An energy consumption control method for an electric heating drying room with adaptive capacity

Abstract

The present application relates to the technical field of industrial oven temperature control, and discloses an electric heating oven energy consumption control method with self-adaptive capability, which comprises the following steps: injecting mutually irrelevant pseudo-random binary sequence excitation signals into each heating zone and fan in the initial cold state stage, and establishing a multiple-input single-output autoregressive model for each measuring point; extracting steady-state gain and equivalent time constant for clustering, dividing the airflow dead zone set and the short-circuit channel set, and calculating the performance index of each heating zone to generate a dynamic energy consumption weight; converting each model into a global incremental state space prediction model; constructing a target function with the dynamic energy consumption weight and solving under constraints; and using closed-loop data to update the model online without disturbance and synchronously refreshing the weight and the prediction model when the trigger condition is met. The present application can automatically perceive the airflow organization characteristics of each batch of mixed materials, and preferentially allocates the heat load to the high-efficiency compensation zone under the premise of ensuring temperature uniformity, thereby reducing the waste of electric energy caused by airflow short circuit.

Description

Technical Field

[0001] This invention relates to the field of industrial drying oven temperature control technology, specifically to an energy consumption control method for an electrically heated drying oven with adaptive capabilities. Background Technology

[0002] Wet porcelain blanks typically refer to specific unsintered blanks such as post insulators, bushings, and transformer insulators. Before high-temperature sintering, the wet blanks must be placed in an electrically heated drying chamber where hot air circulation removes their internal moisture to prevent cracking or deformation due to rapid moisture evaporation after entering the kiln. The electrically heated drying chamber uses electric heating tubes as a heat source, and a variable frequency fan drives hot air to circulate inside the chamber. Through convection heat transfer, heat is transferred to the surface of the blank, gradually removing moisture from the surface inwards.

[0003] Existing electric heating drying oven drying control methods typically pre-set a fixed temperature and humidity control curve based on the type and specifications of the blanks, and execute heating, constant temperature and cooling procedures in segments according to time. However, in the actual production of electric porcelain and electrical appliances, in order to meet delivery cycle requirements, different types and sizes of wet blanks are often mixed in the same drying oven. The differences in height, diameter and wall thickness of these blanks cause the ventilation cross section inside the drying oven to exhibit an unpredictable spatial distribution after loading. The flow distribution of hot air on the cross section shows significant non-uniformity, forming a situation where high-temperature stagnation dead zones and low-temperature short-circuit channels coexist, resulting in an uneven temperature field inside the drying oven. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an adaptive energy consumption control method for an electrically heated drying oven, which solves the problem of uneven temperature field inside the drying oven.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an adaptive energy consumption control method for an electrically heated drying oven, comprising the following steps: During the initial operation phase, pseudo-random binary sequence excitation signals are applied to the input power of each heating zone and the input frequency of the fan, and air temperature response data at each measuring point are collected. The input power, the input frequency, and the air temperature response data are converted into power increments, frequency increments, and temperature increments, and a multi-input single-output autoregressive model is established for each of the measurement points. The steady-state gain and equivalent time constant of each of the multi-input single-output autoregressive models are extracted to divide the airflow dead zone set and short-circuit channel set, and the efficiency index of each heating zone is calculated accordingly to generate dynamic energy consumption weights. Each of the multi-input single-output autoregressive models is converted into a global incremental state-space prediction model, and temperature range constraints and spatial temperature difference constraints are set in the prediction time domain. The objective function is constructed using the dynamic energy consumption weight, and solved based on the global incremental state space prediction model under the constraints of the temperature range and the spatial temperature difference, outputting the target control incremental sequence for each heating zone and the fan. During operation, when the temperature prediction residual exceeds the residual threshold, the multi-input single-output autoregressive models are updated online using closed-loop operation data, and the dynamic energy consumption weights and the global incremental state space prediction model are refreshed simultaneously.

[0006] By adopting the above technical solution, uncorrelated pseudo-random excitations are injected in the initial stage of cold state, and the autoregressive model of each measuring point is identified. Steady-state gain and time constant are extracted for clustering to automatically divide dead zones and short-circuit channels and generate heating zone efficiency weights. Then, in model predictive control, dynamic energy consumption weights are used to limit the heating of high loss zones and spatial temperature difference is used as a hard constraint. This allows the control system to autonomously obtain the airflow organization characteristics of the current mixed materials before each batch of drying begins. Under the premise of ensuring temperature uniformity, the heat load is preferentially allocated to the high-efficiency compensation zone. This improves the problem that traditional electric porcelain blank drying can only operate according to a preset fixed temperature and humidity curve, cannot perceive the airflow distribution differences caused by the mixing of different varieties and sizes, and is forced to use low-temperature slow drying to ensure that cracks are not caused, resulting in a large waste of electricity.

[0007] This invention provides an adaptive energy consumption control method for an electrically heated drying oven. It offers the following advantages: 1. This invention injects uncorrelated pseudo-random excitations and collects distributed temperature responses during the cold phase, identifies the autoregressive model of each measuring point online, extracts steady-state gain and time constant for clustering, automatically divides dead zones and short-circuit channels and generates heating zone efficiency weights, so that the control system can autonomously obtain the airflow organization characteristics of the current mixed load before each batch starts, without the need for manual measurement or preset partitioning.

[0008] 2. This invention introduces dynamic energy consumption weights into the model predictive control objective function, penalizing inefficient heating zones (i.e., heating zones where excessive heat flows into dead zones), and explicitly applies hard constraints on the spatial temperature difference between measurement points. This enables the optimizer to automatically reduce ineffective heating in short-circuit channels while ensuring temperature uniformity, and prioritizes the use of high-efficiency compensation zones to reduce energy waste caused by hot air short circuits.

