Digital exhibition intelligent control method and system based on Internet of Things

Abstract

The invention provides a digital exhibition intelligent control method and system based on the Internet of Things. The method comprises the following steps: acquiring a dynamic system model and a current state deviation of a target controlled object in a digital exhibition environment; predicting the state deviation of the target controlled object at a certain moment in the future by using a graph convolution gating cycle unit network based on the system topology structure and historical state data of the digital exhibition and display environment to obtain a predicted deviation vector; obtaining a nonlinear integral sliding mode surface fused with prediction information by using the current state deviation and the measured deviation vector; a nonlinear disturbance observer is adopted to estimate the lumped uncertainty borne by the system in the operation process on line, and an estimated value of the lumped uncertainty is obtained; forming a total control quantity by an equivalent control item and a robust approaching item; and controlling the target controlled object by using the total control quantity.

Description

Technical Field

[0001] This application belongs to the field of intelligent control, and in particular relates to intelligent control methods and systems for digital exhibitions based on the Internet of Things. Background Technology

[0002] Digital exhibitions are a new form of display. To ensure the smooth operation of digital exhibition systems such as intelligent lighting, interactive devices, and environmental simulators, and to provide a high-quality user experience, precise intelligent control of the core controlled objects is crucial. Sliding mode control has advantages such as insensitivity to external disturbances and fast response speed, and has been widely used in the field of nonlinear system control. Sliding mode control uses a predetermined sliding surface and applies a switching control law to make the system state trajectory reach and move along the sliding surface within a finite time. However, the discontinuous switching terms of the sign function in standard sliding mode control can easily lead to chattering in the system. This high-frequency oscillation will cause waste of control energy and even physical wear and damage to the actuators. To suppress chattering, existing methods often use continuous processing such as boundary layers or saturated functions, but this sacrifices tracking accuracy. Moreover, traditional sliding mode controllers are usually a retrospective compensation mechanism, which only adjusts the feedback based on the current system state deviation and lacks the ability to predict the future of the system. When dealing with rapidly changing systems or systems with significant time delays, control response lag and overshoot problems are prone to occur. How to accurately estimate and feedforward the disturbances in the online system and improve the anti-disturbance performance of the digital exhibition control system is a problem that needs to be solved by existing technologies. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent control method and system for digital exhibitions based on the Internet of Things (IoT), which solves the problems of jitter and time lag when controlling digital exhibition objects. The method includes the following steps: The dynamic system model and current state deviation of the target controlled object in the digital exhibition environment are obtained; based on the system topology and historical state data of the digital exhibition environment, the state deviation of the target controlled object at a certain future moment is predicted using a graph convolutional gated recurrent unit network to obtain the predicted deviation vector. The nonlinear integral sliding surface for fusing prediction information is obtained using the current state deviation and the measured deviation vector; the lumped uncertainty experienced by the system during operation is estimated online using a nonlinear disturbance observer to obtain an estimate of the lumped uncertainty. The equivalent control term is derived based on the dynamic system model and sliding mode conditions. The robust approach term is obtained by using the disturbance compensation term for the lumped uncertainty estimate and the adaptive approach component generated by the fuzzy logic system with the sliding mode variable as input. The total control quantity is composed of the equivalent control term and the robust approach term. The target controlled object is controlled by the total control quantity.

[0004] This invention also provides an intelligent control system for digital exhibitions based on the Internet of Things, comprising the following modules: The prediction deviation calculation module is used to obtain the dynamic system model and current state deviation of the target controlled object in the digital exhibition environment; based on the system topology and historical state data of the digital exhibition environment, the module uses a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a certain future moment, and obtains the prediction deviation vector. The estimation module is used to obtain a nonlinear integral sliding surface for fusing prediction information using the current state deviation and the measured deviation vector; and to obtain an estimate of the lumped uncertainty during system operation by using a nonlinear perturbation observer online. The control module is used to derive the equivalent control term based on the dynamic system model and sliding mode conditions, and to obtain the robust approach term based on the disturbance compensation term for the lumped uncertainty estimate and the adaptive approach component generated by the fuzzy logic system with the sliding mode variable as input. The equivalent control term and the robust approach term constitute the total control quantity. The total control quantity is used to control the target controlled object.

