Cloud-edge collaborative energy device scheduling method and device, electronic device, and storage medium

Abstract

The application discloses a cloud-edge collaborative energy equipment scheduling method and device, electronic equipment and storage medium. The method comprises the following steps: acquiring real-time operation data of an industrial energy equipment group, inputting a scheduling intention in natural language form into a preset large language model to output a structured optimization target; determining corresponding industrial safety constraint conditions according to the real-time operation data and a preset expert rule base; constructing and solving a multi-objective optimization function to generate an initial scheduling scheme; inputting the initial scheduling scheme into a preset digital twin model for simulation and rehearsal, and judging whether it meets a preset safety boundary; if it meets the safety boundary, it is encapsulated as a structured strategy task and issued to an edge controller, so that the edge controller performs secondary safety verification based on a local real-time state and generates a control instruction sequence to drive a bottom equipment. The application realizes safe analysis and global collaborative control of complex intentions through the double blocking mechanism of cloud digital twin and edge secondary verification.

Description

Technical Field

[0001] This invention relates to the fields of industrial internet, edge computing and artificial intelligence control technology, and in particular to a cloud-edge collaborative energy equipment scheduling method, device, electronic device and storage medium. Background Technology

[0002] Intelligent scheduling methods, devices, electronic equipment, and storage media for industrial energy equipment clusters based on large-scale model and edge controller collaboration: Current industrial energy systems (such as energy systems in industrial parks, data centers, cold storage plants, air compressor stations, and energy storage systems) typically contain a massive number of underlying industrial energy devices. Existing control methods mainly rely on fixed logic control by programmable logic controllers (PLCs), independent device controller control, or manual experience-based scheduling and simple rule-based scheduling based on energy management systems (EMS).

[0003] In practical applications, existing technologies have the following significant technical shortcomings: First, traditional controllers and conventional EMS systems can only control local individual devices and cannot comprehensively utilize multi-dimensional data such as time-of-use electricity prices, carbon emission factors, and production plan loads for joint calculations. This results in low overall system energy efficiency for industrial energy equipment groups and makes it impossible to achieve global optimal scheduling.

[0004] Second, existing control systems are unable to parse complex natural language scheduling intentions (such as "reduce energy consumption during tonight's peak hours while ensuring production is not affected"), and the systems lack the technical capabilities for complex task planning and semantic understanding.

[0005] Third, although existing technologies include solutions based on large cloud-based AI models for decision-making, the inherent risks of illusion and parameter out-of-bounds errors in these models make their outputs highly unstable. Allowing the cloud to directly issue control commands to underlying industrial equipment poses serious security risks of exceeding boundaries and unauthorized control. The lack of a secure physical barrier mechanism adapted to millisecond-level responses in industrial settings leads to a severe disconnect between cloud-based intelligent analysis and actual edge control. Summary of the Invention

[0006] To address the technical problems in existing technologies, such as low intelligence levels, inability to handle complex semantic decisions, and low security of direct control of industrial equipment by large cloud models, this invention provides a cloud-edge collaborative energy equipment scheduling method, device, electronic device, and storage medium.

[0007] The first aspect of this invention discloses a cloud-edge collaborative energy device scheduling method, the method comprising: The system acquires real-time operational data of industrial energy equipment clusters and inputs scheduling intentions in natural language form into a pre-defined large language model, which then outputs structured optimization objectives. Based on the real-time operating data and the preset expert rule base, determine the industrial safety constraints corresponding to the structured optimization objective; Based on the structured optimization objective and the industrial safety constraints, a multi-objective optimization function is constructed, and an initial scheduling scheme is generated by solving the multi-objective optimization function. The initial scheduling scheme is input into a preset digital twin model for simulation, the simulation execution results are obtained, and it is determined whether the simulation execution results meet the preset safety boundaries. If the conditions are met, the initial scheduling scheme is encapsulated into a structured policy task and sent to the edge controller, so that the edge controller performs a secondary security check on the structured policy task based on the local real-time state, and generates a control command sequence after the check passes, and controls the industrial energy equipment group to execute the control command sequence.

[0008] Preferably, the structured optimization objective includes: intent type, asset scope, time window, primary optimization objective, secondary optimization objective, hard constraints, soft constraints, and decision variables; The output of the structured optimization objective through the large language model includes: The time information in the scheduling intent is extracted by the large language model, and the time window is determined by combining it with the preset time-of-use electricity price data; The large language model is used to identify the device semantics in the scheduling intent, match the target control device in the industrial energy equipment group, and determine the asset scope. The scheduling intent is mapped into a mathematical expression for the decision variables, thereby generating the main optimization objective and the hard constraints.

[0009] Preferably, the multi-objective optimization function includes an electricity cost item, a carbon emission cost item, an equipment start-up and shutdown penalty item, and an objective deviation penalty item; The construction of a multi-objective optimization function based on the structured optimization objective and the industrial safety constraints includes: Obtain the predicted time-of-use electricity price and predicted carbon emission factor within the time window; The electricity cost item is constructed based on the predicted time-of-use electricity price and the grid-connected power of the industrial energy equipment group; The carbon emission cost item is constructed based on the predicted carbon emission factor and the grid-connected power. The multi-objective optimization function is obtained by weighted summing of the electricity cost item, the carbon emission cost item, the equipment start-up and shutdown penalty item, and the target deviation penalty item.

[0010] Preferably, the structured strategy task includes a unique task identifier, instruction pattern, target curve, control boundary constraints, and digital signature; The instruction mode includes at least one of the following: target curve mode, control instruction sequence mode, boundary constraint update mode, and emergency stop mode; The step of encapsulating the initial scheduling scheme into a structured policy task and distributing it to the edge controller includes: According to the preset communication protocol format, the initial scheduling scheme is converted into the data payload corresponding to the instruction mode; The data payload is signed using a preset encryption algorithm to generate the digital signature; The structured policy task with the digital signature is sent to the edge controller via a preset encrypted communication link.

[0011] Preferably, determining whether the simulation execution result meets the preset safety boundary includes: The predicted state features in the simulation execution results are analyzed, and the predicted state features include at least one of predicted pressure, predicted temperature, predicted current, and predicted state of charge. The predicted state features are compared with the pre-established hard constraint threshold and soft constraint threshold, respectively; If any of the predicted state features exceeds the hard constraint threshold, it is determined that the safety boundary is not met, and the issuance instruction of the initial scheduling scheme is intercepted. If all the predicted state features are within the hard constraint threshold and there are features that exceed the soft constraint threshold, then a downgrade execution strategy or manual review process will be triggered.

