CN116798521B

CN116798521B - Abnormality monitoring method and abnormality monitoring system for immune cell culture control system

Info

Publication number: CN116798521B
Application number: CN202310884421.4A
Authority: CN
Inventors: 梅建勋; 周朋君; 唐淑艳; 陈春兰; 谭慕华
Original assignee: Guangdong Mercells Cell Biotechnology Co ltd
Current assignee: Guangdong Mercells Cell Biotechnology Co ltd
Priority date: 2023-07-19
Filing date: 2023-07-19
Publication date: 2024-02-23
Anticipated expiration: 2043-07-19
Also published as: CN116798521A

Abstract

According to the anomaly monitoring method and system for the immune cell culture control system, the migration learning of migration template culture control data corresponding to the knowledge to be learned can be performed by combining the existing training knowledge of a reference template culture control data sequence, the supervision training of the migration template culture control data sequence can be realized so as to improve the training effectiveness, the target culture control anomaly monitoring network can be generated based on the migration anomaly monitoring network for each end of migration knowledge learning and the anomaly monitoring significant value corresponding to the migration anomaly monitoring network for each end of migration knowledge learning, the obvious condition of the anomaly monitoring result generated by the migration anomaly monitoring network for each end of migration knowledge learning can be combined, the migration effect of the existing knowledge learning is improved, and the anomaly monitoring accuracy of the subsequent target culture control anomaly monitoring network is improved.

Description

Abnormality monitoring method and abnormality monitoring system for immune cell culture control system

Technical Field

The embodiment of the application relates to the technical field of artificial intelligence, in particular to an anomaly monitoring method and system of an immune cell culture control system.

Background

Immune cells refer to cells involved in or associated with an immune response. Including lymphocytes, dendritic cells, monocytes/macrophages, granulocytes, mast cells, and the like. Immune cells can be divided into a variety of classes, and various immune cells play an important role in humans. At present, various abnormal control conditions possibly exist in the process of immune cell culture control, so that the immune cell culture effect is affected, the related technology is used for carrying out subjective judgment by manually observing various parameters in the immune cell culture process or carrying out simple parameter comparison by means of a computer from monitoring of the abnormal control conditions, the accuracy of abnormal monitoring is difficult to ensure by the scheme, and when abnormal category characteristics with different priori knowledge exist, the migration effect of the prior knowledge learning is also difficult to ensure by means of an artificial intelligent model trained by the traditional scheme.

Disclosure of Invention

In order to at least overcome the above-mentioned shortcomings in the prior art, an objective of an embodiment of the present application is to provide an anomaly monitoring method and system for an immune cell culture control system.

According to an aspect of the embodiments of the present application, there is provided an abnormality monitoring method of an immune cell culture control system, including:

Obtaining a plurality of reference template culture control data sequences and a migration template culture control data sequence of the immune cell culture control system, wherein the reference template culture control data sequences and the migration template culture control data sequences comprise a plurality of template culture control data and true abnormal category confidence corresponding to the template culture control data, the reference template culture control data sequences are used for performing network training of the existing training knowledge, and the migration template culture control data sequences are used for performing migration learning of the knowledge to be learned by combining the existing training knowledge of the reference template culture control data sequences;

based on the reference template culture control data sequences, the migration template culture control data sequences and the true abnormal category confidence corresponding to the template culture control data sequences, respectively, carrying out prior knowledge learning on each initialized culture control abnormal monitoring network to generate each culture control abnormal monitoring network after prior knowledge learning;

performing migration knowledge learning on the culture control anomaly monitoring networks after the prior knowledge learning based on the migration template culture control data sequences respectively, generating culture control anomaly monitoring networks after the migration knowledge learning, and outputting migration anomaly monitoring networks for finishing the migration knowledge learning;

Determining abnormal monitoring significance values corresponding to the migration abnormal monitoring networks for ending the migration knowledge learning based on the reference template culture control data sequence and the migration template culture control data sequence;

and generating a target culture control anomaly monitoring network based on the migration anomaly monitoring network for each end of the migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for each end of the migration knowledge learning.

According to one aspect of the embodiments of the present application, there is provided an abnormality monitoring system of an immune cell culture control system including a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement an abnormality monitoring method of an immune cell culture control system in any one of the foregoing possible embodiments.

According to an aspect of embodiments of the present application, there is provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The computer instructions are read from a computer-readable storage medium by a processor of a computer device, and executed by the processor, cause the computer device to perform the methods provided in the various alternative implementations of the above aspects.

In the technical schemes provided by some embodiments of the present application, performing prior knowledge learning on each initialized culture control anomaly monitoring network through a plurality of acquired reference template culture control data sequences, a migration template culture control data sequence and true anomaly class confidence degrees corresponding to template culture control data in each template culture control data sequence, so as to generate each culture control anomaly monitoring network after prior knowledge learning; then, respectively culturing control data sequences based on the migration templates, performing migration knowledge learning on each culture control anomaly monitoring network after the prior knowledge learning, and generating migration anomaly monitoring networks for finishing the migration knowledge learning; determining an abnormal monitoring significance value corresponding to each migration abnormal monitoring network ending the migration knowledge learning based on the reference template culture control data sequence and the migration template culture control data sequence; finally, based on the migration anomaly monitoring network for each end migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for each end migration knowledge learning, a target culture control anomaly monitoring network is generated, so that the migration effect of the existing knowledge learning can be improved by combining the existing training knowledge of the reference template culture control data sequence to perform the migration learning of the migration template culture control data corresponding to the knowledge to be learned, the supervision training of the migration template culture control data sequence can be realized so as to improve the effectiveness of training, and based on the migration anomaly monitoring network for each end migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for each end migration knowledge learning, the target culture control anomaly monitoring network is generated, and the salient condition of the anomaly monitoring result generated by the migration anomaly monitoring network for each end migration knowledge learning can be combined, so that the migration effect of the existing knowledge learning is improved, and the anomaly monitoring accuracy of the subsequent target culture control anomaly monitoring network is improved.

Drawings

For a clearer description of the technical solutions of the embodiments of the present application, reference will be made to the accompanying drawings, which are needed to be activated, for the sake of simplicity, and it should be understood that the following drawings only illustrate some embodiments of the present application, and therefore should not be considered as limiting the scope, and that it is possible for a person skilled in the art to extract other relevant drawings in combination with these drawings without the inventive effort.

FIG. 1 is a schematic flow chart of an abnormality monitoring method of an immune cell culture control system according to an embodiment of the present disclosure;

fig. 2 is a schematic block diagram of an abnormality monitoring system of an immune cell culture control system for implementing the abnormality monitoring method of an immune cell culture control system according to an embodiment of the present application.

Detailed Description

The following description is presented to enable one of ordinary skill in the art to make and use the application and is provided in the context of a particular application and its requirements. It will be apparent to those having ordinary skill in the art that various changes can be made to the disclosed embodiments and that the general principles defined herein may be applied to other embodiments and applications without departing from the principles and scope of the present application. Thus, the present application is not limited to the embodiments described, but is to be accorded the widest scope consistent with the claims.

Fig. 1 is a flowchart of an abnormality monitoring method of an immune cell culture control system according to an embodiment of the present application, and the abnormality monitoring method of the immune cell culture control system is described in detail below.

Step S102, a plurality of reference template culture control data sequences and a migration template culture control data sequence are obtained, wherein the reference template culture control data sequences and the migration template culture control data sequences comprise a plurality of template culture control data and true abnormal category confidence corresponding to the template culture control data.

In an alternative embodiment, the template culture control data is template culture control data with indexes of abnormal categories, the template culture control data may refer to immune cell culture control data samples for performing network training, and specifically may include raw material introduction control data, environmental parameter (such as physical, chemical and biological environment) control data, reactor control data and the like in the immune cell culture control process, and the abnormal categories may refer to categories of abnormal conditions existing in the immune cell culture control process, such as abnormal categories of reactors, abnormal categories of environmental parameters, abnormal categories of raw material introduction and the like.

In an alternative embodiment, the anomaly class confidence may characterize the likelihood that the template culture control data belongs to a certain anomaly class.

In an alternative embodiment, the reference template culture control data sequence refers to a template culture control data sequence corresponding to a culture task similar to a culture task corresponding to the target immune cell culture control data, and the migration template culture control data sequence refers to a template culture control data sequence corresponding to a culture task corresponding to the target immune cell culture control data.

The target immune cell culture control data refers to template culture control data to be identified by a target culture control abnormality monitoring network.

In an alternative embodiment, a template culture control data sequence corresponding to a culture task corresponding to the target immune cell culture control data may be collected and output as a migration template culture control data sequence; and simultaneously collecting template culture control data sequences corresponding to a plurality of culture tasks similar to the culture tasks corresponding to the target immune cell culture control data, and outputting the template culture control data sequences as a plurality of reference template culture control data sequences. Therefore, the follow-up learning of the prior knowledge on each initialized culture control abnormality monitoring network based on the actual abnormality category confidence corresponding to each reference template culture control data sequence, the migration template culture control data sequence and the template culture control data can be facilitated.

In an alternative embodiment, after the template incubation control data sequence is acquired, the template incubation control data in the template incubation control data sequence may also be pre-processed, such as noise feature cleaning, etc.

Step S104, carrying out the prior knowledge learning on each initialized culture control anomaly monitoring network based on each reference template culture control data sequence, each migration template culture control data sequence and the corresponding real anomaly class confidence of each template culture control data respectively, and generating each culture control anomaly monitoring network after the prior knowledge learning.

The culture control abnormality monitoring network can adopt a long-period and short-period memory network.

In an alternative embodiment, each reference template culture control data sequence and each migration template culture control data sequence may be integrated respectively to generate an integrated template culture control data sequence corresponding to each reference template culture control data sequence; based on an integrated template culture control data sequence corresponding to each reference template culture control data sequence, based on an adaptive training algorithm, carrying out prior knowledge learning on each initialized culture control anomaly monitoring network, and generating basic weight information of each initialized culture control anomaly monitoring network; and carrying out back propagation optimization on each initialized culture control anomaly monitoring network based on the basic weight information of each initialized culture control anomaly monitoring network to generate each culture control anomaly monitoring network after the prior knowledge learning. Wherein, the basic weight information of each initialized culture control anomaly monitoring network represents a known preliminary migration of knowledge to be learned. Therefore, the supervision training of the migration template culture control data sequence can be realized by carrying out the prior knowledge learning on each initialized culture control anomaly monitoring network through the real anomaly category confidence degrees corresponding to each reference template culture control data sequence, the migration template culture control data sequence and the template culture control data, so that the training effectiveness is improved.

