expressed as: q+t+l≤100; Among them, Y represents the set of true values ​​under the same conditions, q represents the impact weight value of power quality, t represents the impact weight value of the data transmission and reception time point, l represents the impact weight value of the message length, and the values ​​of q, t, and l are always greater than 0; τ1, τ2, and τ3 are respectively: the predicted delay value set output by the model when any of the variables is changed under the conditions of the same data transmission and reception time point T, message length L, and power quality Q; S5. The training model outputs a basic predicted delay value, further optimizes the predicted delay value according to the influencing factors obtained by feature analysis, obtains optimized predicted delay data, and stores the data in the predicted delay database; S6. Based on the stored data in the prediction delay database, the model is retrained and optimized regularly at a preset time to improve the accuracy and generalization ability of the prediction.

2. The method for acquiring the link delay using a collection terminal as claimed in claim 1, characterized in that: The specific operation process of data preprocessing in step S1 includes: S11. Collect the time delay data of the terminal at different time points, different power quality conditions and different message lengths. These data will be used as the training data set of the machine learning model; S12. Extract power quality parameters, data transmission and reception time points, and message length from each data point as feature X, and delay as target variable Y.

3. The method for acquiring the link delay of a collection terminal according to claim 1, characterized in that: In the S3 step, cross-validation can be used to evaluate the performance of the model during the training process to ensure that it is neither overfitting nor underfitting.

4. The method for acquiring the link delay of a collection terminal according to claim 1, characterized in that: The specific operation steps for training the selected model in step S3 include: S31. For a given input vector X = [Q, T, L], use forward propagation and neural network transformation to obtain the primary neural network model and calculate the predicted delay S32. Use the table to collect the actual value Y of the delay i , which is transformed by the neural network forward propagation to obtain the predicted value of the collected data point And use mean square error (MSE) as the loss function to measure the difference between the predicted value and the true value: Where k is the number of collected data points; S33. Update the parameters of the Adam algorithm model according to the loss function L.

5. The method for acquiring the link delay of a collection terminal according to claim 4, characterized in that: In step S31, the primary neural network model is obtained and the predicted delay is calculated. The specific steps include: (1) Input to the first hidden layer: A [1] =σ(W [1] X+b [1] ) Among them, W [1] and b [1] represents the weight and bias of the first hidden layer, σ(·) represents the activation function; (2) Iterate from the first hidden layer to the nth hidden layer, where the nth hidden layer can be the second hidden layer and any subsequent hidden layer: A [n] =σ(W [n] A [n-1] +b [n] ) Among them, W [n] and b [n] Represents the weight and bias of the nth hidden layer, n is greater than or equal to 2; (3) From the nth hidden layer to the output layer: Where W [n+1] and b [n+1] Represents the weight and bias of the output layer, ω represents the impact factor caused by power quality, time dimension and message length. When v cannot be determined, the ω value is 0; (4) The output layer uses linear activation to directly provide prediction values 6. The method for acquiring the link delay of a collection terminal according to claim 4, characterized in that: The specific operation steps for updating the parameters in step S33 include: (1) Calculate the gradient g of the loss function L with respect to the current parameters t , used to update the weights of the output layer, Among them, W [n+1] It represents the weight of the output layer in the primary neural network model, and t represents the number of gradient descent rounds; (2) Define m0 = 0 and update the first-order moment estimate m of the gradient t (Gradient g t The expectation of ) and the second-order estimate v t (gradient Expectations): m t =β1m t-1 +(1-β1)g t Among them, β1 and β2 are decay rates; The calculation method of the decay rate β1 includes: The calculation method of the decay rate β2 includes: Where k is the number of collected data points; (3) Bias correction is performed on the first-order and second-order moment estimates: in, Expressed as the corrected g t The first moment estimate of Expressed as the corrected g t Estimates of the second moment of Bias correction helped correct the low estimate in the early stages; (4) Update weight parameters: Here, α is the learning rate, ∈=1e-8 is used to avoid division by zero, and this is repeated for all parameters, including weights and biases of other hidden layers.

7. The method for acquiring the link delay of a collection terminal according to claim 4, characterized in that: The value settings of q, t, and l in the influence factor formula in step S4 can be adjusted arbitrarily until the influence factor ω approaches 0 infinitely. At this time, the final values ​​of q, t, and l can be obtained, and the influence factor values ​​in the neural network are updated.

8. The method for acquiring the link delay of a collection terminal according to claim 7, characterized in that: The specific operation steps for further optimization in step S5 include: S51. Input new vector into the trained and optimized model And perform forward propagation through the trained model to calculate the predicted delay Where C represents the basic value of the previous weight revision, and the value of C is obtained by multiple training of the model; S52. Further optimize the predicted delay value based on the influencing factors obtained by feature analysis to obtain the final predicted delay result in:

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 8.

10. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.

Images