Abnormal diffusion molecule communication receiver positioning method and device in vascular environment

By combining the EnformerAttention module and the coordinate-specific processing module, the localization challenges caused by abnormal diffusion and transmitter mobility in the vascular environment are solved, achieving accurate and rapid localization of the target receiver in the vascular environment and improving the robustness and accuracy of the localization model.

CN121144734APending Publication Date: 2025-12-16ZHEJIANG UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511260110.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-04
Publication Date
2025-12-16

AI Technical Summary

Technical Problem

In vascular environments, existing localization methods cannot effectively handle signal transmission challenges caused by anomalous diffusion and transmitter mobility, resulting in inaccurate and unrobust receiver localization.

Method used

A receiver localization model employing the EnformerAttention module and a coordinate-specific processing module, combined with multiple EnformerAttention layers and fully connected layers, achieves accurate target receiver localization by learning position encoding and the sequence of information molecules.

Benefits of technology

Robust and accurate target receiver localization is achieved in complex 3D vascular environments, significantly improving localization accuracy and convergence speed, surpassing traditional Transformer and DNN models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121144734A_ABST
    Figure CN121144734A_ABST
Patent Text Reader

Abstract

The invention discloses a method and a device for positioning an abnormal diffusion molecule communication receiver in a vascular environment, which are applied to a mobile molecule communication system in a first vascular environment, and the mobile molecule communication system in the first vascular environment comprises a mobile emitter, a reference receiver at a fixed position and a target receiver at an unknown position. The emitter releases a preset number of information molecules to the blood vessel environment at each position, and each receiver receives the information molecules. According to the method and device for positioning the abnormal diffusion molecule communication receiver in the vascular environment, the receiver positioning model based on Enformer is adopted to position the position of the target receiver, robust and accurate receiver positioning in a complex three-dimensional scene is achieved, the method and device remarkably exceed traditional Transformer and DNN models in performance, and the method and the device have the advantages that the method and the device are easy to implement, and the method and the device are suitable for popularization and application. And faster convergence, higher precision and stronger robustness are realized.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of molecular communication, and particularly relates to a method and device for positioning an abnormal diffusion molecular communication receiver in a blood vessel environment. BACKGROUND

[0002] Molecular communication (MC) is a communication form that uses biological molecules as information carriers, and has significant application prospects in nanonetworks (IoNT), especially in biomedical fields such as in-vivo nanonetworks and targeted drug delivery systems. However, mobile molecular communication (MMC) in a blood vessel environment faces challenges in signal transmission under complex scenarios such as abnormal diffusion, fluid flow and transmitter mobility.

[0003] Traditional positioning methods usually rely on complex mathematical derivations and analytical formulas, and often perform poorly under the complex and dynamic conditions of the blood vessel environment. Existing deep learning methods mainly focus on the simplified environment assumption of standard Brownian motion, ignoring the key complexity of the abnormal diffusion phenomenon frequently observed in biological media. In addition, most existing works assume that the transmitter is stationary, ignoring the inherent diffusion-driven mobility of nanoscale transmitters naturally exhibited in the blood vessel environment. Therefore, how to achieve accurate and robust receiver positioning in the blood vessel environment considering abnormal diffusion and transmitter mobility is a challenging problem. SUMMARY

[0004] The application aims to solve the problems raised in the background art, and provides a method and device for positioning an abnormal diffusion molecular communication receiver in a blood vessel environment.

[0005] To achieve the above-mentioned purpose, the technical solution adopted by the application is as follows:

[0006] The application provides a method for positioning an abnormal diffusion molecular communication receiver in a blood vessel environment, which is applied to a mobile molecular communication system in a first blood vessel environment. The mobile molecular communication system in the first blood vessel environment includes a mobile transmitter, K-1 fixed-position reference receivers and a target receiver with an unknown position. The transmitter releases information molecules into the blood vessel environment at each position, and each receiver receives the information molecules.

[0007] In an observation period T, the number of information molecules received by the target receiver and all reference receivers in each first time step Δt is formed into a time series, and the time series is taken as the input of a trained receiver positioning model to obtain the position coordinates of the target receiver, wherein Δt is the size of each time period after the observation period T is uniformly divided, and Δt includes the start time point and the end time point.

[0008] Preferably, the receiver positioning model comprises an input signal processing module, an Enformer Attention module and a coordinate-specific processing module connected in sequence.

[0009] Preferably, the time series representation is X counts ∈R K×J , K is the number of all receivers, J is the length of the time series, where Δt = T / J;

[0010] In the input signal processing module, the time series X counts is projected into an embedding space to obtain features X emb ∈R K ×J×d , d is the embedding dimension;

[0011] The features projected into the embedding space are subjected to learnable position encoding, and the formula is as follows:

[0012] E = X emb + P pos ;

[0013] wherein E is the result of the features projected into the embedding space subjected to learnable position encoding, and E ∈R K×J×d , P pos is the learnable position encoding, and P pos ∈R J×d , P pos,j = Embedding(j), j ∈ {1, 2,..., J}, j is a position index.