[0009] 3. During operation, this invention monitors the temperature prediction residual and the sufficiency of window data excitation in real time. Under the triggering of joint conditions, it uses closed-loop data to update model parameters without disturbance, and simultaneously refreshes steady-state gain, region division and dynamic weights, so that the control system can maintain model accuracy after the material shrinks and changes the airflow channel, without introducing additional thermal disturbances throughout the process. Attached Figure Description

[0010] Figure 1This is a flowchart of an energy consumption control method for an electric heating drying room with adaptive capability according to the present invention. Figure 2 A schematic diagram of the process for clustering and dividing airflow dead zones and short-circuit channels and generating performance weights according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the model predictive control rolling optimization solution process provided in an embodiment of the present invention. Detailed Implementation

[0011] The technical solution of the present invention will now be clearly and completely described 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.

[0012] Please see the appendix Figure 1 - Appendix Figure 3 This invention provides an adaptive energy consumption control method for an electrically heated drying oven, comprising the following steps: During the initial operation phase, pseudo-random binary sequence excitation signals are applied to the input power of each heating zone and the input frequency of the fan, and air temperature response data at each measuring point are collected. Furthermore, a pseudo-random binary sequence excitation signal is applied to the input power of each heating zone and the input frequency of the fan, specifically including: Different primitive polynomials are configured for each heating zone and fan; Excitation signals of pseudo-random binary sequences of periodically mutually prime numbers are generated by various primitive polynomials. Each pseudo-random binary sequence excitation signal is superimposed on the corresponding basic bias value of the heating zone and the fan for output; Specifically, this method is applied to the wet blank drying process in the production of electrical porcelain and electrical appliances. The materials to be dried include post insulators, bushings, and transformer porcelain bushings. Due to the matching of production plans and order delivery requirements, different types and sizes of wet blanks are usually mixed in the same drying room. The differences in height, diameter, and wall thickness of these blanks result in a complex ventilation cross section inside the drying room. After loading, the actual distribution of airflow channels cannot be known in advance. The flow distribution of hot air on the cross section shows significant non-uniformity, which causes the problem of inherently uneven temperature field and the coexistence of local hot spots and low temperature dead zones during the drying process. The electric heating drying room is equipped with Each heating zone is independently adjustable and equipped with a variable frequency drive circulating fan, arranged within the workspace. Each temperature sensor measures the air temperature at representative locations. Heating zones are arranged along the height of the drying chamber or in the direction of airflow. Each heating zone consists of a set of electric heating tubes, and the power is independently adjusted via solid-state relays or SCR modules. The circulating fan speed is adjusted by a frequency converter to change the airflow circulation intensity inside the drying chamber. All input and output signals are sampled at a fixed period. Synchronized updates Based on the thermal inertia of the drying chamber, the time is determined to be 1 to 10 seconds in this embodiment; At the start of each batch drying process, while the billet is still cold at ambient temperature, the control system enters the initial excitation stage. First, a primitive polynomial is assigned to each heating zone channel and fan channel. Different channels use different primitive polynomials. Each primitive polynomial corresponds to a linear feedback shift register, the register width of which is equal to the polynomial order. Driven by a clock, the system cyclically generates... and The pseudo-random binary sequence composed of two logic levels has a sequence period length of the primitive polynomial selected by each channel that is coprime to each other, that is, there is no common factor greater than 1 between any two sequence periods. This property makes the cross-correlation function value of the pseudo-random sequence of each channel approach zero in any time period, thus satisfying the condition that the excitation signals are linearly independent when multiple channels are excited at the same time. For the first input channel, its basic bias value is denoted as... The basic offset value for the heating zone channel is taken as 50% of the rated power of the heating zone, and the basic offset value for the fan channel is taken as 50% of the rated frequency of the fan. Excitation amplitude A value of 5% to 15% of the channel's rated value is used to limit the power or frequency fluctuations after superimposed disturbances within the process safety margin. At each sampling time... Linear feedback shift register output This value takes the form of +1 or -1, and is used to synthesize the actual output value of this channel. ; For the heating zone channel As power setpoint Output to the power regulator; for the fan duct, As a frequency given value Output to frequency converter; While the excitation signal is continuously injected, the control system synchronously collects the air temperature at all measuring points through temperature sensors. ,in For the measurement point number, Simultaneously, the reference temperature at each measuring point at the initial moment was recorded. Calculate the temperature increment at each sampling time: ; Calculate the power increment and fan frequency increment for each heating zone: ; ; in, For the first Heating zone at sampling time The power increment, in kilowatts. For the first The base power of the heating zone during the excitation phase; For the wind turbine at the sampling time The frequency increment, in Hertz. The fundamental frequency of the fan during the excitation phase, and the temperature increment. As an output observation, power increment and frequency increment As input observations; All input and output data are stored in the data buffer in chronological order, with a data record length of [length missing]. The period is taken as an integer multiple of the pseudo-random sequence period to ensure that each channel has undergone a complete traversal of the excitation pattern during the data recording period; Through the above process, the control system obtains a set of multi-channel input-output observation data reflecting the actual airflow organization characteristics of the current batch of mixed materials within the safety window when the billet is still in a cold state and has not yet generated drying stress. The data includes distributed temperature dynamic response information caused by power changes in each heating zone, as well as differentiated response information at each measuring point caused by changes in fan frequency leading to changes in convective heat transfer intensity. Moreover, the excitation signals of each channel are statistically independent, forming the data foundation required for the subsequent online identification of the multi-input single-output autoregressive model at each measuring point.

