An intelligent exhibition hall temperature central control system based on an internet of things

Abstract

The application relates to the technical field of intelligent exhibition hall temperature control, and discloses an intelligent exhibition hall temperature central control system based on the Internet of Things, which comprises a data acquisition module, a feature extraction module, a trend prediction module and an optimization control module. The data acquisition module collects real-time temperature data through an Internet of Things temperature sensor node and generates temperature field data with time and space labels through time and space registration. The feature extraction module performs empirical mode decomposition on the temperature field data, extracts intrinsic mode function components and calculates temperature fluctuation feature vectors. The trend prediction module inputs the temperature fluctuation feature vectors into a long short-term memory network model to predict a predicted temperature sequence in a future preset time period. According to the difference between the predicted temperature sequence and a preset comfortable temperature threshold, the optimization control module adopts a multi-objective particle swarm optimization algorithm to iteratively optimize the valve opening degree of air conditioner terminal equipment, thereby generating a valve opening degree control instruction set. The instruction issuing module issues the valve opening degree control instruction set to corresponding actuators through an Internet of Things gateway.

Description

Technical Field

[0001] This invention relates to the field of smart exhibition hall temperature control technology, specifically to a smart exhibition hall temperature control system based on the Internet of Things. Background Technology

[0002] Large exhibition halls have complex spatial structures, and different exhibition areas exhibit significant uneven temperature distribution and dynamic fluctuations due to differences in visitor density, exhibit heat source distribution, and building orientation. Existing exhibition hall temperature control solutions mostly rely on single-point temperature detection and proportional-integral-derivative control or simple threshold triggering logic, directly driving the air conditioning terminal equipment by collecting temperature values ​​from a few discrete points. This control mode can only respond with a lag to temperature deviations that have already occurred, and cannot anticipate the spatiotemporal evolution trend of the temperature field within the exhibition area. The heat load changes brought about by the flow of people within the exhibition hall are random and sudden, causing the fixed-parameter control system to frequently adjust valves, resulting not only in adjustment lag but also exacerbating mechanical wear on the actuators. Temperature data collected by discrete sensors are fragmented in both time and space dimensions, making it difficult to reconstruct the continuous temperature distribution of the exhibition hall, resulting in a single-dimensional information basis for control decisions. Feedback control based on temperature deviations at a single moment cannot distinguish the intrinsic scale characteristics of temperature fluctuations. Faced with the coupled thermal environment of multiple exhibition areas, the lack of means to extract the multi-timescale structural characteristics of temperature data leads to insufficient capture of the essential laws of temperature change, thus affecting the quality of input information for predictive models. Traditional control strategies, when dealing with the temperature regulation needs of multiple exhibition areas, typically treat each area independently or consider only a single performance indicator, neglecting the trade-off between the overall thermal comfort of the exhibition hall and the energy consumption of the air conditioning system. Optimization results under fixed weights or single objectives often emphasize one aspect of performance, failing to effectively balance the sum of temperature deviations across multiple exhibition areas with the total valve adjustment range, making it difficult to obtain a control scheme that takes into account both thermal comfort and equipment stability. Summary of the Invention

[0003] The purpose of this invention is to provide a smart exhibition hall temperature control system based on the Internet of Things, which solves the problem that existing exhibition hall temperature control schemes cannot perceive the spatiotemporal evolution of the temperature field due to reliance on delayed feedback and single-scale data analysis. At the same time, it overcomes the technical defects in multi-exhibition area control that make it difficult to achieve a balance between thermal comfort and valve adjustment costs.

[0004] The objective of this invention can be achieved through the following technical solutions: This invention provides a smart exhibition hall temperature control system based on the Internet of Things (IoT), comprising a data acquisition module, a feature extraction module, a trend prediction module, an optimization control module, and a command issuance module. The data acquisition module collects real-time temperature data through IoT temperature sensor nodes deployed in various exhibition areas of the exhibition hall, and performs spatiotemporal registration on the real-time temperature data according to the geographical location identifiers of each IoT temperature sensor node to generate temperature field data with spatiotemporal tags. Preferably, the three-dimensional spatial coordinates of each IoT temperature sensor node are obtained based on its built-in global positioning module, and these three-dimensional spatial coordinates are attached as geographical location identifiers to the real-time temperature data collected by that node. Kriging interpolation is performed on the real-time temperature data collected by each IoT temperature sensor node at the same time according to its three-dimensional spatial coordinates to obtain continuously spatially distributed temperature grid data within the exhibition hall. The temperature grid data is arranged in chronological order of collection time, and the timestamp and spatial coordinates of each temperature grid point are jointly encoded to generate temperature field data with spatiotemporal tags. By constructing a continuous temperature field through Kriging interpolation, the spatial sampling blind spots caused by sparse sensor deployment are eliminated, enabling temperature data to accurately reflect the local temperature differences in various areas of the exhibition hall, and providing a high-fidelity spatiotemporal data foundation for subsequent feature analysis.

[0005] The feature extraction module performs empirical mode decomposition on the spatiotemporally labeled temperature field data, extracts intrinsic mode function (IMF) components, and calculates the temperature fluctuation feature vector of each exhibition area based on the IMF components. As a technical solution of this invention, temperature time series data corresponding to the target exhibition area is extracted from the spatiotemporally labeled temperature field data; empirical mode decomposition is performed on the temperature time series data, and upper and lower envelopes are constructed by identifying local maxima and local minima, and the mean sequence of the upper and lower envelopes is calculated; the mean sequence is subtracted from the temperature time series data to obtain candidate IMF components, and the envelope construction and subtraction operations are repeated until the IMF determination conditions are met, extracting multiple IMF components; statistical features of the instantaneous energy and instantaneous frequency of each IMF component are calculated, and the statistical features are concatenated to form the temperature fluctuation feature vector of the target exhibition area. Furthermore, a Hilbert transform is performed on each intrinsic mode function component to obtain an analytic signal. The instantaneous energy value is calculated based on the magnitude of the analytic signal, and the instantaneous frequency value is obtained by phase difference. The mean, variance, and skewness of the instantaneous energy and instantaneous frequency are calculated separately, and the statistical features are sequentially concatenated according to the decomposition order of the intrinsic mode function components to form a temperature fluctuation feature vector. The feature depth extracted in this way characterizes the transient energy and frequency changes of temperature fluctuations at different time scales, effectively improving the ability to represent complex temperature dynamics and making subsequent trend predictions closer to the actual temperature evolution law.

