Dynamic monitoring system and method for carbon emission of heavy mechanical equipment

Abstract

The invention relates to the technical field of environmental protection monitoring, and discloses a dynamic monitoring system and method for carbon emission of heavy mechanical equipment, and the system comprises a multi-mode sensing terminal which collects multi-source equipment data in real time; the digital twin server is used for driving the energy flow digital twin model and correcting the energy flow digital twin model on line; the cognition and calculation module is used for recognizing the operation intention and calculating real-time carbon emission in a high-fidelity manner; the collaborative decision-making module is used for generating a carbon quota curve and generating a collaborative strategy under carbon constraint for the equipment cluster based on multi-agent reinforcement learning in a population digital twinning environment; and the strategy execution module visually presents the collaborative strategy to an operator. The method comprises the steps of data acquisition, twin model operation and self-evolution, real-time carbon emission fusion calculation, carbon quota generation and collaborative strategy optimization, and strategy issuing and presentation. According to the invention, through virtual-real fusion and intelligent decision making, the limitation of traditional monitoring is broken through, and accurate quantification and active control of carbon emission are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental protection monitoring, and in particular to a heavy machinery equipment carbon emission dynamic monitoring system and method. BACKGROUND

[0002] At present, with the increasing global attention to environmental protection and sustainable development, the carbon emission problem generated by heavy machinery equipment in the operation process is widely valued. The existing technology has significant limitations in carbon emission monitoring and management. Traditional methods often rely on emission estimation based on empirical factors or static working conditions, or obtain average emission data through periodic manual detection. These methods are difficult to reflect the dynamic carbon emission of heavy machinery equipment in complex and variable actual operation scenarios in real time and accurately.

[0003] Such non-real-time and rough carbon emission data makes it impossible for managers to obtain specific emission data of equipment under different operation modes, load conditions and health states in time, thus missing the opportunity to actively reduce carbon emissions by optimizing operation processes and adjusting operation behavior. In addition, the lack of consideration of individual differences and time-varying characteristics of equipment makes it difficult for traditional monitoring schemes to provide high-fidelity emission evaluation results, which not only affects the fine level of environmental management, but also hinders the scientificity and effectiveness of related decisions.

[0004] Therefore, the present application provides a heavy machinery equipment carbon emission dynamic monitoring system and method to solve the deficiencies of the prior art. SUMMARY

[0005] The purpose of the present application is to provide a heavy machinery equipment carbon emission dynamic monitoring system and method, which solves the problems of insufficient precision in carbon emission monitoring of heavy machinery equipment, lack of adaptive adjustment to dynamic working conditions of equipment, inability to prospectively plan emissions, and lack of collaborative optimization capability in multi-device operation scenarios.

[0006] To solve the above technical problems, the present application provides the following technical solutions.

[0007] The first aspect of the present application provides a heavy machinery equipment carbon emission dynamic monitoring system, which comprises a multi-modal perception terminal, a digital twin server, a cognitive and computing module, a collaborative decision-making module and a strategy execution module.

[0008] The multi-modal perception terminal is deployed on a heavy machinery equipment and is used to collect multi-source real-time data of the heavy machinery equipment. In one specific technical solution, the multi-modal perception terminal comprises: a core operating parameter collection unit configured to collect engine speed and load rate; a dynamic and kinematic parameter collection unit configured to collect angular velocity and acceleration of the heavy machinery equipment; a work load parameter collection unit configured to collect pressure and temperature of a hydraulic system; an acoustic feature parameter collection unit configured to collect acoustic signals during operation of the heavy machinery equipment; and an environmental parameter collection unit configured to collect environmental temperature and atmospheric pressure of a work area of the heavy machinery equipment.

[0009] The digital twin server is in communication connection with the multi-modal perception terminal, and is configured to drive a dedicated energy flow digital twin model to operate based on the multi-source real-time data, and to calculate a twin reality difference degree vector between a predicted physical state output by the energy flow digital twin model and a real physical state reflected by the multi-source real-time data.

[0010] In one preferred technical solution, after calculating the twin reality difference degree vector, the digital twin server is further configured to correct an internal parameter vector of the energy flow digital twin model based on the twin reality difference degree vector by using an online learning algorithm to drive the energy flow digital twin model to evolve automatically. The internal parameter vector comprises a combustion efficiency coefficient and a hydraulic system efficiency coefficient representing individual health conditions of the heavy machinery equipment. The online learning algorithm updates the internal parameter vector by the following steps: first, a loss function based on the twin reality difference degree vector is defined, and a calculation formula of the loss function is as follows: In the formula, L TR is a value of the loss function; Δ TR is the twin reality difference degree vector; represents a square of an L2 norm of the vector.