[0005] This invention introduces a graph convolutional gated recurrent unit network to predict future system state deviations, thereby enabling pre-adjustment of control variables and significantly reducing overshoot during control. The nonlinear sliding surface constructed by fusing predicted information with integral terms not only ensures accurate tracking of the system state but also effectively eliminates steady-state errors. Simultaneously, the nonlinear disturbance observer accurately estimates and compensates for lumped uncertainties encountered during system operation, enhancing the system's disturbance rejection capability and robustness. The fuzzy logic system employed in the reaching law smoothly adjusts the reaching action based on the sliding mode variables, fundamentally suppressing chattering phenomena in traditional sliding mode control, ensuring smooth control output, and comprehensively improving the control performance and operational reliability of the digital display system. Attached Figure Description

[0006] Figure 1 A flowchart of the first embodiment; Figure 2 This is a schematic diagram illustrating the fuzzification of the input s; Figure 3 This is a schematic diagram of a fuzzy rule base. Detailed Implementation

[0007] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0008] In the first embodiment, the present invention provides an intelligent control method for digital exhibitions based on the Internet of Things, such as... Figure 1 As shown, it includes the following steps: S1. Obtain the dynamic system model and current state deviation of the target controlled object in the digital exhibition environment; based on the system topology and historical state data of the digital exhibition environment, use a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a certain future moment, and obtain the predicted deviation vector. The digital exhibition includes multiple controlled objects, such as track lights, interactive robotic arms, and screen rotation controllers. The dynamic system model of the target controlled object, such as the intelligent track lights in the exhibition hall, can be represented as a second-order nonlinear system. Its actual state, such as angle or position, is collected in real time by encoders or vision sensors and compared with a preset desired state to obtain the state deviation *e*. The rate of change of the deviation *de* / dt is calculated by differencing the state deviation *e* between two consecutive sampling times and dividing by the sampling time. The various devices in the exhibition environment are treated as nodes in a graph, and the communication or physical connections between devices are treated as edges, constructing a graph structure. A graph convolutional network layer is used to extract the spatial dependency features between nodes, and its output is fed into a gated recurrent unit layer, which is responsible for learning the dynamic pattern of system state evolution over time. Using historical operating data, including the state of each device and control commands, the network is trained offline using a backpropagation algorithm, enabling it to output predicted values ​​of state deviation for multiple future time steps based on the current and historical states. .

[0009] S2, using the current state deviation and the measured deviation vector, a nonlinear integral sliding surface for fusing prediction information is obtained; a nonlinear disturbance observer is used to estimate the lumped uncertainty experienced by the system during operation to obtain an estimated value of the lumped uncertainty; A nonlinear integral sliding surface s incorporating prediction information is constructed. Preferably, the expression for the nonlinear integral sliding surface, i.e., the sliding variable s, is: Where c1, c2, c3, and γ are preset normal control gains. Control gains c1 and c2 are selected based on the desired closed-loop pole configuration of the system to ensure the stability and convergence speed of the sliding mode dynamics. Integral term Numerical calculations are performed using a cumulative summation method to eliminate steady-state errors. Prediction bias. The predicted value for the most future moment is introduced and smoothed and amplitude-limited using the hyperbolic tangent function tanh. The gain c3 and γ are obtained through simulation experiments to balance the strength and stability of the prediction compensation.

[0010] A finite-time perturbation observer is employed. This observer constructs an auxiliary variable whose dynamic equation includes a nonlinear function term based on the system state and control input. By designing this nonlinear function term, for example, using a p-norm or a higher-order power term based on the sign function, the observer's estimation error of the lumped uncertainty converges to zero in a finite-time period, thus obtaining the perturbation estimate quickly and accurately. .