[0014] Preferably, the Enformer Attention module comprises multiple layers of Enformer Attention layers connected in sequence, and each layer of Enformer Attention layer comprises three parallel Enformer Attention units, each Enformer Attention unit comprises a multi-head self-attention layer, a first layer normalization layer, a convolution layer and a second layer normalization layer connected in sequence, and the output of the multi-head self-attention layer is added to the input of the multi-head self-attention layer to serve as the input of the first layer normalization layer, and the output of the convolution layer is added to the input of the convolution layer to serve as the input of the second layer normalization layer.

[0015] For the first layer of EnformerAttention, the result E after learning the position encoding is taken as the input of each EnformerAttention unit in the first layer of EnformerAttention, and for the remaining layers of EnformerAttention, the output of each EnformerAttention unit in the previous layer of EnformerAttention is taken as the input of each EnformerAttention unit in the next layer of EnformerAttention.

[0016] The EnformerAttention module outputs the learned features on the x-axis, y-axis, and z-axis.

[0017] Preferably, the coordinate-specific processing module includes three parallel processing units, and the learned features on the x-axis, y-axis, and z-axis are taken as the inputs of the three processing units one by one, respectively, and each processing unit includes a fully connected layer connected in sequence with a Flatten layer. Finally, the coordinate-specific processing module outputs the position coordinates of the target receiver on the x-axis, y-axis, and z-axis, respectively, and the position coordinates of the target receiver.

[0018] Preferably, the mobile molecular communication system in the second blood vessel environment used in the training of the receiver positioning model includes a moving transmitter, K'-1 fixed-position reference receivers, and a target receiver with multiple positions, wherein the target receiver at each single position and all reference receivers form a training sample, and each training sample corresponds to an observation period T';

[0019] For each training sample, the number of information molecules received by the target receiver and all reference receivers in each second time step Δt' in the training sample is calculated, J' = T' / Δt', J' is the number of second time steps Δt' in the observation period T' of the training stage, and the time series of the calculated number of information molecules received by the target receiver and all reference receivers in each second time step Δt' in the training sample is used to train the receiver positioning model until the loss function converges, and the loss function L MSE The calculation formula of the loss function L is as follows:

[0020]

[0021] wherein U is the batch size, u is the u-th sample in the batch U, is the true coordinate of the target receiver in the u-th training sample, is the predicted coordinate of the target receiver in the u-th training sample.

[0022] Preferably, calculating the number of information molecules received by the target receiver and all reference receivers in each training sample within each second time step Δt′ includes:

[0023] For each training sample, the displacement of each information molecule within each second time step is Δr. m (ΔX m ΔY m ΔZ m ), ΔX m ΔY m and ΔZ m Let x, y, and z represent the displacements of the information molecules along the x, y, and z axes, respectively, and satisfy the following conditions:

[0024] ΔX m =ΔX diffusion +ΔX Poseuille ;

[0025] ΔY m =ΔY diffusion ;

[0026] ΔZ m =ΔZ diffusion ;

[0027] in,

[0028] ΔX Poiseuille =2v(r) rad )Δt′;

[0029]

[0030] Where, ΔX diffusion ΔY diffusion and ΔZ diffusion It exhibits a Gaussian distribution characteristic, as shown below:

[0031] ΔX diffusion ~N(0,2αD m T′ α-1 Δt′);

[0032] ΔY diffusion ~N(0,2αD m T′ α-1 Δt′);

[0033] ΔZ diffusion ~N(0,2αD m T′α -1 Δt′);

[0034] Where, ΔX poiseuilleThe displacement of information molecules caused by Poiseuille flow is used in the second vascular environment for the receiver localization model training. The mobile molecular communication system is a vascular-like cylindrical structure. v v is the radius of the mobile molecular communication system in the second vascular environment. max v(r) represents the maximum velocity of the center of the mobile molecular communication system in the second vascular environment. rad (r) represents the radial distance r from the central axis of the mobile molecular communication system in the second vascular environment. rad The velocity at that point, ΔX diffusion ΔY diffusion and ΔZ diffusion These represent the displacement components along the x, y, and z axes caused by the diffusion of information molecules within each second time step, respectively. α is the anomaly coefficient, characterizing the degree of deviation from classical diffusion. m Let N be the diffusion coefficient of the information molecule, and let N() represent the normal distribution.