[0011] Then, based on the loss function, the gradient descent method is used to update the internal parameter vector, and an update formula of the internal parameter vector is as follows: In the formula, Θ twin,t+1 is the updated internal parameter vector; Θ twin,t is the internal parameter vector before update; and η is a learning rate. is a gradient of the loss function with respect to the internal parameter vector.

[0012] The cognition and computing module is used to process the multi-source real-time data to identify the operational intent of the heavy machinery, and to calculate the real-time carbon emissions by combining the operational intent, the energy flow digital twin model, and the twin reality difference vector. In a specific technical solution, the cognition and computing module employs a time-series deep learning model to identify the operational intent by analyzing the time series of the multi-source real-time data.

[0013] In a preferred embodiment, when the cognition and calculation module calculates the real-time carbon emissions, the calculation formula for the real-time carbon emissions is as follows: E c =G fuel (X,I,Θ twin )+g corr (Δ TR ); In the formula, E c For real-time carbon emissions; G fuel This is a basic emission calculation function, which uses multi-source real-time data X, operational intent I, and the internal parameter vector Θ of the energy flow digital twin model. twin For input; g corr An anomalous emission correction function is defined by the twin reality difference vector Δ. TR For input.

[0014] The collaborative decision-making module is used to generate a carbon quota curve based on macro-level operational tasks and by calling the energy flow digital twin model for simulation. Then, based on the carbon quota curve and the real-time carbon emissions, a collaborative operational strategy is generated. In a preferred embodiment, the collaborative decision-making module employs a multi-agent reinforcement learning algorithm to generate the collaborative operational strategy under the constraints of the carbon quota curve. The reward function of the multi-agent reinforcement learning algorithm is: R jt =w work ·ΔW-w time ·Δt-P penalty ; In the formula, R jt The combined reward value; ΔW is the effective work completed in a single time step; Δt is the duration of a single time step; w work For work efficiency weighting coefficient; w time P is the time cost weighting coefficient; penalty As a penalty item, when the cumulative carbon emissions of the heavy machinery exceed the upper limit of the carbon quota curve defined at the current moment, the penalty item is a preset positive value.

[0015] Further, the collaborative decision module is configured to connect the energy flow digital twin models of the multiple heavy machinery devices for collaborative work to form a group digital twin system when running the multi-agent reinforcement learning algorithm, and use the group digital twin system as a training environment of the multi-agent reinforcement learning algorithm.

[0016] The policy execution module is deployed on the heavy machinery device and configured to receive and present the collaborative work strategy generated by the collaborative decision module.

[0017] The second aspect of the present application provides a heavy machinery device carbon emission dynamic monitoring method, which is applied to the heavy machinery device carbon emission dynamic monitoring system described above, and includes the following steps: S1, collecting multi-source real-time data of the heavy machinery device; S2, driving a dedicated energy flow digital twin model based on the multi-source real-time data, and calculating a twin reality difference degree vector between a predicted physical state output by the energy flow digital twin model and a real physical state reflected by the multi-source real-time data; S3, processing the multi-source real-time data to identify the work intention of the heavy machinery device, and combining the work intention, the energy flow digital twin model, and the twin reality difference degree vector to calculate a real-time carbon emission amount; S4, generating a carbon quota curve by simulating the energy flow digital twin model based on a macro work task, and then generating a collaborative work strategy according to the carbon quota curve and the real-time carbon emission amount; S5, issuing the collaborative work strategy to a policy execution module deployed on the heavy machinery device for presentation.

[0018] In summary, the present application has at least one of the following beneficial technical effects: 1. The present application constructs a dedicated energy flow digital twin model for each heavy machinery device, calculates a twin reality difference degree vector between a predicted physical state output by the model and a real physical state reflected by multi-source real-time data, and then uses the vector to correct the internal parameter vector of the model online to realize the self-evolution of the model. This way makes the monitoring benchmark of carbon emission dynamically adapt to the individual health state changes of the device due to wear and aging, significantly improving the accuracy and reliability of carbon emission monitoring over a long time span.