In an alternative embodiment, the initialized culture control anomaly monitoring networks can be individually trained based on the integrated template culture control data sequences corresponding to the reference template culture control data sequences to generate basic weight information of the initialized culture control anomaly monitoring networks; if the integrated template culture control data sequences corresponding to the 5 reference template culture control data sequences are available, the basic weight information of the 5 groups of initialized culture control abnormal monitoring networks can be learned; and based on the basic weight information of each initialized culture control abnormality monitoring network, carrying out network weight information learning on each initialized culture control abnormality monitoring network to generate each culture control abnormality monitoring network after the prior knowledge learning.

In an alternative embodiment, a plurality of reference template incubation control data sequences and one migration template incubation control data sequence may be collected, and template incubation control data sequences and verification template incubation control data sequences and cross-time domain template incubation control data sequences of the reference template incubation control data sequences and the migration template incubation control data sequences and cross-time domain template incubation control data sequences are constructed based on the respective reference template incubation control data sequences and migration template incubation control data sequences; identifying a template incubation control data sequence of the reference template incubation control data sequence as a reference template incubation control data sequence, and identifying a template incubation control data sequence of the migration template incubation control data sequence as a migration template incubation control data sequence; the method comprises the steps of verifying a template culture control data sequence and a cross-time domain template culture control data sequence, wherein the verification template culture control data sequence and the cross-time domain template culture control data sequence are used for verifying network performance, the time domain of template culture control data of the verification template culture control data sequence coincides with the time domain of template culture control data of the template culture control data sequence, and the time domain of template culture control data of the cross-time domain template culture control data sequence is behind the time domain of template culture control data of the template culture control data sequence; the cross-time domain template culture control data sequence can better verify the stability of the culture control abnormality monitoring network along with the time domain. In an alternative embodiment, a small part of template culture control data can be extracted from the reference template culture control data sequence and the migration template culture control data sequence respectively, and network importance numerical parameter optimization is performed on the initialized culture control abnormal monitoring network based on the small part of template culture control data of the reference template culture control data sequence and the small part of template culture control data of the migration template culture control data sequence respectively, so as to generate an abnormal monitoring error value of the initialized culture control abnormal monitoring network on the small part of template culture control data of the reference template culture control data sequence and an abnormal monitoring error value of the initialized culture control abnormal monitoring network on the small part of template culture control data of the migration template culture control data sequence; based on the abnormal monitoring error value of the initialized culture control abnormal monitoring network on the small part of template culture control data of the reference template culture control data sequence and the abnormal monitoring error value of the small part of template culture control data of the migration template culture control data sequence, iteratively updating the initialized culture control abnormal monitoring network for set times to generate a culture control abnormal monitoring network corresponding to the existing training knowledge and a culture control abnormal monitoring network corresponding to the knowledge to be learned; extracting a small part of template culture control data from the reference template culture control data sequence again, determining an abnormality monitoring error value on a culture control abnormality monitoring network corresponding to the learned existing training knowledge, and outputting the abnormality monitoring error value as a first training cost value; extracting a small part of template culture control data from the migration template culture control data sequence again, determining an abnormality monitoring error value on a culture control abnormality monitoring network corresponding to training knowledge to be learned, and outputting the abnormality monitoring error value as a second training cost value; optimizing the initialized culture control anomaly monitoring network based on the sum of the first training cost value and the second training cost value, and circulating the operations to perform migration knowledge learning on the optimized culture control anomaly monitoring network until the culture control anomaly monitoring network meets the training termination condition, thereby generating each culture control anomaly monitoring network after the prior knowledge learning. Thus, the shared network weight information transferred to the knowledge to be learned can be learned, and the culture control anomaly monitoring network with the strengthening capability on the knowledge to be learned can be obtained for a plurality of times.

And step S106, respectively carrying out migration knowledge learning on each culture control abnormal monitoring network after the prior knowledge learning based on the migration template culture control data sequence, generating each culture control abnormal monitoring network after the migration knowledge learning, and outputting the migration abnormal monitoring network for each migration knowledge learning.

The culture control abnormal monitoring network after the knowledge is transferred and learned is a single network generated by optimizing the template culture control data of abnormal categories according to a small part of the culture control data sequence of the transfer template, and specifically corresponds to a final transfer network which is known to be aware of the knowledge to be learned, so that the transferred template culture control data can be independently monitored for the abnormal conditions. The culture control abnormal monitoring network after the transfer knowledge learning is called as a transfer abnormal monitoring network for ending the transfer knowledge learning; the migration anomaly monitoring network can perform anomaly monitoring on the migrated template culture control data.

In an alternative embodiment, the training control abnormal monitoring network after the existing knowledge learning can be subjected to the training knowledge learning based on the training control data sequence of the migration template, so as to generate the training control abnormal monitoring network after the training knowledge learning; acquiring a training cost value between an abnormal category confidence coefficient generated by the training control abnormal monitoring network after the transfer knowledge learning and a corresponding real abnormal category confidence coefficient, and performing back propagation optimization on the training control abnormal monitoring network after the transfer knowledge learning based on the training cost value; and the like, the culture control abnormal monitoring network after the transfer knowledge learning can be subjected to multi-round optimization until the culture control abnormal monitoring network after the transfer knowledge learning meets the training termination condition; outputting the current culture control abnormality monitoring network as a migration abnormality monitoring network for ending the migration knowledge learning. Therefore, according to the same migration template culture control data sequence, each migration anomaly monitoring network ending the migration knowledge learning can be obtained, the purpose of optimizing the culture control anomaly monitoring network after the existing knowledge learning is achieved, so that the migration anomaly monitoring network is obtained, the anomaly monitoring performance of the migration anomaly monitoring network on the migration template culture control data is improved, and the accuracy of anomaly class output is improved.

In an alternative implementation manner, the culture control anomaly monitoring networks after the prior knowledge learning can be optimized according to the template culture control data with the anomaly class carried by a small part of the same migration template culture control data sequence, so as to generate a plurality of independent culture control anomaly monitoring networks, and output the plurality of independent culture control anomaly monitoring networks as migration anomaly monitoring networks for ending the migration knowledge learning; each migration anomaly monitoring network ending the migration knowledge learning corresponds to a final migration network known to know the knowledge to be learned.

In an alternative embodiment, the method can also perform migration knowledge learning on the culture control anomaly monitoring network after the existing knowledge learning based on the template culture control data of the migration template culture control data sequence to generate a culture control anomaly monitoring network after the migration knowledge learning; acquiring a global training cost value between an abnormal category confidence coefficient generated by a culture control abnormal monitoring network after the transfer knowledge learning and a corresponding real abnormal category confidence coefficient; and updating the weight information of the culture control anomaly monitoring network after the migration knowledge learning based on the global training cost value, performing iterative optimization on the culture control anomaly monitoring network after the weight information is updated, and outputting the current culture control anomaly monitoring network as the migration anomaly monitoring network for ending the migration knowledge learning when the training termination requirement is met. Thus, according to the same migration template culture control data sequence, each migration anomaly monitoring network ending the migration knowledge learning is obtained.

And S108, determining an abnormality monitoring significance value corresponding to each migration abnormality monitoring network ending the migration knowledge learning based on the reference template culture control data sequence and the migration template culture control data sequence.

The abnormal monitoring significance value corresponding to the migration abnormal monitoring network ending the migration knowledge learning refers to an importance value corresponding to the abnormal category confidence level generated by the migration abnormal monitoring network, and is used for representing the weight of the abnormal category confidence level generated by the migration abnormal monitoring network ending the migration knowledge learning. For example, the greater the anomaly monitoring significance value corresponding to the migration anomaly monitoring network ending the migration knowledge learning, the higher the weight representing the anomaly class confidence level generated by the migration anomaly monitoring network, the greater the importance value representing the anomaly class confidence level generated by the migration anomaly monitoring network; the smaller the corresponding abnormal monitoring significance value of the migration abnormal monitoring network which finishes the migration knowledge learning is, the lower the weight of the abnormal category confidence coefficient generated by the migration abnormal monitoring network is, and the smaller the importance value of the abnormal category confidence coefficient generated by the migration abnormal monitoring network is.

The abnormal monitoring significance value corresponding to each migration abnormal monitoring network ending the migration knowledge learning is determined through the correlation between a reference template culture control data sequence and a migration template culture control data sequence of the migration abnormal monitoring network; for example, the greater the correlation between the reference template incubation control data sequence and the migration template incubation control data sequence of the training migration anomaly monitoring network, which indicates that in the process of training the migration anomaly monitoring network according to the reference template incubation control data sequence, the greater the anomaly monitoring significance value corresponding to the migration anomaly monitoring network is from the known knowledge to be learned to correspond to forward adaptation; the smaller the correlation between the reference template culture control data sequence and the migration template culture control data sequence of the migration anomaly monitoring network is, the smaller the anomaly monitoring significance value corresponding to the migration anomaly monitoring network is required to be.

In an alternative embodiment, the reference template culture control data sequence and the migration template culture control data sequence corresponding to the migration anomaly monitoring network ending the migration knowledge learning can be determined from the reference template culture control data sequence and the migration template culture control data sequence; based on a preset relevance determining network, generating relevance between a reference template culture control data sequence and a migration template culture control data sequence corresponding to each migration anomaly monitoring network ending the migration knowledge learning; searching mapping information of preset relevance and abnormal monitoring significance values based on relevance between a reference template culture control data sequence and a migration template culture control data sequence corresponding to each migration abnormal monitoring network ending the migration knowledge learning, and generating abnormal monitoring significance values corresponding to each migration abnormal monitoring network ending the migration knowledge learning. Wherein the predetermined correlation determination network is a neural network that can determine a correlation between two template culture control data sequences. Therefore, the abnormal monitoring significance value corresponding to the migration abnormal monitoring network for ending the migration knowledge learning is determined, and the weight of the abnormal category confidence coefficient generated by the migration abnormal monitoring network for ending the migration knowledge learning can be combined, so that the forward self-adaption of the existing knowledge is ensured.

