CN-122021848-A - Operation and maintenance strategy reasoning method of model generation rule

CN122021848ACN 122021848 ACN122021848 ACN 122021848ACN-122021848-A

Abstract

The invention relates to the technical field of rule reasoning, in particular to an operation and maintenance strategy reasoning method for generating rules by a model, which comprises the following steps that value sequencing is completed by adopting a Markov process, a quantifiable sequence basis is obtained by an action field in a value evaluation stage, double-sequence constraint is formed by value sequence information and time sequence information in the same action chain, continuous sequential marks are obtained by the action field in a time dimension by adopting a time sequence convolution network to extract time sequence characteristics, and the action linking process depends on factor connection and time sequential marks at the same time, so that the rule numbering linking process is completed under the dual conditions of value continuity and time sequence, and a chain structure and a causal sequence structural element are triggered by actions to carry out cross positioning, so that an output rule link forms consistency basis in terms of action value sequencing, action causal linking and action time evolution, and the rule structure obtains stability, differentiation and reusability in a complex operation and maintenance scene.

Inventors

LI HAOCAI
HE YE
WAN HAIXIN
FAN YANG
ZHANG SHITONG

Assignees

中国大唐集团数字科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251211

Claims (10)

1. The operation and maintenance policy reasoning method of the model generation rule is characterized by comprising the following steps of: Reading three types of data item by item and numbering in sequence through operation and maintenance object real-time state data, alarm real-time grade data and resource real-time occupation data, integrating the numbering contents into a state interval arrangement structure according to the reading sequence, and establishing a state interval forming unit; Step two, based on the state interval forming unit, reading an action field, sequentially adding a recovery time length factor, a risk grade factor and a resource change factor, and sequencing the added action content according to a Markov process to generate an action value sequence frame; Step three, based on the action value sequence frame, matching the state interval numbers of the condition fields item by item, screening action fields, sequentially connecting the screened action fields with corresponding factors, and forming a continuous trigger chain to obtain an action trigger chain structure; Step four, generating time loci item by item for action fields based on the action triggering chain structure, combining a time sequence convolution network to extract time characteristics, comparing front and rear relations, rearranging the compared action fields according to time sequence and forming a causal connection path, and constructing a causal sequence structural element; And fifthly, based on the action triggering chain structure and the causal sequence structural element, the action fields of the two structures are positioned in a crossing way item by item, corresponding rule numbers are determined, the numbers are sequentially connected into rule links according to causal sequences, and rule link compendium is output.
2. The operation and maintenance policy reasoning method of the model generation rule according to claim 1, wherein the state interval forming unit comprises a state interval number set corresponding to operation and maintenance object real-time state data, alarm real-time level data and resource real-time occupation data which are arranged according to a number sequence; the action value sequence box comprises an action field, a recovery time length factor, a risk level factor, a resource variation factor and an action value factor from a Markov decision process; The action triggering chain structure comprises action fields screened according to state interval numbers, and a recovery time length factor, a risk grade factor and a resource change factor which are connected with each action field; the causal sequence structural element comprises a time locus, an action field corresponding to the time locus and a causal connection path formed by extracting time sequence features from a time sequence convolution network; the rule link schema includes a sequence of rule numbers arranged in causal order and rule links corresponding to the rule numbers.
3. The operation and maintenance policy inference method of model generation rule according to claim 1, wherein the specific steps of establishing the state interval constructing unit are: Based on the real-time state data of the operation and maintenance object, the alarm real-time grade data and the resource real-time occupation data, reading the data one by one and recording the data sequence, writing numbers, storing the corresponding content of the numbers, sorting the number records and keeping the sequence continuity to generate a number sequence set; Based on the number sequence set, numbers are extracted and continuous paragraphs are cut one by one, the sequence of the paragraphs is sorted after splicing the paragraphs, the solid of the solidified paragraphs forms interval arrangement records and maintains a continuous structure of the paragraphs, and a state interval forming unit is established.
4. The operation and maintenance policy reasoning method of the model generation rule according to claim 1, wherein the specific steps of generating the action value sequence box are: Based on the state interval forming unit, reading action fields one by one and establishing a recording position according to a field sequence, writing a recovery time length factor into the recording position and writing a risk grade factor into the recording position, writing action value factors from a Markov decision process into the recording position, writing resource variation factors into the recording position, and then writing the resource variation factors into a content arrangement sequence to generate an action additional set; based on the action additional set, extracting additional content according to a recording sequence, reading the difference of adjacent records, writing the difference into corresponding records, moving the recording position according to the difference, and maintaining a sequencing continuous structure after finishing the overall sequencing of the records to obtain an action difference sequence; And reading the sequence content piece by piece based on the action difference sequence, reorganizing the action fields according to the sequence order, synchronously combining the additional factor content, sequentially sorting the reorganization items by the reorganization item marking sequence to form an action arrangement structure, and solidifying the structure content to generate an action value sequence frame.
