An intelligent control system for a drying unit

Abstract

The application relates to the technical field of intelligent control, and discloses an intelligent control system for a drying unit, which comprises a drying unit data acquisition module, a performance analysis module, a target object acquisition module, an object drying attribute acquisition module, a drying unit control strategy acquisition module and an intelligent control module; the performance analysis module and the object drying attribute acquisition module are arranged, which are beneficial to defining drying units with similar performances as the same category, the operation performances of drying unit equipment of the same category are similar, that is, the data such as equipment energy consumption, drying efficiency and drying time when the same object is subjected to the drying operation are similar, therefore, the same control strategy can be adopted when the object is subjected to the drying operation subsequently, the same control parameters are acquired, and the corresponding drying effect is achieved; the drying attribute of the target object is acquired, which lays a foundation for forming a drying strategy and controlling drying process parameters subsequently.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and more specifically to an intelligent control system for a drying unit. Background Technology

[0002] A method for intelligent parameter control of a heat pump dryer unit is disclosed in publication document CN116577995B. This method is applied to an intelligent parameter control system for a heat pump dryer unit. The method configures an N-level drying state cycle based on the basic attributes of the items to be dried; collects data on the items before drying; clusters the items based on the collected data and generates cluster identifiers; configures the N-level drying state cycle as key nodes and configures fuzzy verification nodes between the N-level key nodes; controls the drying of the clustered items based on fitted control curves; performs fuzzy data collection on the items at the fuzzy verification nodes to obtain fuzzy data collection results; performs initial optimization based on the fitted result identifiers; performs data collection on the items at the N-level key nodes to obtain data collection results; performs correction optimization based on the fitted result identifiers; and performs intelligent drying control of the items based on the initial optimization results and correction optimization results. This method solves the technical problem of low parameter control accuracy in heat pump dryers, which leads to poor drying effects. This achieves the technical effect of improving the accuracy of parameter control in heat pump dryers, thereby enhancing the drying quality of heat pump dryers.

[0003] However, for different items to be dried, even if they belong to the same type, the required drying processes are still different, so the parameter requirements for the drying units are still different. Moreover, for large-scale drying tasks, multiple drying units need to work simultaneously. If the drying units are not intelligently controlled, they cannot work alternately, which will shorten the service life of the drying units. The long-term uninterrupted operation of the drying units will reduce the drying efficiency and reduce the working performance of the drying units. Therefore, the lack of adaptive control for multiple drying units makes it impossible to ensure the efficient and stable operation of the drying units, to balance the performance status of each drying unit, and to optimize the energy consumption resources of the drying units.

[0004] Based on this, the present invention proposes a system for dryer units and intelligent control systems, which adopts different operating combinations of dryer units for different drying strategies, thereby balancing the performance status of each dryer unit and extending the service life of each dryer unit. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent control system for a drying unit to solve the problems existing in the background art.

[0006] This invention provides the following technical solution: an intelligent control system for a drying unit, comprising a drying unit data acquisition module, a performance analysis module, a target item acquisition module, an item drying attribute acquisition module, a drying unit control strategy acquisition module, and an intelligent control module;

[0007] The dryer unit data acquisition module is used to collect the operation control data and drying data of n dryer units, mark the n dryer units with digital tags, and transmit the data to the performance analysis module.

[0008] The performance analysis module is used to cluster the n drying units and perform performance index analysis on all drying units within each category, and then transmit the data to the drying unit control strategy acquisition module.

[0009] The target item acquisition module is used to acquire target item data and transmit it to the item drying attribute acquisition module;

[0010] The item drying attribute acquisition module is used to construct an item drying knowledge graph, match target items to obtain the corresponding drying attributes, and transmit them to the dryer unit control strategy acquisition module.

[0011] The dryer unit control strategy acquisition module, based on operation control data and drying data, combined with the drying attributes and performance index of the items, uses an optimized genetic algorithm to obtain the optimal operation category of the dryer unit and forms a drying strategy that is transmitted to the intelligent control module.

[0012] The intelligent control module is used to execute drying strategies and control the operation of corresponding drying units to dry the target items.

[0013] Preferably, the operation control data includes equipment temperature control value, equipment humidity control value, and equipment operating energy consumption. The drying data are the drying process parameters for each drying of items, namely the operating temperature, operating humidity, wind speed, drying time, and drying energy consumption of each drying stage. The drying effect can be represented by the difference in moisture content of the items before and after drying. Each drying stage includes a preheating stage, a constant speed drying stage, a slow-down drying stage, and a cooling stage.

[0014] The target item is the item to be dried, and the target item data includes the target item name, weight, quantity, and initial moisture content;

[0015] The drying attributes refer to the operating temperature, humidity, drying time, and drying energy consumption of each drying stage required to achieve the drying state for different quantities of items.

[0016] The drying strategy refers to the drying process parameters corresponding to the digital tags of each drying unit and the drying attributes of the items, based on the category of the operating drying unit; all drying units of the optimal operating category are selected as the drying units for this operation.

