A Medical Data Caching Method Based on Named Data Networks

Abstract

This invention belongs to the field of communication technology, specifically relating to a medical data caching method based on named data networks. The method includes: a consumer sending an interest packet and its corresponding freshness requirement to a producer via path L; the producer and all forwarders on path L caching the freshness requirement corresponding to the interest packet; the producer generating a data packet corresponding to the interest packet and determining whether to cache it using an LSTM-based freshness caching algorithm; the data packet being returned to the consumer via path L, and each forwarder on path L, upon receiving the data packet, determining whether to cache it using the LSTM-based freshness caching algorithm. This invention can effectively reduce cache redundancy and avoid caching worthless data.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and specifically relates to a medical data caching method based on named data networks. Background Technology

[0002] The Internet of Things (IoT) is a complex technology with global economic and social impacts, and the Internet of Things in healthcare is a significant application of IoT in the medical and health field. Currently, telemedicine systems using IoT primarily utilize sensors or other sensing devices to collect patients' physiological information and transmit this data to a processor via wireless or wired communication. The processor then processes the data and transmits it to doctors, who can use the information to provide timely assistance to patients, improving hospital efficiency. The emergence and application of IoT in healthcare has effectively changed traditional medical service models, ensuring that patients occupy a crucial position in the entire healthcare service field. However, IoT data is inherently content-centric, and the application of 5G technology in the development and implementation of IoT presents further challenges to the traditional TCP / IP architecture.

[0003] Unlike IP-based networks, Named Data Networks (NDN) emphasize the unified identification and naming of content, separating network resources from application locations and better supporting mobility. NDN features intra-network caching, temporarily caching requested data in routers to satisfy subsequent user requests. Through intra-network caching, subsequent user requests do not need to retrieve data from the data source; they can be obtained directly from the cached router, significantly reducing latency and network traffic load. Despite these advantages of applying NDN to IoT, the unique characteristics of medical information mean that existing NDN caching mechanisms are not well-suited to the data characteristics of medical IoT. Therefore, modifications to the current NDN network caching mechanism are needed to better serve remote monitoring systems.

[0004] Unlike traditional IoT data, which is characterized by large fluctuations, rapid obsolescence, and fast update rates, medical data obtained from regularly monitored patients exhibits less fluctuation. Furthermore, historical data such as patient medical records is highly valuable to doctors, aiding in decision-making and enabling effective prediction and prevention of certain diseases. After obtaining historical patient data, doctors assess the patient's condition based on their experience; therefore, the traditional method of producers setting data validity periods is no longer suitable for remote monitoring systems in the medical field. Consequently, caching schemes in remote health monitoring systems need modification. Since remote health monitoring systems are highly doctor-driven, they are represented as consumer-driven systems in NDN. Doctors request historical patient data based on personal experience, preferences, and judgments. Therefore, three points can be derived from remote medical health monitoring systems:

[0005] 1) The doctor specifies the lifespan (freshness of content) of the data packets (data collected by sensors);

[0006] 2) The characteristics of NDN networks provide doctors with opportunities to improve the quality of patient services;

[0007] 3) The distribution of interest packets (doctor requests) in remote health monitoring systems does not conform to a Zipf-like distribution, indicating that doctors are more inclined to request historical data related to the patient's condition.

[0008] Furthermore, since content freshness is an indispensable feature of NDN networks, and consumer-driven freshness requirements are superior to traditional producer-driven freshness mechanisms, NDN network nodes do not need to cache data below the required freshness level, allowing them to cache more valuable data within limited cache capacity. Therefore, given the unique characteristics of medical data, existing caching methods are not well-suited to meet the needs of remote medical health monitoring systems. Summary of the Invention

[0009] To address the aforementioned issues, this invention provides a medical data caching method based on named data networks, comprising M routing nodes, which are divided into consumers, forwarders, and producers.