[0035] At the end of each second time step, the position coordinates of the information molecule are p. m (X m (j′), Y m (j′), Z m (j′)), j′∈{1,2,...,J′}, and satisfy:

[0036] X m (j′)=X m (j′-1)+ΔX m ;

[0037] Y m (j′)=Y m (j'-1)+ΔY m ;

[0038] Z m (j′)=Z m (j′-1)+ΔZ m ;

[0039] Among them, X m (j′), Y m (j′) and Z m (j′) represent the position coordinates of the information molecule in the x-axis, y-axis, and z-axis directions at the end of the j′-th second time step Δt′, respectively. The initial position of each information molecule just released is the position of the transmitter when it releases the molecule, and the position of the transmitter at the end of each second time step is (X TN (j′), Y TN (j′), Z TN (j′)):

[0040] X TN(j′)=X TN (j′-1)+ΔX TN ;

[0041] Y TN (j′)=Y TN (j'-1)+ΔY TN ;

[0042] Z TN (j′)=Z TN (j′-1)+ΔZ TN ;

[0043] Among them, X TN (j′), Y TN (j′) and Z TN (j′) represents the position coordinates of the transmitter TN in the x-axis, y-axis, and z-axis directions at the end of the j′-th second time step Δt′. When j′=1, (X TN (j′-1), Y TN (j'-1), Z TN (j′-1)) is the initial position of the transmitter, ΔX TN ΔY TN and ΔZ TN These represent the displacement components of the transmitter in the x-axis, y-axis, and z-axis directions within each second time step, ΔX. TN ΔY TN and ΔZ TN It also exhibits a Gaussian distribution characteristic, as shown below:

[0044] ΔX TN ~N(0,2αD TN T′ α-1 Δt′);

[0045] ΔY TN ~N(0,2αD TN T′ α-1 Δt′);

[0046] ΔZ TN ~N(0,2αD TN T′ α-1 Δt′);

[0047] Among them, D TN For the diffusion system of the transmitter;

[0048] The target receiver and all reference receivers are spherical absorbers with a radius of r. RNFor each training sample, during the observation period T′ of the training phase, the position coordinates of each information molecule at the end of each second time step are subtracted from the position coordinates of all receivers, and the norm is taken. It is then determined whether the norm is less than or equal to the radius of the corresponding receiver. If it is less than or equal to the radius, the corresponding information molecule is absorbed by the corresponding receiver and removed from the mobile molecular communication system in the second vascular environment. Thus, the number of information molecules received by each receiver in each second time step Δt′ is obtained.

[0049] A device for locating abnormally diffused molecular communication receivers in a vascular environment includes a processor and a memory storing a plurality of computer instructions, characterized in that the computer instructions, when executed by the processor, implement the steps of the method for locating abnormally diffused molecular communication receivers in a vascular environment.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] This invention relates to a method and apparatus for locating abnormal diffusion molecular communication receivers in a vascular environment. It employs an Enformer-based receiver localization model to locate the target receiver, achieving robust and accurate receiver localization in complex 3D scenes. In terms of performance, it significantly surpasses traditional Transformer and DNN models, achieving faster convergence, higher accuracy, and stronger robustness. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the structure of the mobile molecular communication system in the vascular environment in the abnormal diffusion molecular communication receiver positioning method and device of the present invention.

[0053] Figure 2 This is a schematic diagram of the receiver positioning model of the present invention;

[0054] Figure 3 This is a schematic diagram of the experimental results of the positioning error distribution of the present invention;

[0055] Figure 4 This is a schematic diagram showing the experimental results comparing the training loss of the Enformer-based model of this invention with existing Transformer and DNN-based models.

[0056] Figure 5 This is a schematic diagram illustrating the influence of the abnormal diffusion coefficient on the experimental results of this invention;

[0057] Figure 6 This diagram illustrates the experimental results comparing the performance of the Enformer-based model of this invention with existing models based on Transformer and DNN. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0060] like Figures 1-6 As shown, a method for locating abnormally diffused molecular communication receivers in a vascular environment is provided. This method is applied to a mobile molecular communication system in a first vascular environment. The mobile molecular communication system in the first vascular environment includes a mobile transmitter, K-1 reference receivers at fixed positions, and a target receiver at an unknown position. The transmitter releases a preset number of information molecules into the vascular environment at each position (these molecules then move in the blood vessels), and each receiver receives the information molecules.

[0061] It should be noted that, as Figure 1 As shown, the first vascular environment mobile molecular communication system (MMC) is a cylindrical structure resembling a blood vessel. The MMC consists of a mobile transmitter TN and K receivers (of the K receivers, K-1 reference receivers are located at known different positions, and the remaining receiver is the target receiver, whose location is unknown). For tractability calculation, the observation period T is uniformly divided into J discrete time slots, with each slot lasting Δt. The transmitter and all receivers are assumed to be perfectly clock-synchronized, with transmission precision starting from 0. The transmitter's position moves over time, with its initial coordinates set to (0,0,0). Each time the transmitter reaches a new position, it releases new information molecules. The transmitter releases a fixed number λ of information molecules into the vascular environment, which then propagate and move within the blood vessel. The receivers receive the information molecules released by the transmitter.

[0062] Within the observation period T, the number of information molecules received by the target receiver and all reference receivers within each first time step Δt is used to construct a time series (where the number of information molecules received by each receiver in the inference phase time series is set manually, and is calculated in the training phase). This time series is then used as the input to the trained receiver localization model to obtain the position coordinates of the target receiver. Here, Δt is the size of each time period after the observation period T is uniformly divided, and Δt includes the start time point and the end time point (since the time step belongs to a time period).

