Method and device for positioning sending node in mobile molecule communication environment
By using the Transformer-based sending node positioning neural network model in the mobile molecular communication environment, the problem of insufficient positioning accuracy of sending nodes in the prior art is solved, high-precision sending node positioning is realized, and extended to a three-dimensional MMC environment.
Patent Information
- Application Number
- CN202411972064.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-06-03
AI Technical Summary
The prior art is difficult to effectively locate the sending node in a mobile molecular communication environment, especially when deep learning methods are mainly applied to static environments.
A transmission node positioning neural network model based on Transformer is adopted, including a location encoding module, a Transformer encoder module and a linear layer module, and the model is trained to predict the position of the transmission node.
It improves the accuracy of sending node positioning, realizes high-precision positioning in mobile molecular communication systems, has better performance than traditional DNN models, and can be expanded to a three-dimensional MMC environment.
Smart Images
Figure CN120090716A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of molecular communication technology, and particularly relates to a method and device for locating a sending node in a mobile molecular communication environment. Background Art
[0002] Molecular communication (MC) is an emerging communication paradigm that uses biological molecules as a transmission medium to achieve nanoscale communication. It has significant application prospects in fields such as the monitoring and treatment of human health conditions, especially in the communication and control between micro-medical devices.
[0003] A key feature of a mobile molecular communication system (MMC) is that the sending node or the receiving node is constantly in a moving state in the environment under the action of diffusion, and molecules can still be sent during the movement of the sending node to communicate with the receiving node. Therefore, in a mobile molecular communication system, the positioning problem of the sending node or the receiving node is very important. In the research field of node positioning problems, deep learning methods have attracted much attention and have been widely used, aiming to improve the accuracy of positioning. Static node positioning methods based on deep neural networks and positioning methods designed based on the Transformer neural network in an environment with flow velocity have been successively proposed to optimize node positioning.
[0004] Most current research work focuses on using deep learning methods to carry out node positioning in a static environment. Therefore, how to extend deep learning node positioning methods to a mobile molecular communication environment has become a challenging task. Summary of the Invention
[0005] The purpose of this application is to provide a method and device for locating a sending node in a mobile molecular communication environment, and it is also easy to be extended and applied to a three-dimensional MMC system.
[0006] To achieve the above purpose, the technical solution of this application is as follows: A method for locating a sending node in a mobile molecular communication environment, comprising: Construct a sending node positioning neural network model based on Transformer, where the sending node positioning neural network model includes a position encoding module, a Transformer encoder module, and a linear layer module; Generate sample data for training the sending node positioning neural network model, and use the generated sample data to train the sending node positioning neural network model; Construct a sequence of the number of received molecules corresponding to the receiving node in the mobile molecular communication system to be located, and use the trained sending node positioning neural network model to predict the position of the sending node.
[0007] Furthermore, the Transformer encoder module includes a multi-head attention layer, a feed-forward neural network layer, a residual connection, and a normalization layer. Among them, the input of the Transformer encoder module is combined with the output of the feed-forward neural network layer through a residual connection, and then passes through the normalization layer to obtain the final output of the Transformer encoder.
[0008] Furthermore, the encoding method of the position encoding module is as follows: ; where represents the dimension of the position encoding, is the time series index of the number of molecules captured by the corresponding receiving node in the input sequence. 2n and 2n + 1 respectively represent the even index and odd index of the time series of the number of molecules.
[0009] Furthermore, generating the sample data for training the sending node positioning neural network model includes: Simulating the experiment I times. One experiment corresponds to a movement process of the sending node Tx. The total duration of one movement process is discretely divided into time slots . At the beginning of each time slot, the sending node sends a type of molecule different from the previous one, and all receiving nodes capture and absorb the molecules released by the sending node within the current time slot; Set the initial coordinates of the sending node Tx, and set the initial position coordinates of the sending node Tx according to the minimum distance between the sending node Tx and the receiving node ; Discretize the time slot into W time slots, and the time step is ; Use to represent the data sequence composed of the number of molecules received by the receiving node in the m-th time slot of the i-th movement among the W ; For the m-th time slot of the i-th movement, use to represent the absorption molecule time series of K receiving nodes, is represented as: P i,m =[ N R x 1 i,m , N R x 2 i,m , N R x 3 i,m ,…, N R x K i,m ] ; Use C i,mdenote the position of the sending node Tx at the start of the m-th time slot during the i-th movement, and the data samples H of M time slots during the i-th movement i can be expressed as: ; A total of I experiments are simulated, and each experiment corresponds to a movement process of the sending node Tx. Then the entire dataset L I can be expressed as: .