[0019] 2. The application combines the identified equipment operation intention, the self-evolutionary energy flow digital twin model, and the twin reality difference vector as real-time disturbance compensation when calculating real-time carbon emissions. This calculation method breaks through the limitations of traditional methods that rely on only part of the operating parameters, establishing a calculation system that comprehensively reflects the theoretical energy consumption of equipment, the deviation between the driver's operation target and the actual operation, and obtaining high-fidelity real-time carbon emissions.

[0020] 3. The application can generate a carbon quota curve to guide the entire operation process by calling the energy flow digital twin model in the collaborative decision-making module to simulate the macro operation task in advance. This changes carbon emission management from passive and lagging monitoring to proactive and forward planning, providing a scientific quantitative basis and execution target for carbon emission total control in complex task scenarios, enhancing the forward-looking and planning of carbon management.

[0021] 4. The application connects the energy flow digital twin models of multiple devices to form a group digital twin system, and uses it as a training environment for multi-agent reinforcement learning algorithm, finally generating a collaborative operation strategy. This method goes beyond single-device optimization and can seek the optimal solution for operation efficiency and energy consumption from the global perspective of the device cluster while meeting the constraints of the carbon quota curve, achieving collaborative emission reduction among multiple devices and effectively reducing the total carbon footprint of cluster operation. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The figure is a structural diagram of the heavy machinery equipment carbon emission dynamic monitoring system of the application; Figure 2 The figure is a flow chart of the heavy machinery equipment carbon emission dynamic monitoring method of the application; Figure 3 The figure is a self-evolution process diagram of the energy flow digital twin model of the application; Figure 4 The figure is a schematic diagram of generating a collaborative operation strategy based on a carbon quota curve of the application; Figure 5 The figure is a schematic diagram of a carbon quota curve of the application.

[0023] Among them, 10, multi-modal perception terminal; 11, core operating parameter acquisition unit; 12, dynamic and kinematic parameter acquisition unit; 13, operation load parameter acquisition unit; 14, acoustic feature parameter acquisition unit; 15, environmental parameter acquisition unit; 20, digital twin server; 30, cognitive and computing module; 40, collaborative decision-making module; 50, strategy execution module. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings Figure 1 -Appendix Figure 5Further details of the present application are provided below.

[0025] Referring to Figure 1 The embodiment of the present application provides a heavy machinery equipment carbon emission dynamic monitoring system. The system comprises a multi-modal perception terminal 10, a digital twin server 20, a cognitive and computing module 30, a collaborative decision-making module 40 and a strategy execution module 50.

[0026] The multi-modal perception terminal 10 is deployed on the heavy machinery equipment, and its function is to collect multi-source real-time data of the heavy machinery equipment during the operation process. The multi-source real-time data is the data basis for all subsequent calculations and decisions.

[0027] The digital twin server 20 is in communication connection with the multi-modal perception terminal 10, and is used for receiving the multi-source real-time data. The function of the digital twin server 20 is to drive a dedicated energy flow digital twin model to run based on the multi-source real-time data, and to calculate a twin reality difference degree vector between the predicted physical state output by the model and the real physical state reflected by the multi-source real-time data.

[0028] The cognitive and computing module 30 is also in communication connection with the multi-modal perception terminal 10 to receive the multi-source real-time data, and is in communication connection with the digital twin server 20 to receive the twin reality difference degree vector. The function of the cognitive and computing module 30 is to process the multi-source real-time data to identify the operation intention of the equipment, and to combine the operation intention, the energy flow digital twin model and the twin reality difference degree vector to calculate the real-time carbon emission.

[0029] The collaborative decision-making module 40 is in communication connection with the digital twin server 20 and the cognitive and computing module 30. The function of the collaborative decision-making module 40 is to call the energy flow digital twin model to generate a carbon quota curve based on the macro operation task, and to generate a collaborative operation strategy according to the carbon quota curve and the real-time carbon emission calculated by the cognitive and computing module 30.

[0030] The strategy execution module 50 is deployed on the heavy machinery equipment, and is in communication connection with the collaborative decision-making module 40, and is used for receiving and presenting the collaborative operation strategy generated by the collaborative decision-making module 40 to the equipment operator.