[0011] S3. Based on the dynamic system model and sliding mode conditions, the equivalent control term is derived. The robust approach term is obtained by using the disturbance compensation term for the lumped uncertainty estimate and the adaptive approach component generated by the fuzzy logic system with the sliding mode variable as input. The total control quantity is composed of the equivalent control term and the robust approach term. The target controlled object is controlled by the total control quantity.

[0012] Take the time derivative of the expression for the sliding surface s, and let... Substitute these values ​​into the system dynamic model to solve for the equivalent control terms that include the system model parameters. Robust approach term It consists of a disturbance compensation term and a fuzzy adaptive term. The fuzzy logic system uses a sliding mode variable s as the only input, and its input membership function is divided into five fuzzy subsets: negative large, negative small, zero, positive small, and positive large. Based on the set fuzzy rules, if s is positive large, the output is large; if s is zero, the output is small. Defuzzification is performed using the centroid method to obtain an adaptive gain that varies with s. This gain is used to adjust the approach speed, achieving rapid approach when there is a large deviation and smooth transition when there is a small deviation, thereby suppressing chattering.

[0013] In each control cycle, the calculated equivalent control term will be... With robust approach term The signals are summed to obtain the final total control quantity u. This digital control signal is converted into an analog voltage or current signal by a digital-to-analog converter and applied to the actuator of the target controlled object, such as a driver for a DC motor or a power supply module for controlling the brightness of LED lights, thereby achieving precise control of the exhibit.

[0014] In one possible implementation, the method of using a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a future time, and obtaining a predicted deviation vector, includes: The sensors and actuators in the digital exhibition environment are abstracted as network nodes, and the communication relationships and physical connections between nodes are used as edges to construct the adjacency matrix of the system topology. The historical state data sequence of each node over the past N consecutive sampling times is used as the feature matrix of the node; The adjacency matrix and node feature matrix are input into the graph convolutional layer, and the neighbor node information is aggregated through graph convolution operation to obtain the spatial association features of the entire system at the current time. The spatial feature vector output by the graph convolutional layer is fed into the subsequent gated recurrent unit layer in chronological order to learn the dynamic evolution law of the state data in the time dimension, and output the predicted value of the state deviation at future time, thus obtaining the prediction deviation vector.

[0015] In an interactive art installation consisting of multiple collaborative robotic arms and environmentally aware cameras, each robotic arm's joint motor, end effector, and each camera can be considered a network node. The physical connections between the robotic arm joints and the communication bus between the motor controller and the camera data processing unit constitute the edges of the graph, thereby constructing an adjacency matrix A that describes the information flow and physical constraints of the entire system. For example, if joint motor 1 can receive the position information processed by camera 3, then the value of the corresponding element A13 in matrix A is 1, otherwise it is 0.

[0016] For each node, such as a joint motor, its angular position, angular velocity, and output torque data are collected every 10 milliseconds over the past second, forming a data sequence with 100 time steps and 3 features per step, constituting the feature matrix of that node. The adjacency matrix of the entire system, for example, containing 20 nodes, along with the feature matrices of all nodes, is input into a graph convolutional network. Graph convolution operations can fuse information about a node itself and all its neighboring nodes. For example, the movement of a joint depends not only on its own historical state but also on the influence of other connected joints and the cooperating vision system, thereby effectively extracting the complex spatial coupling characteristics of the entire exhibition system at the current moment.

[0017] The system-level spatial feature vectors output by the graph convolutional network at each time step are fed sequentially into a gated recurrent unit (ROU) network. For example, the five spatial feature vectors extracted from the most recent five time steps can be used as the input sequence. Leveraging its unique gating mechanism, the ROU network can learn the evolution patterns of the system state over time, such as the dynamic patterns of the series of processes from visual capture to the approach of the audience to the robotic arm's welcoming gesture. The network ultimately outputs a predicted deviation vector, representing the deviation between the angles of each joint of the robotic arm and the desired angles 200 milliseconds in the future. .

