A method for mining transformation rules of smart home design parameters based on hierarchical dimensionality reduction

Abstract

The present invention discloses a method for mining smart home design parameter conversion rules based on hierarchical dimensionality reduction. The method comprises the following steps: obtaining historical demand conversion order information of smart home customized products, preprocessing the order information, and reducing the dimensionality of the design parameters according to the characteristics of the smart home products; using the preprocessed demand conversion order data as input, setting support and confidence thresholds according to the hierarchical results, mining the data set layer by layer, and generating mapping rules from demand to design parameters at each layer; finally, accelerating the convergence of the rule optimization algorithm based on object collision by formulating a many-to-one probability relative collision rule for the mobile group objects, and obtaining the final mapping rule set. The present invention reduces the dimensionality of each layer of data through hierarchical mining of the data set, thereby achieving accurate conversion of smart home demand to design parameters, and improving the conversion efficiency and accuracy of demand to design parameters by improving the rule optimization algorithm.

Description

Technical Field

[0001] The present invention relates to a technology for converting smart home customized product requirements into design parameters, and in particular to a smart home requirement conversion method based on hierarchical frequent mining. Background Art

[0002] The conversion of customized smart home product requirements often relies on the experience and expertise of experts, transforming requirements into design parameters through analysis, comparison, and judgment. This approach's primary advantage is its high accuracy, making it suitable for smart home products with simple requirements and clear rules. However, it lacks adaptability and intelligence, making it difficult to adapt to complex, personalized needs and uncertainties. Therefore, for smart home products with highly personalized requirements and unclear expressions, demand conversion methods based on data mining offer the advantages of strong adaptability and intelligence. Demand conversion methods that learn patterns and patterns from large amounts of data are more suitable for smart home products, where data volume, variety, and dimensionality are increasing.

[0003] The multifaceted design characteristics of customized smart home products result in large-dimensional order data for historical demand conversion. Existing frequent mining methods suffer from high computational complexity and time-consuming nature when mining large-dimensional datasets, which in turn leads to poor demand conversion results during the product design process. This method addresses this issue by proposing a hierarchical frequent mining method to mine mapping rules from textual requirements to design parameters for customized smart home products. Furthermore, during the hierarchical mining process, a many-to-one probabilistic collision rule is employed to improve the rule optimization algorithm based on object collisions. This allows for the rational and rapid adjustment of support and confidence thresholds, avoiding the need for manual threshold setting and different thresholds for different layers of data. This optimizes the mapping rules and ultimately enables the automatic and effective conversion of customized smart home product requirements to design parameters. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the existing technology and provide a method for converting home customization product requirements based on hierarchical frequent mining. The method reduces the dimension of the data set by layering the design parameters, mines the mapping relationship between the demand text and the design parameters layer by layer, and formulates a many-to-one probability relative collision rule for the mobile group individuals to accelerate the convergence of the rule optimization algorithm based on object collision, accurately mines knowledge in large-dimensional data sets, and completes the efficient and accurate conversion of smart home product demand text into design parameters.

[0005] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0006] 1. A method for mining transformation rules of smart home design parameters based on hierarchical dimensionality reduction

[0007] The following steps are involved:

[0008] (1) Obtain and pre-process the demand text in the historical demand conversion order information of smart home customized products;

[0009] (2) Perform dimensionality reduction on the customized product design parameters in the order information according to the characteristics of smart home products, and stratify the order data based on the dimensionality reduction results;

[0010] (3) Process each layer of order data layer by layer from top to bottom:

[0011] 3.1) Mining the mapping from the design parameters of the current layer to the requirement text based on the fp-growth algorithm;

[0012] 3.2) Using the improved object collision rule optimization algorithm to obtain the optimized support and confidence thresholds of the current layer data, and eliminating mapping rules with substandard support and confidence;

[0013] 3.3) Obtaining the next layer design parameters corresponding to the current layer mapping rule set, and using the order data corresponding to the next layer design parameters as the order data to be processed;

[0014] 3.4) Return to step 3.1) to mine the pending order data until all layers are mined and the final mapping rule set is output.