[0031] Referring to Figure 2 Corresponding to the above system, the present application also provides a heavy machinery equipment carbon emission dynamic monitoring method. The technical logic of the method corresponds to the data flow of the system, and specifically comprises the following steps: S1, collecting multi-source real-time data of the heavy machinery equipment through the multi-modal perception terminal 10; S2, the digital twin server 20 drives a dedicated energy flow digital twin model to run based on the received multi-source real-time data, and calculates a twin reality difference degree vector between the predicted physical state output by the energy flow digital twin model and the real physical state reflected by the multi-source real-time data; S3, the cognition and calculation module 30 processes the multi-source real-time data to identify the operation intention of the equipment, and combines the operation intention, the energy flow digital twin model, and the twin reality difference degree vector to calculate the real-time carbon emission; S4, the collaborative decision-making module 40 generates a carbon quota curve based on the macro operation task and calls the energy flow digital twin model for simulation, and then generates a collaborative operation strategy according to the carbon quota curve and the real-time carbon emission; S5, the collaborative operation strategy is issued to the strategy execution module 50 deployed on the heavy machinery equipment for presentation.

[0032] The specific structure and working process of each part in the embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0033] Reference Figure 1 The detailed structure and working principle of each module in the heavy machinery equipment carbon emission dynamic monitoring system provided by the present application will be described.

[0034] First, the multi-modal perception terminal 10 is described. The multi-modal perception terminal 10 is physically deployed at a specific position of the heavy machinery equipment, and its function is to comprehensively and real-time obtain various physical quantities representing the running state of the equipment and the working environment, and convert these physical quantities into digital signals to form multi-source real-time data. In a specific embodiment, the multi-modal perception terminal 10 includes a core running parameter acquisition unit 11, a dynamic and kinematics parameter acquisition unit 12, a working load parameter acquisition unit 13, an acoustic feature parameter acquisition unit 14, and an environmental parameter acquisition unit 15.

[0035] The core running parameter acquisition unit 11 is connected to the controller area network (CAN) bus of the heavy machinery equipment or directly installs a sensor, which is used to acquire data reflecting the core state of the power system of the equipment. Specifically, the data acquired by the core running parameter acquisition unit 11 is the real-time speed and real-time load rate of the engine.

[0036] The dynamic and kinematics parameter acquisition unit 12 is usually composed of one or more inertial measurement units (IMU), which is used to acquire data reflecting the overall attitude and motion trajectory of the equipment. Specifically, the data acquired by the dynamic and kinematics parameter acquisition unit 12 is the three-axis angular velocity and three-axis acceleration of the heavy machinery equipment.

[0037] The work load parameter acquisition unit 13 acquires data reflecting the load of the device's actuator by arranging pressure and temperature sensors at key nodes of the hydraulic pipeline. Specifically, the data acquired by the work load parameter acquisition unit 13 is the pressure of the heavy machinery device hydraulic system and the temperature of the working oil.

[0038] The acoustic feature parameter acquisition unit 14 acquires acoustic signals of the heavy machinery device under different working conditions by deploying high-fidelity microphones near key noise sources such as the engine compartment and hydraulic pump. The acoustic signals are used to assist in identifying the work behavior and health status of the device.

[0039] The environmental parameter acquisition unit 15 acquires external environmental parameters of the working area where the heavy machinery device is located, which have a direct impact on the engine combustion efficiency. Specifically, the data acquired by the environmental parameter acquisition unit 15 is the environmental temperature and atmospheric pressure of the working site.

[0040] The data acquired by the above-mentioned units, after time stamp synchronization processing, collectively constitutes multi-source real-time data, and is sent to the digital twin server 20 and the cognitive and computing module 30 through the wireless communication module, providing comprehensive and accurate data input for subsequent analysis, calculation and decision-making.

[0041] Referring to Figure 1 The digital twin server 20 is one of the core computing units of the system, which receives multi-source real-time data collected by the multi-modal perception terminal 10, and performs running, correction and output of the energy flow digital twin model.

[0042] The digital twin server 20 first needs to build and run a dedicated energy flow digital twin model corresponding to the physical heavy machinery device. The construction of the model is based on the design parameters, physical mechanism and historical test data of the heavy machinery device, and describes the whole process of energy transfer and loss from engine fuel chemical energy to final mechanical work through a combination of mechanism modeling and data modeling. During system operation, the digital twin server 20 receives the engine speed, load rate, hydraulic system pressure and temperature, etc. as input, drives the energy flow digital twin model to solve in real time, and outputs a series of predicted physical states such as predicted fuel consumption rate, hydraulic system output power, and device running acoustic characteristics.

