Air conditioner control method combining fuzzy detection and ambiguity elimination

Abstract

The invention provides an air conditioner control method combining fuzzy detection and ambiguity elimination, which comprises the following steps: receiving a user instruction, and carrying out text preprocessing on the user instruction to obtain an operation vocabulary; converting the operation vocabulary into an operation vector, and calculating the similarity between the operation vector and a predefined vector; and constructing a mask matrix, and filling the mask matrix according to the similarity. And if all matrix elements in the mask matrix are 0, multi-mode disambiguation is conducted, an operation instruction is obtained, and the air conditioner is controlled to execute the operation instruction. According to the method, the operation instruction selected by the user is obtained by adopting a mode of interacting with the user, so that ambiguity caused by a fuzzy instruction of the NLP system is eliminated. According to the air conditioner control method, the similarity of the vectors is calculated and the mask matrix is constructed on the basis of word segmentation, a deep learning algorithm in the prior art is replaced, and therefore the air conditioner control method does not need to use a large amount of training data for training and has high operation efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning control, and in particular to an air conditioning control method combining fuzzy detection and ambiguity elimination. Background Art

[0002] Existing air conditioning control methods are mainly divided into two categories. The first category is to control the working mode and working temperature of the air conditioner based on the air conditioner remote control. The second category is to receive user voice and convert the user voice into executable instructions for the air conditioner. The air conditioner adjusts the temperature of the indoor environment according to the executable instructions.

[0003] Air conditioning control methods based on user voice usually adopt rule-based NLP (Natural Language Processing) systems and machine learning algorithms. The NLP system is not accurate enough when processing fuzzy instructions, resulting in the inability to control the air conditioner according to user needs. The machine learning algorithm requires a large amount of training data to train the model.

[0004] Therefore, an air conditioning control method is needed that can eliminate the ambiguity caused by ambiguous instructions and does not require a large amount of training data for training. Summary of the invention

[0005] In order to overcome the problems existing in the related art, the purpose of the present invention is to provide an air-conditioning control method combining fuzzy detection and ambiguity elimination, which can eliminate the ambiguity caused by fuzzy instructions and does not require a large amount of training data for training.

[0006] An air conditioning control method combining fuzzy detection and ambiguity elimination, comprising:

[0007] receiving a user instruction, performing text preprocessing on the user instruction, and obtaining an operation vocabulary;

[0008] Converting the operation vocabulary into an operation vector, and calculating the similarity between the operation vector and a predefined vector;

[0009] Constructing a mask matrix, and filling the mask matrix according to the similarity;

[0010] If all matrix elements in the mask matrix are 0, multimodal disambiguation is performed to obtain an operation instruction;

[0011] The air conditioner is controlled to execute the operation instruction.

[0012] In a preferred technical solution of the present invention, the converting the operation vocabulary into an operation vector and calculating the similarity between the operation vector and a predefined vector includes:

[0013] Calculating a normalized cosine distance between the operation vector and a predefined vector;

[0014] Calculating a normalized Euclidean distance between the operation vector and a predefined vector;

[0015] Calculating a normalized Manhattan distance between the operation vector and a predefined vector;

[0016] A weighted sum is performed on the normalized cosine distance, the normalized Euclidean distance, and the normalized Manhattan distance to obtain a similarity between the operation vector and the predefined vector.

[0017] In a preferred technical solution of the present invention, the step of constructing a mask matrix includes:

[0018] Using the user instruction as a row vector of a mask matrix;

[0019] The predefined vocabulary is used as the column vector of the mask matrix.

[0020] In a preferred technical solution of the present invention, filling the mask matrix according to the similarity includes:

[0021] Get the similarity threshold;

[0022] If the similarity is greater than the similarity threshold, filling 1 in the position corresponding to the similarity in the mask matrix;

[0023] If the similarity is less than or equal to the similarity threshold, the position corresponding to the similarity in the mask matrix is ​​filled with 0.

[0024] In a preferred technical solution of the present invention, before obtaining the similarity threshold, the method further includes:

[0025] The similarity threshold is calculated according to the following formula:

[0026] T = mu + k * sigma;

[0027] Wherein, T is the similarity threshold, mu is the average value of the similarity in the historical data, sigma is the standard deviation of the similarity in the historical data, and k is the first coefficient.

[0028] In a preferred technical solution of the present invention, the multimodal disambiguation is performed to obtain an operation instruction, including:

[0029] Selecting a disambiguation mode, wherein the disambiguation mode includes voice interactive disambiguation and image interactive disambiguation;

[0030] If the disambiguation mode is the voice interactive disambiguation, a voice prompt is issued to the user to obtain an operation instruction returned by the user;

[0031] If the disambiguation mode is the image interactive disambiguation, an operation image is displayed on an interactive interface, and the operation instruction corresponding to the operation image clicked by the user is obtained.

[0032] In a preferred technical solution of the present invention, after filling the mask matrix according to the similarity, the method further includes:

[0033] If there are matrix elements that are not 0 in the mask matrix, the air conditioner is controlled to execute the operation instruction.

[0034] In a preferred technical solution of the present invention, the text preprocessing of the user instruction to obtain the operation vocabulary includes:

[0035] Performing text cleaning on the user instruction to obtain a cleaned instruction;

[0036] The cleaned instructions are segmented to obtain operation words.

[0037] In a preferred technical solution of the present invention, the step of performing text cleaning on the user instruction to obtain the cleaned instruction includes:

[0038] Punctuation marks, stop words and special characters in the user instructions are removed to obtain cleaned instructions.

[0039] In a preferred technical solution of the present invention, the user instructions include raising the air conditioning temperature, turning on the air purifier and the fan speed is too slow, and the predefined vocabulary includes temperature, wind speed, mode, on and off.

[0040] The beneficial effects of the present invention are:

[0041] The present invention provides an air conditioning control method combining fuzzy detection and ambiguity elimination, comprising receiving a user instruction, performing text preprocessing on the user instruction, and splitting the user instruction from a sentence or a paragraph into multiple operation words. The operation words are converted into operation vectors, and the similarity between the operation vectors and the predefined vectors is calculated. The higher the similarity, the closer the operation vector is to the predefined vector, the stronger the readability of the operation vector is, and the greater the probability that the air conditioner can execute the operation instruction corresponding to the operation vector. A mask matrix is ​​constructed, and the mask matrix is ​​filled according to the similarity. When the similarity is greater than a similarity threshold, the corresponding matrix element in the mask matrix is ​​filled with 1, indicating that the matrix element is successfully matched. When the similarity is less than or equal to the similarity threshold, the corresponding matrix element in the mask matrix is ​​filled with 0, indicating that the matrix element fails to match. If all matrix elements in the mask matrix are 0, it means that all matrix elements in the mask matrix fail to match, and the air conditioning control instruction represented by the mask matrix at this time has ambiguity and uncertainty. By constructing the mask matrix, the relationship between the operation instruction and the operation word can be determined. Multimodal disambiguation is performed on the mask matrix in which all matrix elements are 0 to obtain operation instructions, control the air conditioner to execute the operation instructions, adjust the environmental parameters, and improve the user's home experience. The present invention uses a method of interacting with the user to obtain the operation instructions selected by the user, thereby eliminating the ambiguity caused by the ambiguous instructions of the NLP system. The present invention calculates the similarity of vectors and constructs a mask matrix based on word segmentation, replacing the deep learning algorithm in the prior art. Therefore, the air conditioning control method of the present invention does not need to use a large amount of training data for training, and has high computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of an air conditioning control method combining fuzzy detection and ambiguity elimination of the present invention;

[0043] Figure 2 is a flow chart of the present invention for calculating the similarity between an operation vector and a predefined vector;

[0044] Figure 3 is a flow chart of filling a mask matrix according to similarity of the present invention;

[0045] Figure 4 It is a flow chart of the present invention for performing text preprocessing on user instructions. DETAILED DESCRIPTION

[0046] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0047] Example 1

[0048] like Figure 1 As shown, this embodiment provides an air conditioning control method combining fuzzy detection and ambiguity elimination, including:

[0049] S1: receiving a user instruction, performing text preprocessing on the user instruction, and obtaining an operation vocabulary.

[0050] S2: Convert the operation vocabulary into an operation vector, and calculate the similarity between the operation vector and a predefined vector.

[0051] S3: construct a mask matrix, and fill the mask matrix according to the similarity.

[0052] S4: If all matrix elements in the mask matrix are 0, multimodal disambiguation is performed to obtain an operation instruction.

[0053] S5: Control the air conditioner to execute the operation instruction.

[0054] Text preprocessing includes text cleaning and word segmentation. Text cleaning is used to detect whether the user instructions conform to the preset format. If not, the user instructions are judged as not meeting the requirements, and the user instructions that do not meet the requirements are removed, while the user instructions that meet the requirements are retained. The user instructions that meet the requirements are segmented to split a sentence or paragraph into multiple operation words.

[0055] The NLP algorithm is used to convert the operation vocabulary into the operation vector, and the operation vocabulary and fuzzy description are predefined to obtain the predefined operation vocabulary and predefined fuzzy description. The predefined operation vocabulary includes temperature, wind speed, wind direction and mode, and the predefined fuzzy description includes comfortable, too hot, too cold and bad air. The predefined operation vocabulary and predefined fuzzy description are converted into predefined vectors, and the similarity between the operation vector and the predefined vector is calculated.

[0056] Construct a mask matrix and fill the mask matrix according to the value of similarity. Set a similarity threshold. If the similarity is greater than the similarity threshold, it means that the similarity between the operation vector and the predefined vector is high at this time, and the air conditioner can execute the operation instruction corresponding to the matrix element, and fill the matrix element with 1. If the similarity is less than or equal to the similarity threshold, it means that the similarity between the operation vector and the predefined vector is low at this time, and the air conditioner may not be able to recognize and execute the operation instruction corresponding to the matrix element, and fill the matrix element with 0.

[0057] After filling the mask matrix, the mask matrix has three forms. The first form is that some matrix elements in the mask matrix are 1 and the other matrix elements are 0. The second form is that all matrix elements in the mask matrix are 1. The third form is that all matrix elements in the mask matrix are 0.

[0058] If all matrix elements in the mask matrix are 0, it means that the operation instruction corresponding to the mask matrix is ​​ambiguous, and the air conditioner cannot execute the operation instruction corresponding to the mask matrix. Therefore, multimodal disambiguation is performed, and multiple options are displayed to the user through voice or image, and the user is prompted to make a selection to obtain the operation instruction selected by the user.

[0059] The air conditioner is controlled to execute the operation instruction, for example, the air conditioner is controlled to start and enter the cooling mode, the cooling temperature of the air conditioner is set to 25°C, and the air outlet speed of the air conditioner is set to level 2.

[0060] The present embodiment provides an air conditioning control method combining fuzzy detection and ambiguity elimination, including receiving user instructions, performing text preprocessing on the user instructions, and splitting the user instructions from a sentence or a paragraph into multiple operation words. The operation words are converted into operation vectors, and the similarity between the operation vectors and the predefined vectors is calculated. The higher the similarity, the closer the operation vector is to the predefined vector, the stronger the readability of the operation vector, and the greater the probability that the air conditioner can execute the operation instruction corresponding to the operation vector. Construct a mask matrix and fill the mask matrix according to the similarity. When the similarity is greater than the similarity threshold, the corresponding matrix element in the mask matrix is ​​filled with 1, indicating that the matrix element matches successfully. When the similarity is less than or equal to the similarity threshold, the corresponding matrix element in the mask matrix is ​​filled with 0, indicating that the matrix element fails to match. If all matrix elements in the mask matrix are 0, it means that all matrix elements in the mask matrix fail to match, and the air conditioning control instruction represented by the mask matrix at this time has ambiguity and uncertainty. Constructing a mask matrix can determine the relationship between the operation instruction and the operation vocabulary. Multimodal disambiguation is performed on the mask matrix in which all matrix elements are 0 to obtain operation instructions, control the air conditioner to execute the operation instructions, adjust the environmental parameters, and improve the user's home experience. The present invention uses a method of interacting with the user to obtain the operation instructions selected by the user, thereby eliminating the ambiguity caused by the ambiguous instructions of the NLP system. The present invention calculates the similarity of vectors and constructs a mask matrix based on word segmentation, replacing the deep learning algorithm in the prior art. Therefore, the air conditioning control method of the present invention does not need to use a large amount of training data for training, and has high computational efficiency.

[0061] Example 2

[0062] This embodiment provides an air conditioning control method combining fuzzy detection and ambiguity elimination. This embodiment only describes the differences from Embodiment 1. Figure 2 As shown, the converting the operation vocabulary into an operation vector and calculating the similarity between the operation vector and a predefined vector includes:

[0063] S21: Calculate the normalized cosine distance between the operation vector and a predefined vector.

[0064] S22: Calculate the normalized Euclidean distance between the operation vector and a predefined vector.

[0065] S23: Calculate the normalized Manhattan distance between the operation vector and a predefined vector.

[0066] S24: performing weighted summation on the normalized cosine distance, the normalized Euclidean distance, and the normalized Manhattan distance to obtain a similarity between the operation vector and the predefined vector.

[0067] The calculation formula of normalized cosine distance is as follows:

[0068] normalized s (A,B) = (cosine(A,B) + 1) / 2;

[0069] cosine(A,B)=(AB) / (||A||*||B||);

[0070] Among them, A represents the operation vector, B represents the predefined vector, and normalized s (A,B) represents the normalized cosine distance between the operation vector A and the predefined vector B, cosine(A,B) represents the cosine distance between the operation vector A and the predefined vector B, . represents the dot product of the vectors, * represents the modulus multiplication operation, ||A|| represents the modulus of the operation vector A, and ||B|| represents the modulus of the predefined vector B.

[0071] The calculation formula of normalized Euclidean distance is as follows:

[0072] normalized e (A,B)=1-(euclidean(A,B) / euclidean max );

[0073]

[0074] Among them, A i is the i-th element of the operation vector A, B i is the i-th element of the predefined vector B, euclidean(A,B) represents the Euclidean distance between the operation vector A and the predefined vector B, euclidean max represents the maximum Euclidean distance, n represents the total number of elements in the operation vector A, normalized e (A, B) represents the normalized Euclidean distance.

[0075] The calculation formula of normalized Manhattan distance is as follows:

[0076] normalized m (A,B)=1-(manhattan(A,B) / manhattan max );

[0077]

[0078] Among them, A i is the i-th element of the operation vector A, B i is the i-th element of the predefined vector B, manhattan(A,B) represents the Manhattan distance between the operation vector A and the predefined vector B, | A i -B i| represents the absolute value of the difference between the i-th element of the operation vector A and the i-th element of the predefined vector B, n represents the total number of elements in the operation vector A, and manhattan max Represents the maximum Manhattan distance, normalized m (A, B) represents the normalized Manhattan distance between the operation vector A and the predefined vector B.

[0079] The normalized cosine distance, normalized Euclidean distance, and normalized Manhattan distance are weighted summed using the following formula:

[0080] composite(A,B)=w1*normalized s (A,B)+w2*normalized e (A,B)+w3*normalized m (A,B);

[0081] w1+w2+w3=1;

[0082] Among them, w1 is the first weight, w2 is the second weight, w3 is the third weight, and composite(A,B) is the similarity between the operation vector A and the predefined vector B.

[0083] Cosine distance is more about distinguishing differences in direction. Cosine distance can overcome the influence of dimension, but may be affected when dealing with extreme values. As the data dimension increases, the accuracy of Euclidean distance decreases, and different features cannot be treated differently, which may exaggerate the role of weak features. Manhattan distance has higher stability in some cases and has higher computational efficiency, but Manhattan distance may not be the shortest distance.

[0084] The cosine distance, Euclidean distance and Manhattan distance are normalized so that they have a unified standard and their values ​​are within a fixed range. The weighted sum of the normalized cosine distance, normalized Euclidean distance and normalized Manhattan distance can comprehensively measure the similarity between two vectors in different aspects, and while highlighting the advantages of different distance indicators, reduce the impact of the disadvantages of different distance indicators. In addition, according to the operation vector A and the predefined vector B, the weights corresponding to the three normalized distances can be set to increase the weights of some distance indicators and reduce the weights of other distance indicators.

[0085] like Figure 3 As shown, the construction of the mask matrix includes:

[0086] S31: Use the user instruction as a row vector of a mask matrix.

[0087] S32: Using the predefined vocabulary as the column vector of the mask matrix.

[0088] As an example, the row vectors of the mask matrix include turning up the air conditioner temperature, turning on the air purifier, and the fan speed is too slow, and the column vectors of the mask matrix include temperature, speed, mode, on, and off.

[0089] Filling the mask matrix according to the similarity comprises:

[0090] S33: Obtain a similarity threshold.

[0091] S34: If the similarity is greater than the similarity threshold, fill 1 in the position corresponding to the similarity in the mask matrix.

[0092] S35: If the similarity is less than or equal to the similarity threshold, fill the position corresponding to the similarity in the mask matrix with 0.

[0093] The similarity threshold is used to measure which matrix elements in the mask matrix have a high enough similarity, so as to determine whether the instruction is clear. If the similarity between the operation vector A and the predefined vector B is greater than the similarity threshold, it means that the operation instruction is clear and the match is successful, and the corresponding matrix element in the mask matrix is ​​filled with 1; if the similarity between the operation vector A and the predefined vector B is less than or equal to the similarity threshold, it means that the operation instruction is ambiguous and the match fails, and the corresponding matrix element in the mask matrix is ​​filled with 0.

[0094] The similarity threshold can be set uniformly or adjusted according to the specific environment and application scenario. Usually, the design of the similarity threshold is combined with past data analysis results and user experience feedback. The range of the similarity threshold is greater than or equal to 0.5 and less than or equal to 0.7. Further fine-tuning is performed based on actual application feedback and data distribution. For example, if the similarity between most operation vectors and predefined vectors is concentrated in this range, the similarity threshold is set to 0.6.

[0095] The optimal value needs to be determined through testing and iteration. Before obtaining the similarity threshold, the following steps are also included:

[0096] The similarity threshold is calculated according to the following formula:

[0097] T = mu + k * sigma;

[0098] Wherein, T is the similarity threshold, mu is the average value of the similarity in the historical data, sigma is the standard deviation of the similarity in the historical data, and k is the first coefficient.

[0099] The larger the value of the first coefficient, the faster the iterative update speed of the similarity threshold; the smaller the value of the first coefficient, the slower the iterative update speed of the similarity threshold. The similarity threshold is positively correlated with the average value of the similarity in the historical data and the standard deviation of the similarity in the historical data, that is, the larger the average value of the similarity in the historical data, the larger the similarity threshold; the larger the standard deviation of the similarity in the historical data, the larger the similarity threshold.

[0100] Historical data refers to the historical data accumulated from 0 when each user uses the system. Historical data includes the processing results of the user's past instructions and the corresponding similarity information. Historical data can be formed by continuously recording the interaction between the user and the system, thereby continuously optimizing the system's responsiveness and data processing capabilities.

[0101] Historical data includes user interaction logs and similarity accumulation. User interaction logs record each user's instructions, system responses, and satisfaction feedback. These data will serve as the basis for training and optimizing the system. Similarity accumulation: The similarity of each user in the historical data can be used to calculate various statistical indicators, such as the mean and standard deviation of similarity. The similarity threshold is iteratively updated based on the mean and standard deviation of similarity.

[0102] In this embodiment, the operation vocabulary is converted into an operation vector, and the similarity between the operation vector and the predefined vector is calculated, including calculating the normalized cosine distance between the operation vector and the predefined vector, calculating the normalized Euclidean distance between the operation vector and the predefined vector, and calculating the normalized Manhattan distance between the operation vector and the predefined vector. The normalized cosine distance, the normalized Euclidean distance and the normalized Manhattan distance are weighted and summed to obtain the similarity between the operation vector and the predefined vector. The weighted summation of the normalized cosine distance, the normalized Euclidean distance and the normalized Manhattan distance can comprehensively measure the similarity between the two vectors in different aspects, and while highlighting the advantages of different distance indicators, reduce the impact of the disadvantages of different distance indicators. In addition, according to the operation vector A and the predefined vector B, the weights corresponding to the three normalized distances can be set, and the weights of some distance indicators are increased and the weights of other distance indicators are reduced.

[0103] Example 3

[0104] This embodiment provides an air conditioning control method combining fuzzy detection and ambiguity elimination. This embodiment only describes the differences from Embodiment 1.

[0105] The performing multimodal disambiguation to obtain an operation instruction includes:

[0106] S41: Select a disambiguation mode, where the disambiguation mode includes voice interactive disambiguation and image interactive disambiguation.

[0107] S42: If the disambiguation mode is the voice interactive disambiguation, a voice prompt is issued to the user to obtain an operation instruction returned by the user.

[0108] S43: If the disambiguation mode is the image interactive disambiguation, an operation image is displayed on an interactive interface, and the operation instruction corresponding to the operation image clicked by the user is obtained.

[0109] Step S42 and step S43 are parallel steps. When it is determined that the operation instruction is clear, if the current wind speed of the air conditioner needs to be adjusted, and the current wind speed of the air conditioner is too slow, the current speed of the air conditioner is increased. As an example, the current wind speed of the air conditioner is 1, and the user's operation instruction is that the air conditioner wind speed is too slow. The system determines that the wind speed expected by the user should be greater than 1. The ideal wind speed range is set to 2-5, the minimum wind speed expected by the user is 2, and the maximum wind speed expected by the user is 5. The current wind speed of the air conditioner is adjusted to a value within the ideal wind speed range. A random adjustment method can be used, that is, a value is randomly selected within the ideal wind speed range as the wind speed expected by the user, or a fixed adjustment method can be used, that is, the user's feedback strength is obtained, and the current wind speed of the air conditioner is adjusted according to the feedback strength. The above-mentioned method for adjusting the current wind speed of the air conditioner can not only ensure that user needs are met, but also enhance the user experience, making the operation of equipment such as air conditioners more intelligent and intuitive.

[0110] Table 1 Blurred mask matrix

[0111]

[0112] As shown in Table 1, when all matrix elements in the mask matrix are 0, it is determined that the operation instruction corresponding to the mask matrix is ​​ambiguous, and voice interactive disambiguation or image interactive disambiguation is adopted at this time.

[0113] Interactive voice disambiguation: When the system detects that the operation instruction is ambiguous, it prompts the user to confirm or provide more information. For example, the system sends a prompt to the user, asking if you want to lower the air conditioner temperature. The system provides the user with multiple options, such as lowering the temperature and increasing the wind speed.

[0114] Image interactive disambiguation: Generate multiple images on the interactive interface and present them to the user. The user selects one of the images, and the air conditioner executes the operation instruction corresponding to the image selected by the user.

[0115] This embodiment adopts a multimodal method for semantic disambiguation, realizes high-reliability home scene command recognition, enhances user interaction experience, effectively solves the problem of ambiguous operation instructions, and improves the system's adaptability to different types of user instructions.

[0116] After filling the mask matrix according to the similarity, the method further includes:

[0117] If there are matrix elements that are not 0 in the mask matrix, the air conditioner is controlled to execute the operation instruction.

[0118] Table 2 Mask matrix with clear meaning

[0119]

[0120] As shown in Table 2, when there are one or more matrix elements with a value of 1 in the mask matrix, it is determined that the operation instruction corresponding to the mask matrix has a clear meaning.

[0121] The user instructions include raising the air conditioning temperature, turning on the air purifier and the fan speed is too slow, and the predefined vocabulary includes temperature, wind speed, mode, on and off.

[0122] The multimodal disambiguation of this embodiment to obtain the operation instruction includes selecting a disambiguation mode, and the disambiguation mode includes voice interactive disambiguation and image interactive disambiguation. If the disambiguation mode is the voice interactive disambiguation, a voice prompt is issued to the user to obtain the operation instruction returned by the user. If the disambiguation mode is the image interactive disambiguation, an operation image is displayed on the interactive interface to obtain the operation instruction corresponding to the operation image clicked by the user. The multimodal method is used for semantic disambiguation to achieve high-reliability home scene instruction recognition. The user interaction experience is enhanced, the problem of ambiguous operation instructions is effectively solved, and the system's adaptability to different types of user instructions is improved.

[0123] Example 4

[0124] This embodiment provides an air conditioning control method combining fuzzy detection and ambiguity elimination. This embodiment only describes the differences from Embodiment 1. Figure 4 As shown, the text preprocessing of the user instruction to obtain the operation vocabulary includes:

[0125] S12: Perform text cleaning on the user instruction to obtain a cleaned instruction.

[0126] S13: performing word segmentation processing on the cleaned instructions to obtain operation words.

[0127] Before step S12, the method further includes step S11: receiving a user instruction.

[0128] The system receives user commands input by voice or text. If the user command is not related to the device, no subsequent processing is performed to ensure the stability of control and reduce the probability of false device wake-up. For example, if the user command obtained by the system is to operate a humidifier, but there is no humidifier in the home environment, the user command is excluded.

[0129] User instructions are in text form. Text cleaning refers to removing punctuation marks, stop words and special characters in user instructions. Word segmentation refers to splitting the cleaned text into multiple operation words, including temperature, wind speed, wind direction and mode.

[0130] The text preprocessing of the user instructions in this embodiment to obtain operation vocabulary includes text cleaning of the user instructions to obtain cleaned instructions. The cleaned instructions are segmented to obtain operation vocabulary. The user instructions are in text form, and text cleaning refers to removing punctuation marks, stop words and special characters in the user instructions. The word segmentation refers to splitting the cleaned text into multiple operation vocabulary, and the operation vocabulary includes temperature, wind speed, wind direction and mode.

[0131] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps set forth in these embodiments do not limit the scope of the application. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a restriction. Therefore, other examples of exemplary embodiments can have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, and therefore, once a certain item is defined in an accompanying drawing, it does not need to be further discussed in the subsequent accompanying drawings.

[0132] It should be understood that spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if the device in the figures is inverted, a device described as "above" or "above" other devices or structures would then be positioned "below" or "below" the other devices or structures. Thus, the exemplary term "above" can include both the orientations of "above" and "below." The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used herein should be interpreted accordingly.

[0133] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. Unless otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of this application.

[0134] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An air conditioning control method combining fuzzy detection and ambiguity elimination, characterized in that: include: receiving a user instruction, performing text preprocessing on the user instruction, and obtaining an operation vocabulary; Converting the operation vocabulary into an operation vector, and calculating the similarity between the operation vector and a predefined vector; Constructing a mask matrix, and filling the mask matrix according to the similarity; If all matrix elements in the mask matrix are 0, multimodal disambiguation is performed to obtain an operation instruction; The air conditioner is controlled to execute the operation instruction.

2. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 1, characterized in that: The converting the operation vocabulary into an operation vector and calculating the similarity between the operation vector and a predefined vector comprises: Calculating a normalized cosine distance between the operation vector and a predefined vector; Calculating a normalized Euclidean distance between the operation vector and a predefined vector; Calculating a normalized Manhattan distance between the operation vector and a predefined vector; A weighted sum is performed on the normalized cosine distance, the normalized Euclidean distance, and the normalized Manhattan distance to obtain a similarity between the operation vector and the predefined vector.

3. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 1, characterized in that: The constructing of the mask matrix comprises: Using the user instruction as a row vector of a mask matrix; The predefined vocabulary is used as the column vector of the mask matrix.

4. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 1, characterized in that: Filling the mask matrix according to the similarity comprises: Get the similarity threshold; If the similarity is greater than the similarity threshold, filling 1 in the position corresponding to the similarity in the mask matrix; If the similarity is less than or equal to the similarity threshold, the position corresponding to the similarity in the mask matrix is ​​filled with 0.

5. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 4, characterized in that: Before obtaining the similarity threshold, the method further includes: The similarity threshold is calculated according to the following formula: T = mu + k * sigma; Wherein, T is the similarity threshold, mu is the average value of the similarity in the historical data, sigma is the standard deviation of the similarity in the historical data, and k is the first coefficient.

6. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 1, characterized in that: The performing multimodal disambiguation to obtain an operation instruction includes: Selecting a disambiguation mode, wherein the disambiguation mode includes voice interactive disambiguation and image interactive disambiguation; If the disambiguation mode is the voice interactive disambiguation, a voice prompt is issued to the user to obtain an operation instruction returned by the user; If the disambiguation mode is the image interactive disambiguation, an operation image is displayed on an interactive interface, and the operation instruction corresponding to the operation image clicked by the user is obtained.

7. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 1, characterized in that: After filling the mask matrix according to the similarity, the method further includes: If there are matrix elements that are not 0 in the mask matrix, the air conditioner is controlled to execute the operation instruction.

8. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 1, characterized in that: The step of performing text preprocessing on the user instruction to obtain an operation vocabulary includes: Performing text cleaning on the user instruction to obtain a cleaned instruction; The cleaned instructions are segmented to obtain operation words.

9. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 8, characterized in that: The step of performing text cleaning on the user instruction to obtain a cleaned instruction includes: Punctuation marks, stop words and special characters in the user instructions are removed to obtain cleaned instructions.

10. The air conditioning control method combining fuzzy detection and ambiguity elimination according to claim 3, characterized in that: The user instructions include raising the air conditioning temperature, turning on the air purifier and the fan speed is too slow, and the predefined vocabulary includes temperature, wind speed, mode, on and off.

Images