[0010] This application also proposes a positioning device for a sending node in a mobile molecular communication environment, including a processor and a memory storing a number of computer instructions. When the computer instructions are executed by the processor, the steps of the positioning method for the sending node in the mobile molecular communication environment are implemented.
[0011] A positioning method and device for a sending node in a mobile molecular communication environment proposed by this application use a Transformer-based model to predict the position of the sending node. This model can also be extended to a three-dimensional MMC environment. The multi-head attention mechanism in this model effectively captures the characteristics of the molecular sequences received by multiple receiving nodes, improves the accuracy of sending node positioning, is superior to the traditional DNN model in performance, realizes high-precision positioning of mobile nodes, and improves the prediction accuracy of the model in a mobile molecular communication system. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 is a schematic diagram of a mobile molecular communication system.
[0013] Figure 2 is a flowchart of the positioning method for the sending node in the mobile molecular communication system of this application.
[0014] Figure 3 is a schematic diagram of the structure of the neural network model for sending node positioning of this application.
[0015] Figure 4 is the experimental result of the average error distance distribution.
[0016] Figure 5 is a comparison chart of the loss trajectories of the model of this application and the DNN model during the training process.
[0017] Figure 6 is a comparison chart of the prediction performance of the model of this application under different numbers of receiving nodes.
[0018] Figure 7 is a comparison chart of the performance when predicting the position of the sending node with different molecular diffusion coefficients. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0019] To make the objectives, technical solutions and advantages of this application more clear and understandable, the following further details this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.
[0020] Figure 1 A mobile molecular communication system (MMC) applicable to this application is shown. The overall shape is two-dimensional and unbounded. The communication topology includes a mobile point sending node (Tx) and K fixed circular receiving nodes , , , …, . The radius of the receiving node is r k , and the coordinates are .
[0021] In the MMC environment, the total duration of a movement process of Tx is discretely divided into time slots . At the beginning of each time slot, the sending node will send a type of molecule different from the previous one, and all receiving nodes capture and absorb the molecules released by the sending node within the current time slot. Let represent the diffusion coefficient of Tx. Then the movement step along the x-axis and the movement step along the y-axis of Tx within one time slot can be calculated using the following normal distribution:
[0022] ); ); At the initial moment of the first time slot, Tx sends N type molecules. In the m-th time slot thereafter, Tx sends N . The sending node and all receiving nodes are clock-synchronized, and within each time slot, all receiving nodes only absorb the molecules released within this time slot and will not absorb molecules of other types.
[0023] However, due to the inherent limitations of using computer simulation to simulate continuous absorption, this application discretizes the time slot into W time slots with a time step of . Then there is:
[0024] ; This application mainly predicts the position of the sending node at the beginning of each time slot by using the sequence of the number of molecules captured by K receiving nodes from the mobile sending node within each time slot. Assume that the current time slot is the m-th time slot in the i-th movement. Then within the time slot, these are ( The sequence of the number of captured molecules is represented by .
[0025] For each molecule of type , the step sizes of its movement along the x-axis and y-axis within are represented by and respectively; The displacements of the molecules along the x and y axes caused by diffusion exhibit a normal distribution characteristic: ; where represents the diffusion coefficient of the information molecules of type transmitted from Tx in the m-th time slot, is a normal distribution with a mean of and a variance of ; Considering that the molecules are released by Tx at the initial moment of the time slot and received by , any molecule will be received by according to the following conditional formula: ; where and represent the coordinate values of the k-th receiving node on the and axes respectively; ( (m, τ), (m, τ)) represents the position of the type molecules at the τ-th moment in the m-th time slot. For the information molecules of type , if they satisfy the above conditional formula, these information molecules will be absorbed and recorded by the receiving node at the τ-th moment in the m-th time slot. Assuming that the current time slot is the m-th time slot in the i-th movement, then the number of molecules captured by within the time slot is the value of . is represented as:
[0026] N R x k i,m =[ N R x k i,m Δ t , N R x k i,m 2Δ t , N R x k i,m 3Δ t ,…, N R x k i,m WΔ t ] .