[0006] The trend prediction module inputs the temperature fluctuation feature vector into a pre-constructed long short-term memory (LSTM) network model to predict the temperature change trend of each exhibition area within a preset time period, thus obtaining a predicted temperature sequence. Preferably, the temperature fluctuation feature vector is input sequentially into the input gate of the LTM network model according to time steps, with the input gate controlling the update weight of each component. A forget gate selectively retains the cell state from the previous time step, obtaining a filtered cell state. An output gate performs a nonlinear transformation between the filtered cell state and the hidden state of the current time step, outputting predicted temperature values ​​for multiple future times. All predicted temperature values ​​are arranged chronologically to generate a predicted temperature sequence. Specifically, the forget gate concatenates the hidden state from the previous time step with the current input, maps it using a weight matrix and activation function to generate a forget gate activation vector, and multiplies it element-wise with the previous cell state to achieve selective forgetting. Utilizing the gating mechanism of the LTM network, the system can capture the long-term dependencies of the temperature fluctuation sequence, avoiding error accumulation in long-term predictions using traditional methods and improving prediction accuracy.

[0007] The optimization control module uses a multi-objective particle swarm optimization algorithm to iteratively optimize the valve opening of the air conditioning terminal equipment corresponding to each exhibition area based on the difference between the predicted temperature sequence and the preset comfort temperature threshold, generating a valve opening control instruction set. In a preferred embodiment of the invention, the predicted temperature value at each moment in the predicted temperature sequence is subtracted from the preset comfort temperature threshold to obtain the temperature deviation value at each moment. Based on the temperature deviation value, the first objective function of the multi-objective particle swarm optimization algorithm is constructed to minimize the sum of squared temperature deviations for each exhibition area, and the second objective function is constructed to minimize the sum of valve opening adjustment amplitudes for all air conditioning terminal equipment. The position vector of each particle in the particle swarm is initialized to a set of candidate valve opening values ​​for each air conditioning terminal equipment, and the velocity vector of each particle is initialized to zero. The first objective function value and the second objective function value are calculated for each particle, and the individual optimal position and the global optimal position of the particle swarm are updated according to the Pareto dominance relationship. The velocity vector and position vector of each particle are updated according to the individual optimal position and the global optimal position. The objective function value is calculated and the optimal position is updated repeatedly until a preset number of iterations is reached. The set of candidate valve opening values ​​corresponding to the global optimal position is used as the valve opening control instruction set. By simultaneously minimizing the temperature deviation in the exhibition area and the valve adjustment range, the frequent operation of the air conditioning terminal equipment is effectively suppressed while ensuring the comfort of the visitors, reducing the energy consumption of the system and achieving a dynamic balance between comfort and energy saving in the exhibition hall environment control.

[0008] The instruction issuing module sends the valve opening control instruction set to the actuators of the corresponding air conditioning terminal devices through the IoT gateway. Further, it extracts the device identifier and corresponding target valve opening value for each air conditioning terminal device from the valve opening control instruction set; encapsulates the target valve opening value into a control frame according to the device identifier, the control frame including a frame header, device address segment, valve opening data segment, and cyclic redundancy check segment; the control frame is broadcast to all actuators of the air conditioning terminal devices via the wireless communication module of the IoT gateway; each actuator continuously listens for the control frames sent by the IoT gateway, and when it receives a control frame, it extracts the cyclic redundancy check segment of the control frame for verification. If the verification passes, it parses the device address segment to obtain the target device address, compares the target device address with the unique identification code burned into the actuator at the factory, and if they match, it determines that the address matches, extracts the valve opening data segment from the control frame and converts it into the target valve opening value, driving the stepper motor inside the actuator to rotate the corresponding angle, thus rotating the valve to the target opening position. This distribution mechanism utilizes broadcast communication and address matching to ensure accurate delivery of control commands in multi-actuator scenarios, and guarantees transmission reliability through cyclic redundancy check. Even in complex electromagnetic environments, it can accurately execute valve opening control and achieve closed-loop distributed control of exhibition hall temperature.

[0009] The beneficial effects of this invention are: Temperature time-series data is adaptively decomposed into multiple intrinsic mode function (EMF) components using empirical mode decomposition (EMD). A Hilbert transform is performed on each component to extract statistical features of instantaneous energy and frequency, including mean, variance, and skewness. These features are then concatenated in the order of decomposition to form a temperature fluctuation feature vector. The fluctuation patterns at different time scales inherent in the temperature signal are separated into independent EMF components. The instantaneous energy statistical features reflect the intensity distribution and asymmetry of temperature fluctuations at different scales, while the instantaneous frequency statistical features reveal the rate of change of fluctuations at each scale. This processing method allows the temperature field dynamics to no longer be presented in the single form of the original time-series numerical values, but rather injected into the prediction model in the form of a multi-scale analytical structure. The prediction model can learn the deep dynamic characteristics of the thermal environment evolution within the exhibition area from the intrinsic modal composition of the temperature signal, enabling the predicted temperature sequence to capture local abrupt changes and periodic fluctuations in the exhibition area's temperature in advance. Based on the deviation between the predicted temperature sequence and the preset comfortable temperature threshold, a first objective function is constructed to minimize the sum of squared temperature deviations in each zone. A second objective function is constructed to minimize the sum of valve opening adjustments for all air conditioning terminal devices. A multi-objective particle swarm optimization algorithm is employed, updating the individual optimal position and the global optimal position according to the Pareto dominance relationship during particle swarm iterations, ultimately outputting a set of valve opening control commands. The two objective functions are processed simultaneously within the multi-objective optimization framework. The particle swarm maintains solution diversity during the search process, and the non-dominated solutions on the Pareto front demonstrate the trade-off boundary between temperature deviation suppression and valve adjustment costs. The generation of valve opening commands no longer relies on a weighted approach of aggregating multiple objectives into a single objective. Instead, the Pareto mechanism preserves the competitive relationship between the two objectives. The valve opening combination corresponding to the global optimal position controls the temperature in each zone towards the comfortable range while avoiding excessively frequent adjustments to the air conditioning terminal devices, suppressing energy loss and mechanical wear caused by repeated valve actions. Attached Figure Description

[0010] The invention will now be further described with reference to the accompanying drawings.