5. The method for operation and maintenance policy reasoning of the model generation rule according to claim 4, wherein the markov decision process comprises: Based on the existing action value factors in the action additional set and the state interval numbers associated with the action fields, reading state transition values from the state transition records corresponding to the state interval numbers one by one; sequentially positioning the reachable states according to the appearance sequence of the action fields, and writing the transfer values of the reachable states into independent transfer storage sites; And reading corresponding value quantities from the action value factor set according to the same sequence number positions, generating state action relation entries according to the value quantities and the contents of the transfer storage sites in a fixed sequence, and storing the entries in a concentrated manner according to the record sequence as a state action sequence which can be used for subsequent sorting processing.
6. The operation and maintenance policy reasoning method of the model generation rule according to claim 1, wherein the specific steps of obtaining the action trigger chain structure are: Based on the action value sequence frame, action fields in the action value sequence frame are sequentially extracted, condition fields are synchronously read, positions of corresponding action fields are positioned according to numbers after the state interval numbers are extracted, the positions are compared, the action fields meeting the numbers are screened, screened contents are sorted according to the extraction sequence, and an action screening sequence is generated; and based on the action screening sequence, extracting action screening sequence contents item by item, connecting a recovery time length factor according to a content sequence, connecting a risk grade factor, connecting a resource variation factor, keeping continuous arrangement of items in a connecting process, finishing to form a chain structure, and solidifying a chain record to obtain an action triggering chain structure.
7. The method for reasoning operation and maintenance strategy of model generation rule according to claim 1, wherein the specific steps of obtaining the causal sequence structural element are as follows: based on the action triggering chain structure, action fields are sequentially extracted, time loci are generated in front of the fields, action sequence positions are recorded after time marks are written in to form a time and action corresponding list, and a time locus sequence is generated; Based on the time locus sequence, reading time loci, combining a time sequence convolution network to extract time sequence features, then comparing the time sequence features, marking a sequence mark, adjusting the sequence of action fields according to the mark, recording a rearrangement index, establishing a time and action corresponding table, and generating a time sequence action set; And based on the time sequence action set, extracting action fields and factors in time sequence, connecting the action fields and the factors, writing the action fields and the factors into a causal path unit, and then adding a connection record to obtain a causal sequence structural element.
8. The operation and maintenance policy inference method of model generation rule according to claim 7, wherein the step of extracting time series features of the time series convolution network comprises: Based on the positions of the time marks recorded in the time locus sequence and the corresponding positions of the action fields, inputting the time marks into a convolution receiving end of a time sequence convolution network one by one according to the appearance sequence, and combining adjacent time marks into a local time segment at the fixed convolution step length of the input trailing edge; sequentially sliding the time slices forward according to the width of the convolution kernel to generate a plurality of groups of time slice response values; and writing the response values of the time segments into corresponding time sequence processing intermediate sites, and keeping the arrangement consistency of the response values according to the sequence in the time sequence sites, so that each action field obtains a time sequence response record corresponding to the time sequence sites at the network output end.
9. The operation and maintenance policy inference method of model generation rules according to claim 7, wherein said step of extracting action fields and extracting factors in chronological order comprises: the method comprises the steps that sequencing results exist in a time sequence action set, each action record is sequentially positioned, and action field content stored in each record is completely read; Splitting the recovery time length factor, the risk level factor and the resource change factor corresponding to the action field in the same record from the record structure one by one, and keeping the original position sequence of the factors in the record unchanged in the splitting process; and marking each record with a sequence number according to the time sequence, and establishing a corresponding relation between the action field and each split factor according to the number to form a combined data row containing the time sequence number, the action field content and the factor content.
10. The operation and maintenance policy inference method of model generation rules according to claim 1, wherein the specific step of outputting the rule link schema is: Based on the action triggering chain structure and the causal sequence structural element, extracting action fields in the two structures, establishing a cross positioning site by using a field serial number, comparing the fields of the positioning site, recording comparison positions, writing rule numbers according to comparison results, and then arranging the number sequence content to generate a rule number sequence; Based on the rule number sequence, reading rule numbers one by one, establishing a connecting site according to the appearance sequence of the numbers, carrying out number connection on the connecting site, recording connection sequence, pushing connection content into a link structure, arranging links, solidifying the link records, and outputting rule link program.