[0063] The receiver localization model includes an input signal processing module, an EnformerAttention module, and a coordinate-specific processing module connected in sequence (from data input to output direction).

[0064] Time series is represented as X counts ∈R K×J K is the total number of receivers, J is the length of the time series, and Δt = T / J, where the time series is:

[0065]

[0066] in, For the (K-1)th reference receiver RN K-1 The number of information molecules received within the j-th first time step. For the target receiver RN K The number of information molecules received within the j-th first time step, j∈{1,2,...,J}.

[0067] In the input signal processing module, the time series X is... counts Projecting onto the embedding space yields feature X. emb ∈R K ×J×d d is the embedding dimension (set to 200 in this implementation);

[0068] The features projected into the embedding space are then encoded at learnable locations, as shown in the following formula:

[0069] E=X emb +P pos ;

[0070] Where E is the result of learningable positional encoding of the features projected into the embedding space, and E∈R K×J×d P pos It is a learnable positional encoding, and P pos ∈R J×d P pos,j=Embedding(j), j∈{1,2,...,J}, where j is the position index, and Embedding() represents the extraction operation.

[0071] The EnformerAttention module includes multiple EnformerAttention layers (eight layers in this embodiment) connected sequentially from the data input to the output direction. Each EnformerAttention layer includes three parallel EnformerAttention units. Each EnformerAttention unit includes a multi-head self-attention layer, a first normalization layer, a convolutional layer (a one-dimensional convolution with a kernel size of 3 and padding of 1) connected sequentially, and a second normalization layer. The output of the multi-head self-attention layer is added to the input of the multi-head self-attention layer and used as the input of the first normalization layer. The output of the convolutional layer is added to the input of the convolutional layer and used as the input of the second normalization layer.

[0072] For the first EnformerAttention layer, the result E after learning the position encoding is used as the input of each EnformerAttention unit in the first EnformerAttention layer. For the remaining EnformerAttention layers, the output of each EnformerAttention unit in the previous EnformerAttention layer is used as the input of each EnformerAttention unit in the next EnformerAttention layer.

[0073] It should be noted that the multi-head self-attention layer operation is defined as follows:

[0074] MultiHead(ξ,ψ,ω)=Concat(head1,...,head l ,...,head h )θ output ;

[0075] Each attention head is calculated as follows:

[0076]

[0077] Where ξ, ψ, and ω represent the query, key, and value matrices, respectively; Concat represents the join operation; and h represents the number of attention heads, which is set to 10 in the coordinate-specific module. It is the output projection matrix, which transforms the connected attention heads to the output dimension. Here, d val The product h·d represents the dimension of the value vector in each attention head. valThis represents the total dimension after connecting the outputs from all h attention heads. and It is the weight matrix of the l-th attention head, where l∈{1,2,...,h}.

[0078] The attention function performs scaled dot product attention:

[0079]

[0080] Where d k It is the dimension of the key vector. This structure allows the model to focus on relevant temporal patterns throughout the sequence, while the convolutional components help preserve local contextual information.

[0081] The EnformerAttention module outputs learned features along the x, y, and z axes (each learned feature ∈ R). K×J×d ).

[0082] The coordinate-specific processing module includes three parallel processing units. The learned features on the x-axis, y-axis, and z-axis are respectively used as the inputs of the three processing units. Each processing unit includes a fully connected Flatten layer connected in sequence (from the data input to the output direction). Finally, the coordinate-specific processing module outputs the position coordinates of the target receiver on the x-axis, y-axis, and z-axis, respectively, and thus the position coordinates of the target receiver.

[0083] In another embodiment, the receiver localization model is trained:

[0084] The mobile molecular communication system used in the second vascular environment for receiver localization model training includes a mobile transmitter, K′-1 reference receivers at fixed locations, and a target receiver with multiple locations (e.g., Q locations). Each target receiver at a single location, together with all reference receivers, constitutes a training sample (a training sample contains one target receiver at a defined location and K′-1 reference receivers at fixed locations, for a total of Q training samples), and the observation period for each training sample is T′. The possible locations of the target receivers are systematically sampled using a grid-based method. To balance computational efficiency and spatial resolution, the location of the target receivers is constrained to a predefined volume.

[0085]

[0086] and These represent the lower and upper bounds of the x-coordinate of the target receiver's sampling volume, respectively. Similarly, and Define the range of the y-coordinate, and and The z-coordinate is specified as a boundary. These boundary parameters collectively define a three-dimensional prism volume within which target receiver locations are systematically sampled. By constraining the sampling space, a practical search domain is established while maintaining computational efficiency. This volume is discretized with a spatial resolution of Δr = 0.07 μm to create a mesh of potential target receiver locations. This approach ensures thorough coverage of the blood vessel cross-section while maintaining computational feasibility.