[0013] The input power, input frequency, and air temperature response data are converted into power increments, frequency increments, and temperature increments, and a multi-input single-output autoregressive model is established for each measurement point. Furthermore, a multi-input single-output autoregressive model is established for each measurement point location, specifically including: The model output data is constructed using the temperature increment at each measuring point at the current sampling time, and a regression vector is constructed using the power increment, frequency increment, and temperature increment of the historical period. Construct a parameter vector containing the coefficients of the model to be identified; The recursive least squares algorithm with a preset forgetting factor is used to estimate the parameter vector online, and the autoregressive coefficients and input coefficients of the multi-input single-output autoregressive model are obtained. Specifically, after completing the initial excitation and data acquisition, the control system uses the obtained temperature increment, power increment and frequency increment data to establish a multi-input single-output autoregressive model for each temperature measurement point, which is used to describe the dynamic response relationship between the temperature change at each measurement point and the power change and fan frequency change in each heating zone. Regarding the first Each temperature measurement point outputs the observations at the current sampling time. Temperature increment The model input consists of two parts: the first part is the measurement point itself in the past... The historical value of temperature increment at each sampling time, i.e. This part reflects the thermal inertial memory effect of temperature changes; the second part is all A heating zone in the past Historical power increment values ​​at each sampling time and the wind turbine in the past The historical value of the frequency increment at each sampling time, i.e. and This part reflects the external excitation effect of changes in the input of each actuator on the temperature response, and the model order. and The choice of product makes and It can cover the main dynamic response period corresponding to the dominant thermal time constant of the drying room; Arrange the above input observations in a fixed order to form the first... Each temperature measurement point at the sampling time regression vector : ; Corresponding to the regression vector, the first is defined as... Parameter vector of each temperature measurement point This vector contains all the model coefficients to be identified: ; in, For the first The first temperature increment at the measuring point Autoregressive coefficient of order, This describes the degree to which the temperature change at the measuring point depends on its own historical values. For the first The power increment in the heating zone is in the first The second lag is related to the third The input coefficient for the temperature increment at the measuring point. , This describes the dynamic transmission characteristics of the power change in the heating zone to the temperature response of the measuring point; For the frequency increment of the wind turbine in the first The second lag is related to the third The input coefficient for the temperature increment at the measuring point. This describes the dynamic effect of changes in fan speed on the thermal response of the measuring point.

[0014] Using the regression vector and parameter vector, the first The multi-input single-output autoregressive model with multiple temperature measurement points can be represented in inner product form: ; in The model residual represents the deviation between the measured temperature increment and the model prediction.

[0015] parameter vector The online estimation employs a recursive least squares algorithm with a forgetting factor, applied at each sampling time during the excitation data acquisition phase. The following recursive updates are performed sequentially. First, the gain vector at the current time step is calculated. : ; In the formula, This is the covariance matrix of the previous time step, with the same dimension as the parameter vector length; As the forgetting factor, in the cold initial learning phase, since the system can be considered time-invariant, A value of 1 ensures that the algorithm applies equal weight to all historical data to obtain the most stable recognition result.

[0016] Update the parameter estimates using the current increment vector and the measured temperature increment: in, This is the updated vector of parameter estimates. This is the parameter estimate vector from the previous time step; the terms within the square brackets represent the prediction error at the current time step, which is the difference between the measured temperature increment and the predicted value based on the model parameters from the previous time step. After completing the parameter estimation update, update the covariance matrix for use in the next time step recursion: ; The initial conditions for the recursive algorithm are set as follows: parameter estimation initial vector. The initial value of the covariance matrix is ​​set to zero, indicating that learning begins from scratch without prior knowledge; the initial value of the covariance matrix is ​​set to... ,in It is the identity matrix. It is a positive number with a relatively large value, typically ranging from 1 to 2. to Larger The value indicates a high degree of uncertainty regarding the initial parameter estimates, which allows the algorithm to have a high learning rate in the early stages of recursion. When the entire data acquisition phase After the data deduction was completed, the first set of data was obtained. The final parameter estimate vector for each temperature measurement point That is, all autoregressive coefficients of the multi-input single-output autoregressive model corresponding to the airflow organization characteristics of the current batch of mixed materials at this measuring point. and input coefficients , ,right The above modeling process was performed on each temperature measurement point to obtain a total of [number] results. The model coefficients form the basis for subsequent airflow feature extraction and global prediction model construction.