[0027] For the m-th time slot of the i-th movement, representing the time series of absorption molecules of K receiving nodes, is represented as: P i,m =[ N R x 1 i,m , N R x 2 i,m , N R x 3 i,m ,…, N R x K i,m ] .
[0028] In one embodiment, as Figure 2 shown, a method for locating a sending node in a mobile molecular communication environment is proposed, including: Step S1, constructing a neural network model for locating a sending node based on Transformer, where the neural network model for locating a sending node includes a position encoding module, a Transformer encoder module, and a linear layer module.
[0029] To accurately predict the position of a mobile sending node in a mobile molecular communication system, this example uses a neural network model for locating a sending node based on Transformer, as Figure 3 shown, including a position encoding module, a Transformer encoder module, and a linear layer module. This model does not use the Transformer decoder part, but instead uses a linear layer to transform and output the coordinates of the sending node. In the model inference stage, the simple structure of the linear layer makes the calculation speed faster. Since there is no need to perform complex operations such as query, key, and value calculations and concatenations in multi-head attention, as well as subsequent normalization operations, the output coordinates can be generated faster, and the position information of the sending node can be updated. During the model training process, the linear layer converges more easily than the decoder, and at the same time, the gradient calculation of the linear layer is relatively simple, and problems such as gradient disappearance or gradient explosion will not occur as in the multi-head attention mechanism in the decoder.
[0030] Figure 3 shows the detailed structure of the neural network model for locating a sending node based on Transformer. For the m-th time slot in the i-th movement, the input sequence is the time series of absorption molecules of K receiving nodes , and its order plays a crucial role in the task.
[0031] Therefore, in this example, the position encoding is added to the input sequence through the position encoding module, represented as and , is the time series index of the number of molecules captured by the corresponding receiving node in the input sequence, where 2n and 2n + 1 represent the even index and odd index of the molecule number time series, respectively.
[0032] Among them, the positional encoding can be calculated as follows: ; where represents the dimension of the positional encoding.
[0033] The data processed by the positional encoding is called the enhanced input, which can provide more feature information of the input sequence, which is beneficial for the model to predict the position of the sending node.
[0034] The Transformer encoder module includes a multi-head attention layer, a feed-forward neural network layer (FNN), a residual connection, and a normalization layer.
[0035] In this example, the multi-head attention layer enhances the model's ability to capture the interdependencies between time series data, which represents the number of molecules received by different receiving nodes. This method significantly improves the prediction accuracy.
[0036] The multi-head attention layer multiplies the time series of the number of molecules received from Tx by all receiving nodes with the parameter matrices , and to obtain the query, key, and value matrices , and .
[0037] The multi-head attention mechanism can be calculated by the following formula: ; Among them, the query matrix (Q) is responsible for calculating the attention weights, which represents the importance of the current position in the attention mechanism. The key matrix (G) provides information representing the importance of each position. It uses the query vector to perform similarity calculations to help determine other positions most relevant to the current position. The value matrix (V) contains the feature information of the current position. The value vector is weighted and averaged by the attention weights to obtain the final representation of the current position, and d is the scaling factor.
[0038] The input of the Transformer encoder module is combined with the output of the feed-forward neural network layer through a residual connection, and then passes through the normalization layer to obtain the final output of the Transformer encoder. It reduces the overfitting of the model, enhances the feature representation ability of the model, makes the model easier to be optimized during training, and avoids the problem of gradient disappearance in the network.
[0039] The normalization layer standardizes the input features, improving the stability and training speed of the model.
[0040] After the Transformer encoder module, there is a linear layer output module, and the coordinates of the output Tx are obtained through the linear transformation of the fully connected layer.