[0015] (4) Based on the final mapping rule set, the customer's needs are converted into corresponding design parameters, and finally the transformation of smart home customized product demand information into design parameters is realized.

[0016] The preprocessing in step (1) is to merge the semantically similar demand texts, specifically:

[0017] The demand texts in the historical demand conversion order information are stored in the database in the form of structured texts. The demand texts are converted into vectors using the Word2vec network model, and the cosine similarity of the demand vectors is calculated. The demand texts corresponding to the cosine similarity greater than the threshold are merged to finally obtain the demand set.

[0018] The process of performing dimensionality reduction processing on the customized product design parameters in the order data in the step (2) is specifically as follows: defining the design parameters from different dimensions according to the characteristics of the smart home products, including but not limited to defining the design parameters in three dimensions of parameter value, parameter name and parameter type, and dividing the design parameters into n levels; wherein the parameter value dimension is the largest, which is the lowest level parameter class; the parameter type dimension is the smallest, which is the highest level parameter class; for numerical parameter values, the parameter values ​​are classified into multiple intervals, and the intervals are used as the storage form of the parameter values ​​to reduce the dimension of the parameter values; the corresponding historical order data are layered according to the design parameters of different dimensions, and a total of L max layer.

[0019] In step 3.1), the design parameters of the current layer are mined for mapping from requirement text to design parameters based on the hierarchical results of step (2):

[0020] 3.1.1) Determine the initial support threshold minS of the current layer based on the current number of layers L L , confidence threshold minC L ; The smaller the number of layers L, the higher the initial support threshold minS L , confidence threshold minC L The bigger;

[0021] 3.1.2) After scanning the pre-processed current layer order data, count the frequency of each item in the item header table, and calculate the frequency according to the initial support threshold minS L Filter out infrequent items in the project; the order data includes project information and project frequency; the project information is the demand text of the customized product and its design parameters in the current layer of order data; the project frequency is the number of times the project information appears in the current layer of order data;

[0022] 3.1.3) Store the current layer order data in a frequent tree structure;

[0023] The item information in the order data is represented as a tree path, and the item frequency is stored in the tree node; the nodes with the same name (tree nodes with the same name as the item header table) are linked to the item header table, and the item frequency tree is generated according to the order of the item header table from bottom to top;

[0024] 3.1.4) Mining frequent item sets from the frequent tree;

[0025] Frequent item sets include frequently occurring requirement texts, frequently occurring design parameters, and frequently occurring requirement texts and design parameters at the same time;

[0026] 3.1.5) Set all design parameters as post-items, generate mapping rules from the requirement text to the design parameters based on the frequent item sets; calculate the confidence of each mapping rule, and use the confidence threshold minC L Eliminate rules with low confidence.

[0027] The object collision rule optimization algorithm improved in step 3.2) is:

[0028] 3.2.1) Encode the n preliminary mapping rules obtained in step 3.1) using the Michigan method to obtain the initial objects obj of the algorithm, obj = 1, 2, ..., n, and the initial position of each object is x obj 0 =(x obj1 0 ,x obj2 0 ,…,xobjD 0 ); then initialize the algorithm parameters, including the maximum number of iterations iter max ; Each object in the algorithm is each mapping rule obtained in step 3.1); x objD 0 is the initial position of the object in the Dth dimension;

[0029] 3.2.2) Iterative optimization process:

[0030] The mapping rule content is inferred based on the position of each object and the Michigan method encoding rule, and then the corresponding fitness value is calculated based on the mapping rule content, and the worst and best global fitness values ​​of this iteration are recorded;

[0031] 3.2.3) Calculate the mass of each object, and then sort the objects in descending order of mass and divide them into two groups, including the stationary object group. Group with moving objects

[0032] The mass of the stationary object group is greater than that of the moving object group;