[0011] Figure 1 This is a schematic diagram of a smart exhibition hall temperature control system based on the Internet of Things. Figure 2 This is a flowchart of the control instruction generation process based on multi-objective particle swarm optimization in the optimization control module; Figure 3 This is a flowchart illustrating the encapsulation, broadcast transmission, and execution of valve opening control commands for air conditioning terminal equipment. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] See Figure 1 This invention provides a smart exhibition hall temperature control system based on the Internet of Things (IoT), comprising a data acquisition module, a feature extraction module, a trend prediction module, an optimization control module, and an instruction issuance module. The data acquisition module collects real-time temperature data through IoT temperature sensor nodes deployed in various exhibition areas of the exhibition hall, and performs spatiotemporal registration of the real-time temperature data according to the geographical location identifiers of each IoT temperature sensor node, generating temperature field data with spatiotemporal labels. The feature extraction module performs empirical mode decomposition on the temperature field data with spatiotemporal labels, extracts intrinsic mode function components (IMFs), and calculates temperature fluctuation feature vectors for each exhibition area based on the IMFs. The trend prediction module inputs the temperature fluctuation feature vectors into a pre-constructed long short-term memory (LSTM) network model to predict the temperature change trend of each exhibition area within a preset future time period, obtaining a predicted temperature sequence. The optimization control module uses a multi-objective particle swarm optimization algorithm to iteratively optimize the valve opening of the air conditioning terminal equipment corresponding to each exhibition area based on the difference between the predicted temperature sequence and a preset comfort temperature threshold, generating a valve opening control instruction set. The instruction sending module sends the valve opening control instruction set to the actuator of the corresponding air conditioning terminal device through the Internet of Things gateway.

[0014] In practical implementation, each IoT temperature sensor node's built-in GPS module acquires its 3D spatial coordinates and attaches these coordinates as a geographic location identifier to the real-time temperature data collected by the IoT temperature sensor node. The IoT temperature sensor node then transmits the real-time temperature data carrying the geographic location identifier to the data acquisition module via the IoT. Kriging interpolation is performed on the real-time temperature data collected by each IoT temperature sensor node at the same time, based on the 3D spatial coordinates, to obtain a continuous spatially distributed temperature grid data within the exhibition hall. The 3D spatial coordinates and real-time temperature data of all IoT temperature sensor nodes at the target time are acquired to construct a sample point set of spatial coordinates and temperature values. Each sample point in the sample point set corresponds to the 3D spatial coordinates and real-time temperature data of one IoT temperature sensor node. The spatial distance between every two sample points in the sample point set is calculated, and a variogram model is fitted based on the spatial distance to determine the nugget value, sill value, and range parameters of the variogram model. The variogram model adopts a spherical model. The nugget value is determined by the temperature measurement error over short distances and spatial variability, and is obtained by fitting the experimental variogram scatter plot using the least squares method. The sill value is the variogram value when stability is achieved, which is equal to the sum of the nugget value and the arch height. The range is the distance at which spatial correlation disappears; when the spatial distance between two sample points exceeds the range, the temperature values ​​of the two sample points no longer have spatial correlation. The exhibition space is divided into regular grids to obtain multiple interpolation grid points. The grid spacing is determined based on the exhibition space size and temperature monitoring accuracy requirements, and is set to 0.5 meters to 2 meters. For each interpolation grid point, sample points within a preset radius centered on the interpolation grid point are selected as reference samples, with the preset radius set to 1.5 to 2 times the range. The Kriging weight coefficients of each reference sample are calculated according to the variogram model, and the ordinary Kriging equations are solved. Simultaneously satisfying the constraints .in, Indicates the number of reference samples. Indicates the first Kriging weight coefficients for each reference sample, Indicates the first The first reference sample and the first The variogram value of the spatial distance between reference samples. Indicates the first The first reference sample and the first Spatial distance between reference samples Represents the Lagrange multipliers. Indicates the first The variogram value of the spatial distance between each reference sample and the grid points to be interpolated. Indicates the first The spatial distance between each reference sample and the grid point to be interpolated. The above equation system is a symmetric positive definite linear equation system, solved using the Chulesky decomposition method to obtain the Kriging weight coefficients for each reference sample. The temperature values ​​of each reference sample are weighted and summed according to their corresponding Kriging weight coefficients to obtain the interpolated temperature values ​​for the grid points to be interpolated. The weighted summation process is repeated to calculate the interpolated temperature values ​​for all grid points to be interpolated, and the interpolated temperature values ​​of all grid points to be interpolated are organized into temperature grid data according to spatial arrangement. The temperature grid data is a two-dimensional or three-dimensional matrix, and the matrix elements are the interpolated temperature values ​​at the grid points. The temperature grid data is arranged in chronological order of acquisition time, and the timestamp of each temperature grid point is jointly encoded with the spatial coordinates of the temperature grid point to generate temperature field data with spatiotemporal labels. The joint encoding method is to concatenate the timestamp and the three-dimensional spatial coordinates into a tuple in the form of (time, x, y, z), which is stored as structured data to form temperature field data with spatiotemporal labels.