[0087] For each training sample, calculate the number of information molecules received by the target receiver and all reference receivers within each second time step Δt′, where J′=T' / Δt′, and J′ is the number of information molecules received within the second time step Δt′ during the observation period T′ in the training phase. Use the time series of the calculated number of information molecules received by the target receiver and all reference receivers within each second time step Δt′ to train the receiver localization model until the loss function converges and L... MSE The calculation formula is as follows (using mean square error):

[0088]

[0089] Where U is the batch size, and u is the u-th sample in batch U. Let U represent the true coordinates of the target receiver in the u-th training sample. Let be the predicted coordinates of the target receiver in the u-th training sample.

[0090] Specifically, the calculation of the number of information molecules received by the target receiver and all reference receivers in each training sample within each second time step Δt′ includes:

[0091] For each training sample, the displacement of each information molecule within each second time step is Δr. m (ΔX m ΔY m ΔZ m ), ΔX m ΔY m and ΔZ m Let x, y, and z represent the displacements of the information molecules along the x, y, and z axes, respectively, and satisfy the following conditions:

[0092] ΔX m =ΔX diffusion +ΔX Poiseuille ;

[0093] ΔY m =ΔY diffusion ;

[0094] ΔZ m =ΔZ diffusion ;

[0095] in,

[0096] ΔX Poiseuille =2v(r) rad )Δt′;

[0097]

[0098] Where, ΔX diffusion ΔY diffusion and ΔZ diffusion It exhibits a Gaussian distribution characteristic, as shown below:

[0099] ΔX diffusion ~N(0,2αD m T′ α-1 Δt′);

[0100] ΔY diffusion ~N(0,2αD m T′ α-1 Δt′);

[0101] ΔZ diffusion ~N(0,2αD m T′ α-1 Δt′);

[0102] Where, ΔX Poiseuille The displacement of information molecules caused by Poiseuille flow is used in the second vascular environment for the receiver localization model training. The mobile molecular communication system is a vascular-like cylindrical structure. v v is the radius of the mobile molecular communication system in the second vascular environment. max v(r) represents the maximum velocity of the center of the mobile molecular communication system in the second vascular environment. rad (r) represents the radial distance r from the central axis of the mobile molecular communication system in the second vascular environment. rad The velocity at that point, ΔX diffusion ΔY diffusion and ΔZ diffusion These represent the displacement components along the x, y, and z axes caused by the diffusion of information molecules, respectively. α is the anomaly coefficient, characterizing the degree of deviation from classical diffusion. m Let N be the diffusion coefficient of the information molecule, and let N() represent the normal distribution.

[0103] At the end of each second time step, the position coordinates of the information molecule are p. m (X m (j′), Y m (j′), Z m (j′)), j′∈{1,2,...,J′}, and satisfy:

[0104] Xm (j′)=X m (j′-1)+ΔX m ;

[0105] Y m (j′)=Y m (j′-1)+ΔY m ;

[0106] Z m (j′)=Z m (j'-1)+ΔZ m ;

[0107] Among them, X m (j′), Y m (j′) and Z m (h′) represent the position coordinates of the information molecule in the x-axis, y-axis, and z-axis directions at the end of the j′-th second time step Δt′, respectively. The initial position of each information molecule is the position where the transmitter releases the molecule, and the position of the transmitter in each second time step is (X... TN (j′), Y TN (j′), Z TN (j′)):

[0108] X TN (j′)=X TN (j′-1)+ΔX TN ;

[0109] Y TN (j′)=Y TN (j′-1)+ΔY TN ;

[0110] Z TN (j′)=Z TN (j′-1)+ΔZ TN ;

[0111] Among them, X TN (j′), Y TN (j′) and Z TN (j′) represents the position coordinates of the transmitter TN in the x-axis, y-axis, and z-axis directions at the end of the j′-th second time step Δt′. When j′=1, (X TN (j′-1), Y TN (j'-1), Z TN (j′-1)) is the initial position of the transmitter (e.g., (0,0,0)), ΔX TN ΔY TN and ΔZ TNThese represent the displacement components of the transmitter in the x-axis, y-axis, and z-axis directions within each second time step, ΔX. TN ΔY TN and ΔZ TN It also exhibits a Gaussian distribution characteristic, as shown below:

[0112] ΔX TN ~N(0,2αD TN T′ α-1 Δt′);

[0113] ΔY TN ~N(0,2αD TN T′ α-1 Δt′);

[0114] ΔZ TN ~N(0,2αD TN T′ α-1 Δt′);

[0115] Among them, D TN For the diffusion system of the transmitter;

[0116] The target receiver and all reference receivers are spherical absorbers with a radius of r. RN For each training sample, within the observation period T' of the training phase, the position coordinates of each information molecule at the end of each second time step are subtracted from the position coordinates of all receivers, and the norm is taken (i.e., at the end of each second time step, the position coordinates of each information molecule in the mobile molecular communication system under the second vascular environment are subtracted from the position coordinates of all receivers (including the target receiver and all reference receivers), and the norm of the subtraction result is taken). It is then determined whether the norm is less than or equal to the radius of the corresponding receiver (which receiver the information molecule is subtracted from, i.e., the radius of which receiver it is compared with). If it is less than or equal to, the corresponding information molecule is absorbed by the corresponding receiver and removed from the mobile molecular communication system under the second vascular environment, thereby obtaining the number of information molecules received by each receiver within each second time step Δt'.