[0033] 3.2.4) Initialize the collision velocities of the two groups of objects; update the velocity and displacement of the stationary group according to the one-to-one relative collision rule, and update the velocity and displacement of the moving group according to the many-to-one probabilistic relative collision rule;

[0034] 3.2.5) Return to step 3.2.2) until the maximum number of iterations iter is reached max Finally, the iteration is stopped and the optimal object (the object with the largest mass) is output;

[0035] 3.2.6) The support value and confidence value corresponding to the optimal object are used as the optimal support threshold and confidence threshold according to and Eliminate the mapping rules whose support and confidence values ​​do not meet the standards in the mapping rule set of step (3) (eliminate the mapping rules whose support and confidence values ​​are less than the optimal support threshold) and confidence threshold mapping rules).

[0036] The fitness value of object x is calculated through the fitness function. The fitness function is composed of support, confidence and promotion indicators. The formula is:

[0037] fit(x)=confidence(x)×lift(x)×log(1+support(x))

[0038] Where support(x) is the support, which is the degree of support for the mapping rule by the transaction. It is the percentage of items A and B appearing simultaneously in the transactions of dataset D and is calculated as Dataset D is the order data of the current layer, item A is the demand text, and item B is the design parameters;

[0039] confidence(x) is the confidence level, which represents the credibility of the mapping rule. It is the probability that event B will occur after event A occurs. It is calculated as

[0040] lift(x) is the lift, which is used to measure whether the mapping rule is useful. It is the probability of event B appearing after event A appears and the popularity of event B is known. It is calculated as

[0041] In the step 3.2.2):

[0042] The mass calculation formula of object x at the iterth iteration is:

[0043]

[0044] Where, fit x iter is the fitness value of object x at the iterth iteration, worst iter is the worst value of global fitness at the iter iteration, best iter is the optimal value of global fitness at the iter-th iteration; x=1,2,…,n;

[0045] It can be seen from the mass calculation formula that the larger the fitness value of an object, the greater its mass and the greater the probability of being classified as a stationary group object, that is, the greater the probability of becoming an individual moving target, providing direction for the individual's movement.

[0046] In the step 3.2.4):

[0047] a) Update the velocity and displacement of the stationary object group according to the one-to-one relative collision rule:

[0048] The velocity update formula for a stationary object is:

[0049]

[0050] Where, is the velocity of the moving group before collision; v i ′ is the velocity of the stationary group after the collision; m i is the mass of the i-th object, which is at rest; It is The mass of an object that is moving; ε is the coefficient of elastic restitution, iter is the current number of iterations, iter max is the maximum number of iterations.

[0051] The displacement update formula for a stationary object is:

[0052]

[0053] Where x′ i , v i ′ are the updated position and updated velocity of the stationary group, r1 is a random number between (-1,1);

[0054] b) Update the velocity and displacement of the moving object group based on the many-to-one probability relative collision rule:

[0055] When the current moving object j collides with each stationary object i, the initial collision velocity of object j, that is, the velocity of the moving group before the collision, is v j =x i -x j , x i is the position of the stationary group before collision, x j is the position of the mobile group before collision;

[0056] The calculation formula for the velocity update of the moving object after the current moving object j collides with one of the stationary objects i is as follows:

[0057]

[0058] Where, v j is the velocity of the moving group before the collision; v j ′ is the velocity of the moving group after the collision; m i is the mass of the i-th object, which is stationary; m j is the mass of the jth object, which is moving;

[0059] The displacement update calculation formula of the moving group objects after the current moving object j collides with one of the stationary objects i is as follows:

[0060]

[0061] Where x′ j ,v j ′ are the updated position and updated speed of the mobile group respectively; ω i is the collision impact factor, calculated as in r2 is a random number between (0,1), representing the collision probability; p ijis the collision probability threshold between the moving object j and the stationary object i, which is calculated as When the distance between two objects is farther, p ij The bigger; fit i is the fitness value of the stationary object i;

[0062] Finally, based on the above rules for updating the displacement of the moving object when two objects collide, we can deduce the overall impact of all stationary objects on the displacement of each moving object, that is, the many-to-one probability relative to the updated displacement of the moving objects after the collision. The calculation method is as follows:

[0063]

[0064] 2. A computer storage medium

[0065] The computer storage medium includes a stored program, wherein when the program is running, the device where the computer storage medium is located is controlled to execute the above-mentioned smart home design parameter conversion rule mining method based on hierarchical dimensionality reduction.