[0018] In one possible implementation, the method of using a nonlinear perturbation observer to estimate the lumped uncertainty experienced by the system during operation to obtain an estimate of the lumped uncertainty includes: Construct a nonlinear state observer of the following form to obtain an estimate of the lumped uncertainty. :

[0019]

[0020] Where p is the internal state variable of the observer, L is the positive definite observer gain matrix, q and dq / dt are the current state and rate of change of the target controlled object, M(q), C(q,dq / dt), and G(q) are the inertia matrix, Coriolis force and centrifugal force matrix, and gravity term in the dynamic system model, respectively, and u is the total control variable.

[0021] The core function of a nonlinear disturbance observer is to estimate and compensate for various unmodeled dynamics and external disturbances in a system in real time. In a scenario of a six-DOF robotic arm interacting with an audience, these uncertainties might include variations in static and dynamic friction at the joints, gearbox clearance, slight load changes, and even air resistance, all of which collectively constitute the lumped uncertainty d. This observer can generate estimates that closely approximate the actual disturbances. It is used to perform feedforward compensation in control and improve control accuracy.

[0022] In the specific implementation of the observer, the observer gain matrix L is defined. For a six-DOF robotic arm, L is designed as a six-by-six diagonal positive definite matrix; for example, its diagonal elements can be set to [20,20,20,10,10,10]. The magnitude of these gain values ​​determines the convergence speed of the observer estimate and its sensitivity to measurement noise. A larger gain speeds up convergence but may amplify noise. The observer acquires the current state of the robotic arm in real time, i.e., the angle q and angular velocity dq / dt of each joint, and uses known dynamic model parameters, such as the inertia matrix M(q), the Coriolis force matrix C(q,dq / dt), and the gravity term G(q), to calculate the theoretical system behavior.

[0023] The observer continuously updates its internal state variable p using the first differential equation. This first equation calculates the difference between the theoretical dynamics generated by the known model and control input u and the actual system dynamics corrected by the gain matrix L; this difference is primarily attributed to the unmodeled disturbance d. The real-time estimate of the lumped uncertainty is then calculated using the second algebraic equation, which combines the internal state p with the system's generalized momentum M(q) multiplied by dq / dt, and scaled by the gain matrix L. For example, if at a certain moment, the calculation is... If the second component is -0.8, it means that the observer estimates that the second joint of the robotic arm is experiencing an additional drag torque of 0.8 Nm.

[0024] In one possible implementation, the equivalent control term derived based on the dynamic system model and sliding mode conditions is specifically as follows: Based on the desired state of the controlled object and its derivatives, combined with sliding mode conditions Dynamic system model of the controlled object Perform the derivation; Depend on The desired error acceleration for sliding mode control can be obtained by solving the problem. ; Combining the definition of error The actual acceleration of the controlled object Represented as ; Substituting the actual acceleration into the dynamic system model and neglecting the lumped uncertainty term d, the equivalent control term is obtained by solving. The expression: Where sech represents the hyperbolic secant function.

[0025] Equivalent control items It is a core component of sliding mode control law, representing the precise control force required to drive the controlled object to move exactly along the desired trajectory under ideal conditions, i.e., without any uncertainties or external disturbances. The derivation logic is based on a core assumption: the system state is already in and needs to be maintained on a predefined sliding surface, mathematically represented by the time derivative of the sliding mode variables. It is always equal to zero. The entire derivation process revolves around this condition, using the system's dynamic equations to solve for the corresponding control quantity.

[0026] The first step in the derivation is to use The condition of zero is used to solve for the error acceleration that the system must possess from the differential equations of the sliding surface. This expected error acceleration is a composite term, which depends not only on the current position error e and velocity error. It also incorporates future error predictions from neural networks. and its rate of change This step establishes an ideal error convergence dynamic. For example, if the current error is large, the calculated expected error acceleration will be adjusted accordingly to drive the error to decrease more quickly.

[0027] Then, by using the error e equal to the desired position Subtracting the actual position q establishes the relationship between the error dynamics and the actual motion of the system. Taking the second time derivative of this definition reveals the actual acceleration of the system. Using the acceleration of the desired trajectory Compared with the expected error acceleration just obtained Let's represent it as such. Substituting this actual acceleration expression into the dynamic model equations of the robotic arm, and temporarily ignoring the disturbance term d, we can solve the equations through algebraic operations. For example, during a certain control cycle, substituting the calculated error values ​​and system state parameters into the final expression yields a six-dimensional torque vector, such as [10.5, 8.2, -5.1, 1.0, 0.5, -0.2] Nm, which is the basic control torque required to maintain ideal motion.