[0117] During training, several advanced optimization strategies were implemented to improve model convergence and generalization. The AdamW optimizer, an extension of the Adam algorithm, was employed, combined with decoupled weight decay regularization. This choice helps mitigate overfitting while maintaining the adaptive learning rate advantage crucial for training deep architectures. The initial learning rate was set to 1×10⁻⁶. -4The learning rate is dynamically adjusted using the ReduceLROnPlateau scheduler. If the validation loss stagnates for 5 consecutive training epochs, the learning rate is reduced by 0.5x. This allows the optimizer to escape local minima and achieve finer convergence. Layer normalization is applied throughout the network, especially after attention and convolution operations. Training is stabilized by normalizing activations within each layer, thus accelerating convergence. Unlike batch normalization, layer normalization is independent of batch size, making it more suitable for sequence processing tasks. Residual connections are incorporated at multiple stages, particularly within the EnformerAttention module. These skip connections facilitate gradient flow in deep networks, addressing the vanishing gradient problem and enabling the model to learn hierarchical representations of molecular absorption patterns. A dropout rate of 0.1 is used in the attention mechanism and 0.2 in coordinate-specific layers. This random deactivation of neurons prevents co-adaptation of feature detectors and enhances robustness. Furthermore, the weight decay parameter of the Ada mW optimizer is set to 1×10⁻⁶. -5 It provides L2 regularization, which adds the squared magnitude of the parameters to the loss function, penalizing large weights and preventing overfitting, further constraining the model parameters.

[0118] In this embodiment, 80% of the samples are used for model training, and 20% of the samples are used for model testing. The learning rate is set to 1×10⁻⁶. -4 Training is performed for a maximum of 200 epochs, with early stopping based on validation performance. Model checkpoints are saved periodically, retaining states with the lowest validation loss. A 5-fold cross-validation strategy is employed: the dataset is divided into five partitions, and the model is trained and evaluated five times, with each iteration using a different partition for validation.

[0119] During model training, mean squared error was chosen as the loss function, with a batch size of 32.

[0120] During the test, the accuracy of the prediction result is evaluated based on the distance (i.e., positioning error) between the predicted coordinates of the target receiver and the corresponding actual coordinates (i.e., true coordinates). The accuracy of the prediction result is evaluated by comparing the distance with a threshold η. If the distance is less than or equal to the threshold η, the prediction is considered accurate. The ratio of the number of accurate predictions to the number of tests is taken as the accuracy rate of the test.

[0121] Training neural network models is a relatively mature technology in this field, and will not be elaborated on here.

[0122] To verify the effectiveness of the abnormal diffusion molecular communication receiver localization method in this vascular environment, a simulation experiment was conducted:

[0123] A molecular absorption time series corresponding to the target receiver in the molecular communication system requiring localization is constructed, and a trained receiver localization model is used to predict the location of the target receiver. Based on the number of information molecules received by the target and reference receivers under the combined effects of anomalous diffusion and Poiseuille flow velocity, the molecular absorption time series corresponding to the target receiver in the molecular communication system requiring localization can be constructed. This time series is input into the trained receiver localization model to obtain the predicted location of the target receiver. To analyze the location prediction performance of the anomalous diffusion molecular communication receiver localization method in this vascular environment, experiments demonstrate the prediction accuracy of the Enformer model, and the Enformer model is compared with models based on Transformer and DNN on the same test dataset. All parameters used in the experiments are shown in Table 1.

[0124] Table 1 Simulation parameter configuration

[0125] Parameter Value Vessel radius R v ]]> 5.0 μm Receiver radius r RN ]]> 2.0 μm / s Molecular diffusion coefficient D m ]] 70 μm 2 / s]] Transmitter diffusion coefficient D TN ]]> 0.1 μm 2 / s]] Anomalous diffusion coefficient α 0.5-1.5 Maximum flow rate v max ]] 400 μm / s Number of released information molecules λ 10000 Number of receivers K 4 Time step Δt 0.001s Total observation time T 0.2s

[0126] To present the statistical results of the proposed model, Table 2 provides a comparative analysis of several statistical metrics between the model on which our method is based (represented by the Enformer model) and existing models based on Transformer and DNN. The mean localization error (localization error being the distance between predicted coordinates and corresponding actual coordinates) is the average localization error of all samples in the test dataset during testing. The results in Table 2 quantitatively highlight and compare the performance characteristics of these three models. Based on the observations in Table 2, it is clear that, at the same threshold η, the accuracy achieved by the Enformer model on which our method is based is significantly higher than that of the Transformer and DNN-based models.