[0012] Preferably, the edge controller performs a secondary security check on the structured policy task based on the local real-time state, and generates a control command sequence after the check passes, including: Obtain the structured policy task and verify its validity period and signature authorization status. The local real-time status of the industrial energy equipment group is obtained, including the underlying hardware interlock status, equipment alarm status, and manual takeover status. When the local real-time state meets the execution conditions, the target control deviation is calculated based on the target curve in the structured strategy task and the local real-time state. Using a preset local closed-loop control algorithm, the target control deviation is converted into a sequence of control commands for each device in the industrial energy equipment group.

[0013] Preferably, the method further includes: During the process of the edge controller controlling the industrial energy equipment group to execute the control command sequence, the feedback status of the industrial energy equipment group is continuously monitored; If the feedback state is detected to trigger a hard interlock condition or an emergency stop command is received, the edge controller cancels the control command sequence and executes a safe shutdown of the device or switches to a preset conservative operation strategy. The status data and failure reason after the undo operation are fed back to the cloud server to trigger the cloud server to update the parameters of the multi-objective optimization function.

[0014] A second aspect of this invention discloses a cloud-edge collaborative energy equipment scheduling device, the device comprising: The intent parsing module is used to acquire real-time operating data of industrial energy equipment groups and input scheduling intents in natural language form into a preset large language model, and output structured optimization objectives through the large language model; The constraint generation module is used to determine the industrial safety constraints corresponding to the structured optimization objective based on the real-time operating data and the preset expert rule base. The optimization solution module is used to construct a multi-objective optimization function based on the structured optimization objective and the industrial safety constraints, and to generate an initial scheduling scheme by solving the multi-objective optimization function; The twin simulation module is used to input the initial scheduling scheme into a preset digital twin model for simulation simulation, obtain the simulation execution results, and determine whether the simulation execution results meet the preset safety boundaries. The cloud-edge collaborative execution module is used to encapsulate the initial scheduling scheme into a structured policy task when the security boundary is met, and send it to the edge controller so that the edge controller can perform a secondary security verification on the structured policy task based on the local real-time status, and generate a control command sequence after the verification is passed, and control the industrial energy equipment group to execute the control command sequence.

[0015] A third aspect of this invention discloses an electronic device, comprising: a processor and a memory; The memory stores a computer program, which, when executed by the processor, implements the cloud-edge collaborative energy device scheduling method as described in any of the above descriptions.

[0016] The fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the cloud-edge collaborative energy device scheduling method as described in any one of the above descriptions.

[0017] Beneficial effects Improved accuracy of scheduling intent parsing and execution: By using a pre-set large language model, unstructured natural language scheduling intents are transformed into structured optimization objectives containing fields such as asset scope, time window, and decision variables. This solves the problem that traditional control systems cannot understand complex task logic, and enables multi-dimensional optimization requirements (such as balancing electricity costs and carbon emissions) to be quantified and converted into a computable mathematical model.

[0018] This invention improves the safety and reliability of industrial equipment control by constructing a dual-interception mechanism of "cloud-based digital twin pre-simulation + edge-based physical verification." In the cloud, the initial scheduling scheme must be simulated using a digital twin model to eliminate high-risk schemes that could lead to out-of-bounds temperature, pressure, and other conditions. At the edge, the edge controller does not directly execute cloud commands but instead uses structured strategy tasks as guidance for local closed-loop control, combining this with secondary safety verification based on the underlying hardware interlocking status. This mechanism fundamentally isolates the illusionary risks of large language models, preventing damage to the industrial field caused by erroneous strategies.

[0019] Improved system control latency in cloud-edge collaboration: By adopting a structured strategy task approach rather than issuing direct action commands, the edge controller gains autonomy. Upon receiving the target curve or control boundary, the edge controller can perform closed-loop calculations locally in millisecond cycles, leveraging the global optimization computing power of the cloud while ensuring the low latency and real-time feedback characteristics necessary for industrial control. Attached Figure Description

[0020] Figure 1 This is a main flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention; Figure 2 This is a first sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention; Figure 3 This is a second sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention; Figure 4 This is a third sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention; Figure 5 This is a fourth sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention; Figure 6 The fifth sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in this embodiment of the invention; Figure 7 The sixth sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in this embodiment of the invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0023] This invention provides a cloud-edge collaborative energy device scheduling method, aiming to solve the technical problems in existing industrial energy systems, such as the lack of global optimization capabilities, the inability to parse complex natural language scheduling intentions, and the potential for security breaches caused by directly issuing instructions from cloud-based AI large language models. Through a collaborative mechanism between the large language model and the edge controller, combined with a pre-set expert rule base and a digital twin model, this invention achieves the transformation from natural language intentions to structured optimization goals, and ensures the safety and reliability of industrial energy device group control through secondary security verification by the edge controller.

[0024] Please see Figure 1 , Figure 1 This is a main flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention. Figure 1 As shown, the cloud-edge collaborative energy device scheduling method includes the following steps: Step S101: Obtain real-time operating data of the industrial energy equipment group, input the scheduling intention in natural language form into the preset large language model, and output the structured optimization target through the large language model; In this step, acquiring real-time operational data of the industrial energy equipment cluster is a fundamental prerequisite for all subsequent optimization and scheduling decisions. Specifically, the industrial energy equipment cluster can include, but is not limited to, various underlying hardware devices deployed in industrial parks, data centers, chiller plants, air compressor stations, energy storage systems, or virtual power plants, such as air compressors, chiller units, water pumps, fans, energy storage battery cabinets, frequency converters, and programmable logic controllers (PLCs). Through industrial Internet of Things (IoT) sensing modules deployed at the industrial site, the system continuously collects real-time operational data from the industrial energy equipment cluster. This real-time operational data includes multi-dimensional physical quantities and status parameters, such as electrical data (voltage, current, active power, reactive power), fluid data (water flow rate, gas flow rate, steam pressure), environmental data (ambient temperature, ambient humidity), equipment status data (start / stop status, fault codes, frequency converter operating frequency), and energy storage status data (state of charge, SOC). During data acquisition, the underlying communication protocol can adopt standard industrial communication protocols such as Modbus, MQTT, BACnet, OPC UA, or IEC104. The collected raw real-time operational data is cleaned and filtered for outliers by the edge gateway before being uploaded to the cloud server for unified storage and analysis.

