A method and system for real-time communication in aerospace distributed network

Abstract

The present invention relates to the field of data communication technology, and discloses a real-time communication method and system for aerospace distributed networks. The present invention obtains the observation value collected by the collection module at the current moment through the publisher end, and obtains the prediction value corresponding to the current moment output by the data prediction model based on the data prediction model associated with the collection module. When it is detected that the error between the observation value and the prediction value exceeds a preset range, the publisher end and the subscriber end will have real communication, and the data prediction model will be updated using the observation value to correct the prediction accuracy of the data prediction model. When it is detected that the error between the observation value and the prediction value does not exceed the preset range, the value of the collection module is predicted by the data prediction model, which will not be affected by communication delays, thereby ensuring the real-time nature of communication.

Description

Technical Field

[0001] The present invention relates to the field of data communication technology, and in particular to a method and system for real-time communication in an aerospace distributed network. Background Art

[0002] For aerospace distributed embedded systems, which include a large number of sensors and multiple computing resources, the complex system is affected by the complexity of aerospace business and the harshness of the operating environment. The aerospace distributed embedded system has highly dynamic characteristics, which are reflected in the dynamics of computing resources and the dynamics of physical processes. Many elements in aerospace embedded systems may change dramatically during operation. The reasons for this change are diverse, including temporary failure of system elements, continuous failure, business changes, communication noise, system reconstruction, etc. Therefore, the dynamic nature of distributed embedded real-time systems puts forward special requirements for the networks and network protocols connecting various system elements to adapt to this dynamic change characteristic.

[0003] At present, most communication protocols use a cache-based approach to improve communication quality. In this approach, the data to be transmitted is placed in the cache, greatly reducing the communication and computing load. For example, for those frequently used objects, once they are transmitted to the target location for the first time, they are saved in the cache; if the object is accessed later, there is no need to initiate a transmission request to the source again, and the value of the object can be directly obtained from the cache. This approach has become a standard communication mode in Web-based service systems. The cache method is more effective for relatively static but frequently accessed systems; however, for systems where the data source changes in real time, the cache method is difficult to meet the dynamic needs of the system. Summary of the invention

[0004] In view of this, the present invention provides a method and system for real-time communication in an aerospace distributed network to solve the problem that related technologies cannot meet the dynamic requirements in a system where data sources change in real time.

[0005] In a first aspect, the present invention provides a method for real-time communication in an aerospace distributed network, which is applied to a publisher end in an aerospace distributed network real-time communication system composed of a publisher end, a subscriber end and a plurality of acquisition modules, wherein the publisher end and the subscriber end rely on data collected by the plurality of acquisition modules, and each acquisition module is associated with a data prediction model for characterizing data features of the acquisition module, and the method comprises:

[0006] Get the observation value collected by the collection module at the current moment;

[0007] Based on the data prediction model associated with the acquisition module, obtain the prediction value corresponding to the current moment output by the data prediction model;

[0008] When it is detected that the error between the observed value and the predicted value exceeds a preset range, the observed value is sent to the subscriber end, so that the subscriber end updates the data prediction model based on the observed value, and stores the observed value for application calls;

[0009] And / or, when it is detected that the error between the observed value and the predicted value does not exceed a preset range, the observed value is discarded so that the subscriber side stores the predicted value output by the data prediction model for application call.

[0010] Thus, the publisher obtains the observation value collected by the acquisition module at the current moment, and based on the data prediction model associated with the acquisition module, obtains the prediction value corresponding to the current moment output by the data prediction model. When it is detected that the error between the observation value and the prediction value exceeds the preset range, the publisher and the subscriber will have real communication, and the data prediction model will be updated using the observation value to correct the prediction accuracy of the data prediction model. When it is detected that the error between the observation value and the prediction value does not exceed the preset range, the value of the acquisition module is predicted by the data prediction model, which will not be affected by communication delays, thereby ensuring the real-time nature of communication.

[0011] This allows the accuracy of the data prediction model to return to the predetermined range.

[0012] In an optional embodiment, the method further includes:

[0013] Obtain the resource utilization of the publisher at different times and the accuracy of the current collection module;

[0014] Based on the pre-acquired target resource utilization and resource utilization and accuracy at different times, the control parameters are calculated to make the resource utilization of the publisher at the current moment reach the target resource utilization; the control parameters are the accuracy of the current acquisition module at the current moment.

[0015] Therefore, by using the pre-acquired target resource utilization and resource utilization at different times and precision, the control parameters that make the resource utilization of the publisher at the current time reach the target resource utilization are calculated, so as to effectively control the resource utilization of the publisher.

[0016] In an optional implementation, the subscriber end includes multiple subscribers, and sending the observation value to the subscriber end includes:

[0017] The observation value is sent to each subscriber on the subscriber side through multicast.

[0018] Thus, the observed value is sent to each subscriber on the subscriber side by multicast, so that each subscriber can obtain the latest value of the acquisition module.

[0019] In a second aspect, the present invention provides a method for real-time communication in an aerospace distributed network, which is applied to a subscriber end in an aerospace distributed network real-time communication system composed of a publisher end, a subscriber end and a plurality of acquisition modules, wherein the publisher end and the subscriber end rely on data collected by the plurality of acquisition modules, and each acquisition module is associated with a data prediction model for characterizing data features of the acquisition module, and the method comprises:

[0020] When receiving the observation value collected by the collection module at the current moment sent by the publisher, the data prediction model associated with the collection module is updated based on the observation value, and the observation value is stored for application program to call;

[0021] And / or, when the observation value sent by the publisher is not received, the data collected by the collection module at the current moment is predicted through the data prediction model to obtain the prediction value corresponding to the current moment, and the prediction value is stored for application call.