[0127] Table 2 Performance Comparison of Different Models

[0128]

[0129] To comprehensively evaluate the prediction accuracy of this method based on the Enformer model, Figure 3 The predicted positioning error distribution is shown in the figure. In N test In the test samples, the number of errors within the error range is N. r The ratio is determined by N. r / N test The histogram of estimated errors is predominantly on the left, indicating that most errors are concentrated in a small range, with approximately 85% of all prediction errors below 0.30 μm. However, some values ​​on the right side of the histogram deviate significantly from the left. These values ​​can be considered outliers, but they represent only a small fraction of the overall data, with the frequency of values ​​exceeding 0.80 μm approaching zero.

[0130] Figure 4 The loss trajectories of our proposed Enformer-based model and existing Transformer and DNN-based models during training are shown. Notably, our Enformer-based model exhibits faster and more stable convergence compared to the Transformer and DNN-based model. Our Enformer-based model demonstrates convergence around the 50th training epoch. However, the Transformer and DNN-based model requires over 100 training epochs to converge. This indicates that our Enformer-based model is more efficient at extracting features from molecular absorption time series and learning the mapping of location information.

[0131] Figure 5 The predictive performance accuracy of our proposed method based on the Enformer model was evaluated under different anomalous diffusion coefficients α. When α values ​​were 0.50 and 0.75, the accuracy exceeded 99.8%. As α increased to 1.00, the accuracy slightly decreased to 95.14%. A more significant decrease occurred for the hyperdiffusion case, with accuracy dropping to 80.71% at α = 1.25 and 58.85% at α = 1.50. This highlights that while our proposed method based on the Enformer model is robust, increased hyperdiffusion adversely affects performance.

[0132] Figure 6 Showing different maximum flow rates v max Impact on positioning accuracy. Accuracy remains high and relatively stable at low to medium flow rates. max The accuracy is 97.45% at 100 μm / s, v max It reaches a peak value of 98.17% at 200 μm / s. When v max When further increased to 300 μm / s, the accuracy slightly decreased to 97.40%. At the highest test speed v... max A more significant decrease in accuracy was observed at 400 μm / s. This indicates that while the model is robust under a range of flow conditions, very high flow rates may pose a greater challenge to maintaining peak localization performance.

[0133] The hybrid attention-convolution mechanism in this application's model can effectively learn the complex relationships between molecular absorption sequences, thereby improving the accuracy of receiver location prediction. Experimental results show that, compared with Transformer- and DNN-based models, the Enformer-based model exhibits significantly higher prediction accuracy, faster convergence speed, and stronger robustness. Furthermore, even with changes in environmental factors, such as strong anomalous diffusion, high flow rates, limited observation time, and transmitter mobility, the Enformer-based model maintains excellent performance.

[0134] In one embodiment, based on a method for locating abnormally diffused molecular communication receivers in a vascular environment, the present invention also provides a device for locating abnormally diffused molecular communication receivers in a vascular environment, including a processor and a memory storing a plurality of computer instructions, characterized in that the computer instructions, when executed by the processor, implement the steps of the method for locating abnormally diffused molecular communication receivers in a vascular environment.

[0135] For specific limitations on the automatic response device that integrates dual word segmentation and iterative feedback, please refer to the limitations on the automatic response method that integrates dual word segmentation and iterative feedback mentioned above, which will not be repeated here.

[0136] The memory and processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory stores a computer program that can run on the processor, and the processor implements the method of the present invention by running the computer program stored in the memory.

[0137] The memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory stores the program, and the processor executes the program upon receiving an execution instruction.

[0138] The processor may be an integrated circuit chip with data processing capabilities. The aforementioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0139] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0140] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for locating abnormally diffused molecular communication receivers in a vascular environment, characterized in that: The abnormal diffusion molecular communication receiver localization method in a vascular environment is applied to a first mobile molecular communication system in a vascular environment, which includes a mobile transmitter. The transmitter has a reference receiver at a fixed location and a target receiver at an unknown location. The transmitter releases information molecules into the vascular environment at each location, and each receiver receives the information molecules. During the observation period Within, each first time step The number of information molecules received by the target receiver and all reference receivers constitutes a time series. This time series is used as input to a trained receiver localization model to obtain the position coordinates of the target receiver. Observation period The size of each time period after uniform division, and It includes the start time point and the end time point.

2. The method for locating abnormally diffused molecular communication receivers in a vascular environment as described in claim 1, characterized in that: The receiver localization model includes an input signal processing module, an Enformer Attention module, and a coordinate-specific processing module connected in sequence.

3. The method for locating abnormally diffused molecular communication receivers in a vascular environment as described in claim 2, characterized in that: The time series is represented as , For the total number of receivers, is the length of the time series, where ; In the input signal processing module, the time series... Projecting onto the embedding space yields the features. , For the embedded dimension; The features projected into the embedding space are then encoded at learnable locations, as shown in the following formula: ; in, The result of learningable positional encoding of features projected into the embedding space, and , For learnable positional encoding, and , , For location index.