[0017] The steady-state gain and equivalent time constant of each multi-input single-output autoregressive model are extracted to divide the airflow dead zone set and short-circuit channel set, and the efficiency index of each heating zone is calculated accordingly to generate dynamic energy consumption weights. Furthermore, the steady-state gain and equivalent time constant of each multi-input single-output autoregressive model are extracted to divide the airflow dead zone set and short-circuit channel set, specifically including: The steady-state gain of each heating zone and fan at each measuring point is calculated using autoregressive coefficients and input coefficients. Solve for the dominant poles of the characteristic equations of each multi-input single-output autoregressive model, and calculate the equivalent time constant of the temperature response at each measuring point. The equivalent time constant and the corresponding steady-state gain at each measurement point are combined into a two-dimensional feature vector; Using the two-dimensional feature vectors of each measuring point location as input data, a clustering algorithm is used to divide each measuring point location into a first cluster and a second cluster. The first cluster is defined as a set of short-circuit channels, and the second cluster is defined as a set of airflow dead zones. Based on this, the efficiency index of each heating zone is calculated, and a dynamic energy consumption weight is generated, specifically including: For each heating zone, the sum of its first steady-state gain at each measuring point within the short-circuit channel set is calculated, and the sum of its second steady-state gain at each measuring point is calculated. The ratio of the sum of the first steady-state gain to the sum of the second steady-state gain is defined as the performance index of the heating zone. The performance indicators of each heating zone are compared with the preset performance thresholds; When the efficiency index of the heating zone is greater than or equal to the efficiency threshold, the dynamic energy consumption weight corresponding to the heating zone is set as the benchmark value. When the efficiency index of the heating zone is less than the efficiency threshold, the dynamic energy consumption weight corresponding to the heating zone is set to a penalty weight greater than the benchmark value, and a dynamic energy consumption weight in the form of a diagonal matrix is ​​generated. Specifically, after obtaining the multi-input single-output autoregressive model coefficients of each temperature measurement point, the control system further extracts the steady-state gain and dynamic time constant reflecting the airflow organization characteristics of the mixed materials in the current batch from these coefficients, and divides the airflow dead zone and short-circuit channel accordingly to generate the dynamic energy consumption weight of each heating zone. For the For each temperature measurement point, the autoregressive coefficients of its multi-input single-output autoregressive model are: The input coefficients are and Under steady-state conditions, the temperature increment no longer changes with time, and all input increments remain constant. Substituting the steady-state conditions into the model equations, we obtain the... Heating zone power for the first Steady-state gain of temperature at the measurement point and the The input channel, i.e., the fan frequency, is related to the... Steady-state gain of temperature at the measurement point : ; ; in, In the first After a unit power change is applied to the heating zone and the system reaches a new steady state, the first... The change in temperature at the measuring point, expressed in degrees Celsius per kilowatt; To determine the first steady state after the wind turbine frequency changes by one hertz and the system reaches a new steady state. The change in temperature at the measuring point, expressed in degrees Celsius per hertz; For the first All heating zones The sum of the input coefficients of order 1, For all fan frequencies The sum of the input coefficients of order 1; The remainder of the sum of the autoregressive coefficients relative to 1, derived from all... Each temperature measurement point and all The steady-state gain of each input channel constitutes the steady-state gain matrix. Dimension is OK List; Simultaneously extracted with the steady-state gain is the equivalent time constant of the temperature response at each measurement point, the first... The characteristic equation of the autoregressive model of the temperature measurement point is composed of its autoregressive coefficients. Solving this characteristic equation yields all discrete-time poles, and the dominant pole with the largest magnitude is selected. Then the equivalent continuous time constant of the temperature response at that temperature measurement point for: ; in, The sampling period is It is the natural logarithm. The mode with the dominant pole, The time required for the temperature at this measuring point to reach approximately 63% of its steady state in response to a step change in the input, expressed in seconds. A larger value indicates greater thermal inertia and slower airflow at that location. A smaller value indicates lower thermal inertia and higher wind speed at that location; The equivalent time constant of each temperature measurement point and steady-state gain at wind turbine frequency The combination forms the two-dimensional feature vector of the measurement point. , will all The feature vectors of each temperature measurement point are used as input data. - Use the mean clustering algorithm for binary classification. -Mean clustering uses Euclidean distance as a similarity measure. It iteratively minimizes the sum of squared distances from each cluster sample to the cluster center, grouping closely spaced measurement points in the feature space into the same cluster. After clustering, two sets of measurement points were obtained. Comparing the eigenvalues ​​of the two cluster centers in the time constant and fan frequency gain dimensions, the first cluster center had a smaller time constant and a larger steady-state gain with respect to fan frequency, indicating strong convective heat transfer and low heat accumulation at this location. The set of measurement points contained in this cluster was defined as the short-circuit channel set. The second cluster center has a large time constant and a small steady-state gain with respect to the fan frequency, indicating that the airflow is sluggish, has high thermal inertia, and heat is easily accumulated at this location. The set of measuring points contained in this cluster is defined as the airflow dead zone set. ; Based on the above regional division results, regarding the first Calculate the short-circuit channel heating efficiency index for each heating zone. : Among them, the numerator is the first heating zone for short-circuit channel collection The sum of the steady-state gains of all temperature measurement points within the range characterizes the ability of the heating zone to transfer heat to the short-circuit channel; the denominator is the first... Heating zone for all The sum of the steady-state gains at each temperature measurement point characterizes the overall heating capacity of the heating zone; It is a very small positive number, usually taking the value of to This is used to prevent abnormal values ​​where the denominator is zero. The value ranges from 0 to 1. A higher value indicates that a higher proportion of the heat input to the heating zone flows to the short-circuit channel, meaning that the heating zone is more effective in improving temperature uniformity. A lower value indicates that more heat flows to the dead zone. This falls within the inefficient heating zone; an efficiency threshold should be set. Its value ranges from 0.3 to 0.5, which determines the efficiency index of each heating zone. With performance threshold Perform a comparison one by one to generate a dynamic energy consumption weight diagonal matrix: ; when When, the corresponding weighting coefficient Take the base value 1, when When, the corresponding weighting coefficient Take penalty weight , The value of is between 2 and 5. This weight matrix will be applied to the objective function of the subsequent model predictive control, imposing an additional cost on the heating power of the inefficient heating zone, i.e., the heating zone where too much heat flows to the dead zone, and guiding the optimization solver to prioritize the use of the efficient compensation zone for heating.