[0028] In one possible implementation, the robust convergence term is obtained by generating an adaptive convergence component based on the disturbance compensation term of the lumped uncertainty estimate and the fuzzy logic system with sliding mode variables as input, specifically as follows: The robust approach term Composed of disturbance compensation term and adaptive approach component The composition, expressed as:

[0029] Among them, disturbance compensation item The adaptive approach component is used to offset the lumped uncertainty estimated by the nonlinear perturbation observer. It is generated by a fuzzy logic system with sliding mode variable s as input, which adjusts the approach speed and suppresses chattering.

[0030] Robust approach term It is a key component in sliding mode control that ensures the system state can converge quickly and stably to the sliding surface from any initial position. It is mainly responsible for handling uncertainties in the system and ensuring the robustness of the control. This component consists of two functionally defined components: direct disturbance feedforward compensation and intelligent adaptive reaching law. The two work together to enable the system to maintain excellent performance even in complex and changing environments.

[0031] Disturbance compensation item It counteracts known perturbations. It utilizes the lumped uncertainty estimate output by the aforementioned nonlinear perturbation observer. This involves introducing a force or torque of equal magnitude but opposite direction into the control signal. For example, if the observer estimates that an unexpected increase in the robot arm's load has resulted in a 1.2 Nm downward torque at the third joint, i.e. If the third component is -1.2, then the disturbance compensation term will actively add a positive torque of 1.2 Nm to the control output of that joint. This direct cancellation method can reduce the tracking error caused by model inaccuracies and external disturbances, and greatly improve the control accuracy of the system.

[0032] Adaptive fuzzy approaching components Mimicking the control experience of human experts, the strength of the approach control action is dynamically adjusted based on the distance of the system state from the sliding surface, i.e., the magnitude and sign of the sliding variable 's'. When the absolute value of 's' is large, it means the system is far from the ideal state. It will output a large control variable to achieve rapid approach. When the absolute value of 's' is very small, it means the system is already very close to the sliding surface. It will significantly reduce its output, such as Figure 2 As shown, this avoids the severe chattering problem caused by sign function switching in traditional sliding mode control, making the control process smoother.

[0033] In one possible implementation, the generation of adaptive approaching components using a fuzzy logic system with sliding mode variables as input specifically involves: Using the sliding mode variable s as input, the universe of discourse is fuzzified into five fuzzy sets: negative large NB, negative small NS, zero ZE, positive small PS, and positive large PB, and described by triangular or Gaussian membership functions. Establish the following fuzzy rule base: If s is NB, then For Zhengda PB; If s is NS, then For the positive PS; If s is ZE, then Zero ZE; If s is PS, then For negative NS; If s is PB, then For negative NB; The fuzzification was defuzzified using a Mamdani-type fuzzy inference engine and the centroid method, resulting in... The output value.

[0034] Fuzzy logic systems convert precise sliding mode variable s values ​​into fuzzy statements and determine the magnitude of the output control quantity through a series of experience-based rules. Figure 3 This represents a possible fuzzy rule base. For example, for a joint of a robotic arm, the actual range of its sliding mode variable *s* might be between -5 and +5. Five fuzzy sets are defined on this universe of discourse, each described by a membership function, such as a triangle with its vertex at zero and its base ranging from -1 to +1 to describe the zero ZE set. When the precise value of *s* is measured to be 0.8 at a certain moment, this value might belong to the zero ZE set with a membership degree of 0.2, and simultaneously belong to the positive small PS set with a membership degree of 0.8.