Step S110, generating a target culture control anomaly monitoring network based on the migration anomaly monitoring network for each end of the migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for each end of the migration knowledge learning.

The target culture control anomaly monitoring network is a final neural network integrated by each migration anomaly monitoring network ending the migration knowledge learning, and can perform anomaly monitoring on the migrated template culture control data, so that the anomaly category of the target culture control data is determined.

In an alternative embodiment, the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for which the migration knowledge learning is finished may be output as the importance value corresponding to the anomaly class confidence level generated by the migration anomaly monitoring network for which the migration knowledge learning is finished; and integrating the migration anomaly monitoring networks ending the migration knowledge learning to generate an integrated monitoring network, and outputting the integrated monitoring network as a target culture control anomaly monitoring network.

In an alternative embodiment, the target immune cell culture control data transferred can be output through the target immune cell culture control anomaly monitoring network, and the anomaly class confidence of the transferred target immune cell culture control data in the transferred anomaly monitoring network ending the transferred knowledge learning is obtained; based on the abnormal monitoring significance values corresponding to the migration abnormal monitoring networks of all the end migration knowledge learning, fusing the abnormal category confidence coefficient of the target immune cell culture control data in the migration abnormal monitoring networks of all the end migration knowledge learning, and generating fusion values of the abnormal monitoring significance values corresponding to the migration abnormal monitoring networks of all the end migration knowledge learning and the abnormal category confidence coefficient; adding the abnormal monitoring significant value corresponding to each migration abnormal monitoring network ending the migration knowledge learning and the fusion value of the abnormal class confidence coefficient to generate the final abnormal class confidence coefficient of the target immune cell culture control data; and determining an abnormality monitoring result of the target immune cell culture control data based on the final abnormality category confidence of the target immune cell culture control data.

Based on the steps, the migration learning of the migration template culture control data corresponding to the knowledge to be learned can be performed by combining the existing training knowledge of the reference template culture control data sequence, the supervision training of the migration template culture control data sequence can be realized so as to improve the training effectiveness, the target culture control anomaly monitoring network is generated based on the migration anomaly monitoring network for ending the migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for ending the migration knowledge learning, the obvious condition of the anomaly monitoring result generated by the migration anomaly monitoring network for ending the migration knowledge learning can be combined, the migration effect of the existing knowledge learning is improved, and the anomaly monitoring accuracy of the subsequent target culture control anomaly monitoring network is improved.

In an alternative embodiment, step S104, respectively, performs the prior knowledge learning on each initialized anomaly monitoring network based on each reference template culture control data sequence, each migration template culture control data sequence, and the true anomaly class confidence corresponding to each template culture control data, so as to generate each anomaly monitoring network after prior knowledge learning, which is implemented by the following embodiments:

Step S202, based on the respective reference template incubation control data sequences, the migration template incubation control data sequences and the true anomaly class confidence corresponding to the respective template incubation control data, constructing a first template incubation control data sequence and a second template incubation control data sequence of the reference template incubation control data sequences and a first template incubation control data sequence and a second template incubation control data sequence of the migration template incubation control data sequences.

The first template culture control data sequence refers to a part of template culture control data extracted from a template culture control data sequence (such as a reference template culture control data sequence or a migration template culture control data sequence) and is used for training a culture control abnormality monitoring network corresponding to the existing training knowledge and a culture control abnormality monitoring network corresponding to the knowledge to be learned; the second template incubation control data sequence refers to a part of template incubation control data re-extracted from remaining template incubation control data of a template incubation control data sequence (such as a reference template incubation control data sequence or a migration template incubation control data sequence) for use as a template incubation control data sequence for determining an anomaly monitoring error value of an incubation control anomaly monitoring network corresponding to existing training knowledge and an anomaly monitoring error value of an incubation control anomaly monitoring network corresponding to knowledge to be learned.

The first template culture control data sequence and the second template culture control data sequence of the reference template culture control data sequence and the first template culture control data sequence and the second template culture control data sequence of the migration template culture control data sequence jointly form a network learning example of the culture control abnormality monitoring network, and the network learning example is equivalent to one sample learning data of the culture control abnormality monitoring network.

In an alternative embodiment, a batch of template incubation control data may be extracted from each reference template incubation control data sequence separately, constituting a first template incubation control data sequence of the reference template incubation control data sequence; re-extracting a batch of template culture control data from the rest template culture control data of each reference template culture control data sequence to form a second template culture control data sequence of the reference template culture control data sequence, thereby obtaining a first template culture control data sequence and a second template culture control data sequence of the reference template culture control data sequence; extracting a batch of template culture control data from the migration template culture control data sequence to form a first template culture control data sequence of the migration template culture control data sequence; and re-extracting a batch of template culture control data from the rest template culture control data of the migration template culture control data sequence to form a second template culture control data sequence of the migration template culture control data sequence, thereby obtaining a first template culture control data sequence and a second template culture control data sequence of the migration template culture control data sequence, and being convenient for learning the capability of distinguishing the template culture control data in the second template culture control data sequence from the first template culture control data sequence when training a culture control abnormality monitoring network.

Step S204, based on the first template culture control data sequence of the reference template culture control data sequence and the first template culture control data sequence of the migration template culture control data sequence, network importance numerical parameter optimization is performed on the initialized culture control abnormality monitoring network, and a culture control abnormality monitoring network after network importance numerical parameter optimization is generated.

The culture control anomaly monitoring network after the network importance numerical parameter optimization refers to a culture control anomaly monitoring network corresponding to the existing training knowledge and a culture control anomaly monitoring network corresponding to the knowledge to be learned.

In an alternative embodiment, each template culture control data in the first template culture control data sequence of the reference template culture control data sequence may be loaded to the initialized culture control anomaly monitoring network respectively, and a global anomaly monitoring error value between an anomaly class confidence coefficient generated by the initialized culture control anomaly monitoring network and a corresponding true anomaly class confidence coefficient is generated; performing preliminary optimization on weight information in the initialized culture control anomaly monitoring network based on the global anomaly monitoring error value, generating an optimized culture control anomaly monitoring network, and outputting the optimized culture control anomaly monitoring network as a culture control anomaly monitoring network corresponding to the existing training knowledge; loading each template culture control data in a first template culture control data sequence of the migration template culture control data sequence to an initialized culture control anomaly monitoring network respectively, and generating a global anomaly monitoring error value between an anomaly category confidence coefficient generated by the initialized culture control anomaly monitoring network and a corresponding real anomaly category confidence coefficient; the weight information in the initialized culture control abnormal monitoring network is updated temporarily based on the global abnormal monitoring error value, an updated culture control abnormal monitoring network is generated, and the updated culture control abnormal monitoring network is output as a culture control abnormal monitoring network corresponding to knowledge to be learned; and taking the culture control abnormality monitoring network corresponding to the existing training knowledge and the culture control abnormality monitoring network corresponding to the knowledge to be learned as the culture control abnormality monitoring network after the network importance numerical parameter is optimized.

Step S206, obtaining abnormal monitoring error values of the culture control abnormal monitoring network after network importance numerical parameter optimization on a second template culture control data sequence of the reference template culture control data sequence and a second template culture control data sequence of the migration template culture control data sequence, updating the initialized culture control abnormal monitoring network based on the abnormal monitoring error values, and generating an updated culture control abnormal monitoring network.

In an alternative embodiment, each template culture control data in the second template culture control data sequence of the reference template culture control data sequence may be respectively input into a culture control anomaly monitoring network corresponding to the existing training knowledge, and a first global training cost value between an anomaly class confidence coefficient generated by the culture control anomaly monitoring network corresponding to the existing training knowledge and a corresponding real anomaly class confidence coefficient is generated; respectively inputting each template culture control data in a second template culture control data sequence of the transfer template culture control data sequence into a culture control anomaly monitoring network corresponding to the knowledge to be learned, and generating a second global training cost value between the anomaly class confidence coefficient generated by the culture control anomaly monitoring network corresponding to the knowledge to be learned and the corresponding real anomaly class confidence coefficient; adding the first global training cost value and the second global training cost value to generate a target global training cost value; and carrying out back propagation optimization on the initialized culture control anomaly monitoring network based on the target global training cost value, generating an optimized culture control anomaly monitoring network, and outputting the optimized culture control anomaly monitoring network as an updated culture control anomaly monitoring network.

Step S208, if the updated culture control anomaly monitoring network does not meet the first training termination requirement, using the updated culture control anomaly monitoring network as an initialized culture control anomaly monitoring network, and returning to execute the first template culture control data sequence based on the reference template culture control data sequence and the first template culture control data sequence of the migration template culture control data sequence, performing network importance numerical parameter optimization on the initialized culture control anomaly monitoring network, and generating the culture control anomaly monitoring network after network importance numerical parameter optimization.

Step S210, if the updated culture control abnormality monitoring network meets the first training termination requirement, outputting the current culture control abnormality monitoring network as the culture control abnormality monitoring network after learning of the prior knowledge.

Wherein the updated abnormal culture control monitoring network meets the first training termination requirement, which means that the iteration number of the updated abnormal culture control monitoring network is greater than the set number or the weight information of the updated abnormal culture control monitoring network is not changed any more.

If the updated culture control abnormality monitoring network does not meet the first training termination requirement, outputting the current culture control abnormality monitoring networks as initialized culture control abnormality monitoring networks, and returning to the execution step S204, and repeatedly executing the step S204 until the updated culture control abnormality monitoring networks do not meet the first training termination requirement until the updated culture control abnormality monitoring networks meet the first training termination requirement.

Therefore, based on the actual anomaly category confidence corresponding to each reference template culture control data sequence, the migration template culture control data sequence and the template culture control data, the prior knowledge learning is carried out on each initialized culture control anomaly monitoring network, and the anomaly monitoring knowledge of a plurality of networks which are primarily trained according to the reference template culture control data sequences can be effectively and adaptively learned.