[0018] Each multi-input single-output autoregressive model is converted into a global incremental state-space prediction model, and temperature range constraints and spatial temperature difference constraints are set in the prediction time domain. Furthermore, temperature range constraints and spatial temperature difference constraints are set within the prediction time domain, specifically including: Obtain the lower and upper limits of the operating temperature, and establish a temperature range constraint that the predicted temperature at each measurement point at each prediction time is between the lower and upper limits of the operating temperature. Set a temperature difference threshold and establish a spatial temperature difference constraint that the absolute value of the predicted temperature difference between any two different measurement points at each prediction time in the prediction time domain is less than or equal to the temperature difference threshold. Specifically, after completing the airflow feature extraction and dynamic energy consumption weight generation, the control system converts the multi-input single-output autoregressive model of each temperature measurement point into a global incremental state-space prediction model, and sets the set of constraints required for model predictive control. State vector of the global incremental state space prediction model The following items are arranged in order: past temperature measurement points Historical value of temperature increment at each sampling time Each heating zone in the past Historical power increment values ​​at each sampling time ; and the wind turbine in the past Historical value of frequency increment at each sampling time State vector The total dimension is The global incremental state-space prediction model has the following standard form: ; ; in, The state vector at the next sampling time. This is the state vector at the current sampling time. To control the increment vector, the dimension is... ,in The input vector consists of the power of each heating zone and the frequency of the fan at the current moment. This is the state transition matrix, and its temperature increment autoregressive sub-block is filled with the autoregressive coefficients of each temperature measurement point. The input increment shift register sub-block is filled with the input increment shifted from the previous time step to the corresponding position in the next time step according to the standard shift. The input matrix is ​​filled with the input coefficients for each temperature measurement point in its sub-blocks. and ; The output matrix is ​​set to 1 for each element corresponding to the current temperature measurement value at each temperature point, and zero for the rest. This represents the measured temperature vector at each temperature measurement point at the current moment. The ambient temperature or the reference temperature at the air inlet. The standard assembly method for constructing a matrix from the state space using the coefficients of an autoregressive model, which is an all-one column vector, is a recognized technique in linear system theory. The prediction time domain of the model predictive controller is denoted as The control time domain is denoted as Both satisfy , The value of makes The time length covers 1 to 2 times the dominant thermal time constant of the drying oven. The value is usually One-third to one-half, after the control time domain, i.e. Within the range, control increment Set to zero, from the current state Starting from this point, the state prediction values ​​for each step in the prediction time domain are obtained by successively substituting the control increment sequence into the state equation and recursively applying it. Then, the temperature prediction values ​​for each step are obtained from the output equation. , ; The construction of the constraint set includes two parts: temperature range constraints and spatial temperature difference constraints. The lower limit of the permissible operating temperature for the drying process of the electrical porcelain blank is obtained. and upper limit of operating temperature The temperature is determined by the minimum temperature at which condensation is avoided on the surface of the green body during the initial drying stage. Determined by the critical temperature at which cracking does not occur on the surface of the billet, for each prediction time in the prediction time domain. and each temperature measurement point Temperature prediction required satisfy This constraint constitutes One linear inequality constraint; The spatial temperature difference constraint is constructed by setting a temperature difference threshold. Its value is determined by the equivalent temperature difference corresponding to the maximum allowable uneven shrinkage stress of the electric porcelain blank during the drying process, and is usually within... to Select from the range. For each prediction time point within the prediction time domain. and any two different temperature measurement points and The absolute value of the predicted temperature difference must satisfy... The absolute value constraint is equivalent to the following pair of linear inequality constraints: and The spatial temperature difference constraint is applied to the optimization problem in the form of a hard constraint, ensuring that the controller must meet the temperature uniformity requirement while pursuing the minimization of energy consumption. In addition to temperature constraints, the constraint set also includes actuator physical limit constraints and single-step rate of change constraints, and the power of each heating zone satisfies... The fan frequency meets ,in For the first The upper limit of the rated power of the heating zone and Single-step rate of change constraint requirements for the allowable operating frequency range of the wind turbine frequency converter. and ,in The switching response rate of solid-state relays or thyristors and the thermal fatigue resistance of heating elements are determined by their respective characteristics. The above constraints are determined by the inverter's acceleration / deceleration parameters and the fan's mechanical inertia. With this establishment, the global incremental state-space prediction model and the complete set of constraints have been constructed, providing a model foundation and constraint boundaries for subsequent rolling optimization of model predictive control with dynamic energy consumption weights.