[0041] It should be noted that it is a relatively mature technology that the sample data is input into the sending node positioning neural network model, position encoding is performed in the model, and then it is input into the multi-head attention through linear transformation in the Transformer encoder module, which will not be elaborated here.
[0042] Step S2: Generate sample data for training the sending node positioning neural network model, and use the generated sample data to train the sending node positioning neural network model.
[0043] To generate the sample data required for training, referring to the relevant inferences of the previous mobile molecular communication system, the sample data is obtained by setting the initial coordinates of the mobile sending node Tx and the coordinates of K fixed receiving nodes.
[0044] For example, set the number of receiving nodes K to 4, and set their coordinates to , , and .
[0045] It should be noted that when generating the mobile sending node Tx, ensure that Tx does not touch any receiving nodes.
[0046] Therefore, for the m-th time slot in the i-th movement, within the time slot, among the W time steps , these captured molecular numbers form a sequence which is . The representation of
[0047] N R x k i,m =[ N R x k i,m Δ t , N R x k i,m 2Δ t , N R x k i,m 3Δ t ,…, N R x k i,m WΔ t ] ; For the m-th time slot of the i-th movement, The absorption molecule time series representing K receiving nodes is represented as: P i,m =[ N R x 1 i,m , N R x 2 i,m , N R x 3 i,m ,…, N R x K i,m ] 。
[0048] Let C i,m represent the position of Tx at the start of the m-th time slot during the i-th movement. The data samples H of M time slots during the i-th movement i can be represented as: ; That is to say, each sample data , corresponds to a position of Tx, and a received molecule sequence of the number of molecules received by K receiving nodes; each H i corresponds to the sample data within all time slots during the i-th movement of the sending node.
[0049] A total of I experiments are simulated, and each experiment corresponds to a movement process of Tx. Then the entire data set L I can be represented as: 。
[0050] Thus, by training the sending node localization neural network model with the above sample data set, the trained sending node localization neural network model can be used to locate the sending node.
[0051] In this embodiment, a total of 67,200 data samples are generated, 80% of which are allocated to the training set and 20% are allocated to the test data set. To locate the sending node, α 1 samples are used in the test data set. α 1 is set to 13400 and the learning rate is 0.0001.
[0052] During the training process, the mean squared error (MSE) loss function is used to evaluate the performance of the model. Let S represent the batch size, S = 64. Let u represent the index of the samples in each training batch S, and let represent the coordinates of the u-th sending node in this batch. The MSE loss value can be calculated from and the predicted coordinates as follows:
[0053] ℒ MSE = 1 S ∑ u=1 S [ (x Tx ,u − x ̂ Tx ,u ) 2 +( y Tx ,u − y ̂ Tx ,u ) 2 ] 。
[0054] During the testing process, is the error distance of the v-th sample in the test dataset. represents the actual coordinates of Tx in the v-th sample of the test dataset. is the predicted coordinate corresponding to Tx in the v-th sample. The calculation method is as follows:
[0055] ; In addition, for each sample obtained, is compared with the threshold and is used to indicate whether the prediction is correct. The calculation method of
[0056] is as follows:
[0057] If equals 1, it is considered that this prediction is correct. Then, the total number of correct predictions can be calculated in the following way:
[0058] is as follows:
[0059] The accuracy rate of the test (denoted as AoP) can be calculated in the following way, where the threshold is used: is as follows:
[0060] Regarding the training of the neural network model, it is a relatively mature technology in this field and will not be elaborated here.
[0061] Step S3: Construct the sequence of the number of received molecules corresponding to the receiving nodes in the mobile molecular communication system to be located, and use the trained neural network model for sending node positioning to predict the position of the sending node.
[0062] According to the number of molecules of each type received by each receiving node under the diffusion effect in each time slot, the sequence of the number of received molecules corresponding to the receiving nodes in the mobile molecular communication system to be located can be constructed. Input it into the trained neural network model for sending node positioning, and thus obtain the prediction result of the position of the sending node.