In an alternative embodiment, step S204 is implemented by performing network importance numerical parameter optimization on the initialized culture control anomaly monitoring network based on the first template culture control data sequence of the reference template culture control data sequence and the first template culture control data sequence of the migration template culture control data sequence to generate a culture control anomaly monitoring network after network importance numerical parameter optimization, by the following examples:

step S302, updating the initialized culture control anomaly monitoring network based on the first template culture control data sequence of the reference template culture control data sequence, and generating a first anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the reference template culture control data sequence.

The first abnormal monitoring error value refers to the sum of abnormal monitoring error values of each template culture control data in a first template culture control data sequence of the reference template culture control data sequence by the initialized culture control abnormal monitoring network.

In an alternative embodiment, template culture control data in a first template culture control data sequence of the reference template culture control data sequence can be loaded into an initialized anomaly monitoring network to generate an anomaly class confidence level generated by the initialized culture control anomaly monitoring network; generating an abnormal monitoring error value of the template culture control data based on the initialized abnormal category confidence coefficient generated by the culture control abnormal monitoring network and the real abnormal category confidence coefficient corresponding to the template culture control data and the cross entropy loss function; combining the steps, obtaining an abnormal monitoring error value of the initialized culture control abnormal monitoring network for each template culture control data in a first template culture control data sequence of a reference template culture control data sequence; and adding the initialized culture control anomaly monitoring network to the anomaly monitoring error values of the template culture control data in the first template culture control data sequence of the reference template culture control data sequence to generate a global anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the reference template culture control data sequence, and outputting the global anomaly monitoring error value as the first anomaly monitoring error value.

Step S304, updating the initialized culture control anomaly monitoring network based on the first template culture control data sequence of the migration template culture control data sequence, and generating a second anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the migration template culture control data sequence.

The second abnormal monitoring error value refers to the sum of abnormal monitoring error values of the initialized culture control abnormal monitoring network to each template culture control data in the first template culture control data sequence of the migration template culture control data sequence.

In an alternative embodiment, template culture control data in a first template culture control data sequence of the migration template culture control data sequence can be loaded into an initialized culture control anomaly monitoring network to generate an anomaly class confidence coefficient generated by the initialized culture control anomaly monitoring network; based on the initialized abnormal class confidence coefficient generated by the culture control abnormal monitoring network and the real abnormal class confidence coefficient corresponding to the template culture control data, generating an abnormal monitoring error value of the template culture control data based on a set loss function, and obtaining the abnormal monitoring error value of the initialized culture control abnormal monitoring network for each template culture control data in a first template culture control data sequence of a migration template culture control data sequence based on the steps; and adding the initialized culture control anomaly monitoring network to the anomaly monitoring error value of each template culture control data in the first template culture control data sequence of the migration template culture control data sequence to generate a global anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the migration template culture control data sequence, and outputting the global anomaly monitoring error value as a second anomaly monitoring error value.

And step S306, carrying out back propagation optimization on the initialized culture control anomaly monitoring network based on the first anomaly monitoring error value, and generating a culture control anomaly monitoring network corresponding to the existing training knowledge.

In an alternative embodiment, the weight information optimization gradient value of the initialized culture control anomaly monitoring network may be determined based on the first anomaly monitoring error value; and optimizing the gradient value based on the weight information of the initialized culture control anomaly monitoring network, performing back propagation optimization on the initialized culture control anomaly monitoring network, generating an optimized culture control anomaly monitoring network, and outputting the optimized culture control anomaly monitoring network as a culture control anomaly monitoring network corresponding to the existing training knowledge.

And step S308, carrying out back propagation optimization on the initialized culture control anomaly monitoring network based on the second anomaly monitoring error value, and generating a culture control anomaly monitoring network corresponding to the knowledge to be learned.

In an alternative embodiment, the weight information optimization gradient value of the initialized culture control anomaly monitoring network may be determined based on the second anomaly monitoring error value; and optimizing the gradient value based on the weight information of the initialized culture control anomaly monitoring network, performing back propagation optimization on the initialized culture control anomaly monitoring network, generating an optimized culture control anomaly monitoring network, and outputting the optimized culture control anomaly monitoring network as a culture control anomaly monitoring network corresponding to the knowledge to be learned.

In an alternative embodiment, the culture control anomaly monitoring network corresponding to the training knowledge and the culture control anomaly monitoring network corresponding to the knowledge to be learned are both culture control anomaly monitoring networks after network importance numerical parameter optimization.

Therefore, the initialized culture control abnormality monitoring network is updated through the first template culture control data sequence of the reference template culture control data sequence and the first template culture control data sequence of the migration template culture control data sequence, and feature vectors of the reference template culture control data sequence and the migration template culture control data sequence can be respectively learned, so that migration from the existing knowledge to the knowledge to be learned is realized.

In an alternative embodiment, in step S206, an abnormal monitoring error value of the culture control abnormal monitoring network after optimizing the network importance numerical parameter is obtained on the second template culture control data sequence of the reference template culture control data sequence and the second template culture control data sequence of the migration template culture control data sequence, and the initialized culture control abnormal monitoring network is updated based on the abnormal monitoring error value, so as to generate an updated culture control abnormal monitoring network, which is implemented by the following embodiments:

Step S402, updating the culture control anomaly monitoring network corresponding to the existing training knowledge based on the second template culture control data sequence of the reference template culture control data sequence, and generating a third anomaly monitoring error value of the culture control anomaly monitoring network corresponding to the existing training knowledge on the second template culture control data sequence of the reference template culture control data sequence.

The third abnormal monitoring error value refers to the sum of abnormal monitoring error values of the template culture control data in the second template culture control data sequence of the reference template culture control data sequence corresponding to the existing training knowledge by the culture control abnormal monitoring network.

In an alternative embodiment, template culture control data in a second template culture control data sequence of the reference template culture control data sequence can be input into a culture control anomaly monitoring network corresponding to the existing training knowledge, so that an anomaly class confidence coefficient generated by the culture control anomaly monitoring network corresponding to the existing training knowledge is generated; based on the abnormal category confidence coefficient generated by the culture control abnormal monitoring network corresponding to the existing training knowledge and the real abnormal category confidence coefficient corresponding to the template culture control data, generating an abnormal monitoring error value of the template culture control data based on a set loss function; based on the steps, abnormal monitoring error values of the culture control data of each template in the second template culture control data sequence of the reference template culture control data sequence by the culture control abnormal monitoring network corresponding to the existing training knowledge can be obtained; adding the abnormal monitoring error values of the template culture control data in the second template culture control data sequence of the reference template culture control data sequence by the culture control abnormal monitoring network corresponding to the existing training knowledge, generating a global abnormal monitoring error value of the culture control abnormal monitoring network corresponding to the existing training knowledge on the second template culture control data sequence of the reference template culture control data sequence, and outputting the global abnormal monitoring error value as a third abnormal monitoring error value.

And step S404, updating the culture control abnormality monitoring network corresponding to the knowledge to be learned based on the second template culture control data sequence of the transfer template culture control data sequence, and generating a fourth abnormality monitoring error value of the culture control abnormality monitoring network corresponding to the knowledge to be learned on the second template culture control data sequence of the transfer template culture control data sequence.

The fourth abnormal monitoring error value refers to the sum of abnormal monitoring error values of the culture control data of each template in the second template culture control data sequence of the migration template culture control data sequence by the culture control abnormal monitoring network corresponding to the knowledge to be learned.

In an alternative embodiment, template culture control data in a second template culture control data sequence of the migration template culture control data sequence can be input into a culture control anomaly monitoring network corresponding to the knowledge to be learned, and an anomaly category confidence coefficient generated by the culture control anomaly monitoring network corresponding to the knowledge to be learned is generated; based on the abnormal category confidence coefficient generated by the culture control abnormal monitoring network corresponding to the knowledge to be learned and the real abnormal category confidence coefficient corresponding to the template culture control data, generating an abnormal monitoring error value of the template culture control data based on a set loss function; based on the steps, abnormal monitoring error values of the culture control data of each template in the second template culture control data sequence of the transfer template culture control data sequence by the culture control abnormal monitoring network corresponding to the knowledge to be learned can be obtained; adding the abnormal monitoring error values of the template culture control data in the second template culture control data sequence of the migration template culture control data sequence by the culture control abnormal monitoring network corresponding to the knowledge to be learned, generating a global abnormal monitoring error value of the culture control abnormal monitoring network corresponding to the knowledge to be learned on the second template culture control data sequence of the migration template culture control data sequence, and outputting the global abnormal monitoring error value as a fourth abnormal monitoring error value.

Step S406, based on the third abnormal monitoring error value and the fourth abnormal monitoring error value, performing back propagation optimization on the initialized culture control abnormal monitoring network, generating an optimized culture control abnormal monitoring network, and outputting the optimized culture control abnormal monitoring network as an updated culture control abnormal monitoring network.

In an alternative embodiment, the third anomaly monitoring error value and the fourth anomaly monitoring error value may be added to generate a target anomaly monitoring error value; determining a weight information optimization gradient value of an initialized culture control anomaly monitoring network based on the target anomaly monitoring error value; and optimizing the gradient value based on the weight information of the initialized culture control anomaly monitoring network, performing back propagation optimization on the initialized culture control anomaly monitoring network, generating an optimized culture control anomaly monitoring network, and outputting the optimized culture control anomaly monitoring network as an updated culture control anomaly monitoring network.

In this embodiment, the updated culture control anomaly monitoring network is subjected to migration knowledge learning based on the second template culture control data sequence of the reference template culture control data sequence and the second template culture control data sequence of the migration template culture control data sequence, so that migration of the existing knowledge to the knowledge to be learned is facilitated, and anomaly monitoring performance of the culture control anomaly monitoring network on the migrated template culture control data is improved.

In an alternative embodiment, the step S106 is implemented by performing the migration knowledge learning on the culture control anomaly monitoring network after the learning of each existing knowledge based on the migration template culture control data sequence, and generating the culture control anomaly monitoring network after the learning of each migration knowledge, respectively, by the following examples:

step S502, obtaining a culture control knowledge vector of template culture control data in the migration template culture control data sequence.