[0022] Therefore, when the observed value collected by the collection module at the current moment is received from the publisher, the data prediction model associated with the collection module is updated based on the observed value, and the prediction accuracy of the data prediction model is corrected. When the observed value sent by the publisher is not received, the data collected by the collection module at the current moment is predicted by the data prediction model to obtain the predicted value corresponding to the current moment, which will not be affected by communication delays, thereby ensuring the real-time nature of communication.

[0023] In an optional implementation, the data prediction model depends on the state vector of the acquisition module; updating the data prediction model associated with the acquisition module based on the observation value includes:

[0024] Update the state vector that the current data prediction model depends on based on the observed value;

[0025] Before updating, the state vector that the data prediction model depends on at the current time t1 is:

[0026]

[0027] After the update, the state vector that the data prediction model depends on at the current time t1 is:

[0028]

[0029] Among them, x t represents the observation value of the acquisition module at any time t, It represents the data change rate of the acquisition module, and v represents the observation value sent by the publisher.

[0030] Therefore, by updating the state vector that the current data prediction model depends on based on the observed value, the efficiency and real-time performance of data communication are guaranteed.

[0031] In an optional implementation, the data collected by the collection module at the current moment is predicted by a data prediction model to obtain a prediction value corresponding to the current moment, including:

[0032] According to the preset cycle, the data collected by the collection module at the current moment is repeatedly predicted by the data prediction model to obtain the prediction value corresponding to the current moment;

[0033] The preset period is smaller than the data valid period corresponding to the acquisition module, and the data valid period is obtained based on the accuracy of the acquisition module and the data change rate.

[0034] Therefore, by repeatedly predicting the data collected by the collection module at the current moment according to a preset cycle, a prediction value corresponding to the current moment is obtained to ensure the freshness and validity of the data.

[0035] In an optional embodiment, the method further includes:

[0036] Predicting the occupancy of the buffer used by the subscriber to store data to obtain a predicted buffer occupancy value;

[0037] Based on the predicted buffer occupancy value, the data receiving speed of the subscriber is adjusted.

[0038] Therefore, by predicting the occupancy of the buffer area used to store data at the subscriber end, the data receiving speed at the subscriber end can be adjusted.

[0039] In a third aspect, the present invention provides an aerospace distributed network real-time communication system, comprising: a publisher end, a subscriber end, and a plurality of acquisition modules; the publisher end and the subscriber end rely on data collected by the plurality of acquisition modules, and each acquisition module is associated with a data prediction model for characterizing data features of the acquisition module;

[0040] The publisher obtains the observation value collected by the collection module at the current moment; based on the data prediction model associated with the collection module, obtains the prediction value corresponding to the current moment output by the data prediction model; when it is detected that the error between the observation value and the prediction value exceeds the preset range, the observation value is sent to the subscriber;

[0041] When the subscriber receives the observation value collected by the collection module at the current moment and sent by the publisher, the subscriber updates the data prediction model associated with the collection module based on the observation value, and stores the observation value for application program to call.

[0042] In an optional implementation, the publisher discards the observed value when detecting that the error between the observed value and the predicted value does not exceed a preset range;

[0043] When the subscriber does not receive the observation value sent by the publisher, it predicts the data collected by the collection module at the current moment through the data prediction model, obtains the prediction value corresponding to the current moment, and stores the prediction value for application calls.

[0044] In an optional implementation, the accuracy of each acquisition module is dynamically adjusted according to a preset quality specification;

[0045] The preset quality specifications include data quality specifications and service quality specifications. The data quality specifications are obtained based on the preset upper limit of accuracy of each acquisition module, and the service quality specifications are obtained based on resource utilization on the publisher side or the subscriber side. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 is a structural block diagram of an aerospace distributed network real-time communication system according to an embodiment of the present invention;

[0048] Figure 2 is a flow chart of a method for real-time communication in an aerospace distributed network according to an embodiment of the present invention;

[0049] Figure 3 is a flow chart of another aerospace distributed network real-time communication method according to an embodiment of the present invention;

[0050] Figure 4 1 is a control schematic diagram of an aerospace distributed network real-time communication system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0052] According to an embodiment of the present invention, a spaceflight distributed network real-time communication system is provided, such as Figure 1 As shown, the aerospace distributed network real-time communication system includes: a publisher end 101, a subscriber end 102 and multiple acquisition modules 103, wherein the publisher end 101 and the subscriber end 102 rely on data collected by multiple acquisition modules 103, and each acquisition module 103 is associated with a data prediction model for characterizing the data characteristics of the acquisition module 103. Exemplarily, the acquisition module 103 can be a sensor, but the present invention is not limited thereto.

[0053] Specifically, taking the aerospace distributed embedded communication system as an example, in the aerospace embedded system, most of the processing nodes (publishers and subscribers) rely on the data collected by the acquisition module 103 of the physical world. These data are often continuous and follow certain physical or mathematical laws, and have corresponding physical or mathematical models. Therefore, the data of these acquisition modules 103 reflect certain semantic characteristics through the model. The semantic characteristics of the data transmitted between the publisher and the subscriber can not only reduce the calculation and communication load of the two communicating parties, but also improve the reliability and real-time performance of the communication. Exemplarily, for a continuous data topic (such as continuous temperature data collected by a temperature sensor), it can be marked as DATA_CONTINUOUS. If a topic data has this mark, then its corresponding acquisition module 103 (such as a temperature sensor) must be associated with a data prediction model (such as a temperature sensor model).