[0066] 3. A processor

[0067] The processor is used to run a program, wherein the program executes the above-mentioned smart home design parameter conversion rule mining method based on hierarchical dimensionality reduction when running.

[0068] The resulting mapping rule set is then used to translate customer requirements into customized smart home products. Based on the customer's input, the corresponding rules are found layer by layer in the mapping rule set. These requirements are then translated into design parameter value ranges based on the mapping rules.

[0069] The beneficial effects of the present invention are:

[0070] This invention uses a layer-by-layer frequent mining approach to reduce the dimensionality of the dataset and the width of the generated frequent tree, enabling rapid knowledge mining of large-dimensional text data and efficiently converting text requirements into design parameters for smart home products. By establishing a many-to-one probabilistic relative collision rule for mobile group individuals, this method accelerates the convergence of the rule-based object collision optimization algorithm and the precise optimization of the conversion rules. This also avoids the drawbacks of manually setting thresholds and requiring different thresholds for different layers of data, making the conversion of text requirements into design parameters for smart home products more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a flowchart of a method for mining smart home design parameter transformation rules based on hierarchical dimensionality reduction;

[0072] Figure 2 Schematic diagram of the improved collision method;

[0073] Figure 3 Schematic diagram of design parameters DETAILED DESCRIPTION

[0074] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0075] The terms used in this invention are for the purpose of describing specific embodiments only and are not intended to limit the invention. The singular forms "a," "the," and "the" used in this invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0076] It should be understood that although the terms "first," "second," "third," etc. may be used in the present invention to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to determining."

[0077] The specific embodiments of the present invention will be described more completely and clearly below with reference to the accompanying drawings. Figure 1 This paper proposes a method for mining smart home design parameter transformation rules based on hierarchical dimensionality reduction, which includes the following steps:

[0078] (1) Obtain and pre-process the demand text in the historical demand conversion order information of smart home customized products; specifically:

[0079] Derive historical demand conversion order information of smart home customized products from user database {R t ,P t},R t is the demand content in each message, P tThe design parameter content in each piece of information is merged. The demand text information with similar semantics is merged. The demand information in the historical demand conversion order data is stored in the database in the form of structured text. The demand text is converted into a vector using the Word2vec network model, and the cosine similarity of the demand vector is calculated; when the cosine similarity is greater than the threshold, the demand items are merged, and finally the preprocessed demand set R is obtained. t ′.

[0080] (2) Perform dimensionality reduction processing on the customized product design parameters in the order data according to the characteristics of smart home products; specifically:

[0081] like Figure 3 As shown in the figure, first, the design parameter classes of different dimensions are defined based on the characteristics of smart home products, that is, the structure of "parameter value-parameter name-parameter type". The parameter type is defined as the first-level design parameter class, and the first-level design parameters include two categories: "appearance parameters" and "function parameters"; the parameter name is defined as the second-level design parameter class, and the parameter name is the design feature name under the current parameter type of a specific home product, so the second-level design parameters include "size", "power", "material", etc.; the parameter value is defined as the third-level design parameter class, and the continuous parameter value is discretized into a suitable interval based on design knowledge. The interval is used as the storage form of the parameter value to reduce the dimension of the parameter value. Therefore, the third-level design parameter includes the parameter value interval under the specific parameter name. Finally, the preprocessed design parameter set P is obtained. t ′.