[0017] Preferably, the specific method by which the performance analysis module performs equipment clustering on the n drying units is as follows:

[0018] Based on the operation control data of the dryer units, a performance similarity matrix U is constructed, U = comm(a,b), where comm(a,b) represents the performance similarity between the dryer unit with digital label a and the dryer unit with digital label b; the formula is expressed as:

[0019]

[0020] Where k1 is the weight parameter, za is the comprehensive feature vector of the dryer unit with digital label a, zb is the comprehensive feature vector of the dryer unit with digital label b, and the comprehensive feature vector is a vectorized representation composed of spliced ​​operation control data; cov(za,zb) is the covariance of za and zb, σ(za) is the standard deviation of za, and σ(zb) is the standard deviation of zb; a = 1, 2, 3, ..., n, b = 1, 2, 3, ..., n;

[0021] Based on the performance similarity matrix, the dryer units are clustered. Each dryer unit is considered as a separate small cluster, with a total of n small clusters, where n is the total number of dryer units. The attraction value between each pair of small clusters is calculated based on the performance similarity matrix. A preset merging threshold d is used to control cluster merging. The small clusters are iteratively merged, and in each iteration, the two small clusters q1 and q2 with the maximum attraction value are found.

[0022] If the attraction value between q1 and q2 is greater than the merging threshold d, then q1 and q2 are merged into a new cluster q, and the attraction value between the new cluster q and all other small clusters is recalculated; otherwise, no iterative merging is performed; this process is repeated until the attraction value between any two small clusters is greater than the merging threshold d, at which point k new clusters are obtained, which are k performance categories.

[0023] Preferably, the formula for calculating the gravitational value is expressed as follows: Where Y(q1,q2) is the gravitational force between q1 and q2, G is the gravitational constant, c1 is the number of drying units in sub-cluster q1, c2 is the number of drying units in sub-cluster q2, Dis(q1,q2) is the distance between sub-clusters q1 and q2; that is, the average distance between all pairs of drying units in the two sub-clusters, which can be calculated using the Euclidean distance formula; ω is a hyperparameter, σ(q1) is the standard deviation of the comprehensive eigenvector of the drying units in sub-cluster q1, σ(q2) is the standard deviation of the comprehensive eigenvector of the drying units in sub-cluster q2; μ(q1) is the mean of the comprehensive eigenvector of the drying units in sub-cluster q1, μ(q2) is the mean of the comprehensive eigenvector of the drying units in sub-cluster q2; comm * (q1,q2) represents the mean similarity between the drying units of small clusters q1 and q2.

[0024] Preferably, the performance index analysis of the dryer unit includes the following steps:

[0025] Step S01: Calculate the temperature control performance index: Among them, JD temp,i Let TW be the temperature control performance index of the i-th dryer unit, N be the total number of operations of the i-th dryer unit, and TW be the total number of operations of the i-th dryer unit. ij TW′ is the actual temperature during the j-th run of the i-th dryer unit. ij Let be the set temperature for the j-th operation of the i-th dryer unit, where j = 1, 2, 3, ..., N; i = 1, 2, 3, ..., n;

[0026] Step S02: Calculate the humidity control performance index: Among them, JD hum,i Let t be the humidity control performance index of the i-th dryer unit, N be the number of operations of the i-th dryer unit, and TS be the t-value. ij TS′ represents the actual humidity during the j-th run of the i-th dryer unit. ij The set humidity is the humidity for the j-th operation of the i-th dryer unit, where j = 1, 2, 3, ..., N; i = 1, 2, 3, ..., n;

[0027] Step S03: Calculate the performance index: Where, δ i Let θ be the performance index of the i-th drying unit. i Let β be the unit energy consumption of the i-th drying unit. i Let λ1, λ2, and λ3 be the other influencing factors of the i-th drying unit, and let λ1, λ2, and λ3 be the corresponding proportional factors.

[0028] Preferably, the drying attribute of the item is obtained in the following way:

[0029] Historical data on dried items is collected, and a knowledge graph of dried items is constructed based on web crawling technology. The knowledge graph of dried items includes entities, entity attribute information, entity relationship information, and entity type.

[0030] The system matches target items with entities in the item drying knowledge graph. Matching methods include string matching and semantic matching. String matching includes exact matching and approximate matching. Exact matching compares the string corresponding to the target item with the string corresponding to the entity in the item drying knowledge graph; if they are identical, a match is made. Approximate matching uses a string similarity algorithm to calculate the similarity between the string corresponding to the target item and the string corresponding to the entity in the item drying knowledge graph, matching the entity with the highest similarity. Semantic matching uses word embedding. Word embedding converts the target item and entities in the item drying knowledge graph into vector form using a word embedding model, and matches them by calculating the similarity between the vectors.

[0031] After a successful match, the entities that are successfully matched in the item drying knowledge graph are marked as matched entities; the matched entities obtained by the two methods are merged, and the entity attribute information corresponding to the matched entities is obtained from the item drying knowledge graph; that is, the drying attributes corresponding to the target item.