[0019] The objective function is constructed using dynamic energy consumption weights. Under the constraints of temperature range and spatial temperature difference, it is solved based on a global incremental state space prediction model, and the target control incremental sequence of each heating zone and fan is output. Furthermore, the objective function is constructed using dynamic energy consumption weights, specifically including: Establish an objective function that includes a weighted energy consumption summation term and a control increment penalty term; In the weighted energy consumption summation term, the predicted power of each heating zone at future time is multiplied by the dynamic energy consumption weight corresponding to the heating zone, and the predicted frequency of the fan is added after the energy consumption coefficient is converted. In the control increment penalty term, a positive definite diagonal penalty matrix is ​​used to weight the change magnitude of the predicted control increment for each heating zone and fan. Output the target control increment sequence for each heating zone and fan, specifically including: In each control cycle, the historical temperature increment values ​​at each measuring point and the historical increment values ​​of each heating zone and fan are extracted to construct the predicted initial state for the current control cycle. The objective control increment sequence that minimizes the objective function is solved using a quadratic programming algorithm. When the temperature range constraint and the spatial temperature difference constraint have no solution, a non-negative relaxation variable is introduced to modify the temperature range constraint into a relaxed form that allows for deviation, and an additional penalty term is added to the objective function for solving. The current time increment in the obtained target control increment sequence is output to each heating zone and fan. Specifically, after completing the construction of the global incremental state space prediction model and constraint set, the control system constructs the objective function using the generated dynamic energy consumption weight in each control cycle, and solves the optimal control increment sequence under all constraints. The control increment at the current moment in the sequence is applied to each heating zone and fan to achieve rolling time-domain optimization control. objective function It consists of two parts: a weighted energy consumption summation term and a control increment penalty term. The weighted energy consumption summation term includes the product of the predicted power and dynamic energy consumption weights of each heating zone, as well as the equivalent energy consumption of the fan. The control increment penalty term is used to suppress frequent adjustments of the actuator. The expression of the objective function is: ; in, To control the step index in the time domain, , To be at the current sampling time For the future Step 1 The decision variable for heating zone power, in kilowatts. For the first The dynamic energy consumption weight of the heating zone in the current control cycle is taken from the diagonal matrix. For the corresponding element, the weight is set to the baseline value of 1 when the performance indicator is greater than or equal to the threshold, and a penalty weight is applied when the performance indicator is less than the threshold. The penalty weight ranges from 2 to 5. For the current moment For the future The decision variable for the frequency of the treadmill fan, in Hertz (Hz). This is the frequency equivalent energy consumption factor for wind turbines, expressed in kilowatts per hertz (kW / Hz), and its value is the rated power of the wind turbine. With rated frequency The ratio, i.e. It is used to convert the fan frequency into an energy consumption dimension comparable to the heating power. For the current moment For the future The control increment vector of the step has a dimension of , The positive definite diagonal penalty matrix has a dimension of . Its diagonal elements are the penalty coefficients for the control increments of each input channel. The faster the actuator responds and the stronger its tolerance to frequent adjustments, the smaller the corresponding penalty coefficient value. At the beginning of each control cycle, the control system first constructs the predicted initial state vector for the current cycle. Extract the data from the sliding data window up to the current time. Historical temperature increment values ​​at each sampling time point Extract the current time before Historical values ​​of power increment in each heating zone at each sampling time and historical values ​​of wind turbine frequency increments The aforementioned historical data are arranged sequentially according to the definition order of the global state vector, constituting... ; Predicted temperature It is recursively calculated from the global incremental state-space prediction model, starting from the initial state. Starting from this point, we sequentially substitute the control increment sequences into the optimization variables. Through state equations Gradually advance Then, from the output equation Obtain the predicted temperature values ​​for each step; The complete form of the optimization problem is to minimize the objective function while satisfying all constraints. The constraints include temperature range constraints. Space temperature difference constraints Actuator limiting constraints and and step size change rate constraint and This problem belongs to the quadratic programming problem with linear constraints, and can be solved using the active set method or the interior point method. When the above set of constraints is not feasible, the temperature range constraint and the spatial temperature difference constraint are relaxed by introducing non-negative relaxation variables. and The temperature range constraint was modified to And add an additional penalty term to the objective function. ,in The relaxation penalty coefficient has a value of [value missing]. The space temperature difference constraint is 10 to 100 times larger than that of the space temperature difference constraint. In a similar way, slack variables are introduced and penalties are added. The modified optimization problem is solved again to ensure that a feasible suboptimal control sequence can be obtained under any operating condition. After solving, the target control increment sequence is obtained. Only the first element in the sequence Acting on the actual system, The former The first component serves as the current increment of power in each heating zone. Each component serves as the current increment of the wind turbine frequency. This is added to the control input from the previous sampling time to obtain the actual input vector applied at the current time. Then, new temperature data is collected, and the sliding window and state vector are advanced by one sampling period to enter the solution process of the next control period. The above rolling time-domain optimization is repeated in each sampling period, so that the control system can continuously use the latest temperature feedback information to correct the model prediction deviation and adjust the heat distribution of each heating zone according to the current effective dynamic energy consumption weight reorganization. During operation, when the temperature prediction residual exceeds the residual threshold, the multi-input single-output autoregressive models are updated online using closed-loop operation data, and the dynamic energy consumption weights and global incremental state space prediction models are refreshed simultaneously. Furthermore, when the temperature prediction residual exceeds the residual threshold, the multi-input single-output autoregressive models are updated online using closed-loop operation data, specifically including: A sliding data window is created during control operation to store the latest input and output data; Calculate the root mean square value of the temperature prediction residual at each measuring point, and determine whether the root mean square value is greater than the residual threshold for multiple consecutive sampling times to determine whether the residual exceeds the limit. An information matrix is ​​constructed using data within a sliding data window, and singular value decomposition is performed. When the minimum singular value of the information matrix is ​​greater than a preset richness threshold, the data incentive sufficiency condition is determined to be met. When both the residual exceeding the limit condition and the data stimulus sufficiency condition are met, the online model update operation is triggered. The autoregressive coefficients and input coefficients are updated online using closed-loop running data within a sliding data window; The dynamic energy consumption weight and global incremental state space prediction model are updated synchronously, specifically including: The steady-state gain and equivalent time constant of each heating zone and fan were recalculated using the updated multi-input single-output autoregressive model; The airflow dead zone set and short-circuit channel set are redefined based on the recalculated steady-state gain and equivalent time constant, the efficiency index is recalculated and the specific value of the dynamic energy consumption weight is updated. The system matrix of the global incremental state-space prediction model is regenerated using the updated autoregressive coefficients and input coefficients. In the next control cycle, the refreshed global incremental state space prediction model and dynamic energy consumption weight execution model are used to solve the predictive control. Specifically, during the drying process, as the green body evaporates moisture and shrinks in volume, the airflow path structure inside the drying chamber gradually deviates from the initial cold state. This leads to a mismatch in the multi-input single-output autoregressive model at each temperature measurement point, manifested as a continuously increasing deviation between the model-predicted temperature and the measured temperature. To maintain model accuracy and control effectiveness, the control system continuously monitors the temperature prediction residual during operation and updates the model online using closed-loop operating data under triggered conditions. It simultaneously refreshes the dynamic energy consumption weights and the global incremental state space prediction model, without injecting any additional thermal disturbance signals into the drying chamber during the entire update process. During model predictive control operation, the control system maintains a length of A sliding data window stores the most recent input and output data in a first-in, first-out manner. The data set stored within the window is... ,in Sampling time The input vector contains the power of each heating zone and the fan frequency; Sampling time The temperature output vector contains the measured temperature at each temperature measurement point and the window length. The value of covers 2 to 3 times the number of sampling periods corresponding to the dominant heating time constant of the drying room, to ensure that the window contains sufficiently rich dynamic response information; The root mean square (RMS) value was used to monitor the temperature prediction residual. At the sampling time, the model from the previous time is used to further predict the current time, resulting in the... Predicted temperature values ​​at temperature measurement points The predicted value is based on time. The global state-space prediction model is obtained in one step recursively to acquire the measured temperature at the current moment. Then, calculate the single-step prediction residual for that measuring point. Summarize all Calculate the root mean square of the predicted residuals for each temperature measurement point: ; Set residual threshold This threshold is determined based on the measurement accuracy of the temperature sensor and the process requirements for model accuracy. In this example, it is taken as 2 to 3 times the measurement error of the temperature sensor. When the root mean square value is continuous The time of each sample is greater than When the residual exceeds the limit, the condition is deemed met. The value is between 5 and 20, used to exclude false triggering caused by occasional measurement noise spikes or brief external disturbances; Data incentive sufficiency assessment for each temperature measurement point Using the data stored within the sliding window Constructing regression vectors from group data The regression vectors are constructed in the same way as during the initial learning period, including the historical temperature increment of the temperature measurement point itself, the historical power increment of each heating zone, and the historical frequency increment of the fan. An information matrix is ​​constructed based on all regression vectors within the window: ; Information matrix Perform singular value decomposition to obtain all singular values, and then take the minimum value among them. As a quantitative indicator of the richness of window data incentives, when the minimum singular value of all measurement points satisfies When the data incentive condition is met, the richness threshold is set. Using the minimum singular value of the information matrix of the initial learning period stimulus data as a reference, 50% to 80% of this reference value is taken; The online model update operation is triggered only when both the residual exceedance condition and the data incentive sufficiency condition are met. If the residual exceeds the limit but the data incentive is insufficient, the data will continue to be accumulated and run in a closed loop until the conditions are met, without forcibly applying additional incentives. The online model update uses a recursive least squares algorithm with a forgetting factor, denoted as . The range of values ​​is to Because the operating system parameters exhibit slow time-varying characteristics due to material shrinkage, Using values ​​less than 1 gives the algorithm greater weight to recent data, enabling it to quickly track changes in airflow channels. The recursive update process utilizes closed-loop data within a sliding window for each measuring point. The parameter vectors are updated sequentially, and the update formulas for the gain vector, parameter estimates, and covariance matrix are completely consistent with those in the initial identification stage; only the forgetting factor needs to be replaced. The initial values ​​for recursion will be the model parameters that are currently valid at the trigger time. Covariance Matrix In short, no additional power or frequency detection signals are injected into the drying chamber during the update process, and all the data used comes from the closed-loop operation data naturally generated during the normal control process of MPC within the window; After the model parameters are updated, a synchronous refresh operation is performed within the same control cycle, using the updated autoregressive coefficients. and input coefficients , Recalculate the steady-state gain of each heating zone and fan at each temperature measurement point. and And the equivalent time constant of the temperature response at each temperature measurement point. Update each temperature measurement point and Combined into a new two-dimensional feature vector, using -Mean clustering algorithm re-divides the set of airflow dead zones and short-circuit channel set Based on the new regional division results, the short-circuit channel heating efficiency index of each heating zone was recalculated. , and the effectiveness threshold Update the dynamic energy consumption weight diagonal matrix after comparison The system matrix of the global incremental state-space prediction model is reassembled using the updated model coefficients, and each element is used. , and And update the weight matrix Applied to the objective function The weighted energy consumption summation term in the equation; All of the above refresh operations are completed within the same control cycle that triggers the update. Starting from the next control cycle, the model predictive controller continues to execute rolling time-domain optimization control using the refreshed global incremental state-space prediction model and dynamic energy consumption weights. The drying process is not interrupted. If the event triggering conditions are met again in subsequent operations, the above-mentioned non-disruptive online update process is executed again, so that the control system tracks the changes in the airflow channel caused by material shrinkage throughout the entire process.