[0025] Simultaneously, the system provides a human-machine interface to receive scheduling intentions in natural language input by operators. These natural language scheduling intentions refer to text or voice information containing unstructured business requests or control expectations, such as "reduce energy consumption during tonight's peak hours while ensuring production is not affected," "prioritize energy storage charging during low-price periods," or "reduce overall carbon emissions from the air compressor station." These natural language scheduling intentions are input into a pre-set large language model. This model, based on a neural network architecture fine-tuned for industrial energy corpora, performs natural language processing (NLP) semantic parsing and entity recognition on the input natural language. The large language model does not directly output action commands for the underlying equipment (such as "start air compressor number 1"), but instead maps the unstructured text into a computer-readable structured data format through slot filling and intention classification mechanisms, i.e., outputting the structured optimization objective. This structured optimization objective provides clear variable definitions, optimization directions, and scope of action for subsequent mathematical model construction.

[0026] Step S102: Based on the real-time operating data and the preset expert rule base, determine the industrial safety constraints corresponding to the structured optimization objective; In this step, to compensate for the lack of physical boundary awareness in large language models within the industrial control field, a pre-defined expert rule base must be introduced to define the boundaries of the structured optimization objectives. This pre-defined expert rule base is stored on a cloud server and contains a large number of logical rules built upon industrial physical laws, equipment mechanical characteristics, and process safety standards. The pre-defined expert rule base includes, but is not limited to, equipment start-up and shutdown constraint rules, pipeline pressure upper and lower limit rules, motor frequency limit rules, battery SOC boundary rules, equipment priority rules, and production cycle time constraint rules.

[0027] Based on the currently acquired real-time operational data and the specific physical scenario involved in the structured optimization objective, the system extracts and determines the corresponding industrial safety constraints from the preset expert rule base. These industrial safety constraints refer to the set of physical boundaries that decision variables must satisfy during mathematical optimization. For example, for an energy storage system, the industrial safety constraints include not only the upper and lower limits of solid-state battery capacity but also the maximum charge / discharge power constraint dynamically related to the current battery temperature (provided by real-time operational data). For an air compressor system, the industrial safety constraints include that the terminal pipeline pressure must not be lower than the minimum pressure threshold required by the process (e.g., 0.62 MPa), the main pipe pressure must not exceed the maximum pressure threshold of the safety relief valve (e.g., 0.85 MPa), and the continuous start-stop interval of a single air compressor must not be lower than the preset safety protection time (e.g., 10 minutes). Through this step, the abstract optimization objective is anchored within a safe physical operating space.

[0028] Step S103: Based on the structured optimization objective and the industrial safety constraints, construct a multi-objective optimization function, and generate an initial scheduling scheme by solving the multi-objective optimization function; In this step, based on the optimization direction determined by the structured optimization objective output in step S101 (e.g., minimizing electricity costs, minimizing carbon emissions, etc.) and the feasible region determined by the industrial safety constraints in step S102, the system constructs a rigorous mathematical programming model on the cloud server, namely the multi-objective optimization function. The multi-objective optimization function is typically a nonlinear programming or mixed-integer programming problem, whose objective is to optimize the overall system efficiency by adjusting the operating parameters of each industrial energy device (as decision variables).

[0029] After constructing the multi-objective optimization function, the system invokes a preset mathematical solver (such as a linear programming algorithm, a mixed-integer programming algorithm, a reinforcement learning algorithm, or a model predictive control algorithm (MPC)) to iteratively optimize the multi-objective optimization function under the strict constraints of the industrial safety conditions. The solver searches for one or more sets of optimal control quantity sequences in the multi-dimensional state space (e.g., power setpoints and start / stop action times for each device within a future period). This set of optimal control quantity sequences constitutes the initial scheduling scheme. The initial scheduling scheme macroscopically satisfies the scheduling intent in the form of natural language, but microscopically it has not yet undergone safety blocking tests by the physical execution engine.

[0030] Step S104: Input the initial scheduling scheme into the preset digital twin model for simulation pre-run, obtain the simulation execution results, and determine whether the simulation execution results meet the preset safety boundaries; In this step, to completely avoid the destructive risks caused by directly issuing calculation results without on-site verification, this invention introduces a digital twin simulation mechanism. The preset digital twin model is a high-fidelity virtual physical model built in the cloud for the industrial energy equipment group, which includes the thermodynamic model, electrodynamic model, and fluid dynamics model of the equipment. The system uses the initial scheduling scheme generated in step S103 as a virtual input and injects it into the preset digital twin model. The preset digital twin model performs simulation pre-runs in virtual space using an accelerated clock or a real-time clock to simulate the dynamic response process of the industrial energy equipment group after receiving the scheduling scheme, including predicting the future fluctuation trend of pipeline pressure, the temperature rise curve of energy storage battery, and the power flow changes of distribution network nodes, thereby obtaining simulation execution results containing various predicted state characteristics.

[0031] After obtaining the simulation execution results, the system rigorously compares them with preset safety boundaries to determine whether the simulation execution results meet the preset safety boundaries. The preset safety boundaries are a set of allowable extreme operating state thresholds for the industrial system (such as the highest safe temperature, the maximum short-circuit current, the minimum gas supply pressure, etc.). If the determination result is that the preset safety boundaries are not met (for example, the simulation pre-run shows that the scheme will cause the exhaust temperature of a compressor to exceed the safety red line of 105°C after 15 minutes), the initial scheduling scheme will be immediately blocked and discarded in the cloud. The system will trigger the solver to readjust the constraint weights and generate a new scheduling scheme again to ensure that no control commands with out-of-bounds risks are issued to the underlying hardware.

[0032] Step S105: If satisfied, the initial scheduling scheme is encapsulated into a structured policy task and sent to the edge controller, so that the edge controller performs a secondary security check on the structured policy task based on the local real-time state, and generates a control command sequence after the check passes, and controls the industrial energy equipment group to execute the control command sequence.

[0033] In this step, if the judgment result of step S104 meets the preset security boundary, the system will not directly send raw data or natural language text to the field device. Instead, it will standardize the data encapsulation of the initial scheduling scheme and convert it into a structured strategy task. The structured strategy task is organized using an industry-standard data format and includes multi-field task description attributes, rather than a simple device power-on / off message. After encapsulation, the cloud server sends the structured strategy task to the edge controller deployed in the industrial field through the industrial IoT channel.