[0010] The medical data caching method based on named data networks includes the following steps:

[0011] S1. The consumer sends an interest packet and the corresponding freshness requirement to the producer through path L; the producer and all forwarders on path L cache the freshness requirement corresponding to the interest packet.

[0012] S2. The producer generates the data packet corresponding to the interest packet, and determines whether to cache the data packet using an LSTM-based freshness caching algorithm, and then executes step S3;

[0013] S3. The data packet is returned to the consumer via path L. Each forwarder on path L receives the data packet and determines whether to cache it using an LSTM-based freshness caching algorithm.

[0014] Furthermore, the freshness of data changes over time, and is defined as the difference between the time the data was generated and the current time; the freshness requirement is a threshold value for the freshness of data specified by the consumer for the requested data.

[0015] Furthermore, the routing nodes j = 1, 2, ..., M maintain a list of sensors. And a list of freshness, This represents the timestamp of the q-th request from routing node i = 1, 2, ..., M for the n = 1, 2, ..., N sensor data; when routing node i sends an interest packet as a consumer, the interest packet carries a freshness requirement P. i,n P i,n This represents the freshness requirement of routing node i for the data from the nth sensor; N is the number of sensors.

[0016] Furthermore, the process of determining whether to cache the current data packet using the LSTM-based freshness caching algorithm includes:

[0017] S11. After receiving the current data packet requested by route node i, route node j retrieves it from sensor list A. j In the process of retrieving the latest timestamp of the current data packet from route node i,... And the timestamp of the previous data packet requested by route node i.

[0018] S12. Calculate the latest timestamp Compared to the previous timestamp The difference between them yields the historical request time interval.

[0019] S13. Historical request time interval θ i,n Perform LSTM processing to obtain the prediction time interval;

[0020] S14. Obtain the freshness requirement P of the current data packet. i,n If freshness requirement P i,n If the time interval is greater than the prediction interval, the current data packet is valid, and routing node j caches the current data packet; if freshness requirement P i,n If the time interval is not greater than the predicted time interval, the current data packet is invalid, and routing node j does not need to cache the data and discards the current data packet.

[0021] Furthermore, the LSTM processing described in step S13 includes:

[0022] f t =σ(W f ·[PI t-1 A t ′]+b f )

[0023] i t =σ(W i ·[PI t-1 A t ′]+b i )

[0024] c t ′=tanh(W c ·[PIt-1 A t ′]+b c )

[0025]

[0026] o t =σ(W o ·[PI t-1 A t ′]+b o )

[0027]

[0028] Among them, f t This represents the output of the forget gate at time t, i t c represents the output of the input gate at time t. t ' represents the current input cell state, c t O represents the cell state at time t. t W represents the output of the output gate at time t. f W represents the forget gate weight matrix. i W represents the input gate weight matrix. c W represents the unit state weight matrix. o b represents the output gate weight matrix. f Indicates the forget gate bias, b i Indicates the input gate bias, b c Indicates the unit state bias, b o σ represents the output gate bias, σ represents the sigmoid function, and PI represents the output gate bias. t-1 A represents the LSTM output at time t-1. t ′={θ i,n} represents the historical request time interval sequence of data generated by the routing node i for the sensor n.

[0029] Furthermore, if freshness requirement P i,n If the time interval is greater than the prediction time interval, the current data packet is valid. The buffering probability of the current data packet is calculated and expressed as:

[0030]

[0031] This represents the cached result of the q-th request for the current data packet by the routing node j, where Q represents the total number of requests; This indicates the probability of caching the data packet generated by sensor n during the Q-th request;

[0032]

[0033] Where B represents the freshness requirement P i,nThe time interval is greater than the prediction time interval, where N represents the number of sensors.

[0034] Furthermore, the information value of the current data packet is calculated based on the buffering probability of the current data packet, and is expressed as:

[0035]

[0036] in, This represents the probability of caching the data packet generated by sensor n during the Q-th request.