[0032] Preferably, the optimal running category is obtained in the following way:

[0033] Step S11: Encode the k performance categories as O, where O represents a chromosome. Obtain the chromosomes and construct the initial population A = {O1, O2, O3, ..., O}. k};

[0034] Step S12: Determine the fitness function;

[0035] Step S13: Perform natural selection on chromosomes in the population;

[0036] Step S14: Perform crossover recombination on chromosomes in the population;

[0037] Step S15: Mutate the chromosomes in the population;

[0038] Step S16: Obtain a new population. The preset population generation number is L, and the fitness threshold is Q, where L is an integer greater than 0 and Q is a real number greater than 0. Repeat steps S13 to S15 until the generation number of the new population is L or a chromosome in the new population has a fitness value greater than or equal to the fitness threshold Q. The loop ends when the chromosome with the highest fitness in the new population is obtained, which is the optimal running category.

[0039] Preferably, the expression for the fitness function is: f r =-ρ r , where f r Let ρ be the fitness corresponding to the r-th chromosome. r The predicted performance degradation value of the dryer unit corresponding to the operating performance category of the r-th chromosome; r = 1, 2, 3, ..., k;

[0040] The method for obtaining the predicted performance degradation value of the dryer unit is as follows: the performance index of each dryer unit in the corresponding operating performance category of the chromosome, along with the corresponding operating control data and drying data, are used as research data. The research data are input into the trained performance prediction model to predict the performance index corresponding to each dryer unit. The difference between the performance index of the previous operation and the predicted performance index is calculated to obtain the performance degradation value of each dryer unit. The average value of the performance degradation values ​​of all dryer units in the operating performance category is taken as the predicted performance degradation value of that operating performance category.

[0041] The specific training method for the performance prediction model is as follows:

[0042] The operation control data and drying data of a dryer unit are used as an analysis data set. The operation control data and drying data of each run are used as elements in an analysis data set. e analysis data sets are collected in advance. For each element in the e analysis data sets, the corresponding performance index after each run is obtained, where e is an integer greater than 1. The elements in each analysis data set and the corresponding performance index are converted into a set of feature vectors; e = 1, 2, 3, ..., n.

[0043] Each set of feature vectors is used as input to the performance prediction model. The performance prediction model outputs a predicted performance index corresponding to each element in the analyzed dataset, and uses the actual performance index value corresponding to each element in the analyzed dataset as the prediction target. The actual performance index value can be obtained through the performance index analysis process. The training objective is to minimize the sum of prediction errors of all elements in the analyzed dataset. The formula for the prediction error is expressed as: ε p =γ p -μ p , where ε p The prediction error is represented by p, where p is the group number of the feature vector corresponding to the element in the analysis dataset, and γ is the prediction error. p Let μ be the prediction performance index corresponding to the element in the p-th set of analytical data. p For the actual performance index corresponding to the element in the p-th set of analytical data, train the performance prediction model until the sum of prediction errors converges and then stop training.

[0044] The performance prediction model is specifically a deep network model.

[0045] The technical effects and advantages of this invention are as follows:

[0046] (1) By providing a performance analysis module and an item drying attribute acquisition module, this invention helps to define dryer units with similar performance as the same category. Dryer units of the same category have similar operating performance, that is, when drying the same item, the equipment energy consumption, drying efficiency, drying time and other data are similar. Although dryer units of different categories adopt the same control strategy, due to the different equipment performance, the equipment energy consumption and drying effect are also different, which will lead to uneven drying of items. Therefore, when drying items in the future, the same control strategy can be adopted and the same control parameters can be obtained to achieve the corresponding drying effect. Obtaining the drying attributes of the target item lays the foundation for the subsequent formation of drying strategy and control of drying process parameters.

[0047] (2) This invention, by providing a dryer unit control strategy acquisition module, facilitates the integration of a trained neural network model into the fitness calculation of the optimized genetic algorithm. This fully leverages the nonlinear and multimodal modeling capabilities of the neural network for operating control data and drying data, effectively capturing complex data relationships. Based on a parallel computing processing mechanism, it improves computational efficiency and evaluates the operating performance of the dryer unit through deep learning technology, thereby enhancing the pertinence and effectiveness of extending the service life of the dryer unit. By fully utilizing operating control data and drying data and capturing complex data relationships through deep learning technology, it can comprehensively and accurately understand the changes in the operating performance of different dryer units, providing customized operating schemes for various types of dryer units, thereby effectively improving the overall operating performance of the dryer units, extending their service life, and ensuring that all dryer units maintain optimal performance, achieving performance balance. Attached Figure Description

[0048] Figure 1 This is a flowchart of the intelligent control system for the drying unit of the present invention. Detailed Implementation

[0049] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The intelligent control system for a drying unit involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] like Figure 1As shown, the present invention provides an intelligent control system for a drying unit, including a drying unit data acquisition module, a performance analysis module, a target item acquisition module, an item drying attribute acquisition module, a drying unit control strategy acquisition module, and an intelligent control module;

[0051] The data acquisition module for the drying units is used to collect the operation control data and drying data of n drying units, mark the n drying units with digital tags, and transmit the data to the performance analysis module. The operation control data includes, but is not limited to, equipment temperature control values, equipment humidity control values, and equipment operating energy consumption. The drying data consists of the drying process parameters for each drying operation, namely, the operating temperature, operating humidity, wind speed, drying time, and drying energy consumption at each drying stage. The drying effect can be represented by the difference in moisture content before and after drying. A larger difference in moisture content before and after drying indicates a better drying effect, while a smaller difference indicates a worse drying effect. Each drying stage includes a preheating stage, a constant-speed drying stage, a decreasing-speed drying stage, and a cooling stage.

