A SPMA protocol design method based on BiLSTM channel load prediction

Abstract

The present invention belongs to the field of communication simulation technology, and specifically discloses a SPMA protocol design method based on BiLSTM channel load prediction, including the following steps: firstly, obtaining the node channel load statistics, and then correcting it to obtain the corrected channel load statistics; constructing a channel load statistics result data set; performing Z-Score standardization on the channel load statistics result data set, and dividing it sequentially according to proportion; reconstructing the input and output of the channel load statistics result data set using a sliding window mode; constructing a BiLSTM neural network channel load prediction model, and training the model after optimizing the network; after the training is completed, using a validation set to evaluate the performance of the model. The present invention adopts the above-mentioned SPMA protocol design method based on BiLSTM channel load prediction to solve the problems of large error in channel load statistics results, poor prediction accuracy, and low channel resource utilization of the existing SPMA protocol.

Description

Technical Field

[0001] The present invention relates to the technical field of communication simulation, and in particular to a SPMA protocol design method based on BiLSTM channel load prediction. Background Art

[0002] Statistical Priority-based Multiple Access (SPMA) is a protocol that provides different QoS supports according to service priorities, reduces network congestion, and guarantees the real-time transmission of high-priority services to a certain extent. Its superior performance has attracted widespread attention from researchers.

[0003] Unlike traditional competitive protocols, SPMA channel occupancy is no longer limited to idle and busy states, but the channel idleness and busyness are further distinguished according to the channel load statistics. In the SPMA protocol, the channel load statistics directly determine whether a node can access the channel for data transmission: when the statistical result is greater than the true value, some priority data packets will be unnecessarily backed off to avoid worsening network congestion, resulting in a waste of channel resources; when the statistical result is less than the true value, too many data packets are connected to the channel for transmission, intensifying competition and also reducing network data transmission efficiency.

[0004] Therefore, channel load statistics is one of the key technologies that affect network performance and has become a hot research topic. However, existing literature uses the weighted average method of historical channel load statistics to estimate the current channel status. This method requires that historical data have strong correlation, but the data traffic of mobile ad hoc networks is bursty. It is difficult to make an accurate estimate of the actual channel load status using the above method. In addition, the weight setting has a great impact on the prediction results.

[0005] The Long Short Term Memory (LSTM) neural network was proposed in 1997 and can effectively solve the gradient vanishing and gradient exploding problems in the traditional Recurrent Neural Network (RNN). At present, LSTM networks have achieved a lot of research results in time series prediction. The Bidirectional Long Short-Term Memory (BiLSTM) neural network is improved on the basis of LSTM, combining bidirectionality and long short-term memory networks, so that the model can process from both forward and backward directions at the same time, which can better capture relevant information in the sequence. Therefore, it is highly feasible to predict channel load based on the BiLSTM model, complete the SPMA protocol improvement, and achieve network performance optimization. Summary of the invention

[0006] The purpose of the present invention is to provide a SPMA protocol design method based on BiLSTM channel load prediction, which solves the problems of large error in channel load statistical results of the existing SPMA protocol, poor prediction accuracy, low channel resource utilization, etc.

[0007] To achieve the above object, the present invention provides a SPMA protocol design method based on BiLSTM channel load prediction, comprising the following steps:

[0008] Step S1: local node statistics cycle The average value of the sum of the number of pulses sent and received at each frequency point within a certain period of time is taken as the node channel load statistic;

[0009] Step S2, correcting the node channel load statistics in step S1 to obtain a corrected channel load statistics result;

[0010] Step S3: Obtain channel load statistics in the network and construct a channel load statistics data set;

[0011] Step S4, performing Z-Score normalization on the channel load statistical result data set, and dividing it sequentially in a ratio of 3:1:1 as a training set, a validation set, and a test set respectively;

[0012] Step S5: reconstruct the input and output of the channel load statistical result data set using a sliding window mode;

[0013] Step S6: construct a BiLSTM neural network channel load prediction model, the model includes an input layer, a BiLSTM layer and an output layer, and use the Adam algorithm to optimize the network;

[0014] Step S7: Set learning rate and the number of training rounds m , train the BiLSTM neural network channel load prediction model;

[0015] Step S8: After the training is completed, the validation set is used to evaluate the performance of the model. The root mean square error RSME is selected as the evaluation indicator. If the value of the root mean square error RSME meets the accuracy requirement, the process proceeds to step S9, otherwise it returns to step S7.