[0054] The publisher 101 obtains the observation value collected by the collection module 103 at the current moment; based on the data prediction model associated with the collection module 103, obtains the prediction value corresponding to the current moment output by the data prediction model; when it is detected that the error between the observation value and the prediction value exceeds the preset range, the observation value is sent to the subscriber 102. When the subscriber 102 receives the observation value collected by the collection module 103 at the current moment sent by the publisher 101, it updates the data prediction model associated with the collection module based on the observation value, and stores the observation value for application calls.

[0055] In addition, when the publisher 101 detects that the error between the observed value and the predicted value does not exceed the preset range, the observed value is discarded. When the subscriber 102 does not receive the observed value sent by the publisher 101, it predicts the data collected by the collection module 103 at the current moment through the data prediction model, obtains the predicted value corresponding to the current moment, and stores the predicted value for application calls.

[0056] In this way, the application can dynamically obtain the value of the acquisition module 103, which may be transmitted by the publisher 101, or calculated according to the corresponding data prediction model. If the value of the acquisition module 103 obtained by the application is a predicted value calculated by the data prediction model, then the subscriber does not request data from the publisher, which enables the system's data distribution service to effectively deal with the problem of data packet loss in an unstable network environment and ensure the semantic correctness of communication. Furthermore, the value of the acquisition module 103 can be estimated by the model without being affected by communication delays, so that the real-time nature of communication is well maintained.

[0057] It should be noted that the publisher end 101 can also update the data prediction model. In the embodiment of the present invention, the data prediction model between the publisher end 101 and the subscriber end 102 is the same. However, in the actual network communication process, there may be differences between the model of the publisher end 101 and the model of the subscriber end 102, and this difference may be caused by network instability. For example, when the data sent from the publisher end 101 to the subscriber end 102 is lost during the communication process, the subscriber end 102 will not update the data prediction model, and the publisher end 101 has completed the model update, so that the models at both ends of the publisher and the subscriber are in different states. If a lot of data is transmitted to the subscriber end 102, then this model deviation will not cause too much impact, because the subsequent update operation will correct this model state error. However, if the application is extremely sensitive to errors, then special measures need to be taken to ensure the accuracy of data transmission. On the basis of the above, the embodiment of the present invention proposes a transmission capacity guarantee mechanism that can be attached. The attachable transmission mechanism allows different topics to select different underlying transmission protocols. When the data of a specific topic is very critical, the topic's criticality flag can be used to determine the use of the TCP transmission protocol to ensure the reliability of data transmission; for non-critical topics, the UDP protocol can be selected as the transmission protocol.

[0058] In some optional embodiments, the accuracy of each acquisition module 103 is dynamically adjusted according to preset quality specifications, wherein the preset quality specifications include data quality specifications and service quality specifications, the data quality specifications are obtained based on the preset upper limit of accuracy of each acquisition module 103, and the service quality specifications are obtained based on the resource utilization of the publisher end 101 or the subscriber end 102.

[0059] It should be noted that correctly setting the upper limit of the precision of the data collected by each collection module 103 is crucial to ensuring the service quality / data quality of the data distribution service communication. The value of the collection module predicted by the subscriber end 102 is determined by the upper limit of the precision of the data prediction model, and the upper limit of the precision of the data prediction model is consistent with the upper limit of the precision of the data collected by the corresponding collection module 103. However, how to correctly set the upper limit of the precision Δ of each collection module 103 is indeed a challenging task. The upper limit of the precision Δ of the collection module 103 cannot be determined in advance either in the design stage or in the deployment stage, because as the operating environment changes, the maximum data quality obtained by the system changes with the running time. If the upper limit of the precision Δ value is set too small, the error between the observed value and the predicted value is easy to exceed the preset range, then the system may face the risk of overload, and the publisher end 101 will pass almost all the observed data to the subscriber end 102. On the contrary, if the Δ value is set too large, the accuracy of the data value predicted by the data prediction model for the collection module 103 will be reduced, and it is difficult to meet the control accuracy requirements of the system. Therefore, the value of the upper limit of the precision Δ of the collection module 103 can be dynamically adjusted through an adaptive control mechanism.

[0060] Specifically, during operation, the actual upper limit of the precision of the data stream of the acquisition module 103 is modified according to the preset quality specification. It should be noted that in the real-time data distribution service, both the data quality and the service quality are set by the user. The data quality can be defined as the upper limit of the precision of the data collected by the acquisition module 103. Assume that the acquisition module 103 is s i , its data flow is stream i , whose upper limit of accuracy is Δ i There is a preset quality specification on both the publisher end 101 and the subscriber end 102, which is used to specify the service quality specification and data quality specification of each acquisition module 103. The preset quality specification is expressed as <Q,U max >, where Q represents the data quality specification requirements, U max Indicates the requirements of the service quality specification.

[0061] The data quality specification requirement Q is as follows:

[0062] Q= <q1,q2,…,q n , γ max >

[0063] Among them, q n Indicates that the data flow is stream n The preset upper limit of accuracy, γ max Indicates the maximum tolerable precision scaling factor. For any data stream i , which corresponds to the label s iThe actual upper limit of the precision of the acquisition module 103 is Δ i =q i ×γ, where γ≤γ max .