[0052] The performance analysis module is used to cluster the n drying units and perform performance index analysis on all drying units within each category, which is then transmitted to the drying unit control strategy acquisition module. The purpose is to ensure that drying units of the same category have similar operating performance, meaning that the energy consumption, drying efficiency, and drying time are similar when drying the same items. Therefore, when drying items subsequently, the same control strategy can be used to obtain the same control parameters, thereby achieving the corresponding drying effect. Even though different categories of drying units use the same control strategy, their different equipment performance leads to different energy consumption and drying effects, resulting in uneven drying of items.

[0053] The target item acquisition module is used to acquire target item data and transmit it to the item drying attribute acquisition module; the target item is the item to be dried, and the target item data includes, but is not limited to, basic information such as the target item name, weight, quantity, and initial moisture content;

[0054] The item drying attribute acquisition module is used to construct an item drying knowledge graph, match the target item to obtain the corresponding drying attributes, and transmit them to the dryer unit control strategy acquisition module. The drying attributes are the drying process parameters such as the operating temperature, humidity, drying time, and drying energy consumption required for each drying stage to reach the drying state when the quantity of the item is different.

[0055] The dryer unit control strategy acquisition module, based on operation control data and drying data, combined with the drying attributes and performance index of the items, uses an optimized genetic algorithm to obtain the optimal operating category of the dryer unit and forms a drying strategy, which is then transmitted to the intelligent control module. All dryer units with the optimal operating category are selected as the dryer units for this operation. The drying strategy consists of the digital tags of each dryer unit corresponding to the operating dryer unit category and the drying process parameters corresponding to the drying attributes of the items. The purpose is to integrate the trained neural network model into the fitness calculation of the optimized genetic algorithm, thereby fully leveraging the nonlinearity and multimodal operation of the neural network. The system's ability to model control and drying data effectively captures complex data relationships. Based on parallel computing mechanisms, it improves computational efficiency and uses deep learning technology to evaluate the operating performance of dryer units, thereby enhancing the targetedness and effectiveness of extending the lifespan of dryer units. By fully utilizing operational control and drying data and capturing complex data relationships through deep learning, it can comprehensively and accurately understand the changes in equipment operating performance of different dryer units, providing customized operating solutions for various types of dryer units. This effectively improves the overall operating performance of dryer units, extends their lifespan, and ensures that all dryer units maintain optimal performance, achieving performance balance.

[0056] The intelligent control module is used to execute the drying strategy, control the operation of the corresponding drying unit to dry the target items, and control the temperature, humidity, wind speed and other parameters of each drying unit.

[0057] In this embodiment, it should be specifically explained that the performance analysis module performs equipment clustering on the n drying units in the following way:

[0058] Based on the operation control data of the dryer units, a performance similarity matrix U is constructed, U = comm(a,b), where comm(a,b) represents the performance similarity between the dryer unit with digital label a and the dryer unit with digital label b; the formula is expressed as:

[0059]

[0060] Where k1 is the weight parameter, za is the comprehensive feature vector of the dryer unit with digital label a, zb is the comprehensive feature vector of the dryer unit with digital label b, and the comprehensive feature vector is a vectorized representation composed of spliced ​​operation control data; cov(za,zb) is the covariance of za and zb, σ(za) is the standard deviation of za, and σ(zb) is the standard deviation of zb; a = 1, 2, 3, ..., n, b = 1, 2, 3, ..., n;

[0061] Based on the performance similarity matrix, the dryer units are clustered. Specifically, each dryer unit is considered as a separate small cluster, with a total of n small clusters, where n is the total number of dryer units. The attraction value between each pair of small clusters is calculated based on the performance similarity matrix. A preset merging threshold d is used to control cluster merging. The small clusters are iteratively merged, and in each iteration, the two small clusters q1 and q2 with the maximum attraction value are found.

[0062] If the attraction value between q1 and q2 is greater than the merging threshold d, then q1 and q2 are merged into a new cluster q, and the attraction value between the new cluster q and all other small clusters is recalculated; otherwise, no iterative merging is performed; this process is repeated until the attraction value between any two small clusters is greater than the merging threshold d, at which point k new clusters are obtained, i.e., k operating performance categories. Drying units in the same category have similar performance, so the same drying unit control strategy can be adopted and the same control parameters can be obtained, thereby ensuring that the items achieve the corresponding drying effect, laying the foundation for subsequent selection of drying units with the same operating performance category for operation control.