[0034] Upon receiving the structured strategy task, the edge controller will not execute it unconditionally. Due to communication delays between cloud simulation and the delivery process, and the possibility of sudden anomalies in the industrial field, the edge controller must perform a secondary safety check on the structured strategy task based on millisecond-level local real-time status data collected at the edge (such as hardware interlock alarms just triggered by sensors, local manual emergency stop signals, etc.). Only after the secondary safety check confirms that the current underlying hardware status fully allows the execution of the task will the edge controller decompose and convert the macroscopic structured strategy task into a sequence of control instructions that can be recognized and executed by various specific devices at the underlying level (for example, converting the macroscopic "target pressure maintained at 0.7MPa" into "set frequency to 42.5Hz" for the inverter and "close contactor to load operation" for a fixed-frequency machine). Finally, the edge controller sends the control instruction sequence to each actuator of the industrial energy equipment group through the local industrial bus, controlling the industrial energy equipment group to execute the control instruction sequence, thereby completing a complete cloud-edge collaborative closed-loop scheduling process. This mechanism achieves a deep integration of global optimization computing power and underlying physical security control through a dual blocking mechanism of "cloud-based digital twin pre-simulation" and "edge-based local secondary verification".

[0035] Please see Figure 2 , Figure 2This is a first sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention. In a preferred embodiment of the present invention, in step S101, the structured optimization objective includes: intent_type, asset_scope, time_window, primary_objective, secondary_objectives, hard_constraints, soft_constraints, and decision_variables. like Figure 2 As shown, the step of outputting a structured optimization objective through the large language model specifically includes the following sub-steps: Step S201: Extract the time information from the scheduling intent using the large language model, and determine the time window by combining it with the preset time-of-use electricity price data; In this step, to transform the time concept from natural language into a precise control cycle, the system utilizes the entity extraction capability of the large language model to identify time-related terms (such as "tonight's peak hours") in the scheduling intent (e.g., "reduce air compressor station energy consumption during tonight's peak hours"). The system performs semantic alignment and knowledge retrieval with the extracted time-of-use electricity price data (including the specific time period distribution of peak, flat, and valley electricity prices) pre-set by the power grid and accessed to the cloud. Through mapping and matching, the system converts the ambiguous time description into a specific timestamp interval, thereby determining the time window (e.g., defined as 17:00 to 20:00). This time window constitutes the integration interval and time domain boundary for subsequent optimization solutions.

[0036] Step S202: Identify the device semantics in the scheduling intent through the large language model, match the target control device in the industrial energy equipment group, and determine the asset scope; In this step, the large language model identifies the physical objects of interest to the user (e.g., "air compressor station") by performing semantic dependency analysis on the scheduling intent. Combining this with a pre-built semantic knowledge graph of industrial equipment assets, the large language model maps the physical objects to a real-world equipment ledger database, identifying all target control devices associated with the physical object (e.g., identifying specific devices associated with "air compressor station" including: No. 1 variable frequency air compressor, No. 2 fixed frequency air compressor, refrigerated dryer, air tank, main pressure sensor, and electricity meter). This set constitutes the asset scope, defining the list of devices whose status can be accessed and retrieved when generating subsequent scheduling schemes, preventing the large language model from overstepping its authority and interfering with devices in irrelevant areas when generating targets.

[0037] Step S203: Map the scheduling intention into a mathematical expression for the decision variables to generate the main optimization objective and the hard constraints.

[0038] In this step, the large language model, based on a pre-defined transformation template mechanism, transforms user requests into standardized structural parameters. First, based on the physical controllable attributes of equipment within the asset's scope, the decision variables (e.g., the start / stop status, loading status, operating frequency, or pressure setting target of each air compressor) are determined. Then, according to the core request of the scheduling intent (e.g., "reduce energy consumption"), it is mapped to a mathematical minimization function description, generating the primary optimization objective (e.g., "minimize electricity cost," i.e., minimize overall electricity cost). Simultaneously, for the safeguard context in the scheduling intent (e.g., "under the condition that production is not affected"), the large language model, through associating historical data tools and expert rules, transforms it into mathematical inequality boundaries, generating the hard constraints (e.g., "end pressure P_end ≥ 0.62MPa" and "main pressure P_main ≤ 0.85MPa"). Through the above sub-steps, the completely unstructured natural language is thoroughly deconstructed into rigorous, structured variables that do not contain any colloquial elements, providing standard input for optimization solutions.

[0039] Please see Figure 3 , Figure 3 This is a second sub-flowchart of a cloud-edge collaborative energy equipment scheduling method provided in an embodiment of the present invention. In a preferred embodiment of the present invention, in step S103, to take into account multiple dimensions of industrial operation indicators, the multi-objective optimization function includes an electricity cost item, a carbon emission cost item, an equipment start-up / shutdown penalty item, and a target deviation penalty item; as shown... Figure 3As shown, the construction of a multi-objective optimization function based on the structured optimization objective and the industrial safety constraints specifically includes the following sub-steps: Step S301: Obtain the predicted time-of-use electricity price and predicted carbon emission factor within the time window; In this step, before constructing the objective function, the system first obtains the sequence of external environmental parameters corresponding to the time window determined in step S201 from the energy data service interface in the cloud. Specifically, this includes: the predicted time-of-use electricity price data (C_e(t)) corresponding to each control step (e.g., every 15 minutes) within the time window, and the predicted carbon emission factor data (EF(t)) for the power grid region during that time period. These time series data are the fundamental multipliers for calculating economic and environmental costs.

[0040] Step S302: Construct the electricity cost item based on the predicted time-of-use electricity price and the grid-connected power of the industrial energy equipment group; In this step, the system uses the grid-connected power (P_grid(t)) drawn from the grid by the industrial energy equipment group as the benchmark decision-related variable. It constructs the electricity cost term by integrating or discretely summing the product of the predicted time-of-use price (C_e(t)) and the grid-connected power at each moment within the time window along the time dimension. The specific mathematical expression can be represented as: \sum [C_e(t) * P_{grid}(t) * \Delta t]. This term aims to guide the solution algorithm to minimize the grid-connected power of the equipment during peak periods with higher time-of-use prices, thereby achieving the economic goal of peak shaving and valley filling.