[0043] After obtaining the predicted physical state of the model output, a key function of the digital twin server 20 is to calculate the twin reality difference degree vector. The vector is used to quantify the deviation between the virtual model world and the physical reality world. The specific calculation process is as follows: The predicted physical states (e.g. predicted acoustic signals, predicted hydraulic system temperature) output by the energy flow digital twin model are compared item by item with the real physical states (e.g. measured acoustic signals, measured hydraulic system temperature) obtained from the multi-modal perception terminal 10 after synchronization processing, the difference between the two is calculated, and these differences are combined into a multi-dimensional vector, which is the twin reality difference degree vector Δ TR .

[0044] Referring to Figure 3 , in order to enable the energy flow digital twin model to reflect the individualized health state degradation of the equipment due to continuous use, the digital twin server 20 also executes a model self-evolution mechanism. Based on the twin reality difference degree vector Δ TR , the internal parameter vector Θ twin of the energy flow digital twin model is corrected online and continuously through an online learning algorithm. The internal parameter vector Θ twin includes the combustion efficiency coefficient and the hydraulic system efficiency coefficient representing the individual health status of the heavy machinery equipment. The process of this online correction is as follows: First, a loss function L TR based on the twin reality difference degree vector Δ TR is defined, whose goal is to minimize the value of the loss function. The calculation formula of the loss function is: In the formula, L TR is the value of the loss function, representing the square of the overall deviation between the model prediction and the physical reality; Δ TR is the twin reality difference degree vector; represents the square of the L2 norm of the vector.

[0045] Then, based on the loss function L TR , the gradient descent method is used to update the internal parameter vector Θ twin . By calculating the gradient of the loss function with respect to the internal parameter vector, the direction of parameter vector adjustment can be determined, so as to reduce the value of the loss function. The update formula of the internal parameter vector is: In the formula, Θ twin,t+1 is the internal parameter vector updated at t+1; Θ twin,t is the internal parameter vector before updating at t; η is a preset learning rate, which determines the step size of each update; is the loss function L TR with respect to the internal parameter vector Θ twinof the gradient. Through this iterative updating process, the energy flow digital twin model can adaptively adjust its internal parameters, so that its output of the predicted state is constantly approaching the real state of the physical device.

[0046] Referring to Figure 1 , the cognitive and computing module 30 receives multi-source real-time data from the multi-modal perception terminal 10, and communicates with the digital twin server 20 to obtain the twin reality difference degree vector. Its core function is to accurately analyze the device behavior and accurately calculate the carbon emissions based on multi-dimensional information, which is specifically realized through two closely linked steps of operation intention recognition and real-time carbon emission fusion calculation.

[0047] First, in terms of operation intention recognition, the goal is to analyze the specific operation action of the heavy machinery device from continuous, high-dimensional data streams. In a specific embodiment, the cognitive and computing module 30 internally solidifies a pre-trained time series deep learning model, such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU) network. The model has been fully learned through an offline training phase before being put into use. The training data includes a large amount of heavy machinery device operation data collected and manually labeled, and each data sequence is clearly labeled with an operation intention label, such as excavation, lifting, rotation, unloading, or idling.

[0048] In real-time operation, the time series deep learning model constructs the multi-source real-time data from the multi-modal perception terminal 10 (including engine speed, load rate, angular velocity, acceleration, hydraulic system pressure and temperature, acoustic signal features) into an input sequence with a preset time length. The model processes the input sequence through its internal recurrent neural network structure to capture the dynamic patterns of each data item over time and the complex nonlinear relationships between data items. The final output layer of the model, such as a Softmax classification layer, will calculate the probability of the current data sequence belonging to each pre-defined operation intention category, and the category with the highest probability will be the final recognition result, i.e. operation intention I. This process realizes the automatic, high-precision real-time decoding of the driver's operation intention.

[0049] After accurately identifying the operation intention I, the cognitive and computing module 30 then performs fusion calculation of real-time carbon emissions. The core of this calculation method is not to rely on a single, static emission factor or a simplified empirical formula, but to dynamically fuse the information of model prediction based on physical mechanism, operation target based on data decoding, and real-time feedback representing real operation deviation, thereby obtaining a high-fidelity real-time carbon emission E cThis method decomposes the total real-time carbon emissions into a base emission reflecting the main operating conditions and equipment health status, and an anomalous emission correction to compensate for unmodeled dynamics and random disturbances; the two are then summed. Specifically, the formula for calculating real-time carbon emissions is: E c =G fuel (X,I,Θ twin )+g corr (Δ TR ); In the formula, each term is defined as follows: E c The final calculated scalar value representing the carbon emission rate of heavy machinery at the current moment is usually expressed in grams per second (g / s) or kilograms per hour (kg / h).