[0064] Quality of Service Specification Requirements max ∈[0, 1], which represents the upper limit of the CPU utilization of the node, where the node can be the publisher node in the publisher end 101 or the subscriber node in the subscriber end 102. In this way, while meeting the target performance, U max Can prevent system overload.

[0065] According to an embodiment of the present invention, an embodiment of a real-time communication method for aerospace distributed networks is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0066] In this embodiment, a real-time communication method for aerospace distributed network is provided. The real-time communication method for aerospace distributed network can be used as follows: Figure 1 The publisher end 101 and the subscriber end 102 are shown. Figure 2 1 is a schematic diagram of the interaction process of the aerospace distributed network real-time communication method according to an embodiment of the present invention, wherein the publisher end 101 is used to execute steps S101 to S104, and the subscriber end 102 is used to execute steps S201 to S202. The specific interaction process between the publisher end 101 and the subscriber end 102 is as follows:

[0067] Step S101, obtaining the observation value collected by the collection module at the current moment.

[0068] Specifically, the acquisition module may be a sensor that can acquire observation values ​​at the current moment, such as temperature, pressure, etc., in real time.

[0069] Step S102, based on the data prediction model associated with the acquisition module, obtain the prediction value corresponding to the current moment output by the data prediction model.

[0070] Specifically, when a publisher p on the publisher side i From a collection module i When a new observation value v is observed at the location, the acquisition module s is queried. i The corresponding data prediction model m i , and then use the data to predict the model m i Calculate the predicted value Here t is the time of observation.

[0071] Step S103, when it is detected that the error between the observed value and the predicted value exceeds a preset range, the observed value is sent to the subscriber end.

[0072] Specifically, if the predicted value If the difference between the observed value and v exceeds the set accuracy range δ, then the data prediction model m i It is impossible to give an effective value for the acquisition module, so it is necessary to use the current observation value v to update the data prediction model m i , and sends the current observation value v to the subscriber. The subscriber can then update the data prediction model based on the observation value and store the observation value for application calls.

[0073] In some optional implementations, the observation value is sent to each subscriber at the subscriber end by multicasting, so that each subscriber can obtain the latest value of the acquisition module.

[0074] Step S201, when receiving the observation value collected by the collection module at the current moment sent by the publisher, the data prediction model associated with the collection module is updated based on the observation value, and the observation value is stored for application calling.

[0075] Specifically, when the subscriber receives the acquisition module s i When a new observation value v is received, the subscriber updates its own collection module s with the observation value v i Data prediction model m i , and then the acquisition module s i The latest observed value v is saved in the buffer Buffer for application access.

[0076] Specifically, the data prediction model depends on the state vector of the acquisition module, and the state vector of the data prediction model at the current moment is updated based on the observed value. Before the update, the state vector of the data prediction model at the current moment t1 is:

[0077]

[0078] After the update, the state vector that the data prediction model depends on at the current time t1 is:

[0079]

[0080] Among them, x t represents the observation value of the acquisition module at any time t, It represents the data change rate of the acquisition module, and v represents the observation value sent by the publisher.

[0081] It should be noted that in order to ensure the efficiency and real-time performance of the data distribution service, a unified data prediction model needs to be provided for similar acquisition modules. In the aerospace distributed network real-time communication system, there are a large number of different physical acquisition modules and virtual acquisition modules. Therefore, the data distribution service needs to process the data streams of a large number of acquisition modules. However, the data prediction models corresponding to many physical acquisition modules are very complex, and the solution of the model is also very complex. Due to the characteristics of the data distribution service, it is unreasonable to introduce complex data models for various acquisition modules, and complex data prediction models are also computationally infeasible. Therefore, the data prediction model corresponding to each acquisition module of the data distribution service in the embodiment of the present invention is obtained by modeling using the state space model of the acquisition module.

[0082] Specifically, for any acquisition module’s state space model x k , its state will evolve according to the following formula:

[0083] x k+1 =F k+1 x k

[0084] Among them, F k+1 is a state transition matrix that transforms x k+1 With x k In association, the state transition matrix F k+1 It can be defined as follows:

[0085]

[0086] Furthermore, since the state of the acquisition module is continuous, it can be described using differential equations. In the data distribution service, a certain quantity of a physical process is generated by a specific acquisition module s i Monitoring, at a given time t, acquisition module s i The state of can be represented as a state vector consisting of two state variables:

[0087]

[0088] Among them, x t Represents the value collected by the acquisition module at time t, It is the differential of x with respect to time t, and is used to indicate the data change rate of the acquisition module.

[0089] The above state space model makes it possible for the acquisition module s to be given a time t. i The state vector is [x t , dev t ] T , then at time t1 its state vector is [x t+(t1-t)dev t , dev t ] T , it can be seen that the value of the acquisition module changes with time, but the differential does not change.

[0090] However, if the predicted value If the difference between the observed value and v exceeds the set accuracy range, the state model needs to be updated so that the state variable is updated to

[0091] Therefore, by updating the state vector that the current data prediction model depends on based on the observed value, the efficiency and real-time performance of data communication are guaranteed.

[0092] In this way, only when the accuracy of the predicted value calculated by the data prediction model cannot be guaranteed, the publisher and the subscriber will have real communication. And once this communication occurs, the observation value of the acquisition module will be used to correct the current state model, so that the prediction accuracy of the data prediction model can return to the predetermined range. This makes it unnecessary for the publisher and the subscriber to have actual communication in the future, thereby reducing the communication overhead of data publishing and improving the real-time performance of data communication.