[0041] Step S303: Construct the carbon emission cost item based on the predicted carbon emission factor and the grid-connected power; In this step, in addition to economic indicators, the system also needs to consider environmental benefits. Based on the predicted carbon emission factor (EF(t)) and the grid-connected power, the system calculates the indirect carbon emissions corresponding to the electrical energy consumed by the equipment, thereby constructing the carbon emission cost item. The specific mathematical expression can be represented as follows: When the scheduling intent includes a sub-optimal objective of reducing carbon emissions, this factor will play a role in the optimization process, prompting the system to use more electricity during periods when the grid's renewable energy absorption rate is high and the carbon emission factor is low, and to use less electricity during periods when the carbon emission factor is high.

[0042] Step S304: The electricity cost item, the carbon emission cost item, the equipment start-up and shutdown penalty item, and the target deviation penalty item are weighted and summed to obtain the multi-objective optimization function.

[0043] In this step, in addition to the aforementioned electricity costs and carbon emissions, to protect the mechanical life of industrial equipment and ensure process stability, the system introduces the equipment start-up / shutdown penalty (used to constrain the daily start-up / shutdown frequency or frequent switching actions of a single piece of equipment) and the target deviation penalty (used to constrain the sum of squared deviations between the actual output pressure / temperature and the target setpoint). The system assigns corresponding weighting coefficients (e.g., ...) to the above four indicators. , , , And, through linear weighted summation, these multiple sub-objectives with mutually interacting game relationships are merged into a unified scalar cost function, namely the multi-objective optimization function. A complete mathematical example is as follows:

[0044] In solving this function, constraints such as the dynamic constraints of the energy storage SOC must be strictly satisfied. ) and dynamic constraints on air pressure ( Hard constraints such as these are applied. Through this step, the system thoroughly mathematizes the complex scheduling intentions, transforming the industrial control problem into a numerical optimization problem that can be precisely solved using operations research algorithms.

[0045] Please see Figure 4 , Figure 4 This is a third sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention. In a preferred embodiment of the present invention, in step S105, to ensure the data integrity, idempotency, and traceability of data sent from the cloud in unreliable public networks or complex factory LANs, the structured strategy task includes a unique task identifier (task_id), command mode (command_mode), target curve (target_profile), control boundary constraints (constraints), and digital signature (signature); the command mode includes at least one of the following: target curve mode, control command sequence mode, boundary constraint update mode, and emergency stop mode; such as... Figure 4 As shown, the step of encapsulating the initial scheduling scheme into a structured policy task and distributing it to the edge controller specifically includes the following sub-steps: Step S401: Convert the initial scheduling scheme into the data payload corresponding to the instruction mode according to the preset communication protocol format; In this step, after generating the initial scheduling scheme, the cloud server first extracts and organizes the initial scheduling scheme, which contains multi-dimensional vectors, into a data payload with a specific structure according to a preset communication protocol specification (such as MQTT protocol or RESTful API specification). The system selects an appropriate instruction mode based on the control depth of the initial scheduling scheme. If the edge device has a high degree of autonomous adjustment capability, the "target curve mode" is selected, that is, the target pressure curve or target SOC curve on the future time series is sent as the data payload, and the edge end calculates the underlying actions itself; if strict control of the action sequence of each device is required, the "control instruction sequence mode" is selected, and a clear power-on / off timing table is used as the data payload; if only the on-site alarm threshold needs to be adjusted, the "boundary constraint update mode" is adopted. At the same time, a globally unique task identifier (task_id) and a link tracing identifier (trace_id) are entered into the data payload to ensure the anti-replay and traceability characteristics of the sent tasks.

[0046] Step S402: Sign the data payload using a preset encryption algorithm to generate the digital signature; In this step, to prevent the structured policy task from being tampered with by malicious man-in-the-middle attacks or subjected to network hijacking and injection of forged control commands during distribution, the cloud server must digitally sign the generated data payload. The system calls the cloud's key management service to obtain the private key of a preset asymmetric encryption algorithm (such as RSA or ECC), performs a hash calculation on the complete data payload containing the task's unique identifier, command mode, and specific control data, and then encrypts the hash digest using the private key to generate the task-specific digital signature. This digital signature is appended to the end field of the structured policy task.

[0047] Step S403: The structured policy task with the digital signature is sent to the edge controller through a preset encrypted communication link.

[0048] In this step, the cloud server uses a pre-defined encrypted communication link (e.g., an MQTT over TLS secure channel based on TLS 1.2 / 1.3 protocol, or a two-way authentication channel based on HTTPS) to send the constructed structured policy task with the digital signature to the edge controller at the target industrial site. During this transmission, it is preferable to configure the MQTT protocol communication quality level to QoS 1 (at least once) or QoS 2 (only once) to ensure that the scheduling policy task is not lost or incorrectly retransmitted in complex industrial weak network environments.

[0049] Please see Figure 5 , Figure 5This is a fourth sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention. In a preferred embodiment of the present invention, in step S104, to ensure that the digital twin pre-simulation can accurately identify and intercept potentially dangerous strategies; as... Figure 5 As shown, determining whether the simulation execution result meets the preset safety boundary specifically includes the following sub-steps: Step S501: Analyze the predicted state features in the simulation execution results, wherein the predicted state features include at least one of predicted pressure, predicted temperature, predicted current, and predicted state of charge; In this step, after the preset digital twin model completes the simulation of the initial scheduling scheme in virtual space, it outputs a simulation execution result containing a large amount of virtual time-series data. The system performs structured analysis on this simulation execution result, extracting a sequence of key indicators characterizing the physical safety status of the industrial equipment group, i.e., the predicted state features. For example, for an air compressor system, the predicted pressure curve of the pipeline network and the predicted exhaust temperature curve of the compressor host are extracted; for an energy storage system, the predicted current curve of the PCS inverter side and the predicted state of charge (SOC) curve of the battery cluster are extracted. These predicted state features can intuitively reflect the physical response of the equipment over a period of time in the future if the scheduling scheme is executed in the real physical world.