[0015] In the specific implementation, temperature time series data corresponding to the target exhibition area is extracted from the temperature field data with spatiotemporal labels. The feature extraction module, based on the spatial coordinate range of the target exhibition area, retrieves all temperature grid points in the spatiotemporally labeled temperature field data whose spatial coordinates fall within the target exhibition area's range, corresponding to all timestamps. Temperature values ​​at different timestamps for the same spatial grid point are arranged chronologically, and the spatial average of the temperature time series data for all spatial grid points within the target exhibition area is taken to generate a temperature time series data representing the overall temperature change of the target exhibition area. Empirical mode decomposition (EMD) is performed on the temperature time series data. All local maxima and minima in the temperature time series data are identified. Cubic spline interpolation is used to interpolate all local maxima to construct the upper envelope of the temperature time series data, and the same function is used to interpolate all local minima to construct the lower envelope. The average values ​​of the upper and lower envelopes at each sampling time are calculated to obtain the mean sequence. The mean sequence is subtracted from the temperature time series data to obtain candidate intrinsic mode function components. Repeat the envelope construction and subtraction operations for the candidate intrinsic mode function (IMF) components. Using the candidate IMF components as new temperature time series data, re-identify all local maxima and minima within the candidate IMF components. Construct the upper and lower envelopes of the candidate IMF components using cubic spline interpolation. Calculate the mean sequences of the upper and lower envelopes of the candidate IMF components. Subtract the mean sequences from the candidate IMF components to obtain the updated candidate IMF components. Repeat the envelope construction and subtraction operations until the IMF determination criteria are met. The IMF determination criteria are: over the entire time range, the sum of the number of local maxima and local minima of the candidate IMF components is equal to or at most differs from the number of zero-crossing points by one; and at any given time point, the average of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero. Additionally, a screening termination threshold is set, and the standardized squared difference (SSD) of the candidate intrinsic mode function (IMF) components obtained from two consecutive screenings is calculated. The SSD is calculated by dividing the sum of the squares of the differences between the two consecutive candidate IMF components by the sum of the squares of the previous candidate IMF components. Screening stops when the SSD is less than a preset threshold. The preset threshold is set to 0.25, which is determined based on the midpoint of the screening termination threshold range of 0.2 to 0.3 in classic research on empirical mode decomposition algorithms. After the IMF determination condition is met, the candidate IMF component at this point is extracted as a single IMF component.The extracted intrinsic mode function (IMF) components are subtracted from the original temperature time series data to obtain the residual sequence. This residual sequence is then used as new temperature time series data, and the entire process of empirical mode decomposition (EMD) is repeated to extract the next IMF component. This process continues until the residual sequence becomes a monotonic function or no more IMF components satisfying the IMF criterion can be extracted, ultimately resulting in multiple IMF components. For each IMF component, the statistical characteristics of instantaneous energy and instantaneous frequency are calculated, and all statistical characteristics are concatenated to form the temperature fluctuation feature vector of the target region. For each IMF component, a Hilbert transform is performed on the target IMF component. The Hilbert transform combines the target IMF component with... Convolution is performed to obtain the Hilbert transform of the target intrinsic mode function (EMF) components. The target EMF components are used as the real part, and their Hilbert transform results are used as the imaginary part to construct the analytic signal of the target EMF components. The instantaneous energy value of each sampling point of the target EMF component is calculated based on the magnitude of the analytic signal. The magnitude of the analytic signal is the square root of the sum of the squares of the real and imaginary parts; the instantaneous energy value is the square of the magnitude of the analytic signal. Differential operation is performed on the phase of the analytic signal to obtain the instantaneous frequency value of each sampling point of the target EMF component. The formula for calculating the instantaneous frequency value is: in, Indicates the first The intrinsic mode function components at the th eigenmode function component in the ... The instantaneous frequency value of each sampling point. This represents the index of the intrinsic mode function component. The sampling point number, The total number of sampling points, ranging from 2 to the target intrinsic mode function components; Indicates the first The intrinsic mode function components at the th eigenmode function component in the ... The instantaneous phase of each sampling point is obtained by calculating the arctangent of the ratio of the real part to the imaginary part of the analytic signal; Indicates the first The intrinsic mode function components at the th eigenmode function component in the ... The instantaneous phase of each sampling point; This represents the sampling time interval for temperature time series data. Its value is the single acquisition interval of the IoT temperature sensor node, which is set to 600 seconds according to the temperature monitoring configuration of the exhibition hall. A constant equal to twice pi is used to convert angular frequency to frequency. The mean, variance, and skewness of the instantaneous energy values ​​corresponding to all instantaneous energy values ​​of the target intrinsic mode function components are calculated to obtain the instantaneous energy statistical characteristics. The mean of the instantaneous energy value is the arithmetic mean of the instantaneous energy values ​​at all sampling points; the variance of the instantaneous energy value is the average of the squares of the differences between the instantaneous energy value at each sampling point and the mean; and the skewness of the instantaneous energy value is the average of the cubes of the differences between the instantaneous energy value at each sampling point and the mean, divided by the cube of the standard deviation of the instantaneous energy value. The mean, variance, and skewness of the instantaneous frequency values ​​corresponding to all instantaneous frequency values ​​of the target intrinsic mode function components are calculated to obtain the instantaneous frequency statistical characteristics. The instantaneous energy statistical characteristics and instantaneous frequency statistical characteristics of each intrinsic mode function component are concatenated in the order of decomposition of the intrinsic mode function components to form a high-dimensional real vector, which is the temperature fluctuation characteristic vector of the target area.