[0016] Step S9: When the data packet with the priority mark arrives at the MAC layer, the queue scheduling module inserts the data packet with the priority mark into the corresponding priority queue according to the data packet priority.

[0017] Preferably, step S1 specifically comprises:

[0018] Define the node channel load statistics period ,Record The number of pulses sent by the local node at each frequency within a certain period of time , then the local node The number of pulses sent in time is:

[0019] ;

[0020] in, Indicates Frequency points, is the total number of frequency hopping frequencies;

[0021] The local node The number of pulses received during this time is:

[0022] ;

[0023] Set the local node statistics period The average value of the sum of the number of pulses sent and received at each frequency point within a certain period of time is taken as the node channel load statistics value, that is:

[0024] .

[0025] Preferably, step S2 specifically comprises:

[0026] Assume that the pulse arrives at the physical layer with a rate of Poisson distribution, the total number of frequency points is , the pulse arrival rate of a single channel is , then the time period Inside The probability of a pulse arriving is:

[0027] ;

[0028] Assume the pulse bit length is , the baseband processing rate is , then the time required to send or receive a single pulse is:

[0029] ;

[0030] In addition, the protection interval length is , then the probability that a node is at the pulse sending or receiving time is:

[0031] ;

[0032] The probability of being in the guard interval is:

[0033] ;

[0034] Set pulse reception without collision as event , then the probability of pulse reception without conflict is:

[0035] ;

[0036] in, , represents the probability of no pulse reception conflict when the node is in the pulse sending time; , which represents the probability of no pulse reception conflict when the node is in the pulse reception time, , which represents the probability of no pulse reception conflict when the node is in the protection interval, and All of them are the first type of error. is the second error; therefore, the corrected channel load statistics are:

[0037] .

[0038] Preferably, step S5 is specifically:

[0039] Processing t In each cycle, t The channel load of the cycle is taken as the output data label y, t - m To t -1 cycle of channel load as input data feature matrix x.

[0040] Preferably, in step S6, a BiLSTM neural network channel load prediction model is constructed by improving the LSTM neural network, and a single neuron structure of the LSTM neural network includes three gating mechanisms, namely a forget gate, an input gate, and an output gate;

[0041] Among them, the calculation formula of the forget gate is as follows:

[0042] ;

[0043] In the formula, is the output of the forget gate, is the activation function, for The hidden layer output at the moment, for Input at any time, , and These are forget gate parameters;

[0044] The input gate calculation formula is as follows:

[0045] ;

[0046] In the formula, is the input gate output, , and These are input gate parameters;

[0047] LSTM neurons pass Determine the cell state at the previous moment Cellular state The degree of influence, while through Determine the cell state update value Cellular state The specific calculation formula is as follows:

[0048] ;

[0049] ;

[0050] In the formula, is the activation function;

[0051] The output gate calculation formula is as follows:

[0052] ;

[0053] In the formula, is the output gate output, , and is the output gate parameter;

[0054] Finally, the output gate outputs and cell state Calculate the hidden layer state , the specific calculation formula is as follows:

[0055] ;

[0056] The above LSTM neuron calculation process is recorded as , then the BiLSTM cell output is:

[0057] ;

[0058] in, for The hidden layer state of the forward LSTM network at any moment, for Hidden layer state of the time-reversed LSTM network.

[0059] Preferably, in step S8, the calculation formula of the root mean square error RSME is:

[0060] ;

[0061] in, is the model prediction value, is the true value, is the number of test set samples.

[0062] Preferably, the queue scheduling module in step S9 adopts a strict priority scheduling principle, that is, only when the high priority queue is empty, the data packets in the low priority queue have the opportunity to be taken out and sent; for the data packets in the same priority queue, the first-in-first-out principle is adopted.