[0050] Step S502: Compare the predicted state features with the pre-established hard constraint threshold and soft constraint threshold respectively; In this step, the system retrieves the equipment safety operation standard benchmark stored in the cloud. This benchmark divides the operating boundaries into absolutely insurmountable hard constraint thresholds (such as equipment overload current limits and destructive overvoltage discharge thresholds) and soft constraint thresholds that are not suitable for long-term operation (such as temperatures slightly above the optimal economic operating range and long-term full-load currents that accelerate equipment insulation aging). The system scans the predicted state characteristics point by point on the time axis and compares and analyzes their values ​​with the hard and soft constraint thresholds one by one.

[0051] Step S503: If any of the predicted state features exceeds the hard constraint threshold, it is determined that the safety boundary is not met, and the issuance instruction of the initial scheduling scheme is intercepted. In this step, during the comparison process, if the system finds that the value of any of the predicted state features at any given time exceeds the absolute safety range defined by the hard constraint threshold (for example, simulation shows that the battery SOC will drop to 18% at the 15th minute of executing the optimization strategy, exceeding the 20% hard constraint lower limit), the system immediately determines that the initial scheduling scheme poses a serious risk of causing an industrial production accident, and the conclusion is that it does not meet the safety boundary. Once this conclusion is reached, the system's internal security gateway mechanism will immediately trigger the interception logic, blocking the cloud-based instruction for the initial scheduling scheme, and absolutely preventing it from entering the encapsulation and distribution process of step S105.

[0052] Step S504: If all the predicted state features are within the hard constraint threshold and there are features that exceed the soft constraint threshold, then trigger the downgrade execution strategy or manual review process.

[0053] In this step, if the system confirms that all predicted state characteristics remain within the safe range of the hard constraint threshold during the comparison process, it indicates that the scheme does not pose an extreme risk of causing a safety accident. However, if the system finds that some characteristics deviate from the ideal operating range and exceed the soft constraint threshold (for example, the predicted operating temperature of the motor reaches 85°C, which is between normal (80°C) and alarm (90°C)), directly issuing the scheme may lead to increased equipment energy consumption or long-term lifespan damage. Therefore, the system automatically triggers a degradation execution strategy (for example, automatically lowering the inverter frequency setting in the scheme by a certain percentage to alleviate the temperature rise), or suspends the initial scheduling scheme and sends a manual review request with a warning message to the maintenance personnel's client. The scheme is then issued only after manual confirmation. This hierarchical verification mechanism effectively balances the system's automation level and operational robustness.

[0054] Please see Figure 6 , Figure 6 This is the fifth sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention. In a preferred embodiment of the present invention, in step S105, the edge controller's ability to ensure on-site control safety even in the event of network outages or communication delays depends on its internally constructed three-layer real-time control architecture; such as... Figure 6 As shown, the edge controller performs a secondary security check on the structured policy task based on the local real-time state, and generates a control command sequence after the check passes, specifically including the following sub-steps: Step S601: Obtain the structured policy task and verify the validity period and signature authorization status of the structured policy task; In this step, after receiving the structured policy task, the task parsing and security verification layer of the edge controller first reads the timestamp and expiration time fields in the task header to verify whether the structured policy task is within its validity period (to avoid control chaos caused by executing delayed policies that have expired due to network latency). Simultaneously, the edge controller uses its locally stored public key to decrypt and compare the hash value of the digital signature attached to the end of the structured policy task to verify its signature authorization status, ensuring that the task was indeed issued by a legitimate cloud platform and has not been tampered with during transmission. Only after both the validity period and signature have been verified can the next step be performed.

[0055] Step S602: Obtain the local real-time status of the industrial energy equipment group, including the underlying hardware interlock status, equipment alarm status, and manual takeover status. In this step, even if signature verification passes, the edge controller will not blindly trust the cloud policy. Because industrial environments are constantly changing, the edge controller must acquire the current local real-time status of the industrial energy equipment group via an underlying fieldbus (such as high-frequency scanning Profibus or real-time industrial Ethernet) with a scan cycle of milliseconds or seconds. This local real-time status includes key safety indicators that directly reflect whether the equipment can operate, specifically: underlying hardware interlock status (such as relay closure status, flow switch status), equipment alarm status (such as primary and secondary fault alarm codes of drives), and manual takeover status (such as whether the knobs on the field control panel are set to "local manual" or "emergency stop"). These statuses are blind spots where cloud-based digital twins struggle to achieve millisecond-level synchronization and must be verified at the edge.

[0056] Step S603: When the local real-time state meets the execution conditions, calculate the target control deviation based on the target curve in the structured strategy task and the local real-time state; In this step, the edge controller comprehensively compares the cloud strategy intent parsed in step S601 with the local real-time status obtained in step S602. If a manual emergency stop signal, a broken underlying water flow interlock, or a first-level alarm state is detected locally, the execution conditions are determined not to be met, the cloud task is directly rejected, and a rejection feedback is returned to the cloud. Only when the local real-time status is completely normal and all execution conditions are met, the supervisory control layer of the edge controller will extract the target curve set in the structured strategy task (e.g., the target water supply pressure curve P_target(t) within the next 10 minutes) and obtain the currently collected feedback value locally (e.g., the current actual pressure of the main pipeline P_actual(k)). The system calculates the target control deviation (error e(k) = P_target(k) - P_actual(k)) for the current control cycle by subtracting the expected value set for the target curve from the current actual value locally.

[0057] Step S604: Using a preset local closed-loop control algorithm, the target control deviation is converted into a sequence of control commands for each device in the industrial energy equipment group.

[0058] In this step, the edge controller invokes its internally integrated preset local closed-loop control algorithm to perform rapid dynamic adjustment calculations on the target control deviation calculated in step S603. The local closed-loop control algorithm can be a classic proportional-integral-derivative (PID) control algorithm, a fuzzy logic control algorithm, or a rule-based logic allocation algorithm. Taking air compressor pressure PID control as an example, the edge controller calculates the frequency adjustment based on the deviation e(k), and the formula can be expressed as: ,in This function ensures that the output regulation is strictly limited to the minimum frequency allowed by the inverter. and maximum frequency Between. The specific adjustment amount is calculated. After the device makes a loading / unloading decision, the device driver layer of the edge controller is responsible for translating these abstract control quantities into communication messages that the underlying hardware can recognize (such as writing a hexadecimal value to a register address), thereby generating the control command sequence and sending the sequence to the field PLC or frequency converter for execution. This process is completely closed-loop at the edge, with fast response speed and strong resistance to network fluctuations.

