System

A server comprising:

• circuitry configured to:
• receive, via a communication interface coupled to a packet-switched network, observation information from a plurality of detection devices deployed in a spatial region, store raw time-series records in a time-series data store indexed by device identifier and timestamp, interpolate missing values, and apply normalization using precomputed scaling parameters to generate feature information formatted as numerical vectors;
• input image data received from at least one imaging device among the plurality of detection devices to a convolutional neural network model to obtain classification scores and category labels, and append the classification scores and the category labels to the feature information;
• input the feature information and external information received via the communication interface to a recurrent neural network model including long short-term memory layers to estimate at least one of a future state and a risk indicator for the spatial region;
• convert a prediction result and the external information into a natural language expression, generate a prompt sentence by applying a template-based generation module that transforms numerical values into human-readable phrases and appends optimization objectives, constraint conditions derived from user preference information, and statistical summaries from historical records, and input the prompt sentence to a transformer-based generative neural network model to obtain a response sentence describing an operation plan;
• parse the response sentence by applying at least one of rule-based text parsing and pattern matching to extract control parameters comprising time information, apparatus category identifiers, setting values, and application conditions, and format the extracted control parameters into device-specific command messages;
• transmit the device-specific command messages via the communication interface to a plurality of actuators to change a physical state of the spatial region, and receive execution result data from the plurality of actuators;
• transmit display information comprising the observation information, the prediction result, the response sentence, and the operation plan to a terminal device via the communication interface, receive intervention instruction information comprising at least one of an approval input, a modification input, and a cancellation input, and regenerate structured control commands based on the intervention instruction information;
• and
• store the execution result data, the intervention instruction information, and the observation information in a database, and update training data for the recurrent neural network model and generation rules for the prompt sentence based on the stored information.

A method performed by a server comprising circuitry, the method comprising:

• receiving, via a communication interface coupled to a packet-switched network, observation information from a plurality of detection devices deployed in a spatial region;
• aggregating the observation information and computing statistical information over predefined time windows, converting the statistical information into feature information formatted as numerical vectors, and inputting the feature information to a trained machine learning model to estimate at least one of a future state and a risk indicator for the spatial region;
• converting a prediction result of the trained machine learning model and external information received via the communication interface into a natural language expression, generating a prompt sentence including the natural language expression, and inputting the prompt sentence to a transformer-based generative neural network model to obtain a response sentence describing an operation plan and a resource allocation policy;
• parsing the response sentence to extract control parameters comprising at least time information, apparatus category identifiers, setting values, and application conditions, and generating, based on the extracted control parameters, structured control commands for a plurality of actuators;
• transmitting the structured control commands via the communication interface to the plurality of actuators to change a physical state of the spatial region;
• transmitting display information comprising at least the observation information, the prediction result, the response sentence, and the operation plan to a terminal device via the communication interface;
• receiving, from the terminal device, intervention instruction information comprising at least one of an approval input, a modification input, and a cancellation input with respect to the operation plan;
• and
• updating the prompt sentence by incorporating preference information derived from the intervention instruction information and past intervention history as contextual information for subsequent input to the transformer-based generative neural network model.