A control method and control system for furniture plate processing equipment

Abstract

The present invention discloses a control method and control system for furniture board processing equipment. The present invention collects data of furniture board processing equipment and forms a multi-depth knowledge vector spectrum based on pyramid knowledge vector mining, which can comprehensively and deeply mine the value of data. The equipment control decision features of different processing tasks are obtained from the knowledge vector spectra of different pyramid depths, which can accurately adapt to the needs of linkage processing tasks and other processing tasks. For linkage processing tasks, the decision features obtained from the first pyramid depth can effectively coordinate the relationship between equipment, avoid adverse effects between equipment, and improve overall processing efficiency. The decision features of the second processing task are obtained from the knowledge vector spectra of other depths, taking into account more complex processing influencing factors, and improving the processing quality. Finally, remote control is achieved by combining the two decision features, reducing the cost of manual intervention, and improving the timeliness and accuracy of control, thereby realizing efficient, high-quality and intelligent furniture board processing.

Description

Technical Field

[0001] The embodiments of the present invention belong to the field of intelligent control technology, and specifically relate to a control method and control system for furniture board processing equipment. Background Art

[0002] In the field of furniture board processing, traditional equipment control methods have many shortcomings. On the one hand, the use of equipment data often stays on the surface, lacks in-depth mining, and cannot fully grasp the various complex factors in the processing process. For example, only focusing on the direct measurement value of a single sensor makes it difficult to understand the equipment coordination relationship and the overall state change of the board hidden behind the data during the processing process. On the other hand, in terms of processing task control, for linkage processing tasks, traditional methods are difficult to accurately coordinate the operating parameters of different equipment, which easily leads to poor coordination between equipment, reduced processing efficiency, and increased defective rate. At the same time, for different processing tasks (such as linkage processing and single process processing), there is a lack of an effective integrated control method, and the control strategy cannot be flexibly adjusted according to the characteristics of the processing task. Moreover, there is a lot of manual intervention, which not only increases costs, but also may lead to untimely and inaccurate control due to human factors. Summary of the invention

[0003] The embodiments of the present invention provide a control method and a control system for furniture board processing equipment, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] An embodiment of the present invention provides a control method for furniture board processing equipment, which is applied to a control system. The method includes: acquiring target sensor monitoring data collected by an Internet of Things sensor installed on the furniture board processing equipment; performing pyramid knowledge vector mining based on the target sensor monitoring data to obtain target sensor monitoring knowledge vector spectra of several pyramid depths; using a target sensor monitoring knowledge vector spectrum of a first pyramid depth among the several target sensor monitoring knowledge vector spectra as a first target sensor monitoring knowledge vector spectrum, obtaining a first equipment control decision feature of a first processing task based on the first target sensor monitoring knowledge vector spectrum, and obtaining a second equipment control decision feature of a second processing task based on target sensor monitoring knowledge vector spectra other than the first target sensor monitoring knowledge vector spectrum; wherein the first processing task includes a linkage processing task; and remotely controlling the furniture board processing equipment in combination with the first equipment control decision feature and the second equipment control decision feature.

[0005] An embodiment of the present invention provides a control system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the above method.

[0006] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0007] The embodiment of the present invention collects data of furniture board processing equipment through IoT sensors, and forms a multi-depth knowledge vector spectrum based on pyramid knowledge vector mining. This method can comprehensively and deeply mine the value of data. The equipment control decision features of different processing tasks are obtained from the knowledge vector spectra of different pyramid depths, which can accurately adapt to the needs of linkage processing tasks and other processing tasks. For linkage processing tasks, the decision features obtained from the first pyramid depth can effectively coordinate the relationship between equipment, avoid adverse effects between equipment, and improve overall processing efficiency. The decision features of the second processing task are obtained from the knowledge vector spectra of other depths, taking into account more complex processing influencing factors, and improving processing quality. Finally, remote control is achieved by combining the two decision features, reducing the cost of manual intervention, and improving the timeliness and accuracy of control, thereby achieving efficient, high-quality and intelligent furniture board processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of a method for controlling furniture plate processing equipment provided by an embodiment of the present invention;

[0009] Figure 2 A schematic diagram of the structure of a control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] Figure 1 A method for controlling furniture board processing equipment is shown, which is applied to a control system. The method includes the following steps 110 to 130.

[0011] Step 110: The control system obtains target sensor monitoring data collected by the IoT sensor installed on the furniture board processing equipment; pyramid knowledge vector mining is performed based on the target sensor monitoring data to obtain target sensor monitoring knowledge vector spectra of several pyramid depths.

[0012] In an embodiment of the present invention, the furniture board processing equipment is equipped with a variety of IoT sensors, which can collect various data related to the processing process. For example, the temperature sensor may collect temperature data of the processing environment every 1 minute, with a collection range of -20°C to 100°C and an accuracy of ±0.5°C; the pressure sensor collects drilling pressure data at a frequency of 5 times per second during the board drilling process, and the pressure measurement range is 0 to 500N with an accuracy of ±1N. These IoT sensors continuously collect data and transmit it to the control system. The control system receives a large amount of raw data from different sensors, which constitute the target sensor monitoring data.

[0013] Furthermore, pyramid knowledge vector mining is a multi-level data mining method. It gradually abstracts data from the original low level to the high level, forming knowledge vector spectra of different depths. This mining method is similar to the structure of a pyramid, which gradually builds up knowledge representations with higher semantic information from a large amount of original data at the bottom.

[0014] For example, the target sensor monitoring data includes data in multiple dimensions such as temperature, pressure, vibration, etc. during the sheet metal processing process, with a total of 1,000 data points (this is just an example number).

[0015] First, at the bottom layer (first pyramid depth), separate feature extraction is performed on the data of each dimension. For example, for temperature data, its statistical features such as mean, variance, maximum and minimum are calculated; for pressure data, its average pressure change rate, pressure peak frequency and other features are calculated. These features are combined to form a vector, which constitutes part of the target sensing monitoring knowledge vector spectrum of the first pyramid depth.

[0016] Then, enter the mining of the second pyramid depth. Combine multiple related vectors of the first pyramid depth, for example, fuse the feature vectors related to temperature and pressure. A weighted summation method can be used. For example, the temperature feature vector is (T=(t_1, t_2, cdots, t_n)), the pressure feature vector is (P=(p_1, p_2, cdots, p_m)), and the elements of the fused vector (F) (f_i=w_1t_i+w_2p_i) (where (w_1) and (w_2) are pre-set weights, for example (w_1=0.6), (w_2=0.4)). In this way, the target sensing monitoring knowledge vector spectrum of the second pyramid depth is constructed.

[0017] As the depth of the pyramid increases, more relevant low-level vectors are continuously fused, and more complex fusion algorithms such as the principal component analysis (PCA) algorithm are used. At a higher pyramid depth, such as the third pyramid depth, multiple vectors of the second pyramid depth are subjected to PCA analysis. The PCA algorithm aims to find the main component directions in the data, project the data onto these main directions, reduce the dimensionality of the data while retaining the main information. For example, after PCA processing, a vector originally composed of 10 features is converted into a vector composed of 3 main component features, thereby obtaining a target sensing monitoring knowledge vector spectrum of the third pyramid depth.

[0018] Through such a step-by-step mining process, the control system obtains target sensing monitoring knowledge vector spectra of several pyramid depths.

[0019] Step 120: The control system uses a target sensor monitoring knowledge vector spectrum of a first pyramid depth among a number of target sensor monitoring knowledge vector spectra as a first target sensor monitoring knowledge vector spectrum, obtains a first equipment control decision feature of a first processing task based on the first target sensor monitoring knowledge vector spectrum, and obtains a second equipment control decision feature of a second processing task based on target sensor monitoring knowledge vector spectra other than the first target sensor monitoring knowledge vector spectrum; wherein the first processing task includes a linkage processing task.

[0020] In step 120, the control system uses the target sensor monitoring knowledge vector spectrum of the first pyramid depth among the plurality of target sensor monitoring knowledge vector spectrums as the first target sensor monitoring knowledge vector spectrum. For example, the knowledge vector spectrum of the first pyramid depth contains the most basic processing equipment state characteristics, such as the statistical characteristics of direct measurement values ​​such as the instantaneous temperature and pressure of each component of the equipment.

[0021] Further, taking the linkage processing task as an example, for example, the linkage processing task involves cutting and drilling of a plate at the same time, when acquiring the first equipment control decision feature from the first target sensor monitoring knowledge vector spectrum, a decision tree algorithm may be used.