[0016] In practice, the core architecture of the Long Short-Term Memory Network model consists of three cascaded cell units: the input gate level, the forget gate level, and the output gate level. The three levels are connected by cell state vectors and hidden state vectors. Before deployment, the Long Short-Term Memory (LSTM) network model was trained offline to learn its parameters. The offline training process was as follows: temperature fluctuation feature vectors generated by the feature extraction module in each exhibition area of ​​the exhibition hall during historical periods were collected as training input samples, and the actual temperature sequences of subsequent preset periods were used as training labels. The preset periods were set to 30 to 120 minutes. A mean squared error loss function was constructed, and the training input samples were input into the initialized LSM network model to obtain temperature prediction values. The mean squared error between the temperature prediction values ​​and the training labels was calculated. The gradients of all weight matrices and bias terms in the LSM network model were calculated using a time-based backpropagation algorithm. The parameters of the LSM network model were updated using the Adam optimizer. The initial learning rate of the Adam optimizer was set to 0.001, the first moment decay coefficient was set to 0.9, the second moment decay coefficient was set to 0.999, the number of training epochs was set to 200, and the number of training samples per batch was set to 32. Training was stopped when the decrease in the loss function value within 10 consecutive training epochs was less than 0.0001, resulting in the pre-constructed LSM network model. The temperature fluctuation feature vector is sequentially input into the input gate of the Long Short-Term Memory (LSTM) network model according to time steps. The temperature fluctuation feature vector is a sequence of data composed of multiple time steps, with each time step corresponding to a specific moment in time. The temperature fluctuation feature vector of the current time step is first concatenated with the hidden state vector output from the previous time step to obtain the input gate concatenation vector. This input gate concatenation vector is multiplied by the input gate weight matrix and then an input gate bias term is added. The dimension of the input gate weight matrix is ​​the sum of the dimensions of the hidden state vector and the temperature fluctuation feature vector multiplied by the dimension of the hidden state vector. The input gate bias term is a vector with the same dimension as the hidden state vector. The result of adding the input gate bias term is mapped to the zero-to-one interval using a sigmoid activation function to generate the input gate activation vector. Simultaneously, the input gate concatenation vector is multiplied by the candidate cell state weight matrix and then a candidate cell state bias term is added. The dimension of the candidate cell state weight matrix is ​​the same as the dimension of the input gate weight matrix, and the candidate cell state bias term is a vector with the same dimension as the hidden state vector. This result is then mapped to the negative-one interval using a hyperbolic tangent activation function to generate the candidate cell state vector. The input gate activation vector is multiplied element-wise with the candidate cell state vector to obtain the candidate cell state vector controlled by the input gate. The cell state from the previous time step is selectively retained using the forget gate of the Long Short-Term Memory network model, resulting in the filtered cell state. The hidden state vector from the previous time step is concatenated with the temperature fluctuation feature vector input at the current time step to obtain the forget gate input vector.The forget gate input vector is multiplied by the forget gate weight matrix, and then a forget gate bias term is added. The dimension of the forget gate weight matrix is ​​the sum of the hidden state vector dimension and the temperature fluctuation feature vector dimension multiplied by the hidden state vector dimension. The forget gate bias term is a vector with the same dimension as the hidden state vector, and its initial value is a vector of all 1s. After being mapped to the zero-to-one interval by a sigmoid activation function, the forget gate activation vector is generated. The forget gate activation vector is then multiplied element-wise with the cell state vector from the previous time step to obtain the intermediate cell state after selective forgetting. The intermediate cell state is then added element-wise with the candidate cell state vector controlled by the input gate to obtain the cell state vector for the current time step. The cell state vector for the current time step is the filtered cell state and is output to the output gate of the Long Short-Term Memory (LSTM) network model. The output gate of the LSM network model performs a nonlinear transformation on the filtered cell state and the hidden state of the current time step to output the temperature prediction values ​​for multiple future time steps. The hidden state vector from the previous time step is concatenated with the temperature fluctuation feature vector input at the current time step to obtain the output gate input vector. The output gate input vector is multiplied by the output gate weight matrix, and then an output gate bias term is added. The output gate weight matrix has the same dimension as the forget gate weight matrix, and the output gate bias term is a vector with the same dimension as the hidden state vector. This is then mapped to the zero-to-one interval using a sigmoid activation function to generate the output gate activation vector. The cell state vector at the current time step is mapped to the negative-to-one interval using a hyperbolic tangent activation function. The mapped cell state vector is then multiplied element-wise with the output gate activation vector to obtain the hidden state vector at the current time step. The hidden state vector at the current time step is mapped to the temperature value space through a fully connected prediction layer. The dimension of the weight matrix of the fully connected prediction layer is the dimension of the hidden state vector multiplied by the number of temperature prediction times within a preset time period. The bias term of the fully connected prediction layer is a vector with a dimension equal to the number of temperature prediction times, which is obtained by dividing the preset time period length by the sampling time interval. The multiple temperature prediction values ​​output by the fully connected prediction layer are advanced according to the time step. At each time step, a set of temperature prediction values ​​corresponding to multiple future times is output. The temperature prediction value output at the last time step is taken as the final prediction result. Arrange all the predicted temperature values ​​in chronological order to generate a predicted temperature sequence. The time span of the predicted temperature sequence is consistent with the preset time period.

[0017] In specific implementation, please refer to Figure 2The optimization control module acquires the predicted temperature sequence output by the trend prediction module. This sequence contains predicted temperature values ​​for each exhibition area within a preset time period, the length of which is consistent with the trend prediction module's settings. The optimization control module stores a preset comfort temperature threshold, pre-set to 25 degrees Celsius based on standards for the preservation of cultural relics or for personnel comfort. The predicted temperature value at each moment in the predicted temperature sequence is subtracted from the preset comfort temperature threshold to obtain the temperature deviation value for each moment. This temperature deviation calculation is performed separately for each exhibition area at each future moment, forming a temperature deviation value matrix. The rows of the temperature deviation value matrix correspond to the exhibition areas, and the columns correspond to the predicted times.

[0018] Based on the temperature deviation values, a first objective function and a second objective function for a multi-objective particle swarm optimization algorithm are constructed. The first objective function is to minimize the sum of squared temperature deviations in each expansion zone, and its expression is: in, This represents a vector consisting of the valve openings of all air conditioning terminal devices, with the vector dimension equal to the total number of air conditioning terminal devices. ; Indicates the total number of exhibition areas; This indicates the total number of times included in the predicted temperature sequence; Indicates the first Individual exhibition area in The predicted temperature value at a future time; The preset comfort temperature threshold is 25 degrees Celsius. The second objective function minimizes the sum of valve opening adjustments across all air conditioning terminal devices. This is defined as the sum of the absolute differences between the current valve opening vector and the candidate valve opening vectors. It is calculated by summing the absolute differences between the candidate valve opening values ​​and the current valve opening value of each air conditioning terminal device. The current valve opening value is obtained from the real-time feedback value from the actuators of the air conditioning terminal devices. If the actuators are not configured with feedback, the target valve opening value issued in the previous control cycle is used as the current valve opening value.

[0019] Initialize the position and velocity vectors of each particle in the particle swarm. The number of particles in the swarm is set to 50. The position vector of each particle represents a set of candidate valve opening values, and the dimension of the position vector is equal to the total number of all air conditioning terminal devices. The value range of each component in the position vector corresponds to the physical range of the valve opening of the air conditioning terminal device, set to 0% to 100%, where 0% represents a fully closed valve and 100% represents a fully open valve. Initially, the value of each component in the position vector of each particle is randomly generated from a uniform distribution between 0% and 100%. The velocity vector of each particle has the same dimension as the position vector and is initialized to zero.