[0020] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling the energy consumption of an electrically heated drying oven with adaptive capabilities, characterized in that, Includes the following steps: During the initial operation phase, pseudo-random binary sequence excitation signals are applied to the input power of each heating zone and the input frequency of the fan, and air temperature response data at each measuring point are collected. The input power, the input frequency, and the air temperature response data are converted into power increments, frequency increments, and temperature increments, and a multi-input single-output autoregressive model is established for each of the measurement points. The steady-state gain and equivalent time constant of each of the multi-input single-output autoregressive models are extracted to divide the airflow dead zone set and short-circuit channel set, and the efficiency index of each heating zone is calculated accordingly to generate dynamic energy consumption weights. Each of the multi-input single-output autoregressive models is converted into a global incremental state-space prediction model, and temperature range constraints and spatial temperature difference constraints are set in the prediction time domain. The objective function is constructed using the dynamic energy consumption weight, and solved based on the global incremental state space prediction model under the constraints of the temperature range and the spatial temperature difference, outputting the target control incremental sequence for each heating zone and the fan. During operation, when the temperature prediction residual exceeds the residual threshold, the multi-input single-output autoregressive models are updated online using closed-loop operation data, and the dynamic energy consumption weights and the global incremental state space prediction model are refreshed simultaneously.

2. The energy consumption control method for an adaptive electric heating drying oven according to claim 1, characterized in that, The application of pseudo-random binary sequence excitation signals to the input power of each heating zone and the input frequency of the fan specifically includes: Different primitive polynomials are configured for each of the heating zones and the fan; The pseudo-random binary sequence excitation signal is generated by generating periodically mutually prime numbers through each of the different primitive polynomials; The pseudo-random binary sequence excitation signals are superimposed on the corresponding basic bias values ​​of the heating zone and the fan for output.

3. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 1, characterized in that, The establishment of the multi-input single-output autoregressive model for each of the aforementioned measurement points specifically includes: The model output data is constructed using the temperature increment at each of the measurement points at the current sampling time, and a regression vector is constructed using the power increment, frequency increment, and temperature increment of the historical period. Construct a parameter vector containing the coefficients of the model to be identified; The parameter vector is estimated online using a recursive least squares algorithm with a preset forgetting factor to obtain the autoregressive coefficients and input coefficients of the multi-input single-output autoregressive model.

4. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 3, characterized in that, The step of extracting the steady-state gain and equivalent time constant of each of the multi-input single-output autoregressive models to divide the airflow dead zone set and short-circuit channel set specifically includes: The steady-state gain of each heating zone and the fan at each measuring point location is calculated using the autoregressive coefficients and the input coefficients. Solve for the dominant poles of the characteristic equations of each of the multi-input single-output autoregressive models, and calculate the equivalent time constant of the temperature response at each of the measurement points; The equivalent time constant and the corresponding steady-state gain at each measurement point location are combined into a two-dimensional feature vector; Using the two-dimensional feature vectors of each measurement point location as input data, a clustering algorithm is used to divide each measurement point location into a first cluster and a second cluster. The first cluster is defined as the short-circuit channel set, and the second cluster is defined as the airflow dead zone set.

5. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 1, characterized in that, The calculation of the efficiency index of each heating zone and the generation of dynamic energy consumption weights specifically include: For each heating zone, the sum of its first steady-state gain with respect to each measuring point position within the short-circuit channel set is calculated, and the sum of its second steady-state gain with respect to each measuring point position is calculated. The ratio of the sum of the first steady-state gain to the sum of the second steady-state gain is defined as the performance index of the heating zone. The performance index of each heating zone is compared with a preset performance threshold. When the performance index of the heating zone is greater than or equal to the performance threshold, the dynamic energy consumption weight corresponding to the heating zone is set as the benchmark value. When the performance index of the heating zone is less than the performance threshold, the dynamic energy consumption weight corresponding to the heating zone is set to a penalty weight greater than the benchmark value, and the dynamic energy consumption weight in the form of a diagonal matrix is ​​generated.

6. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 1, characterized in that, The setting of temperature range constraints and spatial temperature difference constraints within the prediction time domain specifically includes: Obtain the lower limit and upper limit of the operating temperature, and establish a temperature range constraint that the predicted temperature at each measurement point at each prediction time in the prediction time domain is between the lower limit and upper limit of the operating temperature. A temperature difference threshold is set, and a spatial temperature difference constraint is established whereby the absolute value of the predicted temperature difference between any two different measurement points at each prediction time within the prediction time domain is less than or equal to the temperature difference threshold.

7. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 1, characterized in that, The construction of the objective function using the dynamic energy consumption weight specifically includes: Establish the objective function that includes a weighted energy consumption summation term and a control increment penalty term; In the weighted energy consumption summation term, the predicted power of each heating zone at a future time is multiplied by the dynamic energy consumption weight corresponding to the heating zone, and the predicted frequency of the fan calculated by the energy consumption coefficient is added. In the control increment penalty term, a positive definite diagonal penalty matrix is ​​used to weight the change magnitude of the predicted control increment for each heating zone and the fan.

8. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 7, characterized in that, The output of the target control increment sequence for each of the heating zones and the fan specifically includes: In each control cycle, the historical temperature increment values ​​at each of the measuring points, as well as the historical temperature increment values ​​of each of the heating zones and the fan, are extracted to construct the predicted initial state for the current control cycle. The objective control increment sequence that minimizes the objective function is solved using a quadratic programming algorithm. When the temperature range constraint and the spatial temperature difference constraint have no solution, a non-negative relaxation variable is introduced to modify the temperature range constraint into a relaxed form that allows for deviation, and an additional penalty term is added to the objective function for solving. The current time increment in the obtained target control increment sequence is output to each of the heating zones and the fan.

9. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 1, characterized in that, When the temperature prediction residual exceeds the residual threshold, the multi-input single-output autoregressive models are updated online using closed-loop operation data. Specifically, this includes: A sliding data window is created during control operation to store the latest input and output data; Calculate the root mean square value of the temperature prediction residual at each measuring point location, and determine whether the root mean square value is greater than the residual threshold for multiple consecutive sampling times to determine whether the residual exceeds the limit condition. An information matrix is ​​constructed using the data within the sliding data window, and singular value decomposition is performed. When the minimum singular value of the information matrix is ​​greater than a preset richness threshold, the data incentive sufficiency condition is determined to be met. When the residual exceeding the limit condition and the data stimulus sufficiency condition are both satisfied simultaneously, the online model update operation is triggered. The autoregressive coefficients and input coefficients are updated online using the closed-loop operation data within the sliding data window.

10. The energy consumption control method for an electrically heated drying oven with adaptive capability according to claim 9, characterized in that, The synchronous refresh of the dynamic energy consumption weight and the global incremental state space prediction model specifically includes: The steady-state gain and equivalent time constant of each heating zone and the fan are recalculated using the updated multi-input single-output autoregressive model; The airflow dead zone set and short-circuit channel set are redefined based on the recalculated steady-state gain and the equivalent time constant, the efficiency index is recalculated, and the specific value of the dynamic energy consumption weight is updated. The system matrix of the global incremental state-space prediction model is regenerated using the updated autoregressive coefficients and input coefficients. In the next control cycle, the refreshed global incremental state space prediction model and the dynamic energy consumption weight execution model are used for predictive control solution.

Images