CN-122021590-A - Agent-driven differential report generation method, system and storage medium

CN122021590ACN 122021590 ACN122021590 ACN 122021590ACN-122021590-A

Abstract

The invention discloses a method, a system and a storage medium for generating a differential report driven by an agent. The method comprises the steps of monitoring key index data flow in real time, analyzing characteristics of the key index data flow based on a statistical model to generate a state switching instruction, and receiving the instruction to switch among various working states such as routine, early warning and emergency. In response to the state switch, the report generation process flow is dynamically adjusted, the adjustment at least including altering the data sampling frequency, switching the data preprocessing model, and selecting report generation logic that matches the current state. The invention realizes the transition from passive report generation to active and quasi-real-time decision support by introducing event-driven polymorphic working modes, obviously improves the response speed of the system to emergencies and the intelligent level of the decision support, and takes the energy efficiency and the performance of the system into account.

Inventors

WANG XIAO
LIU BIN
WANG YANG
ZHAO HONGLIANG
WANG BANGJUN
LI ZHENG
REN LISHAN
LI NA
Fei Ruixin

Assignees

中煤信息技术(北京)有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. An agent-driven differential report generation method, comprising: Monitoring key index data streams from one or more heterogeneous data sources in real time through a state determining module; the state determining module analyzes the statistical characteristics of the key index data stream based on a predefined statistical model to generate a state switching instruction; The report generating module receives the state switching instruction and switches among a plurality of preset working states at least comprising a normal state, an early warning state and an emergency state according to the instruction; The report generation module responds to the switching of the working state and dynamically adjusts the report generation processing flow, wherein the adjustment at least comprises the steps of changing the data sampling frequency, switching the data preprocessing model and selecting report generation logic matched with the current working state; wherein in the emergency state, the report generation logic is configured to bypass conventional user demand resolution steps and to force a preset emergency report template to be invoked to generate an emergency report.
2. The method of claim 1, wherein the state determination module analyzing statistical features of the key indicator data stream comprises: and continuously calculating and evaluating the key index data stream by adopting a sliding window-based algorithm.
3. The method of claim 2, wherein the sliding window based algorithm is an Exponentially Weighted Moving Average (EWMA) control graph algorithm that identifies trend changes or abnormal fluctuations in data flow by calculating a weighted average of the key indicator data within the sliding window.
4. The method of claim 3, wherein the exponentially weighted moving average control map algorithm is configured to define a weight factor λ between 0 and 1, establish an Upper Control Limit (UCL) and a Lower Control Limit (LCL) based on a statistical standard deviation of historical data of the key indicator, the upper and lower control limits being respectively located at a K times standard deviation of a center line, wherein K is a configurable risk factor, a preset value of 3, and generate the state switching instruction for switching to the early warning state or the emergency state when an EWMA calculated value of consecutive M data points in the key indicator data stream exceeds the upper or lower control limit, wherein M is a preset integer, the value of which is equal to or greater than 2, with the aim of ensuring high confidence of a state switching decision and filtering false alarms caused by transient noise.
5. The method of claim 1, wherein in the contingency state, the report generation logic further comprises executing one or more structured query language (SPARQL) queries in parallel to obtain query results based on a pre-built multi-industry knowledge graph to automatically analyze potential sources of events triggering the contingency state, and associating and extracting corresponding contingency plan information from the knowledge graph based on the query results, and populating the contingency report with the potential sources and contingency plan information.
6. The method of claim 1, wherein the data preprocessing model is an adaptive anomaly detection model configured to: according to the distribution change of recently acquired data, a set of industry weight coefficients for identifying data anomalies are periodically and automatically updated to adapt to the data concept drift caused by the production process or external environment change.
7. The method of claim 1, wherein the report generation process flow further comprises: under the conventional state, analyzing the received user report generation request by adopting a natural language understanding model to obtain an analysis intention; calculating a confidence score of the parsing intention based on a sequence probability output by a final classification layer of the natural language understanding model; and triggering a human-computer interaction clarification process facing the user or generating a default report in a preset format when the confidence score is lower than a preset threshold.
8. An agent-driven differential report generating system, comprising: A state determination module configured to monitor in real time key indicator data streams derived from one or more heterogeneous data sources and to analyze statistical features of the key indicator data streams based on a predefined statistical model to generate a state switch instruction; the report generation module is configured to receive the state switching instruction, switch among a plurality of preset working states at least comprising a normal state, an early warning state and an emergency state according to the instruction, and dynamically adjust the report generation processing flow of the report generation module in response to the switching of the working states, wherein the adjustment at least comprises changing the data sampling frequency, switching the data preprocessing model and selecting report generation logic matched with the current working state; wherein, in the emergency state, the report generating module is configured to bypass the conventional user requirement analyzing step and forcibly call a preset emergency report template to generate an emergency report.
9. The system of claim 8, wherein the report generation module employs an adaptive anomaly detection model as the data preprocessing model, the adaptive anomaly detection model configured to periodically automatically update a set of industry weight coefficients for identifying anomalies in the data based on changes in distribution of recently acquired data to accommodate for data concept drift.
10. A computer readable storage medium having stored thereon a computer program which when executed by a processor implements the method of any of claims 1 to 7.