[0059] Please see Figure 7 , Figure 7 This is the sixth sub-flowchart of a cloud-edge collaborative energy device scheduling method provided in an embodiment of the present invention. As a preferred embodiment of the present invention, in order to form an adaptive iterative closed loop for cloud-edge collaboration, such as... Figure 7 As shown, the method further includes the following sub-steps: Step S701: During the process of the edge controller controlling the industrial energy equipment group to execute the control command sequence, the feedback status of the industrial energy equipment group is continuously monitored; In this step, the scheduling process does not end after the control command sequence is sent to the underlying hardware. The edge controller enters a supervisory operation mode and continues to monitor the physical response results and feedback status of the industrial energy equipment group after executing the control commands at a set high-frequency sampling period. This continuous monitoring ensures that the system has the ability to perceive execution deviations and sudden disturbances in real time.

[0060] Step S702: If the feedback state is detected to trigger a hard interlock condition or an emergency stop command is received, the edge controller cancels the control command sequence and executes a safe shutdown of the device or switches to a preset conservative operation strategy. In this step, if during the execution of the cloud strategy, a sudden change in on-site conditions (e.g., a sudden mechanical jamming of a device causing a surge in current) triggers the device's underlying self-protection logic (triggering a hard interlock condition), or if an on-site operator presses the emergency stop button (receiving an emergency stop command), the edge controller will immediately interrupt the execution of the current strategy, revoking remote control rights from the cloud. The edge controller will issue a highest-priority reset command to cancel the previously generated control command sequence and force the device to execute a safe shutdown sequence, or switch back to a preset conservative operating strategy that does not rely on external networks and cloud computing power (e.g., maintaining minimum safe load operation), thereby ensuring absolute safety in the industrial site.

[0061] Step S703: Feed back the status data and failure reason after the undo operation to the cloud server to trigger the cloud server to update the parameters of the multi-objective optimization function.

[0062] In this step, to enable the system to continuously learn and correct errors, after executing the aforementioned safety reversal or rollback operation, the edge controller packages the rollback action (rolled_back), failure reason code (failure_reason), and detailed state data before and after the reversal operation into a policy execution feedback message, which is then uploaded to the cloud server via an MQTT topic. Upon receiving this feedback, the cloud server analyzes the underlying cause of the hard-locking event and accordingly updates the parameters of the penalty term weights in the multi-objective optimization function, or adds a safety prohibition boundary for this type of situation to the expert rule base. Through this operational feedback and learning mechanism, the system achieves dynamic adaptive adjustment of the cloud model parameters, making the subsequently generated initial scheduling scheme more closely aligned with the actual constraints of the industrial site, gradually improving the overall scheduling robustness and success rate.

[0063] This invention also provides a cloud-edge collaborative energy device scheduling device. The device can be implemented in software, hardware, or a combination of both. The specific logical design of each functional module of the device corresponds one-to-one with the steps in the aforementioned method embodiments, achieving the same beneficial technical effects as the method embodiments.

[0064] Specifically, the device includes: The intent parsing module is used to acquire real-time operating data of the industrial energy equipment group and input the scheduling intent in natural language form into a preset large language model. The large language model then outputs a structured optimization objective. The internal logic of this module corresponds to step S101 in the method of the embodiment, which uses the natural language processing and entity extraction functions of the large language model to transform the abstract objective into structured parameters.

[0065] The constraint generation module is used to determine the industrial safety constraints corresponding to the structured optimization objective based on the real-time operating data and a preset expert rule base. This module corresponds to step S102, defining a feasible set of physical boundaries for subsequent mathematical optimization.

[0066] The optimization and solution module is used to construct a multi-objective optimization function based on the structured optimization objective and the industrial safety constraints, and to generate an initial scheduling scheme by solving the multi-objective optimization function. This module corresponds to step S103 and integrates mathematical solution algorithms such as linear programming and mixed integer programming, and is responsible for calculating the optimal control quantity that satisfies the constraints.

[0067] The digital twin pre-simulation module is used to input the initial scheduling scheme into a preset digital twin model for simulation pre-simulation, obtain the simulation execution results, and determine whether the simulation execution results meet the preset security boundaries. This module corresponds to step S104 and serves as the last line of defense in the cloud, intercepting any potentially dangerous strategies that could lead to physical state out-of-bounds violations.

[0068] The cloud-edge collaborative execution module, when the security boundary is met, encapsulates the initial scheduling scheme into a structured policy task and sends it to the edge controller. The edge controller then performs a secondary security check on the structured policy task based on its local real-time state. Upon successful check, it generates a control command sequence to control the industrial energy equipment group to execute the control command sequence. This module corresponds to step S105, realizing the secure landing and closed-loop execution of cloud-based policies at the edge.

[0069] This invention also provides an electronic device, which can be a computing server cluster deployed in the cloud or an industrial-grade control computer integrated at the edge of an industrial site. The electronic device includes a processor and a memory.

[0070] The memory is used to store computer programs and data. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0071] The processor is the control center of the electronic device, connecting various parts of the entire electronic device through various interfaces and lines. When the computer program is executed by the processor, it can implement all the steps of the cloud-edge collaborative energy device scheduling method described in any of the above embodiments (such as steps S101 to S105), realizing the overall logic of large language model intent parsing, multi-objective optimization, twin simulation pre-playing, and edge security control.

[0072] This invention also provides a computer-readable storage medium storing a computer program. The computer-readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. When the computer program is executed by a processor with data processing capabilities, it can guide the processor to complete various process operations (such as steps S101 to S105 and each sub-step S201 to S703) of the cloud-edge collaborative energy device scheduling method provided in the above embodiments. This storage medium provides physical carrier support for the software distribution and industrial deployment of the large language model and edge controller collaborative control technology provided by this invention.

[0073] In summary, the specific implementation method provided by this invention, through a dual security blocking mechanism of "cloud-based digital twin pre-simulation" and "edge-end physical interlock verification," successfully integrates the intelligent advantages of large language models at the natural language semantic parsing level with the robust advantages of edge controllers at the underlying high-frequency millisecond-level closed-loop control level. This architecture not only solves the technical problems of existing industrial energy systems lacking multi-dimensional factor joint optimization and being unable to handle complex scheduling intentions, but also reduces the inherent output illusion risk and control overreach risks of large language models. It provides a highly secure, low-latency, and continuously feedback-iterative intelligent scheduling technology solution for energy equipment in scenarios such as industrial parks, energy storage systems, and air compressor stations.