4. The method for locating abnormally diffused molecular communication receivers in a vascular environment as described in claim 3, characterized in that: The Enformer Attention module includes multiple Enformer Attention layers connected in sequence, and each Enformer Attention layer includes three parallel Enformer Attention units. Each Enformer Attention unit includes a multi-head self-attention layer, a first normalization layer, a convolutional layer, and a second normalization layer connected in sequence. The output of the multi-head self-attention layer is added to the input of the multi-head self-attention layer and used as the input of the first normalization layer. The output of the convolutional layer is added to the input of the convolutional layer and used as the input of the second normalization layer. For the first Enformer Attention layer, the result after position encoding can be learned. As the input to each Enformer Attention unit in the first EnformerAttention layer, for the remaining Enformer Attention layers, the output of each Enformer Attention unit in the previous Enformer Attention layer is used as the input to each Enformer Attention unit in the next Enformer Attention layer. The Enformer Attention module outputs learned features on the x-axis, y-axis, and z-axis.

5. The method for locating abnormally diffused molecular communication receivers in a vascular environment as described in claim 4, characterized in that: The coordinate-specific processing module includes three parallel processing units. The learned features on the x-axis, y-axis, and z-axis correspond one-to-one as the inputs of the three processing units. Each processing unit includes a fully connected layer of Flatten layers connected in sequence. Finally, the coordinate-specific processing module outputs the position coordinates of the target receiver on the x-axis, y-axis, and z-axis, respectively, and thus the position coordinates of the target receiver.

6. The method for locating abnormally diffused molecular communication receivers in a vascular environment as described in claim 1, characterized in that: The mobile molecular communication system used in the second vascular environment during receiver localization model training includes a mobile transmitter. A reference receiver at a fixed location and a target receiver at multiple locations, wherein the target receiver at each individual location, together with all the reference receivers, constitutes a training sample, and the observation period corresponding to each training sample is 1. ; For each training sample, calculate each second time step. The number of information molecules received by the target receiver and all reference receivers in the training samples, and , The observation period for the training phase Second time step The number, and the calculated second time step The receiver localization model is trained using time series data on the number of information molecules received by the target receiver and all reference receivers in the training samples, until the loss function converges. The calculation formula is as follows: ; in, For batch size, For batch The Middle One sample, For the first The true coordinates of the target receiver in each training sample. For the first The predicted coordinates of the target receiver in each training sample.

7. The method for locating abnormally diffused molecular communication receivers in a vascular environment as described in claim 6, characterized in that: The calculation of each second time step The number of information molecules received by the target receiver and all reference receivers in each training sample, including: For each training sample, the displacement of each information molecule within each second time step is: , , and Let x, y, and z represent the displacements of the information molecules along the x, y, and z axes, respectively, and satisfy the following conditions: ; ; ; in, ; ; in, , and It exhibits a Gaussian distribution characteristic, as shown below: ; ; ; in, The displacement of information molecules caused by Poiseuille flow is used in the receiver localization model training. The mobile molecular communication system in the second vascular environment is a vascular-like cylindrical structure. The radius of the mobile molecular communication system in the second vascular environment. This represents the maximum speed of the center of the mobile molecular communication system in the second vascular environment. Radial distance from the central axis of the mobile molecular communication system in the second vascular environment The speed at that point, , and These represent the displacement components along the x, y, and z axes caused by the diffusion of information molecules within each second time step. The anomaly coefficient characterizes the degree of deviation from classical diffusion. Let be the diffusion coefficient of the information molecule. Represents a normal distribution; At the end of each second time step, the position coordinates of the information molecule are: , And satisfy: ; ; ; in, , and They represent the first time. The second time step The position coordinates of the information molecule in the x, y, and z axes at the end of each time step are given. The initial position of each information molecule just after release is the position of the transmitter when it releases that molecule. The position of the transmitter at the end of each second time step is given. : ; ; ; in, , and Indicates the first The second time step The transmitter at the end of the time point The position coordinates in the x, y, and z axes, when hour, This is the initial position of the transmitter. , and These represent the displacement components of the transmitter in the x-axis, y-axis, and z-axis directions within each second time step. , and It also exhibits a Gaussian distribution characteristic, as shown below: ; ; ; in, For the diffusion system of the transmitter; The target receiver and all reference receivers are spherical absorbers, and the radius of each receiver is... For each training sample, the observation period during the training phase Within each second time step, the position coordinates of each information molecule at the end of the time point are subtracted from the position coordinates of all receivers, and the norm is calculated. It is then determined whether this norm is less than or equal to the radius of the corresponding receiver. If it is less than or equal to the radius, the corresponding information molecule is absorbed by the corresponding receiver and removed from the mobile molecule communication system in the second vascular environment. This process yields the results for each receiver at each second time step. The number of information molecules received internally.

8. A device for locating abnormally diffused molecular communication receivers in a vascular environment, comprising a processor and a memory storing a plurality of computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.