In an alternative embodiment, template culture control data in the sequence of migration template culture control data may be input to a previously configured encoder to generate a culture control knowledge vector of template culture control data in the sequence of migration template culture control data.

Step S504, respectively inputting the culture control knowledge vectors of the template culture control data in the transfer template culture control data sequence into the culture control anomaly monitoring networks after the prior knowledge learning, and generating the anomaly category confidence of the template culture control data in the culture control anomaly monitoring networks after the prior knowledge learning.

In an alternative embodiment, the culture control knowledge vector of the template culture control data in the migration template culture control data sequence may be input into each culture control anomaly monitoring network after the prior knowledge learning, and the culture control knowledge vector of the template culture control data is encoded by an encoder in the culture control anomaly monitoring network after the prior knowledge learning to generate a target culture control knowledge vector of the template culture control data; performing full connection processing on the target culture control knowledge vector of the template culture control data through a full connection unit in the culture control anomaly monitoring network after the prior knowledge learning to generate anomaly class confidence of the template culture control data in the culture control anomaly monitoring network after the prior knowledge learning; based on the steps, the abnormal category confidence of the template culture control data in the culture control abnormal monitoring network after the learning of all the prior knowledge can be obtained.

Step S506, obtaining the subtraction value of the abnormal category confidence coefficient and the corresponding real abnormal category confidence coefficient of the template culture control data in the culture control abnormal monitoring network after the learning of all the prior knowledge.

The subtraction value of the abnormal category confidence coefficient of the template culture control data in the culture control abnormal monitoring network after the prior knowledge learning and the corresponding real abnormal category confidence coefficient can characterize the deviation degree between the abnormal category confidence coefficient of the template culture control data in the culture control abnormal monitoring network after the prior knowledge learning and the corresponding real abnormal category confidence coefficient.

And step S508, determining the training cost value of the culture control anomaly monitoring network after the learning of each existing knowledge based on the subtraction value.

In an alternative embodiment, the subtracted value may be identified as the training cost value of the culture control anomaly monitoring network after the prior knowledge learning, thereby obtaining the training cost value of each of the culture control anomaly monitoring networks after the prior knowledge learning.

Step S510, reversely training the culture control abnormality monitoring networks after the prior knowledge learning based on the training cost values respectively until the culture control abnormality monitoring networks after the prior knowledge learning meet the second training termination requirement.

The second training termination requirement means that the iteration number of the culture control abnormal monitoring network after the prior knowledge learning is greater than the set number, or the weight information of the culture control abnormal monitoring network after the prior knowledge learning is not changed any more.

In an alternative embodiment, a weight information optimization gradient value can be determined based on the training cost value, and the counter-propagation optimization is performed on the culture control anomaly monitoring network after the prior knowledge learning based on the weight information optimization gradient value, so as to generate an optimized culture control anomaly monitoring network; and (3) taking the optimized culture control abnormality monitoring network as the culture control abnormality monitoring network after the prior knowledge learning, and repeatedly executing the steps S504 to S510 until the weight information of the culture control abnormality monitoring network after the prior knowledge learning is not changed.

Step S512, if the culture control anomaly monitoring networks after the prior knowledge learning meet the second training termination requirement, outputting the current culture control anomaly monitoring network after the prior knowledge learning as the culture control anomaly monitoring network after the prior knowledge learning.

Therefore, through transferring the template culture control data sequence, the transfer knowledge learning is carried out on each culture control abnormal monitoring network after the knowledge learning, the purpose of optimizing the culture control abnormal monitoring network after the knowledge learning is achieved, the abnormal monitoring performance of the culture control abnormal monitoring network on the transferred template culture control data is improved, and the abnormal monitoring precision of the target culture control abnormal monitoring network is improved.

In an alternative embodiment, the step S108 is implemented by determining, based on the reference template culture control data sequence and the migration template culture control data sequence, an anomaly monitoring significance value corresponding to the migration anomaly monitoring network for ending the migration knowledge learning, where the anomaly monitoring significance value is implemented by the following examples:

step S602, extracting a reference template culture control data sequence and a migration template culture control data sequence corresponding to the migration anomaly monitoring network ending the migration knowledge learning from the reference template culture control data sequence and the migration template culture control data sequence.

The reference template culture control data sequence and the migration template culture control data sequence corresponding to each migration anomaly monitoring network ending the migration knowledge learning refer to template culture control data sequences for updating each initialized culture control anomaly monitoring network; each migration anomaly monitoring network ending the migration knowledge learning corresponds to a reference template culture control data sequence and a migration template culture control data sequence.

In an alternative embodiment, the network ID of the migration anomaly monitoring network for each end of the migration knowledge learning may be obtained, mapping information of a preset network ID and a reference template culture control data sequence and a migration template culture control data sequence is searched based on the network ID of the migration anomaly monitoring network for each end of the migration knowledge learning, a reference template culture control data sequence and a migration template culture control data sequence corresponding to the network ID of the migration anomaly monitoring network for each end of the migration knowledge learning are generated, and the reference template culture control data sequence and the migration template culture control data sequence of the migration anomaly monitoring network for each end of the migration knowledge learning are output.

And step S604, counting vector cost values between the reference template culture control data sequence and the migration template culture control data sequence corresponding to the migration anomaly monitoring network for ending the migration knowledge learning.

The vector cost value can represent the relevance between the reference template culture control data sequence and the migration template culture control data sequence, can be expressed through characteristic distances, such as Euclidean distance, euclidean distance and the like, and generally, the larger the vector cost value is, the more uncorrelated the reference template culture control data sequence and the migration template culture control data sequence is; the smaller the vector cost value, the more relevant the reference template culture control data sequence and the migration template culture control data sequence are.

In an alternative embodiment, the reference template culture control data sequence and the migration template culture control data sequence of each migration anomaly monitoring network ending the migration knowledge learning can be input into the vector cost determining network, and the vector cost value between the reference template culture control data sequence and the migration template culture control data sequence of each migration anomaly monitoring network ending the migration knowledge learning can be generated; the vector cost determining network is used for analyzing and processing the reference template culture control data sequence and the transfer template culture control data sequence of the transfer anomaly monitoring network for finishing the transfer knowledge learning, and generating vector cost values between the reference template culture control data sequence and the transfer template culture control data sequence of the transfer anomaly monitoring network for finishing the transfer knowledge learning.

In an alternative embodiment, the reference template culture control data sequence and the migration template culture control data sequence of the migration anomaly monitoring network for ending the migration knowledge learning can be further processed based on a preset vector cost determining instruction, so that vector cost values between the reference template culture control data sequence and the migration template culture control data sequence of the migration anomaly monitoring network for ending the migration knowledge learning are generated.

Step S606, searching mapping information of preset vector cost value and importance value based on vector cost value between the reference template culture control data sequence and the migration template culture control data sequence of each migration anomaly monitoring network ending the migration knowledge learning, and determining the importance value corresponding to each migration anomaly monitoring network ending the migration knowledge learning.

The larger the vector cost value between the reference template culture control data sequence and the migration template culture control data sequence of the migration anomaly monitoring network which finishes the migration knowledge learning is, the less correlation between the reference template culture control data sequence and the migration template culture control data sequence is shown, the worse migration effect of the reference template culture control data sequence on the migration template culture control data sequence is shown, the worse significance of the migration anomaly monitoring network obtained through the reference template culture control data sequence and the migration template culture control data sequence is shown, and the smaller the importance value corresponding to the migration anomaly monitoring network is shown; the smaller the vector cost value between the reference template culture control data sequence and the migration template culture control data sequence of the migration anomaly monitoring network which finishes the migration knowledge learning, the more relevant the reference template culture control data sequence and the migration template culture control data sequence are, which indicates that the migration effect of the reference template culture control data sequence on the migration template culture control data sequence is better, the better the significance of the migration anomaly monitoring network obtained through the reference template culture control data sequence and the migration template culture control data sequence is, and the larger the significance value corresponding to the migration anomaly monitoring network is.

In an alternative embodiment, mapping information of a preset vector cost value and an importance value can be obtained, the mapping information of the preset vector cost value and the importance value can be searched based on the vector cost value between a reference template culture control data sequence and a transfer template culture control data sequence of the transfer anomaly monitoring network for ending the transfer knowledge learning, the importance value corresponding to the vector cost value between the reference template culture control data sequence and the transfer template culture control data sequence of the transfer anomaly monitoring network for ending the transfer knowledge learning is generated, and the importance value corresponding to the transfer anomaly monitoring network for ending the transfer knowledge learning is output.

In an alternative embodiment, the reciprocal of the vector cost value between the reference template culture control data sequence and the migration template culture control data sequence of the migration anomaly monitoring network for each end of the migration knowledge learning can be counted, and the importance value corresponding to the migration anomaly monitoring network for each end of the migration knowledge learning can be output.

Step S608, regularized conversion is performed on the importance values corresponding to the migration anomaly monitoring networks for which the migration knowledge learning is finished, so as to generate anomaly monitoring significance values corresponding to the migration anomaly monitoring networks for which the migration knowledge learning is finished.

For example, assuming that there are 4 migration anomaly monitoring networks for ending the migration knowledge learning, the corresponding importance values are K1, K2, K3, and K4, after regularized conversion, the anomaly monitoring significance values corresponding to the migration anomaly monitoring networks for ending the migration knowledge learning are K1/(k1+k2+k3+k4), k2/(k1+k2+k3+k4), k3/(k1+k2+k3+k4), and k4/(k1+k2+k3+k4).