[0063] In this embodiment, it should be specifically noted that the formula for calculating the gravitational value is expressed as follows: Where Y(q1,q2) is the gravitational force between q1 and q2, G is the gravitational constant, c1 is the number of drying units in sub-cluster q1, c2 is the number of drying units in sub-cluster q2, Dis(q1,q2) is the distance between sub-clusters q1 and q2; that is, the average distance between all pairs of drying units in the two sub-clusters, which can be calculated using the Euclidean distance formula; ω is a hyperparameter used to control the overall size; σ(q1) is the standard deviation of the comprehensive eigenvector of the drying units in sub-cluster q1, σ(q2) is the standard deviation of the comprehensive eigenvector of the drying units in sub-cluster q2, reflecting the degree of dispersion within the sub-clusters; μ(q1) is the mean of the comprehensive eigenvector of the drying units in sub-cluster q1, μ(q2) is the mean of the comprehensive eigenvector of the drying units in sub-cluster q2, reflecting the overall level of the sub-clusters; comm * (q1,q2) represents the mean similarity between the drying units of small clusters q1 and q2;

[0064] By adjusting the merging threshold d, the strictness of merging can be controlled, thereby affecting the final number of clusters. The merging threshold d needs to be determined based on the actual operation and energy consumption of the equipment to obtain a reasonable merging threshold, so that the ideal number of clusters k can be obtained in the end.

[0065] In this embodiment, it should be specifically noted that the performance index analysis of the drying unit includes the following steps:

[0066] Step S01: Calculate the temperature control performance index: JD temp,i Let TW be the temperature control performance index of the i-th dryer unit, N be the total number of operations of the i-th dryer unit, and TW be the total number of operations of the i-th dryer unit. ij TW′ is the actual temperature during the j-th run of the i-th dryer unit. ij Let be the set temperature for the j-th operation of the i-th dryer unit, where j = 1, 2, 3, ..., N; i = 1, 2, 3, ..., n;

[0067] Step S02: Calculate the humidity control performance index: JD hum,i Let t be the humidity control performance index of the i-th dryer unit, N be the number of operations of the i-th dryer unit, and TS be the t-value. ij TS′ represents the actual humidity during the j-th run of the i-th dryer unit. ij The set humidity is the humidity for the j-th operation of the i-th dryer unit, where j = 1, 2, 3, ..., N; i = 1, 2, 3, ..., n;

[0068] Step S03: Calculate the performance index: Where, δ i Let θ be the performance index of the i-th dryer unit. i Let β be the unit energy consumption of the i-th drying unit. i Let λ1, λ2, and λ3 be the corresponding proportional factors for the i-th dryer unit, all greater than 0 and less than 1. λ1 and λ2 are greater than λ3. For the dryer unit, humidity and temperature are the most important control parameters; therefore, their corresponding proportional factors are greater than those of other control parameters. These other influencing factors are the performance indices of the dryer unit's other control parameters, calculated using the same method as the temperature control performance index and humidity control performance index. For example, if the other influencing factor is the wind speed performance index, then the formula for calculating the wind speed performance index is: JD wind,i Let N be the wind speed control performance index of the i-th dryer unit, N be the number of operations of the i-th dryer unit, and TF be the wind speed control performance index. ij Let TF′ be the actual wind speed during the j-th operation of the i-th dryer unit. ij The set wind speed is for the j-th operation of the i-th dryer unit, where j = 1, 2, 3, ..., N; i = 1, 2, 3, ..., n; other influencing factors can be selected by those skilled in the art based on the specific model of the dryer unit and the actual drying conditions.

[0069] Analyzing the performance index of the drying units allows us to obtain the changes in the performance index of n drying units, laying the foundation for subsequent optimization of the genetic algorithm.

[0070] In this embodiment, it should be specifically explained that the method for obtaining the drying attribute of the item is as follows:

[0071] Historical data on dried items was collected, and a knowledge graph on item drying was constructed using web crawling technology. This knowledge graph includes entities, entity attribute information, entity relationship information, and entity types. Entities are the dried items, including items that have previously been dried by the drying unit and items that can be dried using the unit but have not been previously dried. Historical dried item data consists of data on items dried by the drying unit, including item name, dried weight, quantity, drying effect, and drying process. Entity attribute information refers to the entity's drying attributes. Entity types define the item category to which the entity belongs, including clothing, food, and medicine, and define common drying attributes and relationships between entities. Entity relationship information represents the semantic connections between different entities, describing various associations and interactions between entities in the knowledge graph.

[0072] The system matches target items with entities in the item drying knowledge graph. Matching methods include string matching and semantic matching. String matching includes exact matching and approximate matching. Exact matching compares the string corresponding to the target item with the string corresponding to the entity in the item drying knowledge graph; if they are identical, a match is made. Approximate matching uses a string similarity algorithm to calculate the similarity between the string corresponding to the target item and the string corresponding to the entity in the item drying knowledge graph, matching the entity with the highest similarity. Semantic matching uses word embedding. Word embedding converts the target item and entities in the item drying knowledge graph into vector form using a word embedding model, and matches them by calculating the similarity between the vectors.

[0073] After a successful match, the entities that are successfully matched in the item drying knowledge graph are marked as matched entities; the matched entities obtained by the two methods are merged, and the entity attribute information corresponding to the matched entities is obtained from the item drying knowledge graph; that is, the drying attributes corresponding to the target item.