[0093] Alternatively, the publisher performs step S104 after performing step S102, and then the subscriber performs step S202.

[0094] Step S104, when it is detected that the error between the observed value and the predicted value does not exceed a preset range, the observed value is discarded.

[0095] Specifically, if the predicted value If the difference between the observed value v and the observed value v is within the set accuracy range δ, then the observed value v is discarded, which means that the current data prediction model m i The observation value of the acquisition module is predicted within a given accuracy range δ, and this prediction can be completed independently by the subscriber side without passing the observed value of the publisher side to each subscriber on the subscriber side. Thus, the subscriber side stores the predicted value output by the data prediction model for application calls.

[0096] Step S202, when no observation value is received from the publisher, the data collected by the collection module at the current moment is predicted by the data prediction model to obtain the prediction value corresponding to the current moment, and the prediction value is stored for application call.

[0097] Specifically, when the observation value sent by the publisher is not received, the data collected by the acquisition module at the current moment is repeatedly predicted by the data prediction model according to the preset period to obtain the prediction value corresponding to the current moment, wherein the preset period is smaller than the data valid period corresponding to the acquisition module, and the data valid period is obtained based on the accuracy of the acquisition module and the data change rate.

[0098] It should be noted that maintaining the freshness of the data collected by the collection module is another difficulty that the data distribution service needs to solve. When the accuracy of the data prediction model meets the set accuracy range, there is no data communication between the publisher and the subscriber; only when the accuracy of the model's predicted value exceeds the set accuracy range, there will be real data communication between the publisher and the subscriber. However, the freshness of the data value of the collection module received last time on the subscriber side will decay with the passage of time. The time period during which the collection module data is valid is called the data valid period or the absolute valid interval avi.

[0099] Specifically, the subscriber must maintain the freshness and accuracy of the observations in the acquisition module when it does not receive an update notification from the publisher. Therefore, the subscriber must periodically update the observations in the buffer before the absolute valid interval avi expires. Since the subscriber maintains the same data prediction model as the publisher, the current state of the model can be used to predict the future state.

[0100] Furthermore, as long as the subscriber does not receive an update notification from the publisher, the subscriber uses each data prediction model to calculate the predicted acquisition module values ​​with a period of P≤avi, and then stores them in the buffer. The data validity period or absolute validity interval avi of each acquisition module can be calculated based on the upper limit of the accuracy of the acquisition module Δ, and the corresponding calculation formula is as follows:

[0101]

[0102] in, Indicates the data change rate of the acquisition module.

[0103] Intuitively speaking, if the data of a collection module changes rapidly, the subscriber end can only update the value of the collection module at a faster speed to ensure the freshness and validity of the data. According to empirical research, in order to maintain the freshness of the data, the update cycle of the collection module data in the embodiment of the present invention should be avi / 2, that is, the data collected by the collection module at the current moment is repeatedly predicted by the data prediction model according to the cycle of avi / 2 to obtain the predicted value corresponding to the current moment, so as to ensure the freshness and validity of the data.

[0104] In this way, the application can dynamically obtain the value of the acquisition module, which may be passed from the publisher or calculated based on the corresponding data prediction model. If the value of the acquisition module obtained by the application is the predicted value calculated by the data prediction model, then the subscriber does not request data from the publisher. This enables the system's data distribution service to effectively deal with the problem of data packet loss in an unstable network environment and ensure the semantic correctness of communication. Furthermore, the value of the acquisition module can be estimated by the model without being affected by communication delays, thereby maintaining the real-time nature of communication.

[0105] The aerospace distributed network real-time communication method provided in this embodiment obtains the observation value collected by the collection module at the current moment through the publisher, and obtains the prediction value corresponding to the current moment output by the data prediction model based on the data prediction model associated with the collection module. When it is detected that the error between the observation value and the prediction value exceeds the preset range, the observation value is sent to the subscriber end, so that the subscriber end updates the data prediction model based on the observation value, and stores the observation value for application call; when it is detected that the error between the observation value and the prediction value does not exceed the preset range, the observation value is discarded, so that the subscriber end stores the prediction value output by the data prediction model for application call.

[0106] This allows the application to dynamically obtain the value of the acquisition module, which may be passed from the publisher or calculated based on the corresponding data prediction model. If the value of the acquisition module obtained by the application is a predicted value calculated by the data prediction model, then the subscriber does not request data from the publisher. This enables the system's data distribution service to effectively deal with the problem of data packet loss in an unstable network environment and ensure the semantic correctness of communication. Furthermore, the value of the acquisition module can be estimated by the model without being affected by communication delays, thereby maintaining the real-time nature of communication.

[0107] Only when the accuracy of the predicted value calculated by the data prediction model cannot be guaranteed, the publisher and the subscriber will have real communication. And once this communication occurs, the observation value of the acquisition module will be used to correct the current state model, so that the prediction accuracy of the data prediction model can return to the predetermined range. This makes it unnecessary for the publisher and the subscriber to have actual communication in the future, thereby reducing the communication overhead of data publishing and improving the real-time performance of data communication.

[0108] In this embodiment, a real-time communication method for aerospace distributed network is provided. The real-time communication method for aerospace distributed network can be used as follows: Figure 1The publisher end 101 and the subscriber end 102 are shown. Figure 3 1 is a schematic diagram of the interaction process of the aerospace distributed network real-time communication method according to an embodiment of the present invention, wherein the publisher end 101 is used to execute steps S301 to S306, and the subscriber end 102 is used to execute steps S401 to S404. The specific interaction process between the publisher end 101 and the subscriber end 102 is as follows:

[0109] Step S301, obtain the observation value collected by the collection module at the current moment. For details, see Figure 2 Step S101 of the illustrated embodiment will not be described in detail here.