In an alternative implementation manner, the weight of the control data sequence training for the migration anomaly monitoring network can be cultivated according to different reference templates, and the anomaly monitoring significance value of each migration anomaly monitoring network can be obtained through self-adaptive learning based on an attention mechanism; in an alternative embodiment, an initial anomaly monitoring significance value of each migration anomaly monitoring network ending the migration knowledge learning can be obtained, template culture control data in a migration template culture control data sequence is input into each migration anomaly monitoring network ending the migration knowledge learning, and an anomaly class confidence coefficient generated by each migration anomaly monitoring network ending the migration knowledge learning is generated; based on initial anomaly monitoring significance values of the migration anomaly monitoring networks for each end of migration knowledge learning, fusing anomaly class confidence degrees generated by the migration anomaly monitoring networks for each end of migration knowledge learning, and generating target anomaly class confidence degrees of template culture control data; obtaining a subtraction value between the target abnormal category confidence coefficient of the template culture control data and the corresponding true abnormal category confidence coefficient, continuously updating the initial abnormal monitoring significance value of the migration abnormal monitoring network for each end of migration knowledge learning based on the subtraction value until the subtraction value between the obtained target abnormal category confidence coefficient of the template culture control data and the corresponding true abnormal category confidence coefficient is smaller than a set value, outputting the current initial abnormal monitoring significance value as the abnormal monitoring significance value of the migration abnormal monitoring network for each end of migration knowledge learning. Thus, the abnormal monitoring significance value of the migration abnormal monitoring network for ending the migration knowledge learning can be learned.

For example, there are 3 migration anomaly monitoring networks, namely a migration anomaly monitoring network a, a migration anomaly monitoring network B, and a migration anomaly monitoring network C, wherein the corresponding initial importance values are m, h, and g, and the confidence of anomaly categories of the corresponding generated template culture control data are x1, x2, and x3, respectively; generating a target anomaly class confidence x' =m×x1+h×x2+g×x3 for the template culture control data based on the 3 migration anomaly monitoring networks; updating the initial anomaly monitoring significance value of each migration anomaly monitoring network ending the migration knowledge learning based on the subtraction value between the target anomaly class confidence x' of the template culture control data and the corresponding real anomaly class confidence x until the obtained subtraction value is smaller than a set value; and outputting the initial abnormal monitoring significance values corresponding to the current 3 migration abnormal monitoring networks as the abnormal monitoring significance values of the 3 migration abnormal monitoring networks. Therefore, the importance value corresponding to the abnormal category confidence coefficient generated by the migration abnormal monitoring network for ending the migration knowledge learning can be effectively determined by determining the abnormal monitoring significance value corresponding to the migration abnormal monitoring network for ending the migration knowledge learning, and the forward self-adaption of the existing knowledge is ensured by combining the weight of the migration abnormal monitoring network for ending the migration knowledge learning.

In an alternative embodiment, in step S604, the calculating the vector cost value between the reference template culture control data sequence and the migration template culture control data sequence corresponding to the migration anomaly monitoring network for ending the migration knowledge learning specifically includes: respectively inputting a reference template culture control data sequence and a migration template culture control data sequence corresponding to the migration anomaly monitoring network for ending the migration knowledge learning into a vector cost network configured in advance, and generating vector cost values between the reference template culture control data sequence and the migration template culture control data sequence corresponding to the migration anomaly monitoring network for ending the migration knowledge learning; the vector cost network is configured in advance, and is used for respectively carrying out full connection output on the reference template culture control data sequence corresponding to the migration anomaly monitoring network ending the migration knowledge learning and the culture control knowledge vector of the template culture control data in the migration template culture control data sequence, so as to obtain the vector cost value between the reference template culture control data sequence corresponding to the migration anomaly monitoring network ending the migration knowledge learning and the migration template culture control data sequence.

Wherein, the vector cost network of the prior configuration can adopt a W distance network. The vector cost value between the reference template culture control data sequence and the transfer template culture control data sequence of each transfer anomaly monitoring network ending the transfer knowledge learning can be obtained through the previously configured vector cost network, so that the subsequent vector cost value between the reference template culture control data sequence and the transfer template culture control data sequence of each transfer anomaly monitoring network is facilitated, the importance value of each transfer anomaly monitoring network ending the transfer knowledge learning is determined, and then the anomaly monitoring significance value of each transfer anomaly monitoring network ending the transfer knowledge learning is determined.

In an alternative embodiment, the training step of the vector cost network configured in advance includes:

step A402, extracting template culture control data sequences of two knowledge learning categories from the template culture control data sequences for a plurality of times to form a plurality of groups of template culture control data sequences; the template culture control data sequences of the plurality of groups are matched with corresponding real vector cost values.

The template culture control data sequence may be a template culture control data sequence consisting of only the reference template culture control data sequence, or may be a template culture control data sequence which is integrated by the reference template culture control data sequence and the migration template culture control data sequence.

In an alternative embodiment, two knowledge learning category template incubation control data sequences, such as a reference template incubation control data sequence 1 and a reference template incubation control data sequence 2, a reference template incubation control data sequence 1 and a reference template incubation control data sequence 1, a reference template incubation control data sequence 3 and a migration template incubation control data sequence 1, etc., may be randomly extracted from all of the reference template incubation control data sequences and the migration template incubation control data sequences at a time, to constitute a pair of template incubation control data sequences.

And step A404, loading each group of template culture control data sequence columns into the initialized vector cost network respectively, and generating vector cost values of each group of template culture control data sequence columns.

Before each group of template culture control data sequence columns are loaded into the initialized vector cost network respectively to generate vector cost values of each group of template culture control data sequence columns, a batch of template culture control data can be randomly extracted from two knowledge learning type template culture control data sequences in each group of template culture control data sequence columns respectively, and two knowledge learning type template culture control data sequences in the template culture control data sequences are reconstructed.

Step A406, determining the training cost value of the vector cost network based on the vector cost value of each group of template culture control data sequence columns and the corresponding real vector cost value.

In an alternative embodiment, the abnormal monitoring error value of each group of the template culture control data sequence columns can be generated based on the vector cost value of each group of the template culture control data sequence columns and the corresponding real vector cost value and the set loss function; and adding the abnormal monitoring error values of the template culture control data sequence columns of each group to generate global abnormal monitoring error values of the vector cost network on the template culture control data sequence columns of each group, and outputting the global abnormal monitoring error values as the training cost value of the vector cost network.

Step A408, training cost value based on the vector cost network reversely trains the vector cost network until the vector cost network meets the third training termination requirement

Step A410, if the vector cost network meets the third training termination requirement, the vector cost network is used as the vector cost network configured in advance.

The fact that the vector cost network meets the third training termination requirement means that the iteration number of the vector cost network is larger than the set number or the weight information of the vector cost network is not changed any more.

In an alternative embodiment, the weight information optimization gradient value of the vector cost network can be determined based on the training cost value of the vector cost network; performing back propagation optimization on the vector cost network based on the weight information optimization gradient value of the vector cost network to generate an optimized vector cost network; taking the updated vector cost network as an initialized vector cost network, and repeatedly executing the steps A404 to A08 to continuously update the weight information of the vector cost network until the weight information of the vector cost network is not changed any more; and outputting the vector cost network with no change of the weight information as a previously configured vector cost network.

In an alternative embodiment, the step a110 generates the target culture control anomaly monitoring network based on the anomaly monitoring network for each end of the migration knowledge learning and the anomaly monitoring significance value corresponding to the anomaly monitoring network for each end of the migration knowledge learning, and specifically includes: integrating each migration anomaly monitoring network ending the migration knowledge learning to generate an integrated monitoring network, and outputting the integrated monitoring network as a target culture control anomaly monitoring network; the abnormal monitoring significance value corresponding to each migration abnormal monitoring network ending the migration knowledge learning represents the importance value of the abnormal category confidence generated by each migration abnormal monitoring network ending the migration knowledge learning.

The template culture control data abnormal category confidence coefficient generated by the target culture control abnormal monitoring network is the sum of the abnormal category confidence coefficient generated by each migration abnormal monitoring network ending the migration knowledge learning and the fusion value of the corresponding abnormal monitoring significance value; for example, assuming that the confidence of the abnormal category generated by the three migration abnormal monitoring networks ending the migration knowledge learning is K1, K2, and K3, and the corresponding abnormal monitoring significance values are h1, h2, and h3, respectively, the confidence of the abnormal category of the template culture control data generated by the target culture control abnormal monitoring network is k1×h1+k2×h2+k3×h3.

In an alternative embodiment, further examples are provided, comprising the steps of:

step A502, obtaining a target culture control abnormality monitoring network based on the abnormality monitoring method of the immune cell culture control system; the target culture control anomaly monitoring network comprises migration anomaly monitoring networks for ending the learning of migration knowledge.

Step A504, inputting the culture control knowledge vector of the migrated target immune cell culture control data into a target culture control anomaly monitoring network, and generating anomaly class confidence of the target immune cell culture control data in the migrated anomaly monitoring network for ending the migration knowledge learning and anomaly monitoring significance values corresponding to the migrated anomaly monitoring network for ending the migration knowledge learning.

In an alternative embodiment, the target immune cell culture control data may be obtained, the target immune cell culture control data may be subjected to regularization conversion to generate regularized conversion features of the target immune cell culture control data, vector encoding may be performed on the regularized conversion features of the target immune cell culture control data to generate a culture control knowledge vector of the target immune cell culture control data; and inputting the culture control knowledge vector of the target immune cell culture control data into a target culture control anomaly monitoring network, and obtaining the anomaly class confidence of the target immune cell culture control data in the migration anomaly monitoring network for ending the migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for ending the migration knowledge learning through the target culture control anomaly monitoring network.

Step A506, based on the abnormal monitoring significance value corresponding to the migration abnormal monitoring network for each end of the migration knowledge learning, fusing the abnormal category confidence level of the target immune cell culture control data in the migration abnormal monitoring network for each end of the migration knowledge learning, and generating the target abnormal category confidence level of the target immune cell culture control data.

In an alternative embodiment, the abnormal category confidence level of the target immune cell culture control data in the migration abnormal monitoring network of each end migration knowledge learning can be fused based on the abnormal monitoring significance value corresponding to the migration abnormal monitoring network of each end migration knowledge learning, so as to generate a fusion value of the abnormal monitoring significance value corresponding to the migration abnormal monitoring network of each end migration knowledge learning and the abnormal category confidence level; and adding the abnormal monitoring significance value corresponding to each migration abnormal monitoring network ending the migration knowledge learning with the fusion value of the abnormal class confidence coefficient to generate the final abnormal class confidence coefficient of the target immune cell culture control data, and outputting the final abnormal class confidence coefficient as the target abnormal class confidence coefficient of the target immune cell culture control data.