Claims

1. A cloud-edge collaborative energy equipment scheduling method, characterized in that, The method includes: The system acquires real-time operating data of industrial energy equipment clusters and inputs scheduling intentions in natural language form into a preset large language model, which then outputs structured optimization objectives. Based on the real-time operating data and the preset expert rule base, determine the industrial safety constraints corresponding to the structured optimization objective; Based on the structured optimization objective and the industrial safety constraints, a multi-objective optimization function is constructed, and an initial scheduling scheme is generated by solving the multi-objective optimization function. The initial scheduling scheme is input into a preset digital twin model for simulation, the simulation execution results are obtained, and it is determined whether the simulation execution results meet the preset safety boundaries. If the conditions are met, the initial scheduling scheme is encapsulated into a structured policy task and sent to the edge controller, so that the edge controller performs a secondary security check on the structured policy task based on the local real-time state, and generates a control command sequence after the check passes, and controls the industrial energy equipment group to execute the control command sequence.

2. The cloud-edge collaborative energy equipment scheduling method according to claim 1, characterized in that, The structured optimization objectives include: intent type, asset scope, time window, primary optimization objective, secondary optimization objective, hard constraints, soft constraints, and decision variables; The output of the structured optimization objective through the large language model includes: The time information in the scheduling intent is extracted by the large language model, and the time window is determined by combining it with the preset time-of-use electricity price data; The large language model is used to identify the device semantics in the scheduling intent, match the target control device in the industrial energy equipment group, and determine the asset scope. The scheduling intent is mapped into a mathematical expression for the decision variables, thereby generating the main optimization objective and the hard constraints.

3. The cloud-edge collaborative energy equipment scheduling method according to claim 1, characterized in that, The multi-objective optimization function includes an electricity cost item, a carbon emission cost item, an equipment start-up and shutdown penalty item, and an objective deviation penalty item. The construction of a multi-objective optimization function based on the structured optimization objective and the industrial safety constraints includes: Obtain the predicted time-of-use electricity price and predicted carbon emission factor within the time window; The electricity cost item is constructed based on the predicted time-of-use electricity price and the grid-connected power of the industrial energy equipment group; The carbon emission cost item is constructed based on the predicted carbon emission factor and the grid-connected power. The multi-objective optimization function is obtained by weighted summing of the electricity cost item, the carbon emission cost item, the equipment start-up and shutdown penalty item, and the target deviation penalty item.

4. The cloud-edge collaborative energy equipment scheduling method according to claim 1, characterized in that, The structured strategy task includes a unique task identifier, instruction pattern, target curve, control boundary constraints, and digital signature. The instruction mode includes at least one of the following: target curve mode, control instruction sequence mode, boundary constraint update mode, and emergency stop mode; The step of encapsulating the initial scheduling scheme into a structured policy task and distributing it to the edge controller includes: According to the preset communication protocol format, the initial scheduling scheme is converted into the data payload corresponding to the instruction mode; The data payload is signed using a preset encryption algorithm to generate the digital signature; The structured policy task with the digital signature is sent to the edge controller via a preset encrypted communication link.

5. The cloud-edge collaborative energy equipment scheduling method according to claim 1, characterized in that, The step of determining whether the simulation execution result meets the preset safety boundary includes: The predicted state features in the simulation execution results are analyzed, and the predicted state features include at least one of predicted pressure, predicted temperature, predicted current, and predicted state of charge. The predicted state features are compared with the pre-established hard constraint threshold and soft constraint threshold, respectively; If any of the predicted state features exceeds the hard constraint threshold, it is determined that the safety boundary is not met, and the issuance instruction of the initial scheduling scheme is intercepted. If all the predicted state features are within the hard constraint threshold and there are features that exceed the soft constraint threshold, then a downgrade execution strategy or manual review process will be triggered.

6. The cloud-edge collaborative energy equipment scheduling method according to claim 1, characterized in that, The edge controller performs a secondary security check on the structured policy task based on the local real-time state, and generates a control command sequence after the check passes, including: Obtain the structured policy task and verify its validity period and signature authorization status. The local real-time status of the industrial energy equipment group is obtained, including the underlying hardware interlock status, equipment alarm status, and manual takeover status. When the local real-time state meets the execution conditions, the target control deviation is calculated based on the target curve in the structured strategy task and the local real-time state. Using a preset local closed-loop control algorithm, the target control deviation is converted into a sequence of control commands for each device in the industrial energy equipment group.

7. The cloud-edge collaborative energy equipment scheduling method according to claim 6, characterized in that, The method further includes: During the process of the edge controller controlling the industrial energy equipment group to execute the control command sequence, the feedback status of the industrial energy equipment group is continuously monitored; If the feedback state is detected to trigger a hard interlock condition or an emergency stop command is received, the edge controller cancels the control command sequence and executes a safe shutdown of the device or switches to a preset conservative operation strategy. The status data and failure reason after the undo operation are fed back to the cloud server to trigger the cloud server to update the parameters of the multi-objective optimization function.

8. A cloud-edge collaborative energy equipment scheduling device, characterized in that, The device includes: The intent parsing module is used to acquire real-time operating data of industrial energy equipment groups and input scheduling intents in natural language form into a preset large language model, and output structured optimization objectives through the large language model; The constraint generation module is used to determine the industrial safety constraints corresponding to the structured optimization objective based on the real-time operating data and the preset expert rule base. The optimization solution module is used to construct a multi-objective optimization function based on the structured optimization objective and the industrial safety constraints, and to generate an initial scheduling scheme by solving the multi-objective optimization function; The twin simulation module is used to input the initial scheduling scheme into a preset digital twin model for simulation simulation, obtain the simulation execution results, and determine whether the simulation execution results meet the preset safety boundaries. The cloud-edge collaborative execution module is used to encapsulate the initial scheduling scheme into a structured policy task when the security boundary is met, and send it to the edge controller so that the edge controller can perform a secondary security verification on the structured policy task based on the local real-time status, and generate a control command sequence after the verification is passed, and control the industrial energy equipment group to execute the control command sequence.

9. An electronic device, characterized in that, Includes: processor and memory; The memory stores a computer program, which, when executed by the processor, implements the cloud-edge collaborative energy device scheduling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the cloud-edge collaborative energy device scheduling method as described in any one of claims 1 to 7.

Images