[0050] G fuel (X,I,Θ twin This is a basic emissions calculation function. Essentially, this function is a mechanism-data hybrid model tightly integrated with an energy flow digital twin model. Its inputs include: Core operating parameters (such as engine speed and load rate) in multi-source real-time data X; the work intention I identified by the time-series deep learning model; And the internal parameter vector Θ of the energy flow digital twin model, which is corrected online by the digital twin server 20. twin (Including combustion efficiency coefficient, etc.)

[0051] The function first calculates the baseline fuel consumption rate at the current speed and load based on the engine's universal characteristic curve and fuel consumption mechanism. Then, it uses the operational intention I as an adjustment factor to correct the baseline fuel consumption rate, because different operational actions (such as pure slewing and loaded lifting) have different energy efficiencies even at the same speed and load. Finally, it utilizes the internal parameter vector Θ... twin The results are scaled to reflect the current individual health status of the equipment. Finally, the function outputs a theoretical carbon emission based on the current operating conditions, operational intent, and equipment health status.

[0052] g corr (Δ TR ) is an anomalous emission correction function. This function is a data-driven regression model, such as a multilayer perceptron (MLP) or a gradient boosting regression tree. The twin-reality difference vector Δ TR As its sole input variable. Δ TRThe vector contains multi-dimensional deviation information between model prediction and physical reality (e.g. unexpected vibration, abnormal temperature rise). The function of this function is to learn and quantify the implicit relationship between real-time deviation and additional carbon emissions. It is used to compensate for additional carbon emissions caused by random disturbances that the physical model fails to fully cover (e.g. sudden changes in ground slope), unmodeled physical effects (e.g. blocked heat sink) or sudden abnormal working conditions. Therefore, the output value of g corr is a real-time and accurate compensation for the basic emissions, ensuring that the final calculation result is highly consistent with the true situation of the physical world.

[0053] Referring to Figure 1 , Figure 4 and Figure 5 , the collaborative decision-making module 40 is the core of the multi-device operation optimization and active planning of carbon emissions. Its operation process includes the generation of carbon quota curve, the generation of collaborative strategy based on multi-agent reinforcement learning, and the algorithm training using group digital twin system.

[0054] First, the collaborative decision-making module 40 performs the generation of the carbon quota curve. This process begins with receiving a macro task, such as the total engineering quantity and planned duration of site leveling or earth excavation. The collaborative decision-making module 40 calls the energy flow digital twin model in the digital twin server 20 to perform forward-looking simulation on the macro task. The simulation decomposes the total task into a series of ordered sub-tasks and simulates the complete process of heavy machinery completing these sub-tasks, thereby predicting the minimum total carbon emissions required to complete the entire macro task. Subsequently, the module non-linearly allocates the predicted total carbon emissions over the entire planned duration according to the phased characteristics of the task (e.g. high energy consumption in the early excavation stage and low energy consumption in the later fine leveling stage), forming a cumulative carbon emission upper limit that changes over time, i.e. the carbon quota curve, as shown in Figure 5 . This curve provides a clear and dynamic global constraint for the subsequent generation of collaborative operation strategies.

[0055] After generating the carbon quota curve, for scenarios involving the collaborative operation of multiple heavy machinery devices, the collaborative decision-making module 40 uses a multi-agent reinforcement learning algorithm to generate a collaborative operation strategy under the strict constraint of the carbon quota curve. The goal of this algorithm is to seek the best way to collaborate between devices to maximize overall operation efficiency without exceeding the carbon emission quota at any time. The algorithm generates specific operation recommendations for each device in the cluster at each decision-making time step. In order to guide the algorithm to converge to the predetermined target, it uses a specific joint reward function, the calculation formula of which is: R jt = w work · ΔW - w time · Δt - P penalty ; In the formula, the definitions of each term are as follows: R jt is the joint reward value obtained by the entire device cluster in a single time step; is the effective work volume, such as the volume of excavated or transported materials, completed by the device cluster collectively in the single time step; is the duration of the single time step; w work is a preset work efficiency weight coefficient for adjusting the proportion of completed work volume in the reward; w time is a preset time cost weight coefficient for adjusting the proportion of time consumption in the reward.