[0035] The fuzzy inference engine calculates the output based on this input and a pre-defined rule base. Since the value of s is 0.8, it also activates the rule that if s is ZE, then... If ZE and s is PS, then For the two rules NS, the system combines the results of these two rules. According to the Mamdani method, the first rule, triggered with a strength of 0.2, results in a zero-ZE output fuzzy set clipped to a height of 0.2; the second rule, triggered with a strength of 0.8, results in a negative small NS output fuzzy set clipped to a height of 0.8. These two clipped output fuzzy sets are then merged into a single, irregularly shaped final output fuzzy set.

[0036] To obtain a precise control value that can be used by the actuator, the merged fuzzy set is defuzzified. The centroid method is used, which involves calculating the x-coordinate of the geometric center or centroid of this irregular region. For example, assuming the output... The universe of discourse ranges from -10 to +10 Nm. Calculations show that the centroid of the aforementioned merged fuzzy set has an x-coordinate of -3.5. Therefore, -3.5 Nm is the final precise value output by the fuzzy logic system at the current moment. This value will be used to construct the robust reaching term, which pushes the system state toward the sliding surface with an appropriate force, thus achieving smooth and adaptive adjustment of the reaching process.

[0037] In a second embodiment, the present invention provides an intelligent control system for digital exhibitions based on the Internet of Things, comprising the following modules: The prediction deviation calculation module is used to obtain the dynamic system model and current state deviation of the target controlled object in the digital exhibition environment; based on the system topology and historical state data of the digital exhibition environment, the module uses a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a certain future moment, and obtains the prediction deviation vector. The estimation module is used to obtain a nonlinear integral sliding surface for fusing prediction information using the current state deviation and the measured deviation vector; and to obtain an estimate of the lumped uncertainty during system operation by using a nonlinear perturbation observer online. The control module is used to derive the equivalent control term based on the dynamic system model and sliding mode conditions, and to obtain the robust approach term based on the disturbance compensation term for the lumped uncertainty estimate and the adaptive approach component generated by the fuzzy logic system with the sliding mode variable as input. The equivalent control term and the robust approach term constitute the total control quantity. The total control quantity is used to control the target controlled object.

[0038] In one possible implementation, obtaining the nonlinear integral sliding surface for fusing prediction information using the current state deviation and the measured deviation vector specifically involves: The expression for the nonlinear integral sliding surface s is: ; Where e is the current state deviation. The prediction bias vector is defined by c1, c2, c3, and γ, which are preset constant control gains.

[0039] In one possible implementation, the method of using a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a future time, and obtaining a predicted deviation vector, includes: The sensors and actuators in the digital exhibition environment are abstracted as network nodes, and the communication relationships and physical connections between nodes are used as edges to construct the adjacency matrix of the system topology. The historical state data sequence of each node over the past N consecutive sampling times is used as the feature matrix of the node; The adjacency matrix and node feature matrix are input into the graph convolutional layer, and the neighbor node information is aggregated through graph convolution operation to obtain the spatial association features of the entire system at the current time. The spatial feature vector output by the graph convolutional layer is fed into the subsequent gated recurrent unit layer in chronological order to learn the dynamic evolution law of the state data in the time dimension, and output the predicted value of the state deviation at future time, thus obtaining the prediction deviation vector.

[0040] In one possible implementation, the method of using a nonlinear perturbation observer to estimate the lumped uncertainty experienced by the system during operation to obtain an estimate of the lumped uncertainty includes: Construct a nonlinear state observer of the following form to obtain an estimate of the lumped uncertainty. :

[0041]

[0042] Where p is the internal state variable of the observer, L is the positive definite observer gain matrix, q and dq / dt are the current state and rate of change of the target controlled object, M(q), C(q,dq / dt), and G(q) are the inertia matrix, Coriolis force and centrifugal force matrix, and gravity term in the dynamic system model, respectively, and u is the total control variable.

[0043] In one possible implementation, the equivalent control term derived based on the dynamic system model and sliding mode conditions is specifically as follows: Based on the desired state of the controlled object and its derivatives, combined with sliding mode conditions Dynamic system model of the controlled object Perform the derivation; Depend on The desired error acceleration for sliding mode control can be obtained by solving the problem. ; Combining the definition of error The actual acceleration of the controlled object Represented as ; Substituting the actual acceleration into the dynamic system model and neglecting the lumped uncertainty term d, the equivalent control term is obtained by solving. The expression: .