[0082] (3) Mining the mapping from each layer of design parameters to the requirement text from top to bottom; specifically:

[0083] First, according to the dimensionality reduction result in step (2), the historical order data is layered and L is set. max =3. The input data of each layer is {R t ′,P t ′}, P of the first layer data t ′ is the first-level design parameter, and P of the second-level data t ′ is the secondary design parameter, and P of the third layer data t ′ is the third-level design parameter. According to the current number of layers L, the initial support threshold minS of the current layer is determined L , confidence threshold minC L , the smaller the number of layers L, the higher the initial support minS L , confidence threshold minC L The bigger.

[0084] Scan the current layer data, collect frequent 1-item sets and their support in the item header table, and calculate the initial support S LFilter out non-frequent items; store the pre-processed current layer order data in a frequent tree structure, generate a header table and an FP tree; order data includes item information and item frequency; the item information is represented by a tree path, and the item frequency is stored in the tree node; items with the same name are linked to the header table; generate a frequent tree for the corresponding items in the order of the header table from bottom to top, take each frequent item in the header table, obtain the conditional pattern base of this element in each path of the tree, and stop iterating when the tree has only the root node; form a frequent item set with the frequent items taken from the header table in each recursive path, and generate a set for each frequent element of the head pointer of each recursive layer, namely a frequent item set; set all design parameters as post-items, and generate mapping rules according to the frequent item set; calculate the confidence of each mapping rule, and use the confidence threshold minC L Eliminate rules with low confidence.

[0085] (4) At the end of each layer of mapping mining, the improved object collision rule optimization algorithm is used to obtain the optimized support and confidence thresholds of the data in this layer; specifically:

[0086] First, the N preliminary mapping rules obtained in step (3) are encoded using the Michigan method. Assume that the number of demand categories in the database and the number of design parameter categories in the current layer are K, and each rule obj contains two parts V K1 and V K2 . V K1 The value of indicates whether the current item appears in this rule, 0 means it does not appear, 1 means it appears; V K2 The value of indicates whether the item is pre-rule or post-rule, 0 indicates pre-rule and 1 indicates post-rule. After encoding, the initial individual obj of the algorithm is obtained (obj=1,2,…,N). The initial position of each particle is x obj 0 =(x obj1 0 ,x obj2 0 ,…,x objD 0 ); Then initialize the algorithm parameters and set the maximum number of iterations iter max =100;

[0087] The fitness function of this algorithm is composed of support, confidence and improvement indicators. The fitness function formula is defined as:

[0088] fit(dx)=confidence(x)×lift(x)×log(1+support(x))

[0089] Where support(x) is the support, which is the support degree of the transaction item set to the rule. It is the percentage of items A and B appearing simultaneously in the transactions in the dataset D. It is calculated as confidence(x) is the confidence level, which is expressed as the credibility of the rule. It is the probability that B will occur after event A occurs. It is calculated as Lift(x) is the lift, which is used to measure whether the rule is useful. It is the probability of event B appearing after event A appears and the popularity of event B is known. It is calculated as

[0090] Set the mass of object i at the iterth iteration to:

[0091]

[0092] Where fit i iter is the fitness value of object i at the iterth iteration, worst iter is the worst value of global fitness at the iter iteration, best iter is the optimal global fitness value at the iterth iteration. From the mass calculation formula, we can see that the larger the fitness value of an object, the greater its mass.

[0093] According to the fitness function value descending order, the population is divided into two parts. The individuals with the largest fitness value in the front are divided into the static group. The initialization speed of the static group before the collision is 0, which is expressed as The individuals with smaller fitness values ​​at the back are divided into a mobile group. Each individual in the mobile group collides with the stationary group. The initial collision velocity of the mobile group is calculated as:

[0094]

[0095] Where v j and x j are the velocity and position vectors of the jth object, respectively. The object is moving; x i is the position vector of the i-th object, which is stationary.