Description

Operation and maintenance strategy reasoning method of model generation rule Technical Field The invention relates to the technical field of rule reasoning, in particular to an operation and maintenance strategy reasoning method for generating rules by a model. Background The technical field of reasoning comprises a calculation method for carrying out knowledge representation, state judgment and decision generation based on a formalized rule set, the technical field generally establishes a rule base, a fact base and a reasoning engine, wherein a plurality of rules expressed in terms of conditions to actions are recorded in the rule base, the current system state parameters are recorded in the fact base, the reasoning engine triggers rules meeting the conditions from the fact base through forward reasoning, backward reasoning, mixed reasoning and conflict resolution strategies to generate a plurality of decision conclusions, and the technical field of rule reasoning realizes policy deduction and policy selection in the complex system in the form of interpretable rules in the scenes of operation and maintenance control, alarm processing, fault diagnosis, policy arrangement and the like. The operation and maintenance strategy reasoning method of model generation rule is a technical scheme of generating multiple operation and maintenance rules by model and outputting operation and maintenance strategy based on rule reasoning in operation and maintenance scene, and the method is based on operation and maintenance object model, operation and maintenance state model and operation and maintenance target model, automatically generates multiple operation and maintenance rules conforming to preset rule template structure, and loads the multiple operation and maintenance rules to reasoning engine to execute reasoning, the method aims to stably generate an executable operation and maintenance rule set from a structured model, further automatically calculate operation and maintenance strategy combinations meeting constraint conditions in a complex operation and maintenance scene, reduce the number of times of rule writing by the method, reduce the experience proportion of operation and maintenance strategy dependence, shorten the average time from fault occurrence to strategy suggestion giving, and promote the uniformity and reusability of multiple operation and maintenance decisions at the rule level. The prior art relies on static rule base record condition and action mapping, a continuous numbering system is lacking among condition fields, state fields and action fields, so that the phenomenon of unstable triggering sequence of rule triggering in a multi-state alternating scene is caused, the prior art mainly maintains rules manually, a unified factor system is lacking from condition to action structure, so that action fields lack comparable basis in different rules, rule conflict and rule overlapping frequently occur in a large operation and maintenance scene, the prior art does not establish unified time sequence record for action generation time, action triggering relation and action influence range in the reasoning process, so that a problem that policy output does not accord with state evolution sequence when an inference engine processes an alarm upgrading chain and a resource occupation change chain occurs, the prior art relies on static formalization rules, the rules are difficult to synchronously reflect state offset when the state of an operation and maintenance object is rapidly changed, the rule response lag occurs when the alarm condition continuously changes, and the policy output is unfavorable for timely correspondence to a fault propagation path in the operation and maintenance scene. Disclosure of Invention The invention aims to solve the defects existing in the prior art, and provides an operation and maintenance strategy reasoning method of a model generation rule. In order to achieve the purpose, the invention adopts the following technical scheme that the operation and maintenance strategy reasoning method of the model generation rule comprises the following steps: Reading three types of data item by item and numbering in sequence through operation and maintenance object real-time state data, alarm real-time grade data and resource real-time occupation data, integrating the numbering contents into a state interval arrangement structure according to the reading sequence, and establishing a state interval forming unit; Step two, based on the state interval forming unit, reading an action field, sequentially adding a recovery time length factor, a risk grade factor and a resource change factor, and sequencing the added action content according to a Markov process to generate an action value sequence frame; Step three, based on the action value sequence frame, matching the state interval numbers of the condition fields item by item, screening action fields, sequentially connecting the screened action