Step A508, determining an abnormality monitoring result of the target immune cell culture control data based on the target abnormality category confidence of the target immune cell culture control data.

Therefore, the transfer learning of the transfer template culture control data corresponding to the knowledge to be learned can be performed by combining the existing training knowledge of the reference template culture control data sequence, the supervision training of the transfer template culture control data sequence can be realized so as to improve the training effectiveness, the target culture control anomaly monitoring network is generated based on the transfer anomaly monitoring network for each end of the transfer knowledge learning and the anomaly monitoring significance value corresponding to the transfer anomaly monitoring network for each end of the transfer knowledge learning, the obvious condition of the anomaly monitoring result generated by the transfer anomaly monitoring network for each end of the transfer knowledge learning can be combined, and the transfer effect of the existing knowledge learning is improved, so that the anomaly monitoring accuracy of the subsequent target culture control anomaly monitoring network is improved.

In an alternative embodiment, further application examples are provided below, specifically comprising the following steps:

step A602, a plurality of reference template culture control data sequences and a migration template culture control data sequence are obtained, wherein the reference template culture control data sequences and the migration template culture control data sequences comprise a plurality of template culture control data and true abnormal category confidence corresponding to the template culture control data.

Step A604, performing prior knowledge learning on each initialized culture control anomaly monitoring network based on each reference template culture control data sequence, each migration template culture control data sequence and the corresponding real anomaly class confidence of each template culture control data respectively, and generating each culture control anomaly monitoring network after prior knowledge learning.

And step A606, respectively carrying out migration knowledge learning on each culture control abnormal monitoring network after the prior knowledge learning based on the migration template culture control data sequence, generating each culture control abnormal monitoring network after the migration knowledge learning, and outputting the migration abnormal monitoring network for each migration knowledge learning.

And step A608, determining the abnormal monitoring significance value corresponding to each migration abnormal monitoring network ending the migration knowledge learning based on the reference template culture control data sequence and the migration template culture control data sequence.

And step A610, generating a target culture control anomaly monitoring network based on the migration anomaly monitoring network for each end of the migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for each end of the migration knowledge learning.

Step A612, detecting an abnormality monitoring instruction of target immune cell culture control data; the abnormality monitoring instruction carries migrated target immune cell culture control data.

Step A614, obtaining the culture control knowledge vector of the migrated target immune cell culture control data.

Step A616, inputting the culture control knowledge vector of the migrated target immune cell culture control data into a target culture control anomaly monitoring network, and generating anomaly class confidence of the target immune cell culture control data in the migrated anomaly monitoring network for ending the migration knowledge learning and anomaly monitoring significance corresponding to the migrated anomaly monitoring network for ending the migration knowledge learning.

Step A618, based on the abnormal monitoring significance value corresponding to the migration abnormal monitoring network for each end of the migration knowledge learning, the abnormal category confidence level of the target immune cell culture control data in the migration abnormal monitoring network for each end of the migration knowledge learning is fused, and the target abnormal category confidence level of the target immune cell culture control data is generated.

Step A620, determining an anomaly monitoring result for the target immune cell culture control data based on the target anomaly class confidence of the target immune cell culture control data.

Step A622, pushing the abnormal monitoring result of the target immune cell culture control data to the terminal.

Therefore, the abnormal class confidence coefficient of the target immune cell culture control data in the migration abnormal monitoring network for finishing the migration knowledge learning is determined through the target immune cell culture control abnormal monitoring network, the abnormal monitoring significance value corresponding to the migration abnormal monitoring network for finishing the migration knowledge learning is combined, the target abnormal class confidence coefficient of the target immune cell culture control data is generated, the abnormal monitoring result of the target immune cell culture control data is further determined, the abnormal class confidence coefficient of the target immune cell culture control data in the migration abnormal monitoring network for finishing the migration knowledge learning and the abnormal monitoring significance value of the abnormal class confidence coefficient generated by the migration abnormal monitoring network for finishing the migration knowledge learning are combined, and therefore the monitoring accuracy of the target immune cell culture control abnormal monitoring network is improved.

Fig. 2 illustrates a hardware structural view of an abnormality monitoring system 100 of an immune cell culture control system for implementing the abnormality monitoring method of an immune cell culture control system according to an embodiment of the present application, and as shown in fig. 2, the abnormality monitoring system 100 of an immune cell culture control system may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In an alternative embodiment, the anomaly monitoring system 100 of the immune cell culture control system may be a single server or a group of servers. The server farm may be centralized or distributed (e.g., the anomaly monitoring system 100 of the immune cell culture control system may be a distributed system). In an alternative embodiment, the anomaly monitoring system 100 of the immune cell culture control system may be local or remote. For example, the anomaly monitoring system 100 of the immune cell culture control system may access information and/or data stored in the machine-readable storage medium 120 via a network. For another example, the anomaly monitoring system 100 of the immune cell culture control system may be directly connected to the machine-readable storage medium 120 to access stored information and/or data. In an alternative embodiment, the anomaly monitoring system 100 of the immune cell culture control system may be implemented on a cloud platform. For example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, or the like, or any combination thereof.

The machine-readable storage medium 120 may store data and/or instructions. In an alternative embodiment, the machine-readable storage medium 120 may store data acquired from an external terminal. In an alternative embodiment, machine-readable storage medium 120 may store data and/or instructions that are used by anomaly monitoring system 100 of an immune cell culture control system to perform or use to perform the exemplary methods described herein. In alternative embodiments, machine-readable storage medium 120 may include mass storage, removable storage, volatile read-write memory, read-only memory, and the like, or any combination thereof. Exemplary mass storage devices may include magnetic disks, optical disks, solid state disks, and the like. Exemplary removable memory may include flash drives, floppy disks, optical disks, memory cards, compact disks, tape, and the like.

In a specific implementation, the plurality of processors 110 execute computer executable instructions stored by the machine-readable storage medium 120, so that the processors 110 may execute the anomaly monitoring method of the immune cell culture control system according to the method embodiment, the processors 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processors 110 may be used to control the transceiving actions of the communication unit 140.

The specific implementation process of the processor 110 may refer to the above-mentioned embodiments of the method executed by the abnormality monitoring system 100 of the immune cell culture control system, and the implementation principle and technical effects are similar, which are not described herein again.

In addition, the embodiment of the application also provides a readable storage medium, wherein computer executable instructions are preset in the readable storage medium, and when a processor executes the computer executable instructions, the abnormality monitoring method of the immune cell culture control system is realized.

Likewise, it should be noted that in order to simplify the presentation disclosed herein and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof. Likewise, it should be noted that in order to simplify the presentation disclosed herein and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof.

Claims

1. A method for monitoring anomalies in an immune cell culture control system, the method comprising:

based on the reference template culture control data sequences, the migration template culture control data sequences and the true abnormal category confidence degrees corresponding to the template culture control data sequences, respectively, carrying out prior knowledge learning on each initialized culture control abnormal monitoring network to generate each culture control abnormal monitoring network after prior knowledge learning, wherein the culture control abnormal monitoring network adopts a long-period and short-period memory network;

Performing migration knowledge learning on the culture control anomaly monitoring networks after the prior knowledge learning based on the migration template culture control data sequence respectively, generating culture control anomaly monitoring networks after the migration knowledge learning, and outputting migration anomaly monitoring networks for finishing the migration knowledge learning, wherein each migration anomaly monitoring network finishing the migration knowledge learning corresponds to a final migration network which is known to know knowledge to be learned;

determining an abnormal monitoring significance value corresponding to the migration abnormal monitoring network for ending the migration knowledge learning based on the reference template culture control data sequence and the migration template culture control data sequence, wherein the abnormal monitoring significance value refers to an importance value corresponding to an abnormal category confidence coefficient generated by the migration abnormal monitoring network and is used for representing the weight of the abnormal category confidence coefficient generated by the migration abnormal monitoring network for ending the migration knowledge learning;

generating a target culture control anomaly monitoring network based on the migration anomaly monitoring network for each end of the migration knowledge learning and the anomaly monitoring significance value corresponding to the migration anomaly monitoring network for each end of the migration knowledge learning;

The step of performing prior knowledge learning on each initialized culture control anomaly monitoring network based on the reference template culture control data sequences, the migration template culture control data sequences and the true anomaly class confidence corresponding to the template culture control data sequences, respectively, to generate each prior knowledge learned culture control anomaly monitoring network, including:

constructing a first template culture control data sequence and a second template culture control data sequence of the reference template culture control data sequence and a first template culture control data sequence and a second template culture control data sequence of the migration template culture control data sequence based on the true anomaly class confidence corresponding to each reference template culture control data sequence, the migration template culture control data sequence and each template culture control data sequence, wherein the first template culture control data sequence refers to a part of template culture control data extracted from the reference template culture control data sequence or the migration template culture control data sequence and is used for training a culture control anomaly monitoring network corresponding to the existing training knowledge and a culture control anomaly monitoring network corresponding to the to-be-learned knowledge, and the second template culture control data sequence refers to a part of template culture control data re-extracted from the rest template culture control data of the reference template culture control data sequence or the migration template culture control data sequence and is used as a template culture control data sequence for determining anomaly monitoring error values of a culture control anomaly monitoring network corresponding to the existing training knowledge and anomaly monitoring error values of a culture control anomaly monitoring network corresponding to the to-be-learned knowledge;

Based on the first template culture control data sequence of the reference template culture control data sequence and the first template culture control data sequence of the migration template culture control data sequence, performing network importance numerical parameter optimization on the initialized culture control anomaly monitoring network to generate a culture control anomaly monitoring network after network importance numerical parameter optimization;

acquiring an abnormal monitoring error value of the culture control abnormal monitoring network after the network importance numerical parameter optimization on a second template culture control data sequence of the reference template culture control data sequence and a second template culture control data sequence of the migration template culture control data sequence, updating the initialized culture control abnormal monitoring network based on the abnormal monitoring error value, and generating an updated culture control abnormal monitoring network;

if the updated culture control anomaly monitoring network does not meet the first training termination requirement, using the updated culture control anomaly monitoring network as an initialized culture control anomaly monitoring network, and returning to execute a first template culture control data sequence based on the reference template culture control data sequence and a first template culture control data sequence of the migration template culture control data sequence, performing network importance numerical parameter optimization on the initialized culture control anomaly monitoring network, and generating operation of the culture control anomaly monitoring network after network importance numerical parameter optimization;