[0110] Step S302: Based on the data prediction model associated with the acquisition module, obtain the prediction value corresponding to the current moment output by the data prediction model. Figure 2 Step S102 of the illustrated embodiment will not be described in detail here.

[0111] Step S303: When it is detected that the error between the observed value and the predicted value exceeds a preset range, the observed value is sent to the subscriber end, so that the subscriber end updates the data prediction model based on the observed value, and stores the observed value for application calls. Figure 2 Step S103 of the illustrated embodiment will not be described in detail here.

[0112] Step S401: upon receiving the observation value collected by the collection module at the current moment from the publisher, the data prediction model associated with the collection module is updated based on the observation value, and the observation value is stored for application program to call. Figure 2 Step S201 of the illustrated embodiment will not be described in detail here.

[0113] Alternatively, the publisher performs step S304 after performing step S302, and then the subscriber performs step S402.

[0114] Step S304: When it is detected that the error between the observed value and the predicted value does not exceed the preset range, the observed value is discarded so that the subscriber can store the predicted value output by the data prediction model for application program to call. Figure 2 Step S104 of the illustrated embodiment will not be described in detail here.

[0115] Step S402: When no observation value is received from the publisher, the data collected by the collection module at the current moment is predicted by the data prediction model to obtain the prediction value corresponding to the current moment, and the prediction value is stored for application program to call. Figure 2 Step S202 of the illustrated embodiment will not be described in detail here.

[0116] The following describes steps S305 to S306 executed by the publisher and steps S403 to S404 executed by the subscriber.

[0117] Step S305, obtaining the resource utilization of the publisher at different times and the accuracy of the current acquisition module.

[0118] Step S306, based on the pre-acquired target resource utilization and resource utilization and accuracy at different times, calculate the control parameters that make the resource utilization of the publisher at the current time reach the target resource utilization; the control parameters are the accuracy of the current acquisition module at the current time.

[0119] Specifically, Figure 4 As shown, the publisher uses an interactive control mode to achieve system adaptation, which is essentially an anti-control structure. In order to effectively control the CPU utilization on the publisher side, a well-defined and effectively verified feedback control theory needs to be used as a basis. In order to evaluate the relationship between the accuracy Δ of the acquisition module and the CPU utilization (resource utilization), the following evaluation method is used: For the CPU utilization U(t) at the kth sampling moment, the CPU utilization and model accuracy Δ at the previous p sampling moments are used as the basis. This relationship can be expressed as the following differential equation in the discrete time domain:

[0120]

[0121] Among them, a i and b i It is an adjustable parameter.

[0122] The above differential equation models the dynamic process on the publisher side. After completing the adaptive control dynamics modeling in the publisher side, a controller needs to be designed based on this model to ensure that the measured CPU utilization U(t) is equal to the target utilization U ref In order to support both average performance evaluation and instantaneous performance evaluation, the PI control (Proportional Integral) method is used to control the publisher. In the PI control method, P controls short-term interactions, and adding I control on the basis of P can achieve long-term stability.

[0123] At the kth sampling point, the PI controller calculates the target utilization U ref The required model accuracy Δ(t), which is the control signal input to the PI controller.

[0124] Δ(t)=Δ(t-1)+(K P +K I )e(t)-KP e(t-1)

[0125] Among them, K P is the P control parameter of the PI controller, K I These are the I control parameters of the PI controller. They have a huge impact on the settling time and stability of the model and can be set according to actual needs.

[0126] When monitoring CPU load (resource utilization), sampling interval selection is another area that needs attention. If the sampling interval of CPU load is too short, the system's metric output may be highly variable, making the controller very sensitive to transient changes in the system. On the contrary, if the sampling interval of CPU load is too long, the control speed will be very slow and unable to correctly capture the dynamic characteristics of the system.

[0127] Therefore, by using the pre-acquired target resource utilization and resource utilization at different times and precision, the control parameters that make the resource utilization of the publisher at the current time reach the target resource utilization are calculated, so as to effectively control the resource utilization of the publisher.

[0128] Step S403: predicting the occupancy of the buffer area used for storing data at the subscriber end to obtain a predicted buffer area occupancy value.

[0129] Step S404: adjusting the data receiving speed of the subscriber end based on the buffer occupancy prediction value.

[0130] Specifically, Figure 4 As shown in the figure, the main purpose of subscriber-side adaptive control is to set the correct data arrival speed (data receiving speed), because the data arrival speed is highly correlated with the application data processing capability and the subscriber-side performance overhead. To control the data arrival speed, it is necessary to notify the publisher to adjust the data sending speed according to the current operating status of the subscriber. Therefore, subscriber-side adaptive control is essentially a feedforward control loop. There may be a long network delay between the data publisher and the subscriber, which makes it impossible to implement an interactive adaptive mechanism on the subscriber side. The state difference between the publisher and the subscriber may cause the publisher to make errors in the corrective behavior based on the current state.

[0131] In order to deal with the synchronization problem between publishers and subscribers caused by network delays, a buffer mechanism is introduced on the subscriber side. The received data and the data estimated by model prediction are placed in the buffer. The potential overflow of the buffer is proactively predicted and notified to the publisher, thereby adjusting the speed at which the publisher sends data and adjusting the data receiving speed on the subscriber side.