[0063] Preferably, after step S9, the following steps are also included:

[0064] Step S10, check all queues from high to low priority, stop at the first non-empty queue, take out the first packet of the queue to determine whether it is expired, if not expired, transfer it to the waiting state; if expired, directly discard it, and re-check in order from high to low priority;

[0065] Step S11: Set the sending threshold of the lowest priority of the data packet to , other high priority thresholds are set according to the priority business ratio using the following formula:

[0066] ;

[0067] in, Indicates priority Threshold, Priority within the historical period Business volume share;

[0068] Step S12: Use the BiLSTM neural network channel load prediction model to predict the channel load value in the current cycle, and perform denormalization on the result;

[0069] Step S13, compare the priority threshold of the data packet to be sent with the channel load statistics result obtained in step S12: when the threshold is greater than the channel load statistics value, directly send the data packet; if the threshold is less than the channel load statistics value, proceed to step S14;

[0070] Step S14: Set the random backoff time length to:

[0071] ;

[0072] in Indicates the critical priority that requires random backoff. express Backoff time for priority packets, is the channel load prediction value in step S12, is a coefficient factor related to priority. The lower the priority, the larger the value. for Priority business volume ratio, Indicates the length of the unit backoff time. Indicates random jitter;

[0073] Step S15: After the backoff timer ends, check whether the data packet has expired. If it has expired, discard it directly and go to step S10 to serve the next data packet.

[0074] If it has not expired, go to step S12 to re-predict the channel load value of the current period;

[0075] If a higher priority data packet arrives during the backoff process, the backoff timing is paused and the process goes to step S10 to give priority to serving the newly arrived high priority data packet.

[0076] Therefore, the present invention adopts the above-mentioned SPMA protocol design method based on BiLSTM channel load prediction, and the beneficial effects are as follows:

[0077] (1) The present invention corrects the channel load statistical error caused by the node using virtual full-duplex technology and the simultaneous arrival of multiple pulses, thereby improving the accuracy of channel load statistics. At the same time, the physical layer channel load based on the local node avoids information interaction and saves channel resources.

[0078] (2) The present invention constructs a BiLSTM channel load prediction model, which achieves high-precision prediction of high-burst traffic in self-organizing networks and effectively improves channel utilization.

[0079] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is an overall flow chart of an embodiment of an SPMA protocol design method based on BiLSTM channel load prediction of the present invention;

[0081] Figure 2 It is a BiLSTM channel load prediction model structure diagram of an SPMA protocol design method embodiment based on BiLSTM channel load prediction of the present invention;

[0082] Figure 3 It is a schematic diagram of a BiLSTM channel load prediction model and a weighted average channel load prediction result of an SPMA protocol design method embodiment based on BiLSTM channel load prediction of the present invention, wherein (a), (b), (c), and (d) respectively represent simulation processes at four different times;

[0083] Figure 4The invention discloses a method for designing an SPMA protocol based on BiLSTM channel load prediction, and shows a data packet sending success rate under CSMA and SPMA protocols. DETAILED DESCRIPTION

[0084] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0085] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0086] As shown in the figure, a SPMA protocol design method based on BiLSTM channel load prediction includes the following steps:

[0087] Step S1: define the node channel load statistics period ,Record The number of pulses sent by the local node at each frequency within a certain period of time , then the local node The number of pulses sent in time is:

[0088] ;

[0089] in, Refers to Frequency points, is the total number of frequency hopping frequencies;

[0090] Similarly, the local node is obtained The number of pulses received during this time is:

[0091] ;

[0092] Set the local node statistics period The average value of the sum of the number of pulses sent and received at each frequency point within a certain period of time is taken as the node channel load statistics value, that is:

[0093] .

[0094] Step S2: In a mobile ad hoc network, especially under high load conditions, the number of received pulses counted by a node is smaller than its true value, that is, there is an error in the load statistics value of step S1.

[0095] There are two main reasons for this error: first, the node uses virtual full-duplex technology, which makes it impossible to receive pulses when sending pulses; second, pulses sent by multiple nodes arrive at the receiving end at the same time, causing conflicts.

[0096] Therefore, it is necessary to correct the node channel load statistics in step S1 to obtain a corrected channel load statistics result to make it closer to the true value and reduce the error introduced, specifically:

[0097] Assume that the pulse arrives at the physical layer with a rate of Poisson distribution, the total number of frequency points is , the pulse arrival rate of a single channel is , then the time period Inside The probability of a pulse arriving is:

[0098] ;

[0099] Assume the pulse bit length is , the baseband processing rate is , then the time required to send or receive a single pulse is:

[0100] ;

[0101] In addition, the protection interval length is , then the probability that a node is at the pulse sending or receiving time is:

[0102] ;

[0103] The probability of being in the guard interval is:

[0104] ;

[0105] Set pulse reception without collision as event , then the probability of pulse reception without conflict is:

[0106] ;

[0107] in, , , , and All of them are the first type of error. is the second error;

[0108] Therefore, the corrected channel load statistics are:

[0109] .