[0074] In this embodiment, it should be specifically explained that the method for obtaining the optimal running category is as follows:

[0075] Step S11: Encode the k performance categories as O, where O represents a chromosome. Obtain the chromosomes and construct the initial population A = {O1, O2, O3, ..., O}. k};

[0076] Step S12: Determine the fitness function;

[0077] Step S13: Perform natural selection on chromosomes in the population;

[0078] Step S14: Perform crossover recombination on chromosomes in the population;

[0079] Step S15: Mutate the chromosomes in the population;

[0080] Step S16: Obtain a new population. The preset population generation number is L, and the fitness threshold is Q, where L is an integer greater than 0 and Q is a real number greater than 0. Repeat steps S13 to S15 until the new population generation number is L or a chromosome in the new population has a fitness value greater than or equal to the fitness threshold Q. The loop ends when the chromosome with the highest fitness in the new population is obtained, which is the optimal running category. For example, if the preset population generation number is 1, the chromosomes in the initial population are subjected to natural selection, crossover, recombination, and mutation to obtain a new population. At this time, the new population generation number is 1, so the loop ends.

[0081] The fitness threshold Q is preset by those skilled in the art based on the algorithm accuracy. The population generation L is obtained by those skilled in the art through multiple genetic algorithms under multiple sets of different operation control data, drying data, and drying attributes. In each genetic algorithm process, when there is a chromosome in the new population with a fitness greater than or equal to the fitness threshold Q, the loop ends and the generation corresponding to the new population is obtained. The largest generation among multiple generations is taken as the population generation L.

[0082] In this embodiment, it should be specifically noted that the expression for the fitness function is: f r =-ρ r , where f r Let ρ be the fitness corresponding to the r-th chromosome. r The predicted performance degradation value of the dryer unit corresponding to the operating performance category of the r-th chromosome; r = 1, 2, 3, ..., k;

[0083] The method for obtaining the predicted performance degradation value of the dryer unit is as follows: the performance index of each dryer unit in the corresponding operating performance category of the chromosome, along with the corresponding operating control data and drying data, are used as research data. The research data are input into the trained performance prediction model to predict the performance index corresponding to each dryer unit. The difference between the performance index of the previous operation and the predicted performance index is calculated to obtain the performance degradation value of each dryer unit. The average value of the performance degradation values ​​of all dryer units in the operating performance category is taken as the predicted performance degradation value of that operating performance category.

[0084] The specific training method for the performance prediction model is as follows:

[0085] The operation control data and drying data of a dryer unit are used as an analysis data set. The operation control data and drying data of each run are used as elements in an analysis data set. e analysis data sets are collected in advance. For each element in the e analysis data sets, the corresponding performance index after each run is obtained, where e is an integer greater than 1. The elements in each analysis data set and the corresponding performance index are converted into a set of feature vectors; e = 1, 2, 3, ..., n.

[0086] Each set of feature vectors is used as input to the performance prediction model. The performance prediction model outputs a predicted performance index corresponding to each element in the analyzed dataset, and uses the actual performance index value corresponding to each element in the analyzed dataset as the prediction target. The actual performance index value can be obtained through the performance index analysis process. The training objective is to minimize the sum of prediction errors of all elements in the analyzed dataset. The formula for the prediction error is expressed as: ε p =γ p -μ p , where ε p The prediction error is represented by p, where p is the group number of the feature vector corresponding to the element in the analysis dataset, and γ is the prediction error. p Let μ be the prediction performance index corresponding to the element in the p-th set of analytical data. p For the actual performance index corresponding to the element in the p-th set of analytical data, train the performance prediction model until the sum of prediction errors converges and then stop training.

[0087] The performance prediction model is specifically a deep network model.

[0088] In this embodiment, it should be specifically noted that the natural selection is carried out using a combination of elite selection and rotation selection. The elite selection method produces F1 offspring chromosomes. For a population of size k, the fitness of the k chromosomes is arranged from highest to lowest, and each of the first F1 chromosomes produces one offspring chromosome. The rotation selection method produces F2 offspring chromosomes, meaning that the k chromosomes produce F2 offspring chromosomes according to their respective rotation probabilities. F1 + F2 = k, thus maintaining a constant population size k while increasing the number of generations.

[0089] The expression for the rotation probability is: Where, ζ r Let r be the rotation probability corresponding to the r-th chromosome;

[0090] In step S14, E chromosomes are randomly selected from the population for crossover recombination to obtain E new chromosomes. The crossover recombination uses the PMX method, which is an existing technology and will not be described in detail here. After chromosome crossover recombination, the fitness of the E new chromosomes is calculated. The fitness of the E new chromosomes is sorted from largest to smallest with the fitness of the E chromosomes to generate a sorting table. The E new chromosomes in the sorting table replace the E chromosomes in the population that are undergoing crossover recombination in ascending order. In this embodiment, E is preferably 0.7k. If the calculated E is not an integer, E is rounded up to ensure that the calculated E is an integer.

[0091] In step S15, the preset mutation probability is V. Based on the mutation probability, k chromosomes in the population are mutated. The mutation method is to randomly select the positions of two genes in the chromosome and exchange the values ​​of the two genes. In this embodiment, V = 0.02 is preferred. The mutation probability is preset by those skilled in the art based on the algorithm efficiency and algorithm accuracy.