[0044] In one possible implementation, the robust convergence term is obtained by generating an adaptive convergence component based on the disturbance compensation term of the lumped uncertainty estimate and the fuzzy logic system with sliding mode variables as input, specifically as follows: The robust approach term Composed of disturbance compensation term and adaptive approach component The composition, expressed as:

[0045] Among them, disturbance compensation item The adaptive approach component is used to offset the lumped uncertainty estimated by the nonlinear perturbation observer. It is generated by a fuzzy logic system with sliding mode variable s as input, which adjusts the approach speed and suppresses chattering.

[0046] In one possible implementation, the generation of adaptive approaching components using a fuzzy logic system with sliding mode variables as input specifically involves: Using the sliding mode variable s as input, the universe of discourse is fuzzified into five fuzzy sets: negative large NB, negative small NS, zero ZE, positive small PS, and positive large PB, and described by triangular or Gaussian membership functions. Establish the following fuzzy rule base: If s is NB, then For Zhengda PB; If s is NS, then For the positive PS; If s is ZE, then Zero ZE; If s is PS, then For negative NS; If s is PB, then For negative NB; The fuzzification was defuzzified using a Mamdani-type fuzzy inference engine and the centroid method, resulting in... The output value.

[0047] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0048] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0049] The method and electronic device for providing product object information provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and its core ideas. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A digital exhibition intelligent control method based on the Internet of Things, characterized in that, Includes the following steps: Acquire the dynamic system model and current state deviation of the target controlled object in the digital exhibition environment; Based on the system topology and historical state data of the digital exhibition environment, a graph convolutional gated recurrent unit network is used to predict the state deviation of the target controlled object at a certain future moment, and a prediction deviation vector is obtained. The nonlinear integral sliding surface for fusing prediction information is obtained using the current state deviation and the measured deviation vector; the lumped uncertainty experienced by the system during operation is estimated online using a nonlinear disturbance observer to obtain an estimate of the lumped uncertainty. The equivalent control term is derived based on the dynamic system model and sliding mode conditions. The robust approach term is obtained by using the disturbance compensation term for the lumped uncertainty estimate and the adaptive approach component generated by the fuzzy logic system with the sliding mode variable as input. The total control quantity is composed of the equivalent control term and the robust approach term. The target controlled object is controlled using the total control quantity.

2. The method according to claim 1, characterized in that, The nonlinear integral sliding surface for obtaining fused prediction information using the current state deviation and the measured deviation vector is specifically as follows: The expression for the nonlinear integral sliding surface s is: ; Where e is the current state deviation. The prediction bias vector is defined by c1, c2, c3, and γ, which are preset constant control gains.

3. The method according to claim 1, characterized in that, The method of using a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a future time, and obtaining a predicted deviation vector, includes: The sensors and actuators in the digital exhibition environment are abstracted as network nodes, and the communication relationships and physical connections between nodes are used as edges to construct the adjacency matrix of the system topology. The historical state data sequence of each node over the past N consecutive sampling times is used as the feature matrix of the node; The adjacency matrix and node feature matrix are input into the graph convolutional layer, and the neighbor node information is aggregated through graph convolution operation to obtain the spatial association features of the entire system at the current time. The spatial feature vector output by the graph convolutional layer is fed into the subsequent gated recurrent unit layer in chronological order to learn the dynamic evolution law of the state data in the time dimension, and output the predicted value of the state deviation at future time, thus obtaining the prediction deviation vector.

4. The method according to claim 1, characterized in that, The method of using a nonlinear perturbation observer to estimate the lumped uncertainty experienced by the system during operation to obtain an estimate of the lumped uncertainty includes: Construct a nonlinear state observer of the following form to obtain an estimate of the lumped uncertainty. : Where p is the internal state variable of the observer, L is the positive definite observer gain matrix, q and dq / dt are the current state and rate of change of the target controlled object, M(q), C(q,dq / dt), and G(q) are the inertia matrix, Coriolis force and centrifugal force matrix, and gravity term in the dynamic system model, respectively, and u is the total control variable.