Description

Agent-driven differential report generation method, system and storage medium Technical Field The invention belongs to the technical field of computers, and particularly relates to an artificial intelligence, a knowledge graph and big data processing technology, in particular to an agent-driven differential report generation method, an agent-driven differential report generation system and a computer-readable storage medium. Background In modern enterprise management, data driven decision support is critical, especially for large group-type enterprises where businesses span multiple industries of coal, electricity, chemical, new energy, etc. The report is used as a core carrier of a data analysis result, and the generation efficiency, accuracy and decision support capability of the report directly influence the operation efficiency and risk management level of a group. Existing report generation techniques mainly exist in one of several modes, namely, a data filling method based on a fixed template. The method presets the chapters, charts and indexes of the report, automatically grabs data through a database interface and fills the data into corresponding positions. Its advantages are simple implementation and high generation speed, but very obvious disadvantages. First, the rigid structure of the template cannot accommodate the differentiated requirements of different management levels (e.g., group headquarters, secondary sub-companies, base production units). The group headquarter pays attention to macro compliance and strategic risks across industries, while the basic unit needs real-time operation data specific to a certain production line and a certain workshop, and the fixed template is difficult to meet analysis requirements of different granularities. Secondly, the method is usually designed aiming at a single service field, and for multi-industry groups, data standards, index naming and service logic differences of different industries are huge (for example, raw coal yield in the coal industry and Internet power in the electric power industry), and effective integration and comparison analysis across industries cannot be realized by simple template filling. Second, self-help analysis reports based on Business Intelligence (BI) tools. The tool gives business personnel a certain flexibility, and the report can be customized in a dragging mode. However, this requires a user to have a high data analysis capability and a deep understanding of the business. For group high-level management, they often need direct, explicit conclusions and suggestions, rather than complex operator interfaces. More importantly, the BI tools themselves do not have domain knowledge and cannot automatically perform compliance verification (e.g., determine whether pollutant emissions meet national standards) or provide decision advice (e.g., recommend corresponding abatement schemes when emissions that exceed standards occur). Thirdly, a preliminary intelligent scheme of knowledge graph package technology is introduced. Some advanced systems begin to attempt to build domain knowledge maps and structure information such as data, standards, cases, etc. However, this is generally limited to a single field, for example, a knowledge graph specific to pollution control may not effectively relate production data, financial data, etc. of enterprises, resulting in analysis dimension aspects, and failure to reveal deep causal relationships between "business activities" and "environmental impact", so that the decision value of reporting is greatly compromised. In summary, the prior art generally faces the following challenges when dealing with complex reporting needs of multi-industry group enterprises: 1. the data island and the suitability are poor, the business data and the environment-friendly data of industries (such as coal, electricity and chemical) of the group are difficult to be effectively fused, and a unified index alignment and association analysis mechanism is lacking, so that the report dimension is single and the report dimension is split. 2. The knowledge decision support is lacking, the report stays at the data listing level, and the automatic compliance verification, risk early warning and intelligent decision suggestion functions based on industry standards, regulation policies and historical successful cases are lacking. 3. The hierarchy requirement is not matched, namely report generation logic is stiff, and the granularity, analysis key point and presentation form of the report cannot be dynamically adjusted according to the specific requirements of different management hierarchies such as groups, secondary enterprises, basic units and the like, so that the report has weak practicability and pertinence. 4. Response lag and passivity most systems adopt a periodic or on-demand report generation mode, for sudden environmental protection pre-warning or emergency events in the production process, the near real-time intel