[0092] In this embodiment, it should be specifically explained that the drying strategy is obtained based on the combination of the optimal operating category and the drying attributes of the item. The optimal operating category refers to each drying unit that performs the drying operation this time. The drying attributes of the item can be obtained by obtaining the drying process parameters, such as the drying time, drying temperature and drying humidity at each stage. After the intelligent control module executes the drying strategy, it runs each drying unit in the optimal category and adjusts the drying process parameters.

[0093] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0094] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent control system for a drying unit, characterized in that: It includes a dryer unit data acquisition module, a performance analysis module, a target item acquisition module, an item drying attribute acquisition module, a dryer unit control strategy acquisition module, and an intelligent control module; The dryer unit data acquisition module is used to collect the operation control data and drying data of n dryer units, mark the n dryer units with digital tags, and transmit the data to the performance analysis module. The performance analysis module is used to cluster the n drying units and perform performance index analysis on all drying units within each category, and then transmit the data to the drying unit control strategy acquisition module. The target item acquisition module is used to acquire target item data and transmit it to the item drying attribute acquisition module; The item drying attribute acquisition module is used to construct an item drying knowledge graph, match target items to obtain the corresponding drying attributes, and transmit them to the dryer unit control strategy acquisition module. The dryer unit control strategy acquisition module, based on operation control data and drying data, combined with the drying attributes and performance index of the items, uses an optimized genetic algorithm to obtain the optimal operation category of the dryer unit and forms a drying strategy that is transmitted to the intelligent control module. The intelligent control module is used to execute the drying strategy and control the operation of the corresponding drying unit to dry the target items.

2. The intelligent control system for a drying unit according to claim 1, characterized in that: The operation control data includes equipment temperature control value, equipment humidity control value, and equipment operating energy consumption. The drying data are the drying process parameters for each drying of items, namely the operating temperature, operating humidity, wind speed, drying time, and drying energy consumption of each drying stage. The drying effect can be represented by the difference in moisture content of the items before and after drying. Each drying stage includes a preheating stage, a constant speed drying stage, a slow-down drying stage, and a cooling stage. The target item is the item to be dried, and the target item data includes the target item name, weight, quantity, and initial moisture content; The drying attributes refer to the operating temperature, humidity, drying time, and drying energy consumption of each drying stage required to achieve the drying state for different quantities of items. The drying strategy refers to the drying process parameters corresponding to the digital tags of each drying unit and the drying attributes of the items, based on the category of the operating drying unit; all drying units of the optimal operating category are selected as the drying units for this operation.

3. The intelligent control system for a drying unit according to claim 1, characterized in that: The specific method by which the performance analysis module performs equipment clustering on the n drying units is as follows: Based on the operation control data of the dryer units, a performance similarity matrix U is constructed, U = comm(a,b), where comm(a,b) represents the performance similarity between the dryer unit with digital label a and the dryer unit with digital label b; the formula is expressed as: Where k1 is the weight parameter, za is the comprehensive feature vector of the dryer unit with digital label a, zb is the comprehensive feature vector of the dryer unit with digital label b, and the comprehensive feature vector is a vectorized representation composed of spliced ​​operation control data; cov(za,zb) is the covariance of za and zb, σ(za) is the standard deviation of za, and σ(zb) is the standard deviation of zb; a = 1, 2, 3, ..., n, b = 1, 2, 3, ..., n; Based on the performance similarity matrix, the dryer units are clustered. Each dryer unit is considered as a separate small cluster, with a total of n small clusters, where n is the total number of dryer units. The attraction value between each pair of small clusters is calculated based on the performance similarity matrix. A preset merging threshold d is used to control cluster merging. The small clusters are iteratively merged, and in each iteration, the two small clusters q1 and q2 with the maximum attraction value are found. If the attraction value between q1 and q2 is greater than the merging threshold d, then q1 and q2 are merged into a new cluster q, and the attraction value between the new cluster q and all other small clusters is recalculated; otherwise, no iterative merging is performed; this process is repeated until the attraction value between any two small clusters is greater than the merging threshold d, at which point k new clusters are obtained, which are k performance categories.

4. The intelligent control system for a drying unit according to claim 3, characterized in that: The formula for calculating the gravitational force value is expressed as follows: Where Y(q1,q2) is the gravitational force between q1 and q2, G is the gravitational constant, c1 is the number of drying units in sub-cluster q1, c2 is the number of drying units in sub-cluster q2, Dis(q1,q2) is the distance between sub-clusters q1 and q2; that is, the average distance between all pairs of drying units in the two sub-clusters, which can be calculated using the Euclidean distance formula; ω is a hyperparameter, σ(q1) is the standard deviation of the comprehensive eigenvector of the drying units in sub-cluster q1, σ(q2) is the standard deviation of the comprehensive eigenvector of the drying units in sub-cluster q2; μ(q1) is the mean of the comprehensive eigenvector of the drying units in sub-cluster q1, μ(q2) is the mean of the comprehensive eigenvector of the drying units in sub-cluster q2; comm * (q1,q2) represents the mean similarity between the drying units of small clusters q1 and q2.