[0096] Next, the speed and displacement of the stationary object group are updated according to the one-to-one relative collision rule, and the speed and displacement of the moving object group are updated according to the many-to-one probability relative collision rule. The improved collision method is as follows Figure 2 As shown. The one-to-one relative collision rule means that the speed and displacement of each stationary object are affected by the collision with its corresponding moving object. Therefore, the speed update formula of the stationary object is:

[0097]

[0098] In the formula is the velocity of the moving group before collision; v i ′ is the velocity of the stationary group after the collision; m i is the mass of the i-th object, which is at rest; It is The mass of an object that is moving; ε is the coefficient of elastic restitution, iter is the current number of iterations, iter max is the maximum number of iterations.

[0099] The displacement update formula for a stationary object is:

[0100]

[0101] Where x′ i , v i ′ are the updated position and updated velocity of the stationary group, and r1 is a random number between (-1,1).

[0102] The probability relative collision rule of many to one is: all stationary objects will have a probability relative collision with the current moving individual, which means that all high fitness individuals will generate different directions and degrees of probability movement direction guidance for low fitness individuals. This rule can accelerate the migration of low fitness individuals to the vicinity of the optimal value. The current moving individual i and each stationary object When a collision occurs, the initial collision velocity of individual i is v j =x i -x j , the speed update calculation formula of the mobile group individual after the current two bodies collide is as follows:

[0103]

[0104] Where, v j is the velocity of the moving group before collision; v j ′ is the velocity of the moving group after the collision; m i is the mass of the i-th object, which is stationary; m j is the mass of the jth object, which is moving.

[0105] The displacement update calculation formula of the mobile group individual after the current two bodies collide is as follows:

[0106]

[0107] Where x′ i , v j ′ are the updated position and updated speed of the mobile group respectively; ω iis the collision impact factor, calculated as in r2 is a random number between (0,1), representing the collision probability, p ij is the collision probability threshold between the moving object j and the stationary object i. The farther the distance between the two objects, the greater the probability of collision. ij The bigger.

[0108] Finally, the overall impact of all stationary group individuals on the displacement of the mobile group individuals is calculated, that is, the displacement of the mobile group after the many-to-one probability collision, that is, the position update of the mobile group individuals in this iteration process. The calculation method is as follows:

[0109]

[0110] After reaching the maximum number of iterations, the individual with the best fitness is obtained, and the optimal support threshold is obtained based on the characteristics of the individual and confidence threshold according to and Eliminate mapping rules whose support and confidence do not meet the standards.

[0111] (5) The resulting mapping rule set is used to transform the demand for customized smart home products; specifically:

[0112] After the customer inputs the demand content, the semantic demand information R of the demand content is calculated first. t According to the mapping rules of the first layer, the demand is converted into the first-level design parameters; then, according to the mapping rules of the second layer, the current first-level design parameters are converted into the second-level design parameters; then, according to the mapping rules of the third layer, the current second-level design parameters are converted into the third-level design parameters, that is, the design parameter value range, and finally the conversion of smart home customized product demand to design parameter value is realized.

[0113] The above embodiments are intended only to illustrate the design concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. The scope of protection of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design concepts disclosed in the present invention are within the scope of protection of the present invention.