[0110] Step S3: Obtain channel load statistics in the network and construct a channel load statistics data set.

[0111] Step S4: Perform Z-Score normalization on the channel load statistical result data set, and divide it sequentially in a ratio of 3:1:1 as a training set, a validation set, and a test set, respectively.

[0112] Step S5: Reconstruct the input and output of the channel load statistics result data set using a sliding window mode, specifically:

[0113] Processing t In each cycle, t The channel load of the cycle is taken as the output data label y, t - m To t -1 cycle of channel load as input data feature matrix x.

[0114] Step S6: construct a BiLSTM neural network channel load prediction model, which includes an input layer, a BiLSTM layer and an output layer, and use the Adam algorithm to optimize the network.

[0115] The present invention constructs a BiLSTM neural network channel load prediction model by improving the LSTM neural network. Figure 2 As shown in the figure, a single neuron structure of the LSTM neural network contains three gating mechanisms, namely the forget gate, input gate, and output gate;

[0116] Among them, the calculation formula of the forget gate is as follows:

[0117] ;

[0118] In the formula, is the output of the forget gate, is the activation function, for The hidden layer output at the moment, for Input at any time, , and These are forget gate parameters;

[0119] The input gate calculation formula is as follows:

[0120] ;

[0121] In the formula, is the input gate output, , and These are input gate parameters.

[0122] LSTM neurons pass Determine the cell state at the previous moment Cellular state The degree of influence, while through Determine the cell state update value Cellular state The specific calculation formula is as follows:

[0123] ;

[0124] ;

[0125] In the formula, is the activation function;

[0126] The output gate calculation formula is as follows:

[0127] ;

[0128] In the formula, is the output gate output, , and are the output gate parameters.

[0129] Finally, the output gate outputs and cell state Calculate the hidden layer state , the specific calculation formula is as follows:

[0130] ;

[0131] The above LSTM neuron calculation process is recorded as , then the BiLSTM cell output is:

[0132] ;

[0133] in, for The hidden layer state of the forward LSTM network at any moment, for Hidden layer state of the time-reversed LSTM network.

[0134] Step S7: Set learning rate and the number of training rounds m , train the BiLSTM neural network channel load prediction model.

[0135] Step S8: After the training is completed, the validation set is used to evaluate the performance of the model. The root mean square error RSME is used as the evaluation indicator. The calculation formula of the root mean square error RSME is:

[0136] ;

[0137] in, is the model prediction value, is the true value, is the number of test set samples.

[0138] If the value of the root mean square error RSME meets the accuracy requirement, the process proceeds to step S9, otherwise the process returns to step S7.

[0139] The RMSE of the model trained in this embodiment is 0.094, and the comparison between the model prediction results and the true value and weighted average prediction results is shown in the figure below. Figure 3 As shown in the figure, the red curve represents the true value of the channel load, the blue curve represents the predicted value of the BiLSTM model, and the green curve represents the weighted average predicted value. The BiLSTM model is closer to the true value than the weighted average method.

[0140] Step S9: When the data packet with the priority mark arrives at the MAC layer, the queue scheduling module inserts the data packet with the priority mark into the corresponding priority queue according to the data packet priority.

[0141] The queue scheduling module adopts the strict priority (SP) scheduling principle, that is, only when the high-priority queue is empty, the data packets in the low-priority queue have the opportunity to be taken out and sent; for data packets in the same priority queue, the first-in-first-out (FIFO) principle is adopted.

[0142] After step S9, the method further includes the following steps:

[0143] Step S10, check all queues from high to low priority, stop at the first non-empty queue, take out the first data packet of the queue to determine whether it is expired, if not expired, put it into the waiting state; if expired, discard it directly, and check again from high to low priority.

[0144] Step S11: Set the sending threshold of the lowest priority of the data packet to , other high priority thresholds are set according to the priority business ratio using the following formula:

[0145] ;

[0146] in, Indicates priority Threshold, Priority within the historical period Business volume share.

[0147] Step S12: Use the BiLSTM neural network channel load prediction model to predict the channel load value in the current cycle, and denormalize the result.

[0148] Step S13: Compare the priority threshold of the data packet to be sent with the channel load statistics result obtained in step S12:

[0149] When the threshold is greater than the channel load statistic, the data packet is sent directly;

[0150] If the threshold is smaller than the channel load statistic, the process proceeds to step S14.