[0020] For each particle, calculate the first objective function value and the second objective function value. For a single particle, use the particle's current position vector as the input set of valve opening values. Substitute this set of valve opening values ​​into the exhibition hall's thermodynamic model to calculate the simulated temperature sequence of each exhibition area within a preset time period. The thermodynamic model is an exhibition hall temperature response model built based on energy conservation and heat transfer equations. The model inputs include the valve opening of the air conditioning terminal equipment, outdoor temperature, exhibition hall envelope parameters, and internal heat load. The model outputs temperature change curves for each exhibition area. Substitute the model output temperature sequence and the preset comfort temperature threshold into the first objective function to calculate the first objective function value. Sum the absolute values ​​of the differences between each component value in the particle's position vector and the corresponding current valve opening value of the air conditioning terminal equipment to obtain the second objective function value. Repeat this process to obtain the first and second objective function values ​​for all particles.

[0021] The Pareto dominance relation is used to update the individual optimal position of each particle and the global optimal position of the particle swarm. The Pareto dominance relation is defined as follows: for particles A and B, if the first objective function value of particle A is not greater than the first objective function value of particle B, and the second objective function value of particle A is not greater than the second objective function value of particle B, and at least one of the inequalities is strictly less than, then particle A is said to dominate particle B. For each particle, the objective function value vector of the particle's current position is compared with the objective function value vector of the particle's individual optimal position. If the particle's current position dominates the particle's individual optimal position, then the particle's individual optimal position is replaced by the particle's current position; if the particle's individual optimal position dominates the particle's current position, then the particle's individual optimal position remains unchanged; if the two do not dominate each other, then the particle's current position and the particle's individual optimal position are randomly selected as the new particle's individual optimal position. For the global optimal position of the particle swarm, the individual optimal positions of all particles are combined into a candidate set. The candidate set is sorted in a non-dominated manner to divide the frontier layer. In the first frontier layer, the crowding distance of the individual optimal position of each particle is calculated. The crowding distance is the normalized value of the sum of the distances between adjacent individuals after sorting according to the objective function values. The optimal position of the individual particle with the largest crowding distance is randomly selected as the global optimal position of the particle swarm.

[0022] The velocity and position vectors of each particle are updated based on the individual particle's optimal position and the global optimal position of the particle swarm. For the ... For the nth particle, update the particle velocity vector of the nth particle. The method for determining the velocity component is as follows: The 3rd dimension of the particle's current velocity vector is... The dimensional component is multiplied by the inertia weight, which has a value of 0.7298; then the product of the first learning factor and the first random number is added, and multiplied by the dimensional component of the particle's optimal position. The dimensional component and the particle's current position The difference between the dimensional components, with the first learning factor taking the value 1.4962, and the first random number being a uniformly distributed random number between 0 and 1; plus the product of the second learning factor and the second random number, multiplied by the global optimal position of the particle swarm. The dimensional component and the particle's current position The difference between the dimensional components, the second learning factor takes the value 1.4962, and the second random number is a uniformly distributed random number between 0 and 1. Update the dimensional component of the particle position vector. The method for using the dimensional component is: to convert the particle's current position vector into its dimensional component. The dimensional component plus the updated velocity vector of the particle Dimensional components. When a certain component of the updated particle position vector exceeds the range of 0% to 100%, the component value exceeding the boundary is corrected to the boundary value, and the corresponding velocity component is multiplied by -0.5 for bounce processing.

[0023] The process of repeatedly calculating the first and second objective function values ​​for each particle, updating the individual optimal position and the global optimal position of the particle, and updating the particle velocity vector and position vector continues until a preset number of iterations is reached (set to 100). After completing the iterations, the set of candidate valve opening values ​​corresponding to the global optimal position of the particle swarm is used as the valve opening control instruction set. This set includes the device identifier and the corresponding target valve opening value for each air conditioning terminal device.

[0024] In specific implementation, please refer to Figure 3 The instruction issuing module obtains the valve opening control instruction set from the optimization control module. This instruction set is a data table, where each row contains the device identifier of the air conditioning terminal equipment and the corresponding target valve opening value. The device identifier is a unique number assigned to the air conditioning terminal equipment during production, and the target valve opening value is a real number ranging from 0% to 100%. The instruction issuing module iterates through the valve opening control instruction set, extracting the device identifier and corresponding target valve opening value for each air conditioning terminal equipment.

[0025] The target valve opening value is encapsulated into a control frame according to the device identifier. The control frame adopts a fixed-length binary data frame format, including a frame header, a device address field, a valve opening data field, and a cyclic redundancy check (CRC) segment. The frame header occupies 1 byte and is fixed as the hexadecimal number 0xA5, used to identify the start of a control frame. The device address field occupies 2 bytes and is used to store the device identifier of the air conditioning terminal device. The device identifier is converted into a 16-bit unsigned integer and filled into the device address field, with the high byte first and the low byte last. The valve opening data field occupies 1 byte and converts the target valve opening value into an 8-bit unsigned integer. The conversion method is to divide the target valve opening value by 100% and multiply by 255, then round the result to the nearest integer, so that 0% valve opening corresponds to the binary value 0 and 100% valve opening corresponds to the binary value 255. The CRC segment occupies 2 bytes and uses the CRC-16-CCITT standard to generate the checksum. The generator polynomial is as follows: The initial value is 0xFFFF. Cyclic redundancy check is performed on all bytes of the frame header, device address segment, and valve opening data segment in the order of transmission to obtain a 16-bit check value. The high byte of the check value is filled into the cyclic redundancy check segment with the low byte at the end.

[0026] The encapsulated control frame is broadcast to the actuators of all air conditioning terminal devices via the wireless communication module of the IoT gateway. The IoT gateway is equipped with a ZigBee wireless communication module operating in the 2.4GHz band and using the IEEE 802.15.4 protocol. The IoT gateway hands the control frame to the ZigBee wireless communication module, which assembles the control frame into a ZigBee broadcast frame. The destination address of the ZigBee broadcast frame is set to the broadcast address 0xFFFF, and the payload of the ZigBee broadcast frame is filled with all the bytes of the control frame. The ZigBee wireless communication module transmits the ZigBee broadcast frame on a designated channel. The actuators of the air conditioning terminal devices integrate ZigBee receiving modules, and all actuators simultaneously receive the same broadcast frame.