5. The intelligent control system for a drying unit according to claim 4, characterized in that: The performance index analysis of the drying unit includes the following steps: Step S01: Calculate the temperature control performance index: Among them, JD temp,i Let TW be the temperature control performance index of the i-th dryer unit, N be the total number of operations of the i-th dryer unit, and TW be the total number of operations of the i-th dryer unit. ij TW′ is the actual temperature during the j-th run of the i-th dryer unit. ij Let be the set temperature for the j-th operation of the i-th dryer unit, where j = 1, 2, 3, ..., N; i = 1, 2, 3, ..., n; Step S02: Calculate the humidity control performance index: Among them, JD hum,i Let t be the humidity control performance index of the i-th dryer unit, N be the number of operations of the i-th dryer unit, and TS be the t-value. ij TS′ represents the actual humidity during the j-th run of the i-th dryer unit. ij The set humidity is the humidity for the j-th operation of the i-th dryer unit, where j = 1, 2, 3, ..., N; i = 1, 2, 3, ..., n; Step S03: Calculate the performance index: Where, δ i Let θ be the performance index of the i-th drying unit. i Let β be the unit energy consumption of the i-th drying unit. i Let λ1, λ2, and λ3 be the other influencing factors of the i-th drying unit, and let λ1, λ2, and λ3 be the corresponding proportional factors.

6. The intelligent control system for a drying unit according to claim 5, characterized in that: The specific method for obtaining the drying attribute of the item is as follows: Historical data on dried items is collected, and a knowledge graph of dried items is constructed based on web crawling technology. The knowledge graph of dried items includes entities, entity attribute information, entity relationship information, and entity type. The system matches target items with entities in the item drying knowledge graph. Matching methods include string matching and semantic matching. String matching includes exact matching and approximate matching. Exact matching compares the string corresponding to the target item with the string corresponding to the entity in the item drying knowledge graph; if they are identical, a match is made. Approximate matching uses a string similarity algorithm to calculate the similarity between the string corresponding to the target item and the string corresponding to the entity in the item drying knowledge graph, matching the entity with the highest similarity. Semantic matching uses word embedding. Word embedding converts the target item and entities in the item drying knowledge graph into vector form using a word embedding model, and matches them by calculating the similarity between the vectors. After a successful match, the entities that are successfully matched in the item drying knowledge graph are marked as matched entities; the matched entities obtained by the two methods are merged, and the entity attribute information corresponding to the matched entities is obtained from the item drying knowledge graph; that is, the drying attributes corresponding to the target item.

7. The intelligent control system for a drying unit according to claim 6, characterized in that: The specific method for obtaining the optimal running category is as follows: Step S11: Encode the k performance categories as O, where O represents a chromosome. Obtain the chromosomes and construct the initial population A = {O1, O2, O3, ..., O}. k }; Step S12: Determine the fitness function; Step S13: Perform natural selection on chromosomes in the population; Step S14: Perform crossover recombination on chromosomes in the population; Step S15: Mutate the chromosomes in the population; Step S16: Obtain a new population. The preset population generation number is L, and the fitness threshold is Q, where L is an integer greater than 0 and Q is a real number greater than 0. Repeat steps S13 to S15 until the generation number of the new population is L or a chromosome in the new population has a fitness value greater than or equal to the fitness threshold Q. The loop ends when the chromosome with the highest fitness in the new population is obtained, which is the optimal running category.

8. The intelligent control system for a drying unit according to claim 7, characterized in that: The fitness function is expressed as: f r =-ρ r , where f r Let ρ be the fitness corresponding to the r-th chromosome. r The predicted performance degradation value of the dryer unit corresponding to the operating performance category of the r-th chromosome; r = 1, 2, 3, ..., k; The method for obtaining the predicted performance degradation value of the dryer unit is as follows: the performance index of each dryer unit in the corresponding operating performance category of the chromosome, along with the corresponding operating control data and drying data, are used as research data. The research data are input into the trained performance prediction model to predict the performance index corresponding to each dryer unit. The difference between the performance index of the previous operation and the predicted performance index is calculated to obtain the performance degradation value of each dryer unit. The average value of the performance degradation values ​​of all dryer units in the operating performance category is taken as the predicted performance degradation value of that operating performance category. The specific training method for the performance prediction model is as follows: The operation control data and drying data of a dryer unit are used as an analysis data set. The operation control data and drying data of each run are used as elements in an analysis data set. e analysis data sets are collected in advance. For each element in the e analysis data sets, the corresponding performance index after each run is obtained, where e is an integer greater than 1. The elements in each analysis data set and the corresponding performance index are converted into a set of feature vectors; e = 1, 2, 3, ..., n. Each set of feature vectors is used as input to the performance prediction model. The performance prediction model outputs a predicted performance index corresponding to each element in the analyzed dataset, and uses the actual performance index value corresponding to each element in the analyzed dataset as the prediction target. The actual performance index value can be obtained through the performance index analysis process. The training objective is to minimize the sum of prediction errors of all elements in the analyzed dataset. The formula for the prediction error is expressed as: ε p =γ p -μ p , where ε p The prediction error is represented by p, where p is the group number of the feature vector corresponding to the element in the analysis dataset, and γ is the prediction error. p Let μ be the prediction performance index corresponding to the element in the p-th set of analytical data. p For the actual performance index corresponding to the element in the p-th set of analytical data, train the performance prediction model until the sum of prediction errors converges and then stop training. The performance prediction model is specifically a deep network model.

Images