Knowledge inference engine system and method of implementation

Patent metadata

Field	Value
Publication number	US-11526787-B1
Application number	US-202217842917-A
Country	US
Kind code	B1
Filing date	Jun 17, 2022
Priority date	Jun 18, 2021
Publication date	Dec 13, 2022
Grant date	Dec 13, 2022

Abstract

The present invention discloses a knowledge inference engine system and a method of implementation, relating to the field of wired communication networking technology. The system includes a data generation module, a stream partitioning model, an offline scheduling module, an online scheduling module, a scheduling solution base and a historical information base module, the data generation module used to generate a dataset and divide it into a plurality of partitions, the stream partitioning model used to partition the dataset, the offline scheduling module used to generate a scheduling solution for static network requirements, the online scheduling module used to rapidly generate a scheduling solution for a new TT stream, the scheduling solution base used to store a result of the partitioning, an iterative scheduling order for the partitions and the offline scheduling solution, the historical information base used to update and store relevant data information and performance indicators. The method of implementation proposed in the present invention combines and collaboratively operates offline iterative partition scheduling and an online incremental backtracking algorithm, addressing the problem that existing deterministic scheduling methods cannot simultaneously ensure scalability and schedulability.

First claim

The invention claimed is:

1. A knowledge inference engine system, characterized in comprising a data generation module, a stream partitioning model, an offline scheduling module, an online scheduling module, a scheduling solution base and a historical information base module;
the data generation module used to generate a large unlabeled dataset and a small labeled dataset and equally divide the datasets into a plurality of partitions;
the stream partitioning model used to, upon receipt of the datasets, based on a stream relevance metric value obtained from knowledge inference, partition the datasets and calculate an iterative scheduling order for individual partitions;
the offline scheduling module used to generate a scheduling solution for static network requirements;
the online scheduling module used to, when in time-sensitive networking, there is newly-added traffic, through reasonably adjusting an old stream scheduling solution, rapidly generate a scheduling solution for a new TT stream;
the scheduling solution base used to store partitioning results and an iterative scheduling order for the individual partitions, store the offline scheduling solution and the online scheduling solution, and enable, when there is newly-added traffic, the online scheduling module to immediately invoke the offline scheduling solution as a basis for scheduling;
the historical information base used to, based on a historical scheduling solution and the partitioning results in the scheduling solution base, update and store relevant data stream attributes, a scheduling label, a partitioning label and like data information and delay, jitter, cycle period and like performance indicators;
the stream partitioning model comprising a preprocessing unit, a learning unit and a partitioning unit;
the preprocessing unit used to normalize the labeled and unlabeled datasets and establish dependency relationships between data streams;
the learning unit used to conduct unsupervised learning, semi-supervised training and priori knowledge computation on the datasets, obtaining a relevance metric between the data streams;
the partitioning unit used to, based on the relevance, partition the data streams.

2. The knowledge inference engine system as in claim 1 , characterized in that the learning unit, through unsupervised learning and semi-supervised collaborative training on the datasets, infers conflict relationships between the TT streams, thereby obtaining a data driving-based relevance metric between the TT streams, and, from time slot occupancy, number of shared end systems and inter-stream dependency relationships, calculates a relevance metric based on priori stream attributes.

3. A method of implementing a knowledge inference engine, characterized in that the method comprises the steps of:
Step 1: inputting a dataset to be scheduled, and by a data generation module, importing information from a historical information base, according to requirements of a practical application, generating a stream characterization dataset consisting of different scheduling instances, labeling schedulable ones of the scheduling instances and equally dividing the stream characterization dataset into a plurality of partitions;
Step 2: by a stream partitioning model, upon receipt of the dataset from the data generation module, from knowledge inference, obtaining partitioning results of the dataset;
Step 3: importing the partitioning results to an offline scheduling module, storing them in the historical information base, and in an offline scheduling phase, constructing an offline scheduling constraint set for each partition, constructing an iterative scheduling algorithm and outputting an offline scheduling solution;
Step 4: from the offline scheduling solution which is output, using associated configuration software, generating configuration files for switch ports and terminal devices, configuring them at each of switches and terminal devices, and storing offline scheduling results in a scheduling solution base;
Step 5: when in the time-sensitive networking, the networking requirements experience a dynamic change, i.e., there is a new scheduling instance being input, by an online scheduling module, importing the partitioning results and the offline scheduling results from the scheduling solution base, according to the stream partitioning model, identifying the partition to which the newly-added TT stream belongs, constructing a dynamic scheduling constraint set, designing a heuristic algorithm, conducting incremental scheduling on the newly-added TT stream on a basis of a previous scheduling solution, if there is no feasible scheduling solution, identifying old TT streams belonging to the same partition as the newly-added TT stream, adjusting their scheduling solution until a feasible solution is obtained, and outputting the online scheduling solution;
Step 6: from the online scheduling solution which is output, using associated configuration software, generating configuration files for the switch ports and the terminal devices, configuring them at each of the switches and the terminal devices, storing the online scheduling results in the scheduling solution base, by the scheduling solution base, importing all scheduling result information to the historical information base and updating it.

4. The method of implementing a knowledge inference engine as in claim 3 , characterized in that, each of the scheduling instances in the dataset generated in Step 1 possesses a stream with different attributes and an inter-stream dependence graph.

5. The method of implementing a knowledge inference engine as in claim 4 , characterized in that, in a partitioning process in Step 2, the higher stream relevance between TT streams, the higher a probability of occurrence of conflicts and dependencies, and the more they tend to be divided into a same partition.

6. The method of implementing a knowledge inference engine as in claim 5 , characterized in that, a iterative scheduling algorithm in Step 3 constructs an inter-stream dependency constraint that, assuming there is one child task and one father task in a network, a TT stream of the child task has to be sent in dependence on a TT stream of the father task and an injection time of the child task has to be started only when the father task has arrived, depicting inter-stream dependency characteristics.

7. The method of implementing a knowledge inference engine as in claim 6 , characterized in that, the iterative scheduling algorithm in Step 3 constructs an inter-partition scheduling constraint that, according to an iterative scheduling order that has been assigned during the partitioning of the dataset, each partition undergoes, on a basis of partition(s) for which scheduling has been previously completed, scheduling solution generation, improving scheduling scalability.

8. The method of implementing a knowledge inference engine as in claim 7 , characterized in that, the online scheduling phase in Step 5 utilizes the offline scheduling results and the offline scheduling solution.

9. The method of implementing a knowledge inference engine as in claim 8 , characterized in that, the online scheduling solution in Step 5 employs incremental scheduling for newly-added traffic.

Assignees

• Univ Shanghai Jiaotong

Inventors

• Chen Cailian
• Xu Qimin
• Wang Shouliang
• Zhu Shanying
• Xu Lei
• Guan Xinping

Classifications

• G06F30/18 Primary

  Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling (circuit design at the physical level G06F30/39; network planning tools for wireless communication networks H04W16/18) · CPC title

• G06N5/04 Primary

  Inference or reasoning models · CPC title

• G06F2111/04

  Constraint-based CAD · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06F2111/08

  Probabilistic or stochastic CAD · CPC title