[0132] In some optional implementations, based on the buffer mechanism, queuing theory related techniques are used to predict the buffer occupancy and the delay caused by the input load.

[0133] Specifically, for a subscriber, let its target buffer occupancy be b occ , the expected buffer size b size is the maximum buffer size b max Multiply by b occ , that is, b size =b max ×b occ .

[0134] At any sampling point n, the average size of the system's buffer L(n) is:

[0135] L(n)=λ(n)W(n)

[0136] Among them, λ(n) represents the data arrival rate, and W(n) represents the average delay of data processing. This formula is independent of the probability distribution involved, so there is no need to make any assumptions about the probability distribution of data flow arrival or data processing in the acquisition module.

[0137] Therefore, by predicting the occupancy of the buffer area used to store data at the subscriber end, the data receiving speed at the subscriber end can be adjusted.

[0138] See again Figure 4 As shown in the figure, in order to achieve the expected service quality and data quality, the real-time data distribution service adopts a dual-loop adaptive control strategy control architecture. The key points of this architecture are the dynamic nature of the control and the dynamic evolution process in two independent time scales. The publisher can immediately obtain the latest value of the acquisition module data and can timely perceive the service quality and data quality violations of the communication. Therefore, the service quality and data quality adaptive control of the publisher should be performed on a smaller time scale.

[0139] However, for the subscriber side, the time scale required for adaptive adjustment will be longer. First, the propagation of global load is relatively slow. It needs to be transmitted from the publisher side to the subscriber side. The publisher side also needs to filter it according to data quality requirements. Only when the accuracy violation occurs will the subscriber side be notified to make adjustments. Secondly, there is a communication delay between the publisher side and the subscriber side. It is very difficult for the subscriber side to make adaptive adjustments immediately according to the load changes; due to the communication delay, the corrective actions taken by the publisher side based on the current state may be incorrect. Based on the above factors and problems, the control loop on the subscriber side is proactive and needs to be performed on a longer time scale. Proactive control is mainly based on the buffer on the subscriber side. Each subscriber monitors the usage of the buffer, proactively predicts the possible overflow of the buffer, and reports to the publisher in a larger time period.

[0140] The following table is the definition of the adaptive control strategy provided by the embodiment of the present invention. In this adaptive control strategy, the publisher and the subscriber have great differences in control mechanism, monitoring variables, control input, sampling interval, etc. In terms of control mechanism, the publisher adopts interactive control, while the subscriber adopts proactive control; in terms of monitoring variables, the publisher monitors CPU utilization; the subscriber monitors the speed of data input or processing; in terms of control input, the publisher's input is the upper limit of the accuracy of the acquisition module Δ, and the subscriber monitors the upper limit of the CPU utilization of the publisher.

[0141] Table 1 Adaptive control strategy for data distribution service

[0142]

[0143] The aerospace distributed network real-time communication method provided by the embodiment of the present invention provides a real-time and reliable data distribution service in a network environment with explosive load and unstable network, so that the communication process has semantic perception ability, can perform proactive or interactive communication mode adjustment according to the quality of data, use the feedback control mechanism on the publisher side to adjust the model accuracy to reach the expected utility limit, use the proactive feedforward mechanism on the subscriber side to correctly set the speed at which the publisher provides data, and form an organic and adaptive semantic perception data distribution control loop. It can provide a robust real-time communication means in the environment of load mutation and unstable network, and deal with the unreliable and uncertain problems caused by temporary failure, continuous failure, business change, communication noise, system reconstruction, etc. of system elements in the communication process.

[0144] The aerospace distributed network real-time communication method according to an embodiment of the present invention is further described in detail below with reference to a specific application example.

[0145] As shown in Algorithm 1, when a publisher p on the publisher side i From a collection module i When a new observation value v is observed at the location, the acquisition module s is queried. i The corresponding data prediction model m i , and then use the data to predict the model m i Calculate the predicted value Here t is the time of observation. If the predicted value If the difference between the observed value v and the observed value v is within the set accuracy range δ, then the observed value v is discarded, which means that the current data prediction model m i The prediction of the observed value of the acquisition module within a given accuracy range δ can be done independently by the subscriber side without passing the observed value of the publisher side to each subscriber on the subscriber side. However, if the predicted value If the difference between the observed value and v exceeds the set accuracy range δ, then the data prediction model m i It is impossible to give an effective value for the acquisition module, so it is necessary to use the current observation value v to update the data prediction model m i , and delivers the current observation value v to all subscribers in a multicast manner.

[0146]

[0147] Only when the accuracy of the predicted value calculated by the data prediction model cannot be guaranteed, the publisher and the subscriber will have real communication. And once this communication occurs, the observation value of the acquisition module will be used to correct the current state model, so that the prediction accuracy of the data prediction model can return to the predetermined range. This makes it unnecessary for the publisher and the subscriber to have actual communication in the future, thereby reducing the communication overhead of data publishing and improving the real-time performance of data communication.

[0148] As shown in Algorithm 2, when the subscriber receives the acquisition module s i When a new observation value v is received, the subscriber updates its own collection module s with the observation value v i Data prediction model m i , and then the acquisition module s i The latest observed value v is saved in the buffer Buffer for application access.

[0149]

[0150]

[0151] In this way, the application can dynamically obtain the value of the acquisition module si, which may be passed from the publisher or calculated based on the corresponding data prediction model mi. If the value of the acquisition module obtained by the application is the predicted value calculated by the data prediction model mi, then the subscriber does not request data from the publisher, which enables the system's data distribution service to effectively deal with the problem of data packet loss in an unstable network environment and ensure the semantic correctness of communication. Furthermore, the value of the acquisition module can be estimated by the model without being affected by communication delays, thereby maintaining the real-time nature of communication.