[0063] To evaluate the location prediction performance of the Transformer-based neural network model for the sending node in a mobile molecular communication system, experiments were conducted to demonstrate the prediction accuracy of the Transformer model. Meanwhile, on the same test dataset, a comparative analysis was carried out between the Transformer model and the DNN-based model. All the parameters used in the experimental process are shown in Table 1:
[0064] Table 1 Parameter Value #timg# 0.1#timg# I 16,000 #timg# 5 #timg# 0.1#timg# #timg# 100 #timg# #timg# #timg# #timg# #timg# #timg# <![CDATA[r k > 2#timg# #timg# #timg# To visually present the statistical results of the proposed model, Table 2 conducts a comparative analysis of the proposed model and the DNN-based model in several statistical metrics. Among them, the error is determined by the distance between the predicted coordinates and the corresponding actual coordinates, denoted as , and the average error is the mean of all samples in the test dataset. The results in Table 2 clearly show and compare the performance characteristics of these two models in a quantitative form. By observing Table 2, it can be significantly found that when the value is the same, the accuracy achieved by the model of this application is higher than that of the DNN-based model.
[0065] Table 2 Model Transfomer Model DNN Model Accuracy#timg# #timg# #timg# Accuracy#timg# #timg# #timg# Accuracy#timg# #timg# #timg# Mean Error Example#timg# 0.86 1.98 To better evaluate the prediction performance of the Transformer-based model, α r is defined as the number of samples within the error range of the error distances of α 1 test samples. The error values are integer multiples of 0.05 µm from 0 µm to 2.5 µm, and the ratio calculation method is α r / α 1 . As Figure 4 shown, most of the prediction errors from the test samples are concentrated on the left side of the histogram. This distribution indicates that the predictions of most test samples are normal and consistent with the statistical results in Table 2. On the contrary, on the right side of the histogram, only a few data points have large prediction errors. These instances can be considered reasonable anomalies because they only represent a small part of the test dataset. The overall trend of the histogram first rises and then falls. The rising speed is significantly faster than the falling speed. This is because the error distances of most test samples are relatively small and concentrated on the left side of the histogram. Although there are some rising phenomena during the falling process, the rising speed is very small and does not affect the overall falling trend.
[0066] Figure 5Shows the changes in the loss function during the training of two models on the same dataset. Notably, the initial loss function of the Transformer-based model is smaller than that of the DNN-based model. The loss function of the Transformer-based model converges faster, converging around 40 epochs, while the DNN-based model converges around 60 epochs. Additionally, when the Transformer-based model converges, its loss function value is significantly smaller than that of the DNN-based model.
[0067] Figure 6 Evaluated the impact of the number of receiving nodes on the prediction accuracy of the sending node's position in the MMC environment. When the threshold is = 1.2 µm, the prediction accuracy with only one receiving node is significantly lower, at 26.73%. However, as the number of receiving nodes gradually increases, the accuracy improves significantly. For example, with 4 receiving nodes, the prediction accuracy reaches 95.41%, and with 5 receiving nodes, the prediction accuracy further increases to 95.97%. When K > 4, the improvement in prediction accuracy is very small. The result behind this trend is that each receiving node captures molecular signals of different time series in each time slot. In the case of fewer than 4 receiving nodes, adding more receiving nodes can increase the amount of available data for predicting the Tx position. When using 4 receiving nodes, the prediction accuracy has reached a high level. Therefore, the incremental gain in accuracy for each receiving node will decrease. Thus, to obtain higher prediction accuracy while saving resources, 4 receiving nodes is the optimal choice.
[0068] Figure 7 Shows the impact of the molecular diffusion coefficient on the prediction accuracy. Set to 100 、500 and 1000 . Under the condition of = 0.6 µm, the prediction accuracies of the model are 81.63%, 83.77%, and 86.19% respectively. It can be seen that when the value increases, the prediction accuracy of the model is also higher. This is because as the diffusion coefficient increases, the number of molecules collected by the receiving nodes increases. This enhancement effectively reduces the risk of communication loss between the sending node and the receiving node due to insufficient molecules arriving at the receiving node within one time slot.