If the updated culture control abnormality monitoring network meets the first training termination requirement, outputting the current culture control abnormality monitoring network as the culture control abnormality monitoring network after learning of all the existing knowledge;

the method comprises the steps of performing migration knowledge learning on the culture control anomaly monitoring network after the prior knowledge learning based on the migration template culture control data sequences respectively to generate culture control anomaly monitoring networks after the migration knowledge learning, and comprises the following steps:

acquiring a culture control knowledge vector of template culture control data in the migration template culture control data sequence;

loading culture control knowledge vectors of template culture control data in the migration template culture control data sequence to the culture control anomaly monitoring networks after the prior knowledge learning respectively, and generating anomaly class confidence coefficients of the template culture control data in the culture control anomaly monitoring networks after the prior knowledge learning;

obtaining the subtraction value of the abnormal category confidence coefficient of the template culture control data in the culture control abnormal monitoring network after each prior knowledge study and the corresponding real abnormal category confidence coefficient;

Determining the training cost value of the culture control anomaly monitoring network after each existing knowledge study based on the subtraction value;

reversely training the training control anomaly monitoring networks after the prior knowledge learning based on the training cost values respectively until the training control anomaly monitoring networks after the prior knowledge learning meet the second training termination requirement;

and if the culture control abnormality monitoring networks after the prior knowledge learning meet the second training termination requirement, outputting the current culture control abnormality monitoring network after the prior knowledge learning as the culture control abnormality monitoring network after the prior knowledge learning.

2. The method for monitoring anomalies in an immune cell culture control system according to claim 1, wherein the generating a network-importance-value-parameter-optimized culture control anomaly monitoring network based on the first template culture control data sequence of the reference template culture control data sequence and the first template culture control data sequence of the migration template culture control data sequence comprises:

Performing network importance numerical parameter optimization on the initialized culture control anomaly monitoring network based on a first template culture control data sequence of the reference template culture control data sequence, and generating a first anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the reference template culture control data sequence, wherein the first anomaly monitoring error value refers to a global anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the reference template culture control data sequence;

performing network importance numerical parameter optimization on the initialized culture control anomaly monitoring network based on a first template culture control data sequence of the migration template culture control data sequence, and generating a second anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the migration template culture control data sequence, wherein the second anomaly monitoring error value refers to a global anomaly monitoring error value of the initialized culture control anomaly monitoring network on the first template culture control data sequence of the migration template culture control data sequence;

Based on the first anomaly monitoring error value, performing back propagation optimization on the initialized culture control anomaly monitoring network to generate a culture control anomaly monitoring network corresponding to the existing training knowledge;

and carrying out back propagation optimization on the initialized culture control anomaly monitoring network based on the second anomaly monitoring error value to generate a culture control anomaly monitoring network corresponding to the knowledge to be learned.

3. The abnormality monitoring method of an immune cell culture control system according to claim 2, wherein the acquiring the abnormality monitoring error value of the culture control abnormality monitoring network after the network importance numerical parameter optimization on the second template culture control data sequence of the reference template culture control data sequence and the second template culture control data sequence of the migration template culture control data sequence updates the initialized culture control abnormality monitoring network based on the abnormality monitoring error value, and generating an updated culture control abnormality monitoring network includes:

updating the culture control anomaly monitoring network corresponding to the existing training knowledge based on the second template culture control data sequence of the reference template culture control data sequence, and generating a third anomaly monitoring error value of the culture control anomaly monitoring network corresponding to the existing training knowledge on the second template culture control data sequence of the reference template culture control data sequence, wherein the third anomaly monitoring error value refers to a global anomaly monitoring error value of the culture control anomaly monitoring network corresponding to the existing training knowledge on the second template culture control data sequence of the reference template culture control data sequence;

Updating the culture control anomaly monitoring network corresponding to the knowledge to be learned based on the second template culture control data sequence of the transfer template culture control data sequence, and generating a fourth anomaly monitoring error value of the culture control anomaly monitoring network corresponding to the knowledge to be learned on the second template culture control data sequence of the transfer template culture control data sequence, wherein the fourth anomaly monitoring error value refers to a global anomaly monitoring error value of the culture control anomaly monitoring network corresponding to the knowledge to be learned on the second template culture control data sequence of the transfer template culture control data sequence;

and carrying out counter-propagation optimization on the initialized culture control anomaly monitoring network based on the third anomaly monitoring error value and the fourth anomaly monitoring error value, generating an optimized culture control anomaly monitoring network, and outputting the optimized culture control anomaly monitoring network as the updated culture control anomaly monitoring network.

4. The method for monitoring anomalies in an immune cell culture control system according to claim 1, wherein the determining anomaly monitoring significance values corresponding to the migration anomaly monitoring network for each end of migration knowledge learning based on the reference template culture control data sequence and the migration template culture control data sequence comprises:

Extracting a reference template culture control data sequence and a migration template culture control data sequence corresponding to the migration anomaly monitoring network for ending the migration knowledge learning from the reference template culture control data sequence and the migration template culture control data sequence;

calculating vector cost values between a reference template culture control data sequence and a migration template culture control data sequence corresponding to each migration anomaly monitoring network ending migration knowledge learning, wherein the vector cost values represent the relevance between the reference template culture control data sequence and the migration template culture control data sequence, and the greater the vector cost values are expressed through characteristic distances, the less relevant the reference template culture control data sequence and the migration template culture control data sequence are; the smaller the vector cost value is, the more relevant the reference template culture control data sequence and the migration template culture control data sequence are represented;

searching mapping information of preset vector cost values and importance values based on vector cost values between the reference template culture control data sequence and the transfer template culture control data sequence of each transfer abnormality monitoring network ending the transfer knowledge learning, and determining the importance value corresponding to each transfer abnormality monitoring network ending the transfer knowledge learning;

Carrying out regularized conversion on importance values corresponding to the migration anomaly monitoring networks for ending the migration knowledge learning, and generating anomaly monitoring significance values corresponding to the migration anomaly monitoring networks for ending the migration knowledge learning;

the calculating the vector cost value between the reference template culture control data sequence and the migration template culture control data sequence corresponding to each migration anomaly monitoring network ending the migration knowledge learning comprises the following steps:

respectively inputting a reference template culture control data sequence and a migration template culture control data sequence corresponding to each migration anomaly monitoring network ending the migration knowledge learning into a vector cost network configured in advance, and generating a vector cost value between the reference template culture control data sequence and the migration template culture control data sequence corresponding to each migration anomaly monitoring network ending the migration knowledge learning, wherein the vector cost network configured in advance adopts a W distance network;

the vector cost network is configured in advance, and is used for respectively carrying out full connection output on the reference template culture control data sequence corresponding to the migration anomaly monitoring network ending the migration knowledge learning and the culture control knowledge vector of the template culture control data in the migration template culture control data sequence, so as to obtain the vector cost value between the reference template culture control data sequence corresponding to the migration anomaly monitoring network ending the migration knowledge learning and the migration template culture control data sequence;

The training step of the vector cost network of the prior configuration comprises the following steps:

extracting template culture control data sequences of two knowledge learning categories from the template culture control data sequences for a plurality of times to form a plurality of groups of template culture control data sequences; the template culture control data sequences of the multiple groups are matched with corresponding real vector cost values;

loading each group of template culture control data sequences into an initialized vector cost network respectively to generate vector cost values of the template culture control data sequences;

determining the training cost value of the vector cost network based on the vector cost value of each group of template culture control data sequences and the corresponding real vector cost value;

reversely training the vector cost network based on the training cost value of the vector cost network until the vector cost network meets a third training termination requirement;

and if the vector cost network meets the third training termination requirement, taking the vector cost network as the vector cost network of the prior configuration.

5. The method for monitoring anomalies in an immune cell culture control system according to claim 1, wherein the generating a target culture control anomaly monitoring network based on the transition anomaly monitoring network for each end of transition knowledge learning and anomaly monitoring significance values corresponding to the transition anomaly monitoring network for each end of transition knowledge learning comprises:

Integrating the migration anomaly monitoring networks ending the migration knowledge learning to generate an integrated monitoring network, and outputting the integrated monitoring network as a target culture control anomaly monitoring network; and the abnormal monitoring significance value corresponding to the migration abnormal monitoring network for each end of the migration knowledge learning represents the importance value of the abnormal category confidence level generated by the migration abnormal monitoring network for each end of the migration knowledge learning.

6. The method for monitoring abnormalities of an immune cell culture control system according to claim 1, characterized in that said method comprises:

acquiring a target culture control abnormality monitoring network; the target culture control anomaly monitoring network comprises migration anomaly monitoring networks for ending the learning of migration knowledge;

loading culture control knowledge vectors of the migrated target immune cell culture control data into the target culture control anomaly monitoring network, and generating anomaly class confidence levels of the target immune cell culture control data in the migration anomaly monitoring networks of all the end migration knowledge learning and anomaly monitoring significance values corresponding to the migration anomaly monitoring networks of all the end migration knowledge learning;

based on the abnormal monitoring significance value corresponding to the migration abnormal monitoring network for each end of migration knowledge learning, fusing the abnormal class confidence of the target immune cell culture control data in the migration abnormal monitoring network for each end of migration knowledge learning, and generating the target abnormal class confidence of the target immune cell culture control data;

And determining an abnormality monitoring result of the target immune cell culture control data based on the target abnormality class confidence of the target immune cell culture control data.

7. A computer readable storage medium having stored therein machine executable instructions that are loaded and executed by a processor to implement the method of anomaly monitoring of an immune cell culture control system of any one of claims 1-6.

8. An anomaly monitoring system for an immune cell culture control system, comprising a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement the anomaly monitoring method for an immune cell culture control system of any one of claims 1-6.