[0037] The beneficial effects of this invention are:

[0038] This solution primarily applies LSTM to the content storage repository of NDN router nodes. By calculating the historical data requested by doctors, it predicts new requests and determines whether data should be cached in the limited cache space based on the prediction results. This effectively reduces cache redundancy and avoids caching worthless data. When doctors initiate content requests, more valuable data can be obtained, providing better care for patients. Attached Figure Description

[0039] Figure 1 This invention relates to a remote medical scenario;

[0040] Figure 2 This is a flowchart of the method of the present invention;

[0041] Figure 3 This is a flowchart illustrating the algorithm execution of the present invention;

[0042] Figure 4 This is a schematic diagram of the LSTM processing of the present invention;

[0043] Figure 5 This is a schematic diagram illustrating the application of LSTM to the NDN node of this invention. Detailed Implementation

[0044] 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.

[0045] In traditional telemedicine systems, significant network latency occurs when consumers retrieve data remotely. Named Data Networks (DND) feature internal caching, allowing historical request data to be stored in intermediate routers, significantly reducing data retrieval latency and network traffic load. Therefore, this invention, combining the internal caching characteristics of NDN networks, proposes a medical data caching method based on Named Data Networks. Figure 1 The telemedicine scenario shown in the NDN network includes M routing nodes. Each routing node has a triple identity as a consumer, forwarder, and producer. That is, each routing node can send interest packets as a consumer, generate data packets as a producer, and transmit data for consumers and producers as a forwarder.

[0046] In the medical scenario of this invention, doctors, nurses, and other medical stakeholders act as consumers. Consumers create interest packages based on their own medical experience and preferences, and then send these interest packages through the network to obtain patients' health data. Patients act as producers, monitoring their own health data by wearing various instruments and sending data packets containing their health data that match the interest packages back to consumers through the network. At the same time, there may also be forwarders between consumers and producers. These forwarders can cache data packets and change their identities to producers or consumers in subsequent processes.

[0047] In DND networks, the freshness of data requested by consumers is a crucial factor, related to whether the forwarder caches the data. In the medical field, patient health data is constantly changing, making the freshness of data requested by doctors extremely important. Freshness is not a static value; it changes over time. This invention defines data freshness as the difference between the data's creation time and the current time. When a consumer sends an interest packet, they specify a freshness requirement for that packet; this requirement is a data freshness threshold specified by the consumer for the requested data.

[0048] This invention proposes a medical data caching method based on named data networks, such as... Figure 2 As shown, it includes the following steps:

[0049] S1. The consumer sends an interest packet and the corresponding freshness requirement to the producer through path L; the producer and all forwarders on path L cache the freshness requirement corresponding to the interest packet.

[0050] S2. The producer generates the data packet corresponding to the interest packet, and determines whether to cache the data packet using an LSTM-based freshness caching algorithm, and then executes step S3;

[0051] S3. The data packet is returned to the consumer via path L. Each forwarder on path L receives the data packet and determines whether to cache it using an LSTM-based freshness caching algorithm.

[0052] Specifically, the routing nodes j = 1, 2, ..., M maintain a list of sensors. A freshness list. The sensor list records the timestamps of interest packets sent by all routing nodes (excluding itself) when acting as consumers, arranged in chronological order; among them, This represents the timestamp of the q-th request from routing node i = 1, 2, ..., M for the n = 1, 2, ..., N sensor data; the freshness list is used to record the data freshness. When routing node i sends an interest packet as a consumer, the interest packet carries a freshness requirement P. i,n P i,n This represents the freshness requirement of routing node i for the data from the nth sensor; N is the number of sensors.

[0053] Preferably, the process of determining whether to cache the current data packet based on the LSTM-based freshness caching algorithm assumes that the current data packet contains data from sensor n requested by routing node i in the qth iteration, such as... Figure 3 As shown, it includes:

[0054] S11. After receiving the current data packet requested by route node i for the qth time, route node j retrieves data from its own sensor list A. j Get the timestamp of the current data packet requested by the i-th routing node. and the timestamp of the (q-1)th request of the current data packet by routing node i.