Claims

1. A method for mining smart home design parameter transformation rules based on hierarchical dimensionality reduction, characterized in that: The following steps are involved: (1) Obtain and pre-process the demand text in the historical demand conversion order information of smart home customized products; (2) Perform dimensionality reduction on the customized product design parameters in the order information according to the characteristics of smart home products, and stratify the order data based on the dimensionality reduction results; (3) Process each layer of order data layer by layer from top to bottom: 3.1) Mining the mapping from the design parameters of the current layer to the requirement text based on the fp-growth algorithm; 3.2) Using the improved object collision rule optimization algorithm to obtain the optimized support and confidence thresholds of the current layer data, and eliminating mapping rules with substandard support and confidence; 3.3) Obtaining the next layer design parameters corresponding to the current layer mapping rule set, and using the order data corresponding to the next layer design parameters as the order data to be processed; 3.4) Return to step 3.1) to mine the pending order data until all layers are mined and the final mapping rule set is output; (4) According to the final mapping rule set, the customer's needs are converted into corresponding design parameters, and finally the transformation of smart home customized product demand information into design parameters is realized; The object collision rule optimization algorithm improved in step 3.2) is: 3.2.1) Encode the n preliminary mapping rules obtained in step 3.1) using the Michigan method to obtain the initial objects obj of the algorithm, obj = 1, 2, ..., n, and the initial position of each object is x obj 0 =(x obj1 0 ,x obj2 0 ,…,x objD 0 ); then initialize the algorithm parameters, including the maximum number of iterations iter max ; Each object in the algorithm is each mapping rule obtained in step 3.1); objD 0 is the initial position of the object in the Dth dimension; 3.2.2) Iterative optimization process: The mapping rule content is inferred based on the position of each object and the Michigan method encoding rule, and then the corresponding fitness value is calculated based on the mapping rule content, and the worst and best global fitness values ​​of this iteration are recorded; 3.2.3) Calculate the mass of each object, then sort the objects in descending order of mass and divide them into two groups: group i for stationary objects and group j for moving objects. 3.2.4) Initialize the collision velocities of the two groups of objects; update the velocity and displacement of the stationary group according to the one-to-one relative collision rule, and update the velocity and displacement of the moving group according to the many-to-one probabilistic relative collision rule; 3.2.5) Return to step 3.2.2) until the maximum number of iterations iter is reached max Then stop the iteration and output the optimal object; 3.2.6) The support value and confidence value corresponding to the optimal object are used as the optimal support threshold and confidence threshold according to and Eliminate the mapping rules whose support and confidence do not meet the standards in the mapping rule set in step (3).

2. The method for mining smart home design parameter conversion rules based on hierarchical dimensionality reduction according to claim 1 is characterized in that: The preprocessing in step (1) is to merge the semantically similar demand texts, specifically: The demand texts in the historical demand conversion order information are stored in the database in the form of structured texts. The demand texts are converted into vectors using the Word2vec network model, and the cosine similarity of the demand vectors is calculated. The demand texts corresponding to the cosine similarity greater than the threshold are merged to finally obtain the demand set.

3. The method for mining smart home design parameter conversion rules based on hierarchical dimensionality reduction according to claim 1 is characterized in that: The process of performing dimensionality reduction processing on the customized product design parameters in the order data in step (2) is specifically as follows: Define design parameters from different dimensions based on the characteristics of smart home products, including but not limited to defining design parameters in three dimensions: parameter value, parameter name, and parameter type, and classify design parameters into n levels; for numerical parameter values, classify the parameter values ​​into multiple intervals, and use intervals as the storage format for parameter values; According to the design parameters of different dimensions, the corresponding historical order data is divided into L max layer.

4. The method for mining smart home design parameter transformation rules based on hierarchical dimensionality reduction according to claim 1, characterized in that: In step 3.1), the design parameters of the current layer are mined for mapping from requirement text to design parameters based on the hierarchical results of step (2): 3.1.1) Determine the initial support threshold minS of the current layer based on the current number of layers L L , confidence threshold minC L ; 3.1.2) After scanning the pre-processed current layer order data, count the frequency of each item in the item header table, and calculate the frequency according to the initial support threshold minS L Eliminate infrequent items from the project; The order data includes item information and item frequency; item information is the demand text of the customized product and its design parameters in the current layer of order data; item frequency is the number of times the item information appears in the current layer of order data; 3.1.3) Store the current layer order data in a frequent tree structure; The item information in the order data is represented by a tree path, and the item frequency is stored in the tree node; Link the nodes with the same name with the item header table, and generate a frequent tree of items according to the order of the item header table from bottom to top; 3.1.4) Mining frequent item sets from the frequent tree; Frequent item sets include frequently occurring requirement texts, frequently occurring design parameters, and frequently occurring requirement texts and design parameters at the same time; 3.1.5) Set all design parameters as post-items, generate mapping rules from the requirement text to the design parameters based on the frequent item sets; calculate the confidence of each mapping rule, and use the confidence threshold minC L Eliminate rules with low confidence.