[0069] This application proposes a sender node location predictor for a Transformer model in an MMC environment, and this model can also be extended to a three-dimensional MMC environment. The multi-head attention mechanism in this model effectively captures the features of the molecular sequences received by multiple receiving nodes, improving the positioning accuracy of the sender node. To verify the superiority of the model, a comparative experiment was conducted with a DNN-based model, and the results show that the model of this application achieved better performance. In addition, even when environmental factors change, the Transformer-based model still maintains stability.
[0070] In another embodiment, this application also proposes a sender node positioning device in a mobile molecular communication environment, including a processor and a memory storing a number of computer instructions. When the computer instructions are executed by the processor, the steps of the sender node positioning method in the mobile molecular communication environment are implemented.
[0071] For the specific limitations of the sender node positioning device in the mobile molecular communication environment, reference can be made to the limitations of the sender node positioning method in the mobile molecular communication environment described above, which will not be elaborated here. The above-mentioned sender node positioning device in the mobile molecular communication environment can be implemented in whole or in part through software, hardware, and their combination. It can be embedded in the processor of a computer device in hardware form or be independent of it, or be stored in the memory of a computer device in software form for the processor to call and execute the corresponding operations above.
[0072] The memory and the processor are directly or indirectly electrically connected to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines. The memory stores a computer program that can run on the processor, and the processor realizes the method in the embodiments of the present invention by running the computer program stored in the memory.
[0073] Among them, the memory can be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. Among them, the memory is used to store the program, and the processor executes the program after receiving the execution instruction.
[0074] The processor may be an integrated circuit chip with the ability to process data. The above-mentioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0075] The above embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.
Claims
1. A method for locating a sending node in a mobile molecular communication environment, characterized in that: The sending node positioning method in the mobile molecular communication environment comprises: Constructing a Transformer-based sending node positioning neural network model, wherein the sending node positioning neural network model includes a position encoding module, a Transformer encoder module, and a linear layer module; Generate sample data for training the sending node positioning neural network model, and use the generated sample data to train the sending node positioning neural network model; Construct a sequence of the number of received molecules corresponding to the receiving node in the mobile molecular communication system that needs to be positioned, and use the trained sending node positioning neural network model to predict the position of the sending node.
2. The method for locating a sending node in a mobile molecular communication environment according to claim 1, characterized in that: The Transformer encoder module includes a multi-head attention layer, a feedforward neural network layer, a residual connection and a normalization layer, wherein the input of the Transformer encoder module is combined with the output of the feedforward neural network layer through a residual connection, and then passes through the normalization layer to obtain the final output of the Transformer encoder.
3. The method for locating a sending node in a mobile molecular communication environment according to claim 1, characterized in that: The encoding method of the position encoding module is as follows: ； in represents the dimension of the position encoding, is the time series index of the number of molecules captured by the corresponding receiving node in the input sequence, 2n and 2n+1 represent the even index and odd index of the time series of the number of molecules, respectively.
4. The method for locating a sending node in a mobile molecular communication environment according to claim 1, characterized in that: The generating of sample data for training the sending node positioning neural network model comprises: The simulation experiment is repeated I times. One experiment corresponds to a movement process of the sending node Tx. The total duration of one movement process is discretely divided into Time slots At the beginning of each time slot, the sending node will send a different type of molecule than the previous one, and all receiving nodes capture and absorb the molecules released by the sending node in the current time slot; Set the initial coordinates of the sending node Tx, according to the sending node Tx and the receiving node The minimum distance between them is used to set the initial position coordinates of the sending node Tx; Time slot Discretized into W time slots, the time step is ; use Receiving node In the mth time slot of the ith movement, W A data sequence consisting of the number of molecules received in ; For the mth time slot of the ith movement, use represents the absorption molecule time series of K receiving nodes, The representation is: ； Using C i,m To represent the position of the transmitting node Tx at the beginning of the mth time slot during the i-th movement, the data sample H of the Mth time slot in the i-th movement i It can be expressed as: ； A total of I experiments are simulated, each experiment corresponds to a movement process of the sending node Tx, so the total data set L I It can be expressed as: 。 5. A sending node positioning device in a mobile molecular communication environment, comprising a processor and a memory storing a plurality of computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method described in any one of claims 1 to 4 are implemented.