[0027] Each actuator, upon receiving a control frame, parses the device address segment. It executes the target valve opening value in the valve opening data segment only if the address in the device address segment matches its own address. Each actuator continuously listens for control frames sent by the IoT gateway. The actuator's internal ZigBee receiver module performs carrier sensing on a designated channel. When a signal conforming to the ZigBee broadcast frame format is detected, the signal is demodulated and despread, and the ZigBee broadcast frame payload is extracted to obtain the original control frame. Upon receiving a control frame, the cyclic redundancy check (CRC) segment of the control frame is extracted for verification. The verification method is as follows: the CRC is recalculated for the frame header, device address segment, and valve opening data segment of the control frame, using the same calculation method as the sender. The calculated check value is compared with the CRC segment in the control frame. If the two check values ​​are identical, the verification passes; if they differ, the verification fails, and the control frame is discarded. After successful verification, the actuator parses the device address segment, extracting a 16-bit unsigned integer from the device address segment of the control frame in high-byte-first, low-byte-last order to obtain the target device address. The target device address is compared with the unique identifier burned into the actuator at the factory. The unique identifier is a 16-bit unsigned integer that is embedded in the electrically erasable programmable read-only memory inside the microcontroller during the actuator's manufacturing process. If the target device address and the unique identifier are exactly equal, the address is determined to match; if they are not equal, the address is determined to be mismatched, and the actuator ignores the control frame.

[0028] After address matching, the actuator extracts the valve opening data segment from the control frame and converts the binary value in the valve opening data segment into the target valve opening value. The conversion method is to convert a byte of binary value into an integer from 0 to 255, divide the integer by 255, and then multiply by 100% to obtain the target valve opening value.

[0029] The actuator drives an internal stepper motor to rotate by a corresponding angle based on the target valve opening value, causing the valve to rotate to the target opening position. The microcontroller inside the actuator outputs pulse signals to the stepper motor driver. The stepper motor is a two-phase hybrid stepper motor with a step angle of 1.8 degrees. The relationship between the stepper motor rotation angle and the target valve opening value is linear; the total rotation angle of the stepper motor from fully closed to fully open is 90 degrees. Multiplying the target valve opening value by 90 degrees gives the target angle the stepper motor needs to rotate. Dividing the target angle by the step angle of 1.8 degrees and rounding to the nearest integer gives the number of pulses the microcontroller needs to output. The microcontroller sends pulse signals to the stepper motor driver according to the calculated number of pulses. The stepper motor driver drives the stepper motor to rotate according to the pulse signals. The stepper motor output shaft is connected to the valve stem via a coupling. When the stepper motor rotates, it drives the valve to rotate, bringing the valve to the opening position corresponding to the target valve opening value.

[0030] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A smart exhibition hall temperature control system based on the Internet of Things, characterized in that, include: The data acquisition module collects real-time temperature data through IoT temperature sensor nodes deployed in various exhibition areas of the exhibition hall, and performs spatiotemporal registration of the real-time temperature data according to the geographical location identifier of each IoT temperature sensor node to generate temperature field data with spatiotemporal tags. The feature extraction module performs empirical mode decomposition on the temperature field data with spatiotemporal labels, extracts intrinsic mode function components, and calculates the temperature fluctuation feature vector of each exhibition area based on the intrinsic mode function components. The trend prediction module inputs the temperature fluctuation feature vector into a pre-built long short-term memory network model to predict the temperature change trend of each exhibition area within a preset time period, and obtains the predicted temperature sequence. The optimization control module uses a multi-objective particle swarm optimization algorithm to iteratively optimize the valve opening of the air conditioning terminal equipment corresponding to each exhibition area based on the difference between the predicted temperature sequence and the preset comfort temperature threshold, and generates a valve opening control instruction set. The instruction issuing module sends the valve opening control instruction set to the actuator of the corresponding air conditioning terminal device through the Internet of Things gateway.

2. The smart exhibition hall temperature control system based on the Internet of Things according to claim 1, characterized in that, The process involves collecting real-time temperature data through IoT temperature sensor nodes deployed in various exhibition areas of the exhibition hall, and performing spatiotemporal registration of the real-time temperature data based on the geographical location identifiers of each IoT temperature sensor node to generate temperature field data with spatiotemporal tags. Specifically: The three-dimensional spatial coordinates of each IoT temperature sensor node are obtained based on the built-in global positioning module, and the three-dimensional spatial coordinates are attached to the real-time temperature data collected by the node as a geographical location identifier. The real-time temperature data collected by each IoT temperature sensor node at the same time is processed by Kriging interpolation according to its three-dimensional spatial coordinates to obtain temperature grid data with continuous spatial distribution in the exhibition hall. The temperature grid data is arranged in chronological order of acquisition time, and the timestamp and spatial coordinates of each temperature grid point are jointly encoded to generate temperature field data with spatiotemporal labels.

3. The smart exhibition hall temperature control system based on the Internet of Things according to claim 2, characterized in that, Empirical mode decomposition (EMD) is performed on the spatiotemporally labeled temperature field data to extract intrinsic mode function (IMF) components. Then, based on these IMF components, the temperature fluctuation feature vectors for each region are calculated. Specifically: Extract the temperature time series data corresponding to the target exhibition area from the temperature field data with spatiotemporal labels; Empirical mode decomposition is performed on the temperature time series data. Upper and lower envelopes are constructed by identifying local maxima and local minima, and the mean sequence of the upper and lower envelopes is calculated. The candidate intrinsic mode function components are obtained by subtracting the mean sequence from the temperature time series data. The envelope construction and subtraction operations are repeated on the candidate intrinsic mode function components until the intrinsic mode function determination conditions are met, and multiple intrinsic mode function components are extracted. Calculate the statistical characteristics of the instantaneous energy and instantaneous frequency of each intrinsic mode function component, and concatenate the statistical characteristics to form the temperature fluctuation feature vector of the target exhibition area.

4. The smart exhibition hall temperature control system based on the Internet of Things according to claim 3, characterized in that, The temperature fluctuation feature vector is input into a pre-constructed long short-term memory network model to predict the temperature change trend of each exhibition area within a preset time period, resulting in a predicted temperature sequence, specifically: The temperature fluctuation feature vector is sequentially input into the input gate of the long short-term memory network model according to the time step, and the update weight of each component in the temperature fluctuation feature vector is controlled through the input gate. The cell state of the previous time step is selectively retained by the forgetting gate of the long short-term memory network model to obtain the filtered cell state. The output gate of the Long Short-Term Memory network model is used to perform a nonlinear transformation between the filtered cell state and the hidden state at the current time step, and output the temperature prediction values ​​for multiple future times. Arrange all the predicted temperature values ​​in chronological order to generate a predicted temperature sequence.