[0056] P penalty is a penalty term, which is taken as a preset positive value when the cumulative carbon emissions of any heavy machinery device at the current time exceed the upper limit of the carbon quota defined by the carbon quota curve at the time; otherwise, the value of the penalty term is zero. The design of the penalty term ensures that the algorithm seeks high efficiency while strictly complying with the carbon emission constraints. penalty

[0057] In order to enable the multi-agent reinforcement learning algorithm to be trained efficiently and safely, the present application also provides a group digital twin training environment. Specifically, the collaborative decision-making module 40 connects and synchronizes the energy flow digital twin models of each heavy machinery device involved in the collaborative work, which are respectively exclusive and online corrected, to form a group digital twin system in a virtual space. This system completely reproduces the physical interaction and energy consumption process of multiple devices in the same work space. The collaborative decision-making module 40 uses this group digital twin system as the training environment for the multi-agent reinforcement learning algorithm. The agents of the algorithm perform tens of thousands of simulated work cycles in this virtual environment, optimizing their collaborative strategies through continuous trial and error and learning, without the need for physical devices, which greatly improves the efficiency and safety of algorithm training and enables it to converge to a high-performance collaborative work strategy before deployment.

[0058] Referring to Figure 1 , the strategy execution module 50 is physically deployed in the cab of the heavy machinery device, such as integrated on the vehicle-mounted control panel, which is in communication connection with the collaborative decision-making module 40 for receiving and analyzing the collaborative work strategy generated thereby.

[0059] ​The specific implementation form of the policy execution module 50 is a man-machine interface integrated with a display screen. The core function of this module is to convert the data-based collaborative operation strategy instructions output by the collaborative decision-making module 40 into specific and visual operation instructions and state information, and present them to the device operator to guide him to perform the optimal operation.

[0060] In a specific embodiment, the content presented on the man-machine interface of the policy execution module 50 is divided into several functional areas, specifically including: The operation instruction area is used to display direct and quantitative operation suggestions to the operator. For example, in the form of numbers or graphical instruments, the recommended engine speed range, the recommended hydraulic lever movement speed or stroke range under the current working condition are clearly indicated.

[0061] The carbon emission monitoring area is a key man-machine interaction window for realizing closed-loop control of carbon emissions. It displays the real-time carbon emissions calculated by the cognitive and computing module 30 in a graphical manner in real time, and compares it with the carbon quota curve generated by the collaborative decision-making module 40 on the same screen. For example, a dynamic progress bar is used to show the percentage of the cumulative carbon emissions of the current task in the total quota, and an instrument panel is used to display the instantaneous emission rate. When the instantaneous emission rate approaches or exceeds the upper limit at the current time defined by the carbon quota curve, this area can give a warning through color change (for example, from green normal state to yellow pre-warning state or red over-limit state) or sound prompt.

[0062] The collaborative operation situation area is enabled in the scenario of multi-device collaborative operation. Through a simplified two-dimensional or three-dimensional overhead view, the relative position, orientation and planned action (such as movement path or operation range) of the device and other collaborative operation devices are displayed in real time. This function provides the device operator with the global situation information necessary for performing collaborative actions.

[0063] In the above manner, the policy execution module 50 effectively transmits the complex optimization decision results in the system background to the front-end operator, guiding him to complete the operation task efficiently and safely under the premise of meeting the carbon emission constraints.

[0064] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic monitoring system for carbon emissions from heavy machinery, characterized in that, The system includes: Multimodal sensing terminals are used to collect multi-source real-time data from heavy machinery and equipment; A digital twin server is used to drive a dedicated energy flow digital twin model based on the multi-source real-time data, and to calculate the twin reality difference vector between the predicted physical state output by the energy flow digital twin model and the real physical state reflected by the multi-source real-time data. The cognition and computing module is used to process the multi-source real-time data to identify the operating intention of the heavy machinery and to calculate the real-time carbon emissions by combining the operating intention, the energy flow digital twin model, and the twin reality difference vector. The collaborative decision-making module is used to generate a carbon quota curve based on the macro-operation task and by calling the energy flow digital twin model. Then, based on the carbon quota curve and the real-time carbon emissions, a collaborative operation strategy is generated. The strategy execution module is used to receive and present the collaborative operation strategy generated by the collaborative decision-making module.

2. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 1, characterized in that, The multimodal sensing terminal includes a core operating parameter acquisition unit, a dynamic and kinematic parameter acquisition unit, a workload parameter acquisition unit, an acoustic feature parameter acquisition unit, and an environmental parameter acquisition unit; wherein... The core operating parameter acquisition unit is used to collect engine speed and load rate; The dynamic and kinematic parameter acquisition unit is used to acquire the angular velocity and acceleration of the heavy machinery. The work load parameter acquisition unit is used to acquire the pressure and temperature of the hydraulic system; The acoustic feature parameter acquisition unit is used to acquire acoustic signals during the operation of the heavy machinery. The environmental parameter acquisition unit is used to collect the ambient temperature and atmospheric pressure of the operating area of ​​the heavy machinery equipment.

3. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 1, characterized in that, After calculating the twin reality difference vector, the digital twin server is also used to correct the internal parameter vector of the energy flow digital twin model online through an online learning algorithm based on the twin reality difference vector, so as to drive the self-evolution of the energy flow digital twin model; the internal parameter vector includes the combustion efficiency coefficient and the hydraulic system efficiency coefficient, which characterize the individual health status of the heavy machinery equipment.

4. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 3, characterized in that, The online learning algorithm includes the following steps: Define a loss function based on the twin-reality difference vector, and the formula for calculating the loss function is as follows: In the formula, L TR The value of the loss function; Δ TR The vector representing the difference between twins and reality; This indicates taking the square of the L2 norm of the vector; Based on the loss function, the internal parameter vector is updated using gradient descent. The update formula for the internal parameter vector is as follows: In the formula, Θ twin,t+1 For the updated internal parameter vector; Θ twin,t The vector of internal parameters before the update; η is the learning rate; This is the gradient of the loss function with respect to the internal parameter vector.

5. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 1, characterized in that, The cognition and computing module employs a time-series deep learning model to identify the operational intent of the heavy machinery by analyzing the time series of the multi-source real-time data.

6. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 1, characterized in that, When the cognition and calculation module calculates real-time carbon emissions, the formula for calculating real-time carbon emissions is as follows: E c =G fuel (X,I,Θ twin )+g corr (D TR ); In the formula, E c For real-time carbon emissions; G fuel This is a basic emission calculation function, which uses multi-source real-time data X, operational intent I, and the internal parameter vector Θ of the energy flow digital twin model. twin For input; g corr An anomalous emission correction function is defined by the twin reality difference vector Δ. TR For input.

7. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 1, characterized in that, The collaborative decision-making module employs a multi-agent reinforcement learning algorithm to generate the collaborative operation strategy under the constraints of the carbon quota curve.

8. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 7, characterized in that, The reward function of the multi-agent reinforcement learning algorithm is: R jt =w work ·ΔW-w time ·Δt-P penalty ; In the formula, R jt The combined reward value; ΔW is the effective work completed in a single time step; Δt is the duration of a single time step; w work For work efficiency weighting coefficient; w time P is the time cost weighting coefficient. penalty As a penalty item, when the cumulative carbon emissions of the heavy machinery equipment exceed the upper limit of the carbon quota curve defined at the current moment, the penalty item is a preset positive value.

9. The carbon emission dynamic monitoring system for heavy machinery equipment according to claim 7, characterized in that, When running the multi-agent reinforcement learning algorithm, the collaborative decision-making module connects the energy flow digital twin models of multiple heavy machinery equipment working collaboratively to form a group digital twin system, and uses the group digital twin system as the training environment for the multi-agent reinforcement learning algorithm.

10. A method for dynamic monitoring of carbon emissions from heavy machinery, applied to the system described in any one of claims 1-9, characterized in that, The method includes the following steps: S1. Collect multi-source real-time data from heavy machinery and equipment; S2. Based on the multi-source real-time data, drive a dedicated energy flow digital twin model to run, and calculate the twin reality difference vector between the predicted physical state output by the energy flow digital twin model and the real physical state reflected by the multi-source real-time data. S3. Process the multi-source real-time data to identify the operating intention of the heavy machinery and equipment, and combine the operating intention, the energy flow digital twin model, and the twin reality difference vector to calculate the real-time carbon emissions. S4. Based on the macro-level operation task and by calling the energy flow digital twin model to perform simulation to generate a carbon quota curve, and then generate a collaborative operation strategy based on the carbon quota curve and the real-time carbon emissions. S5. The collaborative operation strategy is distributed to the strategy execution module deployed on the heavy machinery equipment for presentation.

Images