[0151] Step S14: Set the random backoff time length to:

[0152] ;

[0153] in Indicates the critical priority that requires random backoff. express Backoff time for priority packets, is the channel load prediction value in step S12, is a coefficient factor related to priority. The lower the priority, the larger the value. for Priority business volume ratio, Indicates the length of the unit backoff time. Indicates random jitter;

[0154] Step S15: After the backoff timer ends, check whether the data packet has expired. If it has expired, discard it directly and go to step S10 to serve the next data packet.

[0155] If it has not expired, go to step S12 to re-predict the channel load value of the current period;

[0156] If a higher priority data packet arrives during the backoff process, the backoff timing is paused and the process goes to step S10 to give priority to serving the newly arrived high priority data packet.

[0157] The comparison of the data packet sending success rate of the SPMA protocol and the CSMA protocol designed in this embodiment under the same conditions is shown in the figure below: Figure 4 shown.

[0158] The blue curve in the figure is the packet sending success rate under the CSMA protocol, and the red curve is the test result under the SPMA protocol. The packet sending success rate under the two protocols remains unchanged as the number of packets to be sent increases, and then begins to decrease. This is because at the beginning, the number of packets to be sent has not reached system saturation, and the system can still send packets with a high success rate. However, as the number of packets continues to increase, the system begins to saturate, channel conflicts intensify, and the number of backoffs during packet sending increases. Therefore, in the same time, the number of packets sent successfully decreases, resulting in a decrease in the packet sending success rate. When the packet sending success rate drops to 90%, the number of packets to be sent that the CSMA protocol allows is about 170, while the SPMA protocol is about 500, indicating that the use of the SPMA protocol can enable the system to have a higher communication capacity. In addition, under the same number of packets to be sent, the packet sending success rate of the SPMA protocol is always better than that of the CSMA protocol, proving the effectiveness of the designed SPMA protocol.

[0159] Therefore, the present invention adopts the above-mentioned SPMA protocol design method based on BiLSTM channel load prediction, analyzes the errors in channel load statistics according to the working mechanism of the SPMA protocol, and corrects them; proposes a channel load prediction model based on the BiLSTM network, improves the accuracy of channel load prediction, and thus effectively improves the channel utilization efficiency.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A SPMA protocol design method based on BiLSTM channel load prediction, characterized in that: The following steps are involved: Step S1: Count the local node period The average value of the sum of the number of pulses sent and received at each frequency point within a certain period of time is taken as the node channel load statistic; Step S2, correcting the node channel load statistics in step S1 to obtain a corrected channel load statistics result; Step S3: Obtain channel load statistics in the network and construct a channel load statistics data set; Step S4, performing Z-Score normalization on the channel load statistical result data set, and dividing it sequentially in a ratio of 3:1:1 as a training set, a validation set, and a test set respectively; Step S5: Reconstruct the input and output of the channel load statistical result data set using a sliding window mode; Step S6: construct a BiLSTM neural network channel load prediction model, the model includes an input layer, a BiLSTM layer and an output layer, and use the Adam algorithm to optimize the network; Step S7: Set learning rate and the number of training rounds m , train the BiLSTM neural network channel load prediction model; Step S8: After the training is completed, the validation set is used to evaluate the performance of the model. The root mean square error RSME is selected as the evaluation indicator. If the value of the root mean square error RSME meets the accuracy requirement, the process proceeds to step S9, otherwise it returns to step S7. Step S9: When the data packet with the priority mark arrives at the MAC layer, the queue scheduling module inserts the data packet with the priority mark into the corresponding priority queue according to the data packet priority.