[0022] The decision tree is constructed based on a large amount of historical processing data and corresponding control decisions. For example, historical data shows that when the temperature characteristic value of the cutting equipment is in a certain range (such as [30℃, 40℃]) and the pressure characteristic value of the drilling equipment is in another range (such as [100N, 150N]), in order to ensure the smooth progress of the linkage processing task, the cutting speed should be set to 1 meter per minute and the drilling depth should be increased by 0.5 cm each time.

[0023] According to the characteristic values ​​such as current temperature and pressure in the first target sensor monitoring knowledge vector spectrum, a search is performed in the decision tree to obtain the first equipment control decision features of the corresponding first processing task (linked processing task), such as the speed control value of the cutting equipment, the depth control value of the drilling equipment, etc.

[0024] Next, the second equipment control decision feature of the second processing task can be obtained based on the target sensor monitoring knowledge vector spectrum other than the first target sensor monitoring knowledge vector spectrum. For example, the knowledge vector spectrum of the second pyramid depth or higher pyramid depth contains more comprehensive information, such as the overall deformation trend of the plate during processing, the mutual influence relationship between different processing links, etc.

[0025] For example, the second processing task is to grind the surface of the plate. A neural network algorithm is used to obtain the second equipment control decision feature based on these knowledge vector spectra. A multi-layer perceptron (MLP) neural network is constructed. The number of input layer nodes is determined according to the dimension of the knowledge vector spectrum, for example, 20 nodes; the hidden layer is set to 2 layers, with 10 and 5 nodes in each layer respectively; the number of output layer nodes is 3, representing the speed, grinding pressure and grinding time of the grinding equipment respectively.

[0026] A large amount of historical processing data (e.g., 1,000 sets of data) is used to train the neural network, which includes knowledge vector spectrum data of different pyramid depths and the corresponding optimal control parameters of the grinding equipment. After the training is completed, the current target sensor monitoring knowledge vector spectrum other than the first target sensor monitoring knowledge vector spectrum is input into the neural network, and the neural network outputs the second equipment control decision features of the second processing task, such as the grinding equipment speed is 1,000 revolutions per minute, the grinding pressure is 50N, and the grinding time is 10 minutes.

[0027] Step 130: The control system combines the first device control decision feature and the second device control decision feature to remotely control the furniture board processing equipment.

[0028] Through step 130, the control system combines the obtained first device control decision feature with the second device control decision feature. For example, when comprehensively controlling furniture board processing equipment, parameters such as cutting speed and drilling depth in the first device control decision feature and grinding equipment parameters in the second device control decision feature need to be considered in coordination. If the cutting speed is fast, it may cause the surface roughness of the board to increase, so the grinding time and pressure need to be adjusted accordingly in the control decision of the grinding equipment. This combination of features is achieved through pre-set rules or mathematical models. For example, using a linear weighted method, assume that the control decision feature of the first device is (C_1=(c_{11}, c_{12}, cdots, c_{1n})), the control decision feature of the second device is (C_2=(c_{21}, c_{22}, cdots, c_{2m})), and the combined control decision feature is (C=(w_3C_1+w_4C_2)) (where (w_3) and (w_4) are weights, such as (w_3=0.4), (w_4=0.6)).

[0029] During the remote control process, the control system uses a suitable communication protocol to transmit the combined control decision features to the furniture board processing equipment to achieve remote control. For example, using the industrial Ethernet protocol (such as Profinet), the control system packages the control instructions into a data packet that complies with the Profinet protocol. The data packet contains the control parameters of the cutting equipment, drilling equipment, and grinding equipment. The data packet is transmitted to each corresponding processing equipment through the industrial Ethernet network in the workshop.

[0030] After the furniture board processing equipment receives the control command from the control system, each device operates according to the command. For example, the cutting equipment adjusts the cutting speed according to the command, the drilling equipment adjusts the drilling depth, and the grinding equipment adjusts the speed, pressure, time and other parameters. During the execution process, the equipment will continue to collect data through the IoT sensor and feed it back to the control system. The control system will continuously repeat the above steps 110-130 according to the new target sensor monitoring data to achieve continuous optimization control of the furniture board processing equipment to ensure processing quality and efficiency.

[0031] Through the implementation of the above steps 110 to 130, the present invention can effectively utilize the sensor monitoring data of the furniture panel processing equipment, and through pyramid knowledge vector mining, equipment control decision feature acquisition and remote control processing, realize intelligent and precise control of the furniture panel processing equipment, improve the quality and efficiency of furniture panel processing, and reduce the cost and error rate of manual intervention.

[0032] In the above steps 110 to 130, the target sensor monitoring data, the target sensor monitoring knowledge vector spectrum, the first device control decision feature and the second device control decision feature are the core technical features of the embodiments of the present invention. These core technical features are further explained below.

[0033] 1. The data sources and types of target sensor monitoring data are described as follows

[0034] 1) Sensor types and collected data examples

[0035] Temperature sensor: In furniture board processing equipment, temperature sensors are distributed in different key locations, such as near cutting tools and around motors of drilling equipment. These temperature sensors can collect temperature data of equipment components during processing. For example, when cutting tools cut boards at high speed, the temperature may rise due to frictional heat. The temperature sensor may collect data at a frequency of once per second, and the data range may be between -10℃ and 200℃, with an accuracy of ±0.3℃.

[0036] Pressure sensor: Pressure sensors play an important role in the processes of pressing and drilling the sheet metal. In the pressing process, the pressure sensor is installed under the pressing plate of the pressing equipment to measure the pressure applied to the sheet metal. Its range may be 0-1000N, with an accuracy of ±2N and a collection frequency of once every 0.5 seconds. In the drilling process, the pressure sensor is installed on the fixture of the drill bit to measure the axial pressure during drilling to determine the state of the drilling.

[0037] Vibration sensor: installed on the base and key transmission parts of the processing equipment. For example, when the saw blade of the cutting equipment is unbalanced or worn, it will cause the vibration of the equipment to change. The vibration sensor can collect the vibration acceleration data of the equipment in three axes (x, y, z), with a measurement range of ±5g (g is the acceleration of gravity), an accuracy of ±0.05g, and a collection frequency of 10 times per second.

[0038] Displacement sensor: used in the cutting size control of the plate and the positioning of some automated processing procedures. For example, when cutting the plate, the linear displacement sensor can accurately measure the displacement of the cutting tool relative to the plate. Its measurement range depends on the processing size of the equipment, which may be 0-5 meters, with an accuracy of ±0.1 mm and a collection frequency of 1 time every 0.1 seconds.

[0039] 2) Data preprocessing and organization

[0040] Data cleaning: Since the sensor may be affected by environmental interference or its own faults, the collected data may contain abnormal values. For example, when the temperature sensor is suddenly subjected to strong external electromagnetic interference, it may generate a temperature value far beyond the normal range. Through data cleaning algorithms, such as the statistical 3σ principle (for example, if the data follows a normal distribution, data beyond the mean ±3 times the standard deviation is considered an abnormal value), these abnormal values ​​can be removed or corrected.

[0041] Data standardization: Data collected by different sensors have different dimensions and numerical ranges. In order to facilitate subsequent data analysis and mining, the data needs to be standardized. For example, using the Z-score standardization method, for a data set collected by a sensor (X={x_1, x_2, cdots, x_n}), the standardized data set (Z={(z_1, z_2, cdots, z_n)}) is (z_i=frac{x_i-mu}{sigma}) ((mu) is the mean of the data set (X) and (sigma) is the standard deviation of the data set (X)).

[0042] Data organization form: The target sensor monitoring data after preprocessing is organized in the form of time series. For example, for each sensor, a data sequence is formed in the order of acquisition time. The data sequences of all sensors are combined to form a multi-dimensional time series data matrix. For example, there are (m) sensors in total, and each sensor collects (n) data points, then the dimension of this data matrix is ​​(mtimesn).

[0043] 2. Target Sensing Monitoring Knowledge Vector Spectrum

[0044] 1) The significance of the pyramid levels

[0045] The pyramid structure of the target sensor monitoring knowledge vector spectrum reflects the abstract process from raw data to high-level knowledge. At the first pyramid depth, the original sensor monitoring data is mainly subjected to exemplary feature extraction and combination to obtain a preliminary knowledge representation. For example, for temperature and pressure data, their basic statistical features such as mean and variance are calculated at the first pyramid depth, and these features are combined into a vector. This vector is the most basic knowledge representation, reflecting the basic state of the device at a certain moment.