5. The method according to claim 1, characterized in that, The equivalent control term derived based on the dynamic system model and sliding mode conditions is as follows: Based on the desired state of the controlled object and its derivatives, combined with sliding mode conditions Dynamic system model of the controlled object Perform the derivation; Depend on The desired error acceleration for sliding mode control can be obtained by solving the problem. ; Combining the definition of error The actual acceleration of the controlled object Represented as ; Substituting the actual acceleration into the dynamic system model and neglecting the lumped uncertainty term d, the equivalent control term is obtained by solving. The expression: 。 6. The method according to claim 4, characterized in that, The robust approach term is obtained by using the disturbance compensation term for the lumped uncertainty estimate and the adaptive approach component generated by the fuzzy logic system with sliding mode variables as input. Specifically: The robust approach term Composed of disturbance compensation term and adaptive approach component The composition, expressed as: Among them, the disturbance compensation item is This is used to offset the lumped uncertainty estimated by the nonlinear perturbation observer; the adaptive approaching component It is generated by a fuzzy logic system with sliding mode variable s as input, which adjusts the approach speed and suppresses chattering.

7. The method according to claim 6, characterized in that, The method of generating adaptive approaching components using a fuzzy logic system with sliding mode variables as input specifically involves: Using the sliding mode variable s as input, the universe of discourse is fuzzified into five fuzzy sets: negative large NB, negative small NS, zero ZE, positive small PS, and positive large PB, and described by triangular or Gaussian membership functions. Establish the following fuzzy rule base: If s is NB, then For Zhengda PB; If s is NS, then For the positive PS; If s is ZE, then Zero ZE; If s is PS, then For negative NS; If s is PB, then For negative NB; The fuzzification was defuzzified using a Mamdani-type fuzzy inference engine and the center-of-gravity method, resulting in... The output value.

8. A digital exhibition intelligent control system based on the Internet of Things, characterized in that, Includes the following modules: The prediction deviation calculation module is used to obtain the dynamic system model and current state deviation of the target controlled object in the digital exhibition environment; based on the system topology and historical state data of the digital exhibition environment, the module uses a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a certain future moment, and obtains the prediction deviation vector. The estimation module is used to obtain a nonlinear integral sliding surface for fusing prediction information using the current state deviation and the measured deviation vector; and to obtain an estimate of the lumped uncertainty during system operation by using a nonlinear perturbation observer online. The control module is used to derive the equivalent control term based on the dynamic system model and sliding mode conditions, and to obtain the robust approach term based on the disturbance compensation term for the lumped uncertainty estimate and the adaptive approach component generated by the fuzzy logic system with the sliding mode variable as input; the total control quantity is composed of the equivalent control term and the robust approach term. The target controlled object is controlled using the total control quantity.

9. The system according to claim 8, characterized in that, The nonlinear integral sliding surface for obtaining fused prediction information using the current state deviation and the measured deviation vector is specifically as follows: The expression for the nonlinear integral sliding surface s is: ; Where e is the current state deviation. The prediction bias vector is defined by c1, c2, c3, and γ, which are preset constant control gains.

10. The system according to claim 8, characterized in that, The method of using a graph convolutional gated recurrent unit network to predict the state deviation of the target controlled object at a future time, and obtaining a predicted deviation vector, includes: The sensors and actuators in the digital exhibition environment are abstracted as network nodes, and the communication relationships and physical connections between nodes are used as edges to construct the adjacency matrix of the system topology. The historical state data sequence of each node over the past N consecutive sampling times is used as the feature matrix of the node; The adjacency matrix and node feature matrix are input into the graph convolutional layer, and the neighbor node information is aggregated through graph convolution operation to obtain the spatial association features of the entire system at the current time. The spatial feature vector output by the graph convolutional layer is fed into the subsequent gated recurrent unit layer in chronological order to learn the dynamic evolution law of the state data in the time dimension, and output the predicted value of the state deviation at future time, thus obtaining the prediction deviation vector.

Images