5. The method for mining smart home design parameter conversion rules based on hierarchical dimensionality reduction according to claim 1 is characterized in that: The fitness value of object x is calculated through the fitness function. The fitness function is composed of support, confidence and promotion indicators. The formula is: fit(x)=confidence(x)×lift(x)×log(1+support(x)) Where support(x) is the support, which is the degree of support for the mapping rule by the transaction. It is the percentage of items A and B appearing simultaneously in the transactions of dataset D and is calculated as Dataset D is the order data of the current layer, item A is the demand text, and item B is the design parameters; confidence(x) is the confidence level, which represents the credibility of the mapping rule. It is the probability that event B will occur after event A occurs. It is calculated as lift(x) is the lift, which is used to measure whether the mapping rule is useful. It is the probability of event B appearing after event A appears and the popularity of event B is known. It is calculated as 6. The method for mining smart home design parameter transformation rules based on hierarchical dimensionality reduction according to claim 1, characterized in that: In the step 3.2.2): The mass calculation formula of object x at the iterth iteration is: Where, fit x iter is the fitness value of object x at the iterth iteration, worst iter is the worst value of global fitness at the iter iteration, best iter is the optimal value of global fitness at the iter-th iteration; x = 1, 2, …, n.

7. The method for mining smart home design parameter transformation rules based on hierarchical dimensionality reduction according to claim 1, characterized in that: In the step 3.2.4): a) Update the velocity and displacement of the stationary object group according to the one-to-one relative collision rule: The velocity update formula for a stationary object is: Where, is the velocity of the moving group before collision; v i ′ is the velocity of the stationary group after the collision; m i is the mass of the i-th object, which is at rest; It is The mass of an object that is moving; ε is the elastic recovery coefficient, iter is the current number of iterations, iter max is the maximum number of iterations; The displacement update formula for a stationary object is: Where x′ i , v i ′ are the updated position and updated velocity of the stationary group, r1 is a random number between (-1,1); b) Update the velocity and displacement of the moving object group based on the many-to-one probability relative collision rule: When the current moving object j collides with each stationary object i, the initial collision velocity of object j, that is, the velocity before the collision, is v j =x i -x j , x i is the position of the stationary group before collision, x j is the position of the mobile group before collision; The calculation formula for the velocity update of the moving object after the current moving object j collides with one of the stationary objects i is as follows: Where, v j is the velocity of the moving group before the collision; v j ′ is the velocity of the moving group after the collision; m i is the mass of the i-th object, which is at rest; m j is the mass of the jth object, which is moving; The displacement update calculation formula of the moving group objects after the current moving object j collides with one of the stationary objects i is as follows: Where x′ j ,v j ′ are the updated position and updated speed of the mobile group respectively; ω i is the collision impact factor, calculated as in r2 is a random number between (0,1), representing the collision probability; p ij is the collision probability threshold between the moving object j and the stationary object i, which is calculated as When the distance between two objects is farther, p ij The bigger; fit i is the fitness value of the stationary object i; Finally, based on the above rules for updating the displacement of the moving object when two objects collide, we can deduce the overall impact of all stationary objects on the displacement of each moving object, that is, the many-to-one probability relative to the updated displacement of the moving objects after the collision. The calculation method is as follows:

8. A computer storage medium, characterized in that The computer storage medium includes a stored program, wherein when the program is running, the device where the computer storage medium is located is controlled to execute the smart home design parameter conversion rule mining method based on hierarchical dimensionality reduction as described in any one of claims 1 to 7.

9. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the smart home design parameter conversion rule mining method based on hierarchical dimensionality reduction as described in any one of claims 1 to 7.

Images