[0152] As shown in Algorithm 3, when the subscriber does not receive updated data from the publisher, the data of the collection module will also be updated. The data collected by the collection module at the current moment will be repeatedly predicted through the data prediction model according to the preset period to obtain the prediction value corresponding to the current moment to ensure the freshness and validity of the data.

[0153]

[0154] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A space distributed network real-time communication method, applied to the publisher end of a space distributed network real-time communication system composed of a publisher end, a subscriber end and a plurality of acquisition modules, characterized in that: The publisher and the subscriber rely on data collected by multiple collection modules, each collection module is associated with a data prediction model for characterizing data features of the collection module, and the subscriber maintains a data prediction model that is the same as that of the publisher. The method includes: Get the observation value collected by the collection module at the current moment; Based on the data prediction model associated with the acquisition module, obtaining a prediction value corresponding to the current moment output by the data prediction model; When it is detected that the error between the observed value and the predicted value exceeds a preset range, the observed value is sent to the subscriber end, so that the subscriber end updates the data prediction model based on the observed value, and stores the observed value for application program call; And / or, when it is detected that the error between the observed value and the predicted value does not exceed a preset range, the observed value is discarded so that the subscriber side stores the predicted value output by the data prediction model for application call.

2. The method according to claim 1, characterized in that The method further comprises: Obtain the resource utilization of the publisher at different times and the accuracy of the current collection module; According to the pre-acquired target resource utilization and resource utilization and accuracy at different times, the control parameters are calculated so that the resource utilization of the publisher at the current time reaches the target resource utilization; the control parameters are the accuracy of the current acquisition module at the current time.

3. The method according to claim 1, characterized in that The subscriber end includes a plurality of subscribers, and sending the observation value to the subscriber end includes: The observation value is sent to each subscriber at the subscriber end by multicasting.

4. A space distributed network real-time communication method, applied to the subscriber end of a space distributed network real-time communication system composed of a publisher end, a subscriber end and a plurality of acquisition modules, characterized in that: The publisher and the subscriber rely on data collected by multiple collection modules, each collection module is associated with a data prediction model for characterizing data features of the collection module, and the subscriber maintains a data prediction model that is the same as that of the publisher. The method includes: When receiving the observation value collected by the collection module at the current moment and sent by the publisher, the data prediction model associated with the collection module is updated based on the observation value, and the observation value is stored for application program to call; And / or, when the observation value sent by the publisher is not received, the data collected by the acquisition module at the current moment is predicted by the data prediction model to obtain the prediction value corresponding to the current moment, and the prediction value is stored for application call.

5. The method according to claim 4, characterized in that The data prediction model depends on the state vector of the acquisition module; and updating the data prediction model associated with the acquisition module based on the observation value includes: Based on the observed value, the state vector on which the data prediction model depends at the current moment is updated; Before updating, the state vector that the data prediction model depends on at the current time t1 is: After the update, the state vector that the data prediction model depends on at the current time t1 is: Among them, x t represents the observation value of the acquisition module at any time t, represents the data change rate of the acquisition module, and v represents the observation value sent by the publisher.

6. The method according to claim 4, characterized in that The step of predicting the data collected by the collection module at the current moment by using the data prediction model to obtain a prediction value corresponding to the current moment includes: According to a preset cycle, the data collected by the collection module at the current moment is repeatedly predicted by the data prediction model to obtain a prediction value corresponding to the current moment; The preset period is smaller than the data validity period corresponding to the acquisition module, and the data validity period is obtained based on the accuracy of the acquisition module and the data change rate.

7. The method according to claim 4, characterized in that The method further comprises: Predicting the occupancy of the buffer used by the subscriber to store data to obtain a predicted buffer occupancy value; Based on the buffer occupancy prediction value, the data receiving speed of the subscriber is adjusted.

8. An aerospace distributed network real-time communication system, characterized in that: The system comprises: a publisher, a subscriber and a plurality of collection modules; the publisher and the subscriber rely on data collected by the plurality of collection modules, each collection module is associated with a data prediction model for characterizing the data features of the collection module, and the subscriber maintains a data prediction model identical to that of the publisher; The publisher obtains the observation value collected by the collection module at the current moment; based on the data prediction model associated with the collection module, obtains the prediction value corresponding to the current moment output by the data prediction model; when it is detected that the error between the observation value and the prediction value exceeds a preset range, the observation value is sent to the subscriber; When the subscriber receives the observation value collected by the collection module at the current moment and sent by the publisher, the subscriber updates the data prediction model associated with the collection module based on the observation value, and stores the observation value for application calling.

9. The system according to claim 8, characterized in that When the publisher detects that the error between the observed value and the predicted value does not exceed a preset range, the publisher discards the observed value; When the subscriber does not receive the observation value sent by the publisher, the subscriber predicts the data collected by the collection module at the current moment through the data prediction model, obtains the prediction value corresponding to the current moment, and stores the prediction value for application call.

10. The system according to claim 8, characterized in that The accuracy of each acquisition module is dynamically adjusted according to a preset quality specification; The preset quality specification includes a data quality specification and a service quality specification. The data quality specification is obtained based on a preset upper limit of accuracy of each of the acquisition modules, and the service quality specification is obtained based on resource utilization of the publisher or subscriber.

Images