5. The smart exhibition hall temperature control system based on the Internet of Things according to claim 4, characterized in that, Based on the difference between the predicted temperature sequence and the preset comfort temperature threshold, a multi-objective particle swarm optimization algorithm is used to iteratively optimize the valve opening of the air conditioning terminal equipment corresponding to each exhibition area, generating a valve opening control instruction set, specifically: The temperature deviation value at each moment is obtained by subtracting the predicted temperature value from the preset comfort temperature threshold in the predicted temperature sequence. Based on the temperature deviation value, the first objective function of the multi-objective particle swarm optimization algorithm is to minimize the sum of squared temperature deviations in each exhibition area, and the second objective function is to minimize the sum of valve opening adjustment amplitudes of all air conditioning terminal devices. The position vector of each particle in the particle swarm is initialized to the set of candidate valve opening values ​​for each air conditioning terminal device, and the velocity vector of each particle is initialized to zero. For each particle, calculate its first objective function value and second objective function value, and update the individual optimal position of each particle and the global optimal position of the particle swarm according to the Pareto dominance relationship. The velocity vector and position vector of each particle are updated based on the individual optimal position and the global optimal position. The objective function value is calculated and the optimal position is updated repeatedly until the preset number of iterations is reached. The set of valve opening candidate values ​​corresponding to the global optimal position is used as the valve opening control instruction set.

6. The smart exhibition hall temperature control system based on the Internet of Things according to claim 5, characterized in that, The valve opening control command set is sent to the actuator of the corresponding air conditioning terminal device through the Internet of Things gateway, specifically as follows: Extract the device identifier and corresponding target valve opening value of each air conditioning terminal device from the valve opening control command set; The target valve opening value is encapsulated into a control frame according to the device identifier. The control frame includes a frame header, a device address field, a valve opening data field, and a cyclic redundancy check field. The control frame is broadcast to the actuators of all air conditioning terminal devices via the wireless communication module of the IoT gateway. After receiving the control frame, each actuator parses the device address segment and executes the target valve opening value in the valve opening data segment only if the address in the device address segment matches the actuator's own address.

7. The smart exhibition hall temperature control system based on the Internet of Things according to claim 2, characterized in that, The real-time temperature data collected by each IoT temperature sensor node at the same time is processed by Kriging interpolation according to its three-dimensional spatial coordinates to obtain continuous spatially distributed temperature grid data within the exhibition hall, specifically: Obtain the three-dimensional spatial coordinates of all IoT temperature sensor nodes at the target time and their corresponding real-time temperature data, and construct a sample point set of spatial coordinates and temperature values; Calculate the spatial distance between every two sample points in the sample point set, and fit a variogram model based on the spatial distance to determine the nugget value, sill value, and range parameter of the variogram model; The exhibition space is divided into regular grids to obtain multiple interpolation grid points. For each interpolation grid point, sample points within its preset radius are selected as reference samples. The Kriging weight coefficients for each reference sample are calculated based on the variogram model. The temperature values ​​of each reference sample are then weighted and summed according to the Kriging weight coefficients to obtain the interpolated temperature value of the grid point to be interpolated. The interpolated temperature values ​​of all grid points to be interpolated are organized into temperature grid data according to their spatial arrangement.

8. The smart exhibition hall temperature control system based on the Internet of Things according to claim 3, characterized in that, The step of calculating the statistical characteristics of the instantaneous energy and instantaneous frequency of each intrinsic mode function component, and then concatenating these statistical characteristics into a temperature fluctuation feature vector for the target exhibition area, specifically involves: Perform a Hilbert transform on each intrinsic mode function component to obtain the analytic signal of that intrinsic mode function component, and calculate the instantaneous energy value of each sampling point based on the magnitude of the analytic signal; The phase of the analytical signal is differentially analyzed to obtain the instantaneous frequency value of each sampling point; The mean, variance, and skewness of all instantaneous energy values ​​corresponding to each intrinsic mode function component are calculated to obtain the instantaneous energy statistical characteristics; The mean, variance, and skewness of all instantaneous frequency values ​​corresponding to each intrinsic mode function component are calculated to obtain the statistical characteristics of the instantaneous frequency. The instantaneous energy and instantaneous frequency statistical characteristics of each intrinsic mode function component are concatenated sequentially according to the decomposition order of the intrinsic mode function components to form a temperature fluctuation feature vector.

9. A smart exhibition hall temperature control system based on the Internet of Things according to claim 4, characterized in that, The long short-term memory network model selectively retains the cell state from the previous time step using a forgetting gate, resulting in a filtered cell state, specifically: The hidden state from the previous time step is concatenated with the temperature fluctuation feature vector input at the current time step to obtain the forget gate input vector. The forget gate input vector is multiplied by the forget gate weight matrix, and then a forget gate bias term is added. After passing through the activation function, it is mapped to the zero-to-one interval to generate the forget gate activation vector. The forget gate activation vector is multiplied element-wise with the cell state of the previous time step to obtain the intermediate cell state after selective forgetting. The intermediate cell states are output as filtered cell states to the input and output gates of the Long Short-Term Memory network model.

10. A smart exhibition hall temperature control system based on the Internet of Things according to claim 6, characterized in that, After receiving the control frame, each actuator parses the device address segment and executes the target valve opening value in the valve opening data segment only if the address in the device address segment matches the actuator's own address. Specifically: Each actuator continuously listens for control frames sent by the IoT gateway. When a control frame is received, the cyclic redundancy check segment of the control frame is extracted for verification. If the verification passes, the device address segment is parsed to obtain the target device address. Compare the target device address with the unique identification code burned into the actuator at the factory. If the two match, the address is determined to be a match. After address matching, the valve opening data segment is extracted from the control frame, and the binary value in the valve opening data segment is converted into the target valve opening value. The stepper motor inside the actuator is driven to rotate by a corresponding angle according to the target valve opening value, thereby rotating the valve to the target opening position.

Images