[0055] S12. Calculate the timestamp Compared to the previous timestamp The difference between them yields the historical request time interval.

[0056] S13. Historical request time interval θ i,n Perform LSTM processing to obtain the prediction time interval;

[0057] S14. Obtain the freshness requirement P of the current data packet. i,n If freshness requirement P i,n If the time interval is greater than the prediction interval, the current data packet is valid, and routing node j caches the current data packet; if freshness requirement P i,n If the time interval is not greater than the predicted time interval, the current data packet is invalid and should be discarded.

[0058] Specifically, the LSTM processing flow is as follows: Figure 4As shown, LSTM uses two gates to control the content of the unit state c. One is the forget gate, which determines the cell state c from the previous time step. t-1 How many cell states c are retained up to the current time step? t The other is the input gate, which determines the network's input A at the current moment. t How many cell states c are saved to the current time step? t LSTM uses output gates to control the current cell state c. t How much output is given to the current output value PI of the LSTM? t In the middle. Among them:

[0059] Forget Gate Formula:

[0060] f t =σ(W f ·[PI t-1 A t ′]+b f (1)

[0061] Among them, f t W represents the output of the forget gate at time t. f The weight matrix of the forget gate, b f PI represents the bias term of the forget gate. t-1 A represents the LSTM output at time t-1. t ′={θ i,n} represents the sequence of historical request time intervals; σ represents the sigmoid function, defined as:

[0062]

[0063] Input gate formula:

[0064] i t =σ(W i ·[PI t-1 A t ′]+b i (3)

[0065] Among them, i t W represents the output of the input gate at time t. i Let b represent the input gate weight matrix. i This indicates the input gate bias.

[0066] Based on the previous LSTM output PI t-1 The current LSTM input A′ is used to calculate the cell state c that describes the current input. t ′:

[0067] c t ′=tanh(W c ·[PI t-1 A t ′]+b c (4)

[0068] The output activation function tanh is defined as follows:

[0069]

[0070] Calculate the cell state c at the current moment. t It is determined by the cell state c from the previous time step. t-1 Multiply by the element-wise output f of the forget gate at the current time step. t Then use the current input cell state c t 'Element-wise multiplication of the input gate at time t' output i t Then, sum the two products to produce:

[0071]

[0072] in This indicates element-wise multiplication.

[0073] The above calculations establish the LSTM's current memory c. t ′ and long-term memory c t-1 Combined together, they form a new unit state c. t Because of the forget gate, information from a very long time ago can be preserved; because of the input gate, currently irrelevant content is prevented from entering memory. The output gate controls the influence of long-term memory on the current output.

[0074] o t =σ(W o ·[PI t-1 A t ′]+b o (7)

[0075] The final output, i.e., the prediction time interval, is determined by both the output gate and the cell state.

[0076]

[0077] Specifically, if freshness requirement P i,n If the time interval is greater than the prediction time interval, the current data packet is valid, and the buffering probability of the current data packet is calculated. Represented as:

[0078]

[0079]

[0080] in, This represents the cached result of the q-th request for the current data packet by the routing node j, where Q represents the total number of requests; This represents the caching probability of the current data packet, that is, the caching probability of the data packet generated by sensor n at the Qth request. Specifically, the current data packet refers to the data packet containing the requested sensor n data at the latest Qth request. At this time, the total number of times this data packet has been requested is increased by one, from Q-1 to Q, and the corresponding caching probability also changes accordingly.

[0081] In a medical setting, doctors, as consumers, are less concerned with the probability of caching and more focused on the size of the cached information. They only care about data with large amounts of information. This way, even if a node caches a lot of data, doctors can quickly remove useless data based on the amount of information when requesting data. This invention uses an information content formula to represent the amount of cached information.