[0046] As the depth of the pyramid increases, the abstraction level of knowledge continues to increase. At the second pyramid depth, multiple related vectors at the first pyramid depth are fused. This fusion is not just an exemplary combination, but is based on a deeper physical relationship or processing logic. For example, the vector reflecting the temperature and pressure of the cutting equipment is fused with the vector reflecting the deformation of the sheet to reflect the relationship between the equipment status and the sheet processing effect during the cutting process.

[0047] At a higher pyramid depth, such as the third pyramid depth, multiple vectors of the second pyramid depth will be further integrated to consider more global factors, such as the collaborative working status of the entire processing production line, energy transfer and material flow between different processing equipment, etc., so as to obtain a more comprehensive and semantic knowledge vector spectrum.

[0048] 2) Example of feature extraction and fusion method

[0049] Feature extraction of the first pyramid depth: For temperature data, in addition to calculating the mean and variance, the rate of change of temperature can also be calculated. For example, in a period of time from (t_1) to (t_2), the temperature data sequence is (T={T_{t1}, T_{t2}, cdots, T_{tn}}), and the temperature change rate is (r_T=frac{T_{tn}-T_{t1}}{t_2-t_1}). For pressure data, calculate the frequency of pressure peaks. By performing a peak detection algorithm on the pressure data, count the number of times the pressure peak occurs in a certain time window, and then divide it by the length of the time window to get the pressure peak frequency. Combine these temperature and pressure features into a vector, such as (V_1=(overline{T}, sigma_T, r_T, overline{P}, sigma_P, f_P)) (where (overline{T}) is the temperature mean, (sigma_T) is the temperature variance, (overline{P}) is the pressure mean, (sigma_P) is the pressure variance, and (f_P) is the pressure peak frequency), which is part of the target sensing monitoring knowledge vector spectrum of the first pyramid depth.

[0050] Feature fusion at the second pyramid depth: At the second pyramid depth, for example, to fuse the temperature-pressure vector (V_1) and the vibration vector (V_v=(overline{V}, sigma_V, f_V)) at the first pyramid depth (where (overline{V}) is the vibration mean, (sigma_V) is the vibration variance, and (f_V) is the vibration frequency). A fusion method based on the entropy weight method can be used. First, the information entropy of each feature is calculated. For example, for the temperature mean (overline{T}), its information entropy (H(overline{T})) is calculated according to the Shannon entropy formula (H(X)=-sum_{i=1}^{n}p(x_i)logp(x_i)) (where (p(x_i)) is the probability of the feature value (x_i) appearing in the data set). Then, the weight of each feature is calculated according to the information entropy, weight (w_i=frac{1-H(x_i)}{sum_{j=1}^{m}(1-H(x_j))})) (where (m) is the total number of features). Finally, the features in (V_1) and (V_v) are weighted and summed according to the weights to obtain the fusion vector in the target sensing monitoring knowledge vector spectrum of the second pyramid depth.

[0051] 3. First Equipment Control Decision Features

[0052] 1) Association with linkage processing tasks

[0053] Equipment coordination in linked processing tasks: In linked processing tasks, such as the process of cutting and milling a plate at the same time, there is a close coordination relationship between the equipment. The cutting speed of the cutting equipment will affect the processing allowance and processing accuracy of the milling equipment. If the cutting speed is too fast, it may cause the plate cutting surface to be uneven, which in turn affects the milling accuracy of the milling equipment on the plate edge. At the same time, the milling force of the milling equipment will also affect the stability of the cutting equipment.

[0054] Generation of decision features based on the first target sensor monitoring knowledge vector spectrum: The first equipment control decision feature is generated based on the first target sensor monitoring knowledge vector spectrum, which reflects the basic status information of the equipment. For example, in a linkage processing task, if the temperature characteristic value of the cutting equipment in the first target sensor monitoring knowledge vector spectrum is high, it means that the cutting equipment may be in an overloaded or tool-worn state. According to the pre-established control strategy model, control decision features such as reducing the cutting speed and increasing the tool penetration depth of the milling equipment (to share part of the processing load of the cutting equipment) may be obtained.

[0055] For example, a fuzzy logic control method is used to generate the first equipment control decision feature. The temperature, pressure, vibration and other characteristics of the cutting equipment are used as fuzzy input variables, and fuzzy sets are defined, such as the fuzzy set of temperature is "low", "medium", "high", and the fuzzy set of pressure is "small", "medium", "large", etc. A fuzzy rule base is established based on expert experience and historical data. For example, "If the temperature is high and the pressure is large, the cutting speed is reduced to 80% of the original, and the cutting depth of the milling equipment is increased by 0.5 mm." Through fuzzy reasoning and defuzzification processes, specific first equipment control decision features are obtained, such as the specific value of the cutting speed, the value of the cutting depth of the milling equipment, etc.

[0056] 2) Parametric representation of control decision features

[0057] Parameter meaning and value range: The first device control decision feature is expressed in a parameterized form. These parameters directly correspond to the control instructions of each device in the linkage processing task. For example, for cutting equipment, the control decision features may include cutting speed (v_c) (value range is 0-5 m / s), cutting tool feed rate (f_c) (value range is 0-1 mm / time), etc. For milling equipment, it may include milling speed (v_m) (value range is 1000-5000 rpm), milling tool cutting depth (d_m) (value range is 0-5 mm), etc. The values ​​of these parameters are obtained according to the data in the first target sensor monitoring knowledge vector spectrum through the above decision-making method, and must meet the working range and processing technology requirements of the equipment.

[0058] 4. Second device control decision characteristics

[0059] 1) Compatibility with the second processing task

[0060] Characteristics of the second processing task: For example, the second processing task is the surface polishing process of the plate. This process has different requirements from the previous linkage processing tasks (cutting and milling). Surface polishing needs to consider factors such as the roughness and hardness of the plate surface and the processing marks left by the previous processing steps. In the polishing process, the speed, polishing pressure and polishing time of the polishing equipment are key control parameters.

[0061] Decision basis based on target sensor monitoring knowledge vector spectrum: The second equipment control decision feature is obtained based on the target sensor monitoring knowledge vector spectrum in addition to the first target sensor monitoring knowledge vector spectrum. These knowledge vector spectra contain more information about the overall state of the plate and the impact of previous processing steps. For example, the roughness estimate of the plate surface (obtained by analyzing the relationship between data such as vibration and pressure and the surface roughness of the plate) and the hardness information of the plate (inferred from the force and displacement data during cutting and milling) are obtained from the knowledge vector spectrum with a higher pyramid depth.

[0062] Based on this information, a neural network algorithm is used to generate the second equipment control decision features. For example, a convolutional neural network (CNN) is constructed, and the input layer receives features related to the surface state of the plate extracted from the knowledge vector spectrum, such as roughness estimation, hardness value, and some feature vectors reflecting the influence of previous processing procedures. After being processed by the convolution layer, pooling layer, and fully connected layer, the output layer outputs the control parameters of the polishing equipment, such as the speed of the polishing equipment (v_p) (the range is 500-3000 rpm), polishing pressure (p_p) (the range is 10-100N), and polishing time (t_p) (the range is 1-10 minutes).

[0063] 2) Impact of decision-making characteristics on processing quality

[0064] Processing quality indicators and control relationship: In the plate surface polishing process, the processing quality indicators mainly include surface roughness (R_a) (in microns) and surface gloss. The control decision characteristics such as the speed, pressure and time of the polishing equipment directly affect these processing quality indicators. For example, in general, a higher polishing speed and appropriate polishing pressure can effectively reduce the surface roughness and improve the surface gloss in a shorter polishing time, but if the speed is too high or the pressure is too high, it may cause scratches on the plate surface, affecting the processing quality.

[0065] By continuously adjusting the second equipment control decision characteristics and combining the quality inspection data after actual processing (such as data measured by a roughness meter and a gloss meter), decision models such as neural networks are optimized to ensure that the second equipment control decision characteristics can enable the polishing process to achieve the best processing quality effect.