2. A SPMA protocol design method based on BiLSTM channel load prediction according to claim 1, characterized in that, Step S1 is specifically as follows: Define the node channel load statistics period ,Record The number of pulses sent by the local node at each frequency within a certain period of time , then the local node The number of pulses sent in time is: ； in, Indicates Frequency points, is the total number of frequency hopping frequencies; The local node The number of pulses received during this time is: ； Set the local node statistics period The average value of the sum of the number of pulses sent and received at each frequency point within a certain period of time is taken as the node channel load statistics value, that is: 。 3. A SPMA protocol design method based on BiLSTM channel load prediction according to claim 2, characterized in that, Step S2 is specifically as follows: Assume that the pulse arrives at the physical layer with a rate of Poisson distribution, the total number of frequency points is , the pulse arrival rate of a single channel is , then the time period Inside The probability of a pulse arriving is: ； Assume the pulse bit length is , the baseband processing rate is , then the time required to send or receive a single pulse is: ； In addition, the protection interval length is , then the probability that a node is at the pulse sending or receiving time is: ； The probability of being in the guard interval is: ； Set pulse reception without collision as event , then the probability of pulse reception without conflict is: ； in, , represents the probability of no pulse reception conflict when the node is in the pulse sending time; , which represents the probability of no pulse reception conflict when the node is in the pulse reception time, , which represents the probability of no pulse reception conflict when the node is in the protection interval, and All of them are the first type of error. is the second error; therefore, the corrected channel load statistics are: 。 4. A SPMA protocol design method based on BiLSTM channel load prediction according to claim 3, characterized in that: Step S5 is specifically as follows: Processing t In each cycle, t The channel load of the cycle is taken as the output data label y, t - m To t -1 cycle of channel load as input data feature matrix x.

5. A SPMA protocol design method based on BiLSTM channel load prediction according to claim 4, characterized in that, In step S6, a BiLSTM neural network channel load prediction model is constructed by improving the LSTM neural network. A single neuron structure of the LSTM neural network includes three gating mechanisms, namely a forget gate, an input gate, and an output gate. Among them, the calculation formula of the forget gate is as follows: ； In the formula, is the output of the forget gate, is the activation function, for The hidden layer output at the moment, for Input at any time, , and These are forget gate parameters; The input gate calculation formula is as follows: ； In the formula, is the input gate output, , and These are input gate parameters; LSTM neurons pass Determine the cell state at the previous moment Cellular state The degree of influence, while through Determine the cell state update value Cellular state The specific calculation formula is as follows: ； ； In the formula, is the activation function; The output gate calculation formula is as follows: ； In the formula, is the output gate output, , and is the output gate parameter; Finally, the output gate outputs and cell state Calculate the hidden layer state , the specific calculation formula is as follows: ； The above LSTM neuron calculation process is recorded as , then the BiLSTM cell output is: ； in, for The hidden layer state of the forward LSTM network at any moment, for Hidden layer state of the time-reversed LSTM network.

6. A SPMA protocol design method based on BiLSTM channel load prediction according to claim 5, characterized in that: In step S8, the calculation formula of the root mean square error RSME is: ； in, is the model prediction value, is the true value, is the number of test set samples.

7. A SPMA protocol design method based on BiLSTM channel load prediction according to claim 6, characterized in that: The queue scheduling module in step S9 adopts a strict priority scheduling principle, that is, only when the high priority queue is empty, the data packets in the low priority queue have the opportunity to be taken out and sent; for the data packets in the same priority queue, the first-in-first-out principle is adopted.

8. A SPMA protocol design method based on BiLSTM channel load prediction according to claim 7, characterized in that: After step S9, the method further includes the following steps: Step S10, check all queues from high to low priority, stop at the first non-empty queue, take out the first packet of the queue to determine whether it is expired, if not expired, transfer it to the waiting state; if expired, directly discard it, and re-check in order from high to low priority; Step S11: Set the sending threshold of the lowest priority of the data packet to , other high priority thresholds are set according to the priority business ratio using the following formula: ； in, Indicates priority Threshold, Priority within the historical period Business volume share; Step S12: Use the BiLSTM neural network channel load prediction model to predict the channel load value in the current cycle, and perform denormalization on the result; Step S13, compare the priority threshold of the data packet to be sent with the channel load statistics result obtained in step S12: when the threshold is greater than the channel load statistics value, directly send the data packet; if the threshold is less than the channel load statistics value, proceed to step S14; Step S14: Set the random backoff time length to: ； in Indicates the critical priority that requires random backoff. express Backoff time for priority packets, is the channel load prediction value in step S12, is a coefficient factor related to priority. The lower the priority, the larger the value. for Priority business volume ratio, Indicates the length of the unit backoff time. Indicates random jitter; Step S15: After the backoff timer ends, check whether the data packet has expired. If it has expired, discard it directly and go to step S10 to serve the next data packet. If it has not expired, go to step S12 to re-predict the channel load value of the current period; If a higher priority data packet arrives during the backoff process, the backoff timing is paused and the process goes to step S10 to give priority to serving the newly arrived high priority data packet.

Images