[0082] Specifically, the information value of the current data packet is calculated based on the buffering probability of the current data packet, and is expressed as:

[0083]

[0084] Specifically, the process of this invention is as follows: Figure 5 As shown, when an NDN node receives an interest packet, it first checks its own storage container (CS). If the CS contains a data packet that satisfies the interest packet, it directly returns the data packet to the consumer. If the CS does not contain a data packet that satisfies the interest packet, it forwards the interest packet to the producer through the forwarding information table (FIB). The NDN node also has a pending information table (PIT) to record the forwarding interfaces for unsatisfied interest packets. These forwarding interfaces are used to forward data packets returned by other nodes.

[0085] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for caching medical data based on a named data network, characterized by, M routing nodes are included, and the M routing nodes are divided into consumers, forwarders and producers; The medical data caching method based on the named data network comprises the following steps: S1. The consumer sends an interest packet and a freshness requirement corresponding to the interest packet to the producer through a path L; the producer and all forwarders on the path L cache the freshness requirement corresponding to the interest packet; The size of the data freshness changes over time, and the size of the data freshness is defined as the difference between the generation time of the data and the current time; the freshness requirement is a data freshness threshold specified by the consumer for the requested data; S2. The producer generates a data packet corresponding to the interest packet, and judges whether to cache the data packet through an LSTM-based freshness caching algorithm, and then executes step S3; S3. The data packet returns to the consumer through the path L, and each forwarder on the path L receives the data packet and judges whether to cache the data packet through the LSTM-based freshness caching algorithm; The process of judging whether to cache the current data packet by the LSTM-based freshness caching algorithm comprises: S11. After receiving the current data packet, the routing node j obtains from the sensor list A j the latest timestamp requesting the current data packet and the previous timestamp requesting the current data packet ; S12. Calculate the difference between the latest timestamp and the previous one, resulting in the historical request time interval ;​​ S13. performing LSTM processing on the historical request time intervals to obtain a predicted time interval; S14. Obtain the freshness requirement P of the current data packet. i,n If freshness requirement P i,n If the time interval is greater than the prediction interval, the current data packet is valid, and routing node j caches the current data packet; if freshness requirement P i,n If the time interval is not greater than the predicted time interval, the current data packet is invalid and there is no need to cache it. If the freshness requirement P i,n If the prediction time interval is greater than the prediction time interval, the current data packet is valid, and the cache probability of the current data packet is calculated, and the calculation formula is: wherein, represents the cache result of the qth request of the current data packet by the routing node j, and Q represents the total number of requests; represents the cache probability of the data packet generated by the sensor n at the Qth request. where B represents the freshness requirement P i,n greater than the prediction time interval, N represents the number of sensors.

2. The method for medical data caching based on named data networking according to claim 1, characterized in that, A routing node j = 1, 2, …, M maintains a list of sensors and a list of data freshness; represents the time stamp of the qth request of the nth = 1, 2, …, N sensor data by the routing node i = 1, 2, …, M; the routing node i sends an interest packet as a consumer, and the interest packet carries a freshness requirement P i,n , P i,n represents the freshness requirement of the nth sensor data by the routing node i; N is the number of sensors. 3.The method of claim 1, wherein, The LSTM processing in step S13 comprises: wherein, represents the output of the forget gate at time t, represents the output of the input gate at time t, represents the current input cell state, represents the cell state at time t, represents the output of the output gate at time t, represents the forget gate weight matrix, represents the input gate weight matrix, represents the cell state weight matrix, represents the output gate weight matrix, represents the forget gate bias, represents the input gate bias, represents the cell state bias, represents the output gate bias, represents the sigmod function, represents the LSTM output at time t-1, represents the historical request time interval sequence of routing node i to the data generated by sensor n.

4. The method of claim 1, wherein, The information value of the current data packet is calculated through the caching probability of the current data packet, which is represented as: wherein, represents the probability of caching the data packet produced by sensor n at the Qth request.

Images