[0066] The embodiment of the present invention collects data of furniture board processing equipment through IoT sensors, and forms a multi-depth knowledge vector spectrum based on pyramid knowledge vector mining. This method can comprehensively and deeply mine the value of data. The equipment control decision features of different processing tasks are obtained from the knowledge vector spectra of different pyramid depths, which can accurately adapt to the needs of linkage processing tasks and other processing tasks. For linkage processing tasks, the decision features obtained from the first pyramid depth can effectively coordinate the relationship between equipment, avoid adverse effects between equipment, and improve overall processing efficiency. The decision features of the second processing task are obtained from the knowledge vector spectra of other depths, taking into account more complex processing influencing factors, and improving processing quality. Finally, remote control is achieved by combining the two decision features, reducing the cost of manual intervention, and improving the timeliness and accuracy of control, thereby achieving efficient, high-quality and intelligent furniture board processing.

[0067] It is worth mentioning that the specific implementation method of the above-mentioned technical solution of remote control processing combined with device control decision characteristics can also be as follows.

[0068] 1. Overall architecture of remote control processing

[0069] System composition and connection mode: In the control system of intelligent furniture board processing equipment, there are main parts such as control system, furniture board processing equipment and remote monitoring platform. The control system is connected to the furniture board processing equipment through the Internet of Things, which can obtain the data collected by the IoT sensors on the equipment and control the equipment. At the same time, a communication connection is also established between the control system and the remote monitoring platform to realize remote monitoring and control operations. For example, the control system and the processing equipment can be connected using industrial Ethernet (such as Ethernet / IP) or wireless communication protocols (such as Wi-Fi, ZigBee, etc.). Industrial Ethernet provides high-speed and stable data transmission, which is suitable for the transmission of large amounts of data, such as real-time operating status data of processing equipment, equipment control decision characteristics, etc. Wi-Fi facilitates the access and networking of equipment, and ZigBee is suitable for low-power, low-rate sensor networks, which can be used to transmit some status information or control instructions of the equipment. The control system and the remote monitoring platform can be connected through the Internet, and a secure communication protocol (such as HTTP protocol encrypted by SSL / TLS) is used to ensure the security of data transmission.

[0070] Data interaction and processing flow: When the control system performs pyramid knowledge vector mining based on the target sensor monitoring data, obtains several pyramid-depth target sensor monitoring knowledge vector spectra, and further obtains the first device control decision feature and the second device control decision feature, these feature data will be used for remote control processing. First, the control system integrates and analyzes these feature data to ensure the accuracy and integrity of the data. For example, the rationality of the linkage processing task related parameters in the first device control decision feature (such as the cutting speed of the cutting device, the drilling depth of the drilling device, etc.) and the subsequent processing task parameters in the second device control decision feature (such as the speed and pressure of the plate surface polishing, etc.) are checked to exclude abnormal values ​​that may be generated during data collection or mining. Then, the control system packages the processed first device control decision feature and second device control decision feature data and encapsulates them according to the predetermined communication protocol format. For example, the data is encapsulated into a message format that conforms to the MQTT (MessageQueuingTelemetryTransport) protocol, which contains information such as device identification, control decision feature data, and timestamp. MQTT is a lightweight publish / subscribe message transmission protocol suitable for communication between IoT devices. Its low bandwidth occupancy and reliable message delivery mechanism can meet the needs of remote control of furniture panel processing equipment.

[0071] 2. Remote control operation based on the control decision feature of the first device

[0072] Remote control of linkage processing tasks: In linkage processing tasks, the first device control decision feature plays a key role. For example, when the first device control decision feature indicates that the cutting speed of the cutting device needs to be adjusted, the control system sends a control instruction to the cutting device through a remote communication link. Assuming that the current cutting speed of the cutting device is (v_1) (unit: m / s), and the cutting speed adjustment value in the first device control decision feature is (Deltav), then the new cutting speed (v_2=v_1+Deltav). The control system sends the control instruction containing the (v_2) value to the cutting device. The same is true for the drilling device. If the drilling depth in the first device control decision feature needs to be changed, assuming that the current drilling depth is (d_1) (unit: mm), the adjustment value is (Deltad), and the new drilling depth (d_2=d_1+Deltad), the control system sends the corresponding control instruction to the drilling device. This remote control based on the first device control decision feature can adjust the operating parameters of each device in the linkage processing task in real time to ensure the coordination and efficiency of linkage processing.

[0073] Application innovation of algorithms and models: In the process of determining the control decision characteristics of the first equipment, innovative data mining methods such as pyramid knowledge vector mining and multi-round knowledge vector mining processing are used. These methods can mine information closely related to the linkage processing task from complex target sensor monitoring data, thereby obtaining accurate equipment control decision characteristics. For example, in the multi-round knowledge vector mining processing, by fusing and analyzing the knowledge vector spectra of different rounds, the feature representation of data at different levels is considered. Compared with the traditional method of making control decisions based only on raw data, it can more comprehensively reflect the actual operating status and processing requirements of the equipment. In addition, when remote control is performed based on the control decision characteristics of the first equipment, an adaptive control algorithm is used. For example, when the tool of the cutting equipment is worn, the data collected by the sensor is reflected in the control decision characteristics of the first equipment, and the control system can adaptively adjust the cutting speed according to the pre-established tool wear-cutting speed relationship model (such as a linear or nonlinear model established through a large amount of experimental data). Let the tool wear amount be (w) (unit: mm), and the cutting speed adjustment model be (v=f(w)), where (f) is the functional relationship obtained by fitting the experimental data. This model-based adaptive control is an innovative remote control method that can improve the processing quality of coordinated processing tasks and the service life of equipment.

[0074] 3. Remote control operation based on the control decision feature of the second device

[0075] Remote control of subsequent processing tasks: The second device control decision feature is mainly used for subsequent processing tasks, such as surface treatment of plates (polishing, grinding, etc.). Taking the polishing task as an example, when the speed of the polishing device in the second device control decision feature needs to be adjusted, assuming that the current speed of the polishing device is (n_1) (unit: revolutions per minute), and the speed adjustment value in the second device control decision feature is (Deltan), then the new speed (n_2=n_1+Deltan). The control system sends the control instruction containing the (n_2) value to the polishing device. The adjustment of the polishing pressure is also similar. If the current polishing pressure is (p_1) (unit: Newton), the adjustment value is (Deltap), and the new polishing pressure (p_2=p_1+Deltap), the control system sends the corresponding control instruction to the polishing device. This remote control based on the second device control decision feature can ensure that subsequent processing tasks are carried out according to the predetermined processing requirements and improve the processing quality of the plate.

[0076] In the process of obtaining the second device control decision feature, a unique knowledge vector mining and derivative knowledge vector mining method is used. For example, when the derivative knowledge vector mining is performed based on the first sensor monitoring knowledge vector spectrum of several pyramid depths to obtain the second sensor monitoring knowledge vector spectrum of several pyramid depths, advanced mathematical tools such as wavelet transform are used. This method can mine deeper feature information in the data, which is more helpful to accurately determine the second device control decision feature compared with the traditional simple data processing method.

[0077] When remote control is performed based on the control decision feature of the second device, an intelligent optimization algorithm can also be used. For example, in order to minimize the polishing time while meeting the surface roughness requirements of the plate, a genetic algorithm can be used to optimize the control parameters of the polishing equipment. Suppose the objective function is (T=g(n,p)) (where (T) is the polishing time, (n) is the rotation speed, (p) is the pressure, and (g) is a functional relationship established based on experiments), and the genetic algorithm is used to search for the optimal (n) and (p) values ​​in the feasible parameter space, and this optimal value is used as the control decision feature of the second device for remote control. The application of this intelligent optimization algorithm is the innovation of the embodiment of the present invention in remotely controlling subsequent processing tasks, which can improve processing efficiency and resource utilization.

[0078] (IV) Collaborative remote control of the first device control decision feature and the second device control decision feature

[0079] Overall coordination of the processing flow: During the entire furniture board processing process, the first equipment control decision feature and the second equipment control decision feature need to work together for remote control. For example, the cutting and drilling operations in the linkage processing task will affect the surface quality and dimensional accuracy of the board, and the subsequent polishing operation needs to be based on the state of the board after the previous processing. If the cutting speed is too fast, it may cause the surface roughness of the board to increase. In this case, the speed and pressure of the polishing equipment need to be adjusted accordingly in the second equipment control decision feature. Assume that there is a relationship between the cutting speed (v) and the polishing equipment speed (n) and pressure (p) (n=h_1(v)), (p=h_2(v)) (where (h_1) and (h_2) are functional relationships established based on the processing technology and experimental data). When the cutting speed in the first equipment control decision feature changes, the control system will adjust the polishing equipment speed and pressure in the second equipment control decision feature according to these functional relationships, and then send the corresponding control instructions to the equipment to achieve coordinated control of the entire processing flow.

[0080] It can be seen that this collaborative remote control mechanism is the innovation of the embodiment of the present invention. Traditional furniture board processing equipment control often controls each processing task in isolation, lacking overall consideration of the entire processing flow. The embodiment of the present invention achieves collaborative control by establishing a correlation between the control decision feature of the first device and the control decision feature of the second device, using a method based on data mining and model building.

[0081] For example, when determining this association relationship, by analyzing a large amount of historical processing data, a data mining algorithm (such as an association rule mining algorithm) is used to find the implicit relationship between the parameters of the linkage processing task and the parameters of the subsequent processing task. Assume that the mined association rule is: if the cutting speed of the cutting equipment is within a certain interval ([v_{min}, v_{max}]), and the drilling depth of the drilling equipment is within the interval ([d_{min}, d_{max}]), then the speed of the polishing equipment should be within the interval ([n_{min}, n_{max}]), and the pressure should be within the interval ([p_{min}, p_{max}]). This association rule establishment based on data mining provides a basis for collaborative remote control, enabling the entire furniture board processing process to be carried out more efficiently and stably, improving product quality and production efficiency.

[0082] Based on steps 110 to 130, the method also includes: if the first device control decision feature and the second device control decision feature meet the set device control decision conditions, the target sensor monitoring data, the first device control decision feature and the second device control decision feature are reported to the remote monitoring middle-office system, so that the remote monitoring middle-office system performs automatic production line scheduling based on the received target sensor monitoring data and the first device control decision feature and the second device control decision feature of each target sensor monitoring data.

[0083] In the embodiment of the present invention, setting the device control decision condition is the basis for determining whether to report the target sensor monitoring data, the first device control decision feature and the second device control decision feature to the remote monitoring middle station system. This condition may be a comprehensive logical judgment, including multiple considerations.

[0084] For example, from the perspective of the stability of the equipment operation status, for the key parameters in the first equipment control decision feature (such as the cutting speed (v_c) of the cutting equipment in the linkage processing task, the drilling pressure (p_d) of the drilling equipment) and the key parameters in the second equipment control decision feature (such as the speed (v_p) and pressure (p_p) of the surface polishing equipment), the fluctuation range is set not to exceed a certain threshold. For example, the fluctuation threshold of the cutting speed (v_c) is ±0.1 m / s, the fluctuation threshold of the drilling pressure (p_d) is ±5N, the fluctuation threshold of the polishing equipment speed (v_p) is ±50 rpm, and the fluctuation threshold of the pressure (p_p) is ±2N.

[0085] At the same time, from the perspective of processing quality indicators, the surface roughness (R_a) of the plate after processing must be less than a certain value (such as (R_a<5) microns), and the dimensional accuracy error of the plate must be within ±0.5 mm, etc. These conditions are set based on a large amount of experimental data and actual production experience to ensure the normal operation of the equipment and product quality.

[0086] Furthermore, with regard to the condition setting example based on statistical analysis, statistical analysis methods can be used to determine these thresholds. For example, historical processing data is analyzed to collect the distribution of equipment operating parameters under normal production conditions. For example, the historical data of cutting speed (v_c) is fitted with a normal distribution to obtain its mean (mu_{v_c}) and standard deviation (sigma_{v_c}). According to the 3σ principle, (mu_{v_c}pm0.1sigma_{v_c}) is used as the preliminary range of the fluctuation threshold, and then fine-tuned according to the acceptable fluctuation range in actual production, and finally determined to be ±0.1 m / s.

[0087] It can be understood that when the first device control decision feature and the second device control decision feature meet the set device control decision conditions, the control system begins to report the target sensor monitoring data, the first device control decision feature and the second device control decision feature to the remote monitoring middle platform system. First, the target sensor monitoring data is sorted and arranged in order of sensor type and acquisition time. For example, the temperature sensor data is arranged in chronological order into a series (T={t_1, t_2, cdots, t_n}), the pressure sensor data is arranged as (P={p_1, p_2, cdots, p_m}), etc.

[0088] For the first device control decision feature and the second device control decision feature, their parameters are organized in a structured form. For example, the first device control decision feature can be represented as a structure (C_1={v_c, p_d, cdots}), where (v_c) is the cutting speed, (p_d) is the drilling pressure, etc.; the second device control decision feature is represented as (C_2={v_p, p_p, cdots}). Then these data are packaged into a data packet, and the format of the data packet can be encapsulated in a custom binary format or based on a general data format such as XML, JSON, etc.

[0089] In the process of data reporting, select a suitable communication protocol to ensure reliable data transmission. For example, the MQTT (Message Queuing Telemetry Transport) protocol is used. MQTT is a lightweight publish / subscribe message transmission protocol suitable for communication between IoT devices and servers. The control system, as a publisher, publishes the packaged data to a specific topic, and the remote monitoring middle system, as a subscriber, subscribes to the topic to receive data.

[0090] The QoS (Quality of Service) mechanism in the MQTT protocol ensures the reliability of data transmission. There are three levels of QoS: 0, 1, and 2. For the data reporting of the embodiment of the present invention, if QoS=1 is selected, it means that the message is transmitted at least once, there may be duplicate messages, but it is ensured that the message will not be lost. During the transmission process, the control system sends the data packet to the MQTT proxy server, and the proxy server forwards the data to the remote monitoring middle system according to the subscription relationship.

[0091] This embodiment also involves the automated production line scheduling processing of the remote monitoring middle-office system. After the remote monitoring middle-office system receives the data reported by the control system, it first performs data parsing. If the data is encapsulated in JSON format, for example, a JSON data containing target sensor monitoring data, first device control decision characteristics and second device control decision characteristics may be as follows: `{"temperature":[25,26,27],"pressure":[100,105,110],"first_control":{"cutting_speed":2.5,"drilling_pressure":150},"second_control":{"polishing_speed":1500,"polishing_pressure":30}}`. The remote monitoring middle-office system uses the JSON parser to parse it into an internal data structure for subsequent processing.

[0092] The remote monitoring middle station system performs automatic production line scheduling based on the received target sensor monitoring data and the first equipment control decision feature and the second equipment control decision feature of each target sensor monitoring data. For example, a rule-based expert system is used for scheduling decision.

[0093] The rule base may include the following rules: If the operating temperature of multiple devices (such as cutting equipment and drilling equipment) continues to rise (based on the temperature data in the target sensor monitoring data, for example, the temperature data collected for three consecutive times is more than 2°C higher than the previous temperature), and the cutting speed and drilling pressure in the control decision characteristics of the first device are close to the upper limit (such as the cutting speed reaches the upper limit of 4.5 m / s - 0.1 m / s, and the drilling pressure reaches the upper limit of 450 N - 5 N), then adjust the processing sequence of the production line, suspend the operation of the two devices first, start the cooling equipment to cool them, and postpone the subsequent surface polishing process to avoid processing when the equipment is in poor condition.

[0094] This rule can be expressed by a logical expression: Let (T_{cut}) be the temperature sequence of the cutting equipment, (T_{drill}) be the temperature sequence of the drilling equipment, (v_{cut}) be the cutting speed, (p_{drill}) be the drilling pressure, (T_{cut}[i]) represents the temperature collected by the cutting equipment for the (i)th time, and (T_{drill}[i]) is similar. The rule is: ((for alliin{1, 2, 3}, T_{cut}[i]>T_{cut}[i-1]+2landT_{drill}[i]>T_{drill}[i-1]+2)land(v_{cut}geq4.4landp_{drill}geq445)) → "Pause the cutting and drilling equipment, start the cooling equipment, and postpone the polishing process".

[0095] In addition to rule-based scheduling, optimization algorithms can also be used to further optimize production line scheduling. For example, genetic algorithms can be used to optimize the scheduling sequence of equipment to improve production efficiency.

[0096] First, the parameters such as the start-up time and processing time of each device in the production line (such as cutting equipment, drilling equipment, polishing equipment, etc.) are encoded as chromosomes. For example, the chromosome is represented as (X=(x_1, x_2, cdots, x_n)), where (x_i) represents the relevant parameters of a certain device. The fitness function (f(X)) can be defined as the number of qualified products completed within a certain period of time.

[0097] Then, through the selection, crossover and mutation operations of the genetic algorithm, the population is continuously evolved to find the chromosome with the highest fitness, that is, the best equipment scheduling solution. For example, the initial population has 100 chromosomes. After 100 generations of evolution, the chromosome with the highest fitness is obtained. The best scheduling order and time arrangement of the equipment are determined based on this chromosome.

[0098] With this design, first of all, by setting equipment control decision conditions and reporting data when the conditions are met, valuable data can be screened out for centralized management and analysis, avoiding unnecessary data transmission and improving the efficiency of data transmission and processing. For example, by setting the fluctuation threshold of equipment operating parameters, data is reported only when the equipment operating status may affect production. Secondly, the remote monitoring middle-end system performs automated production line scheduling based on comprehensive data, which improves the intelligence of the entire production process. Rule-based scheduling decisions can quickly respond to changes in equipment status and processing conditions, such as adjusting the processing sequence according to equipment temperature and operating parameters. At the same time, the application of optimization algorithms further optimizes production line scheduling and improves production efficiency. For example, genetic algorithms are used to find the best equipment scheduling sequence, so that more qualified products can be produced within a certain period of time.

[0099] In a replaceable embodiment, pyramid knowledge vector mining is performed based on the target sensor monitoring data to obtain target sensor monitoring knowledge vector spectra of several pyramid depths, including: performing multiple rounds of knowledge vector mining processing based on the target sensor monitoring data to obtain first sensor monitoring knowledge vector spectra of several pyramid depths; performing pyramid knowledge vector mining based on the first sensor monitoring knowledge vector spectra of several pyramid depths to obtain target sensor monitoring knowledge vector spectra of several pyramid depths.

[0100] Among them, each round of knowledge vector mining processing includes the following contents: pyramid knowledge vector mining is performed based on the first sensor monitoring data to obtain a first implicit sensor monitoring knowledge vector spectrum; wherein, the first sensor monitoring data is the sensor monitoring data of the attention index set in the sensor monitoring data to be processed corresponding to the current round of pyramid knowledge vector mining; the sensor monitoring data to be processed corresponding to the first pyramid knowledge vector mining is the target sensor monitoring data; the first implicit sensor monitoring knowledge vector spectrum and the second implicit sensor monitoring knowledge vector spectrum are fused with attention index to obtain a third implicit sensor monitoring knowledge vector spectrum; the second implicit sensor monitoring knowledge vector spectrum is the sensor monitoring data in the sensor monitoring data to be processed except the first sensor monitoring data; pooling knowledge vector mining is performed on the third implicit sensor monitoring knowledge vector spectrum to obtain an original sensor monitoring knowledge vector spectrum; wherein the original sensor monitoring knowledge vector spectrum is the sensor monitoring data to be processed corresponding to the next round of pooling knowledge vector mining; the first sensor monitoring knowledge vector spectrum of several pyramid depths includes the original sensor monitoring knowledge vector spectrum obtained by at least two rounds of knowledge vector mining in multiple rounds of knowledge vector mining processing.

[0101] In the above embodiment, the overall technical idea of ​​multi-round knowledge vector mining processing is as follows: This multi-round knowledge vector mining processing method is intended to gradually and more deeply mine knowledge vector spectra with different levels and semantic information from the target sensor monitoring data. Through multiple rounds of processing, each round performs specific operations on the data, and gradually constructs a pyramid-structured knowledge vector spectrum, so that the final target sensor monitoring knowledge vector spectrum can more comprehensively and accurately reflect the status of the furniture board processing equipment and related information of the processing process. Its principle is based on the staged processing of data, each round processes the data according to different strategies, and the processing of the next round is based on the results of the previous round, thereby realizing the gradual abstraction from raw data to high-level knowledge.

[0102] Furthermore, the specific content of each round of knowledge vector mining processing includes the following three steps.

[0103] 1) Based on the first sensor monitoring data, pyramid knowledge vector mining is performed to obtain the first implicit sensor monitoring knowledge vector spectrum

[0104] Data selection: In each round of knowledge vector mining processing, the first sensor monitoring data must be determined first. The first sensor monitoring data is the sensor monitoring data with a set attention index in the sensor monitoring data to be processed corresponding to the current round of pyramid knowledge vector mining. For example, the set ratio is 30%. For example, the sensor monitoring data to be processed in the current round is a data set (D) containing 1,000 data points (the data points here can be a combination of measurement values ​​from different sensors). According to the attention index (this index can be based on the importance assessment of the data, such as data related to the key operating parameters of the equipment has a higher attention index), 300 data points are selected from (D) as the first sensor monitoring data (D_1).

[0105] Pyramid knowledge vector mining: For this part of data (D_1), pyramid knowledge vector mining is performed. Taking an exemplary two-layer pyramid structure as an example, in the first layer, basic statistical features such as mean (mu_i) and variance (sigma_i) ((i=1, 2, cdots, n)) are calculated for each data dimension in (D_1) (for example, (D_1) has (n) dimensions, such as different sensor data dimensions such as temperature and pressure). These features form a vector (V_1=(mu_1, sigma_1, mu_2, sigma_2, cdots, mu_n, sigma_n)), which is the first implicit sensor monitoring knowledge vector spectrum of the first pyramid depth.

[0106] At the second pyramid depth, the relevant features in the vector (V_1) of the first layer are fused. For example, a weighted summation method is used, such as fusing the features related to temperature and pressure, the temperature feature vector is (T=(t_1, t_2)) (such as (t_1=mu_{temperature}), (t_2=sigma_{temperature})), the pressure feature vector is (P=(p_1, p_2)), and the elements of the fused vector (F) are (f_i=w_1t_i+w_2p_i) (where (w_1=0.6), (w_2=0.4)). In this way, the first implicit sensor monitoring knowledge vector spectrum of the second pyramid depth is constructed.

[0107] 2) The first implicit sensor monitoring knowledge vector spectrum and the second implicit sensor monitoring knowledge vector spectrum are fused with attention indicators to obtain the third implicit sensor monitoring knowledge vector spectrum

[0108] Determination of the second implicit sensor monitoring knowledge vector spectrum: The second implicit sensor monitoring knowledge vector spectrum is the sensor monitoring data other than the first sensor monitoring data in the sensor monitoring data to be processed. Continuing with the above example, the 700 data points other than (D_1) (300 data points) in the sensor monitoring data to be processed (D) constitute the second sensor monitoring data (D_2). A similar pyramid knowledge vector mining is performed on (D_2) to obtain the second implicit sensor monitoring knowledge vector spectrum (V_2).

[0109] Attention index fusion: Attention index fusion is to fuse (V_1) and (V_2) according to the attention index of their data. For example, the attention index is a weight vector related to the degree of influence of the data on the operating status of the device. For each element in (V_1) and (V_2), a weighted sum is performed according to its corresponding attention index weight. Suppose (V_1=(v_{11}, v_{12}, cdots, v_{1m})), its attention index weight vector is (W_1=(w_{11}, w_{12}, cdots, w_{1m})); (V_2=(v_{21}, v_{22}, cdots, v_{2m})), its attention index weight vector is (W_2=(w_{21}, w_{22}, cdots, w_{2m})). Then the elements of the third implicit sensor monitoring knowledge vector spectrum (V_3) (v_{3i}=w_{1i}v_{1i}+w_{2i}v_{2i})((i=1,2,cdots,m)). This fusion method can comprehensively consider the knowledge vector spectrum obtained by mining different parts of the data and make reasonable combinations according to the importance of the data.

[0110] 3) Perform pooled knowledge vector mining on the third implicit sensor monitoring knowledge vector spectrum to obtain the original sensor monitoring knowledge vector spectrum.

[0111] Principle of pooling knowledge vector mining: Pooling is an operation of data dimensionality reduction. When mining the pooling knowledge vector of the third implicit sensor monitoring knowledge vector spectrum (V_3), for example, the maximum pooling operation is used. For example, (V_3) is a two-dimensional vector matrix (which can be regarded as a matrix composed of multiple feature vectors) with a dimension of (ptimesq). In the maximum pooling operation, (V_3) is divided into several small areas (for example, a small area of ​​(2times2)), and the maximum value is selected in each small area as the result after pooling. Let (V_3=begin{bmatrix}v_{311}&v_{312}v_{321}&v_{322}end{bmatrix}), and the result after maximum pooling is (begin{bmatrix}max(v_{311},v_{312})max(v_{321},v_{322})end{bmatrix}). In this way, pooled knowledge vector mining is performed on (V_3) to obtain the original sensor monitoring knowledge vector spectrum. This original sensor monitoring knowledge vector spectrum will be used as part of the sensor monitoring data to be processed in the next round of knowledge vector mining processing, thereby realizing the cycle and progression of multiple rounds of knowledge vector mining processing.

[0112] On the basis of the above content, pyramid knowledge vector mining can be performed based on the first sensor monitoring knowledge vector spectra of several pyramid depths to obtain target sensor monitoring knowledge vector spectra of several pyramid depths.

[0113] After completing multiple rounds of knowledge vector mining to obtain several pyramid-depth first sensor monitoring knowledge vector spectra, these first sensor monitoring knowledge vector spectra are used as the basis for pyramid knowledge vector mining again. This process is similar to the pyramid knowledge vector mining in each round of knowledge vector mining, but the data at this time is the first sensor monitoring knowledge vector spectrum after multiple rounds of screening and processing.

[0114] For example, after multiple rounds of knowledge vector mining, three first sensor monitoring knowledge vector spectra (V_{f1}), (V_{f2}), and (V_{f3}) with different pyramid depths are obtained (corresponding to the original sensor monitoring knowledge vector spectra obtained in different rounds of mining). When pyramid knowledge vector mining is performed to construct the target sensor monitoring knowledge vector spectrum, first, at the first pyramid depth, the corresponding features in (V_{f1}), (V_{f2}), and (V_{f3}) are combined. For example, the elements representing the temperature mean in the three vectors are combined into a new temperature mean vector, and the elements of the pressure variance are combined into a new pressure variance vector, etc.

[0115] At higher pyramid depths, such as the second pyramid depth, more complex fusion algorithms are used, such as the principal component analysis (PCA) algorithm. The vectors combined at the first pyramid depth are used as input, and the PCA algorithm is used to find the main component directions in the data, and the data is projected onto these main directions, reducing the dimension of the data while retaining the main information. For example, after PCA processing, the vector originally composed of 10 features is converted into a vector composed of 3 main component features, thereby obtaining the target sensing monitoring knowledge vector spectrum of the second pyramid depth. Continuing this process, target sensing monitoring knowledge vector spectra of multiple pyramid depths can be constructed.

[0116] It can be seen that through multiple rounds of knowledge vector mining processing, data information can be mined step by step and in depth. For example, in each round, data is selected for mining according to the set proportion of attention indicators, making the mining process more targeted. When fusing the knowledge vector spectra of different parts of the data, attention indicator fusion can reasonably combine information according to the importance of the data. Pooled knowledge vector mining reduces the data dimension while retaining key information, thereby improving data processing efficiency. After multiple rounds of mining, pyramid knowledge vector mining is performed again to construct the target sensor monitoring knowledge vector spectrum, which can more comprehensively and accurately extract different levels of knowledge from the target sensor monitoring data, which helps to better perform subsequent equipment control decisions and monitoring management operations.

[0117] In some examples, at least some of the attention indicators in the first sensory monitoring data based on two consecutive rounds of pyramid knowledge vector mining have different distribution labels of attention indicators in the corresponding sensory monitoring data to be processed.

[0118] In this example, the attention index is a quantitative index used in this technical solution to measure the importance or relevance of sensor monitoring data. It reflects the importance of data for mining the knowledge vector spectrum related to furniture board processing equipment. For example, when monitoring furniture board processing equipment, data directly related to the core components of the equipment (such as the temperature of the cutting tool, the pressure of the drilling equipment, etc.) may be given a higher attention index because these data have a key impact on the operating status, processing quality and efficiency of the equipment.

[0119] The determination of the attention index can be based on a variety of factors, such as the correlation between sensor data and equipment failure modes, sensitivity to processing accuracy, etc. For example, through the analysis of historical data, if the correlation between the change of a certain sensor data and the occurrence of equipment failure is as high as 80%, then the attention index of the data may be set to a higher value, such as 0.8 (for example, the attention index ranges from 0 to 1).

[0120] In addition, the distribution label is an identifier used to describe the distribution characteristics of the attention index in the sensor monitoring data to be processed. It can reflect the distribution pattern of the attention index of different data parts (such as data collected by different sensors or data collected by the same sensor in different time periods). For example, in a round of pyramid knowledge vector mining, the sensor monitoring data to be processed contains data from temperature sensors, pressure sensors, and vibration sensors. The distribution label can be expressed as a distribution pattern of the attention indicators of these sensor data, such as the attention indicator of the temperature sensor data presents a normal distribution, the attention indicator of the pressure sensor data presents a uniform distribution, and the attention indicator of the vibration sensor data presents an exponential distribution.

[0121] Based on this, in two consecutive rounds of pyramid knowledge vector mining, at least part of the attention indicators in the first sensor monitoring data have different distribution labels in the corresponding attention indicators in the sensor monitoring data to be processed. This means that when selecting the first sensor monitoring data in each round, the selection basis will change.

[0122] For example, in the first round of pyramid knowledge vector mining, the sensor monitoring data to be processed is the overall equipment monitoring data, for example, there are 1,000 data points in total. The distribution label of its attention index is a normal distribution, that is, the attention index of most data is concentrated near the middle value, and a few data have higher or lower attention indexes. According to this distribution label, when selecting the first sensor monitoring data, the part of the data with a higher attention index may be selected, such as selecting 30% of the data points with a higher attention index (ie, 300 data points). These data points may mainly come from sensors related to the current critical operating status of the equipment (such as the temperature and pressure data of the cutting tool in the ongoing cutting process).

[0123] In the second round of pyramid knowledge vector mining, the sensor monitoring data to be processed is the original sensor monitoring knowledge vector spectrum obtained after the first round of mining or the related data after some transformation. The distribution label of the attention index in this round becomes uniformly distributed, that is, the attention index of each data point is relatively average. At this time, when selecting the first sensor monitoring data, different data points or data combinations may be selected according to the new uniform distribution label. For example, according to the new distribution label, a data combination of every several data points is selected as the first sensor monitoring data, which is different from the way of selecting data points with higher attention indicators according to the normal distribution label in the first round.

[0124] It can be seen that at different levels of pyramid knowledge vector mining, the difference in the distribution labels of attention indicators will lead to different mining focuses. Take the two-layer pyramid knowledge vector mining as an example, in the mining of the first layer of pyramid depth.

[0125] In the first round of mining, since the first sensor monitoring data (such as data related to key equipment operating parameters) are selected based on normal distribution labels, mining at this level may focus more on direct feature extraction of these key data. For example, for the cutting tool temperature data (the first sensor monitoring data selected in the first round), its basic statistical features such as mean, variance, maximum and minimum are calculated to obtain a preliminary understanding of the temperature change of the cutting tool.

[0126] In the second round of mining, the first sensor monitoring data selected based on the uniform distribution label may pay more attention to the overall distribution characteristics of the data in the first layer of pyramid depth mining. For example, the statistical characteristics such as the median and quartile of the data are calculated to grasp the central trend and dispersion of the overall data. Because the data selected at this time is based on the uniform distribution label, the overall distribution characteristics of the data become more important in mining.

[0127] Furthermore, when the first implicit sensor monitoring knowledge vector spectrum and the second implicit sensor monitoring knowledge vector spectrum obtained in each round of mining are fused, the difference in attention indicator distribution labels will also affect the fusion method.

[0128] For example, in the first round of mining, since the first sensor monitoring data is data with a higher attention index selected based on the normal distribution label, when it is fused with the second implicit sensor monitoring knowledge vector spectrum (i.e., sensor monitoring data other than the first sensor monitoring data), the first implicit sensor monitoring knowledge vector spectrum may be given a higher weight. Suppose the first implicit sensor monitoring knowledge vector spectrum is (V_1=(v_{11}, v_{12}, cdots, v_{1m})), and its weight vector is (W_1=(w_{11}, w_{12}, cdots, w_{1m})), and the second implicit sensor monitoring knowledge vector spectrum is (V_2=(v_{21}, v_{22}, cdots, v_{2m})), and its weight vector is (W_2=(w_{21}, w_{22}, cdots, w_{2m})). Since the importance of the first sensor monitoring data in the first round of mining is higher, it is possible to set (w_{1i}=0.7), (w_{2i}=0.3) ((i=1, 2, cdots, m)) for weighted sum fusion.

[0129] In the second round of mining, since the attention indicator distribution labels become evenly distributed, more balanced weights may be used during fusion. For example, (w_{1i}=0.5) and (w_{2i}=0.5) are set for fusion. This fusion method that adjusts the difference in attention indicator distribution labels between different rounds enables more flexible and accurate integration of data from different sources and features during multiple rounds of pyramid knowledge vector mining, and gradually builds a more representative and effective target sensor monitoring knowledge vector spectrum.

[0130] For example, when selecting the first sensor monitoring data according to the attention indicator distribution label, a sampling algorithm based on probability distribution can be used. For example, when the distribution label of the attention indicator is a normal distribution, the probability density function of the normal distribution is assumed to be (f(x)=frac{1}{sqrt{2pi}sigma}e^{-frac{(x-mu)^2}{2sigma^2}}), where (mu) is the mean and (sigma) is the standard deviation.

[0131] To select data points with a higher attention index of (p) (such as (p=0.3)), the cumulative distribution function (F(x)=int_{-infty}^{x}f(t)dt) can be calculated, and then the selected data points can be determined according to the value of (F(x)). For example, the value of (F(x)) is divided from 0 to 1, and the data points corresponding to the (F(x)) value in the interval of (1-p) to 1 are selected as the first sensor monitoring data.

[0132] When the first implicit sensor monitoring knowledge vector spectrum (V_1) and the second implicit sensor monitoring knowledge vector spectrum (V_2) are fused in the first round of mining, the third implicit sensor monitoring knowledge vector spectrum (V_3) is obtained by weighted summation as described above, whose elements (v_{3i}=w_{1i}v_{1i}+w_{2i}v_{2i})((i=1, 2, cdots, m)), where (w_{1i}=k_1)((k_1=0.7)), (w_{2i}=1-k_1=0.3).

[0133] In the second round of mining, (w_{1i}=k_2)((k_2=0.5)), (w_{2i}=1-k_2=0.5). This formula that adjusts weights based on round and attention index distribution label differences can reasonably combine knowledge vector spectra based on the importance of data in different rounds of mining, so that the fused results can better reflect the essential characteristics of the data and mining needs.

[0134] Based on the above example, the difference in the distribution labels of the attention indicators in two consecutive rounds of pyramid knowledge vector mining can change the basis for data selection in each round, thereby mining more comprehensive information. Different distribution labels enable each round to have different focuses when mining at different levels, and can deeply mine data features from multiple angles. When fusion of knowledge is carried out, the weights are adjusted according to the difference in distribution labels, which can more reasonably integrate the data mined from different rounds, improve the accuracy and effectiveness of the fusion results, and help to build a more representative target sensor monitoring knowledge vector spectrum, thereby providing a more accurate decision-making basis for the control and management of furniture board processing equipment.

[0135] In an optional technical solution, pyramid knowledge vector mining is performed based on the first sensor monitoring knowledge vector spectra of several pyramid depths to obtain target sensor monitoring knowledge vector spectra of several pyramid depths, including: derivative knowledge vector mining is performed based on the first sensor monitoring knowledge vector spectra of several pyramid depths to obtain second sensor monitoring knowledge vector spectra of several pyramid depths; wherein the second sensor monitoring knowledge vector spectrum of the second pyramid depth is the first sensor monitoring knowledge vector spectrum with the largest feature recognition; the second sensor monitoring knowledge vector spectrum of the remaining pyramid depths is obtained by performing derivative knowledge vector mining on the second sensor monitoring knowledge vector spectra of the marked pyramid depths of the superior and subordinate and integrating the first sensor monitoring knowledge vector spectra of the same pyramid depth; pooling knowledge vector mining is performed based on the second sensor monitoring knowledge vector spectra of several pyramid depths to obtain the target sensor monitoring knowledge vector spectra of several pyramid depths; wherein the target sensor monitoring knowledge vector spectrum of the first pyramid depth is the second sensor monitoring knowledge vector spectrum of the first pyramid depth; the target sensor monitoring knowledge vector spectrum of the remaining pyramid depths is obtained by performing pooling knowledge vector mining on the target sensor monitoring knowledge vector spectra of the unmarked pyramid depths of the superior and subordinate and integrating the second sensor monitoring knowledge vector spectra of the same pyramid depth.

[0136] In the next step, the first sensor monitoring knowledge vector spectrum with several pyramid depths is subjected to derivative knowledge vector mining to obtain the second sensor monitoring knowledge vector spectrum with several pyramid depths, including: the first sensor monitoring knowledge vector spectrum to be processed is subjected to derivative knowledge vector mining to obtain the sensor monitoring derivative knowledge vector spectrum; wherein the first sensor monitoring knowledge vector spectrum to be processed based on the first round of derivative knowledge vector mining is the first sensor monitoring knowledge vector spectrum obtained by the last round of pooling knowledge vector mining; the sensor monitoring derivative knowledge vector spectrum is integrated with the first sensor monitoring knowledge vector spectrum with the same pyramid depth to obtain the second sensor monitoring knowledge vector spectrum; and the second sensor monitoring knowledge vector spectrum is used as the first sensor monitoring knowledge vector spectrum to be processed corresponding to the next round of derivative knowledge vector mining.

[0137] Furthermore, the method of performing pooled knowledge vector mining based on the second sensor monitoring knowledge vector spectrum at several pyramid depths to obtain the target sensor monitoring knowledge vector spectrum at several pyramid depths includes: performing pooled knowledge vector mining on the second sensor monitoring knowledge vector spectrum to be processed to obtain a sensor monitoring pooled knowledge vector spectrum; wherein the second sensor monitoring knowledge vector spectrum obtained by the last round of derived knowledge vector mining is the second sensor monitoring knowledge vector spectrum to be processed of the first round of pooled knowledge vector mining; obtaining a target sensor monitoring knowledge vector spectrum by integrating the sensor monitoring pooled knowledge vector spectrum with the second sensor monitoring knowledge vector spectrum at the same pyramid depth; and using the target sensor monitoring knowledge vector spectrum as the second sensor monitoring knowledge vector spectrum to be processed corresponding to the next round of pooled knowledge vector mining.

[0138] In a preferred design scheme, the target sensor monitoring knowledge vector spectrum is obtained by integrating the sensor monitoring pooling knowledge vector spectrum with the second sensor monitoring knowledge vector spectrum of the same pyramid depth, including: interacting the sensor monitoring pooling knowledge vector spectrum with the second sensor monitoring knowledge vector spectrum of the same pyramid depth to obtain a first sensor monitoring interaction knowledge vector spectrum; determining the attention confidence coefficients corresponding to each attention indicator in the first sensor monitoring interaction knowledge vector spectrum according to the first sensor monitoring interaction knowledge vector spectrum, and interacting each attention indicator of the second sensor monitoring knowledge vector spectrum according to each attention confidence coefficient to obtain a second sensor monitoring interaction knowledge vector spectrum; determining the processing environment sensor monitoring knowledge vector spectrum corresponding to the second sensor monitoring interaction knowledge vector spectrum, and determining the target sensor monitoring knowledge vector spectrum based on the processing environment sensor monitoring knowledge vector spectrum and the second sensor monitoring interaction knowledge vector spectrum.

[0139] In the above embodiment, firstly, derived knowledge vector mining is performed based on the first sensor monitoring knowledge vector spectrum of several pyramid depths to obtain the second sensor monitoring knowledge vector spectrum of several pyramid depths.

[0140] 1) The first round of derived knowledge vector mining

[0141] Determination of the first sensor monitoring knowledge vector spectrum to be processed: In the first round of derivative knowledge vector mining, the first sensor monitoring knowledge vector spectrum to be processed is the first sensor monitoring knowledge vector spectrum obtained by the last round of pooling knowledge vector mining. For example, after multiple rounds of previous operations, the first sensor monitoring knowledge vector spectrum is obtained, and after a certain round of pooling knowledge vector mining, a specific first sensor monitoring knowledge vector spectrum (V_{f1}) is obtained. This (V_{f1}) becomes the first sensor monitoring knowledge vector spectrum to be processed in the first round of derivative knowledge vector mining.

[0142] Derivative knowledge vector mining process: Derivative knowledge vector mining is performed on this first sensor monitoring knowledge vector spectrum to be processed ($V_{f1}$) to obtain a sensor monitoring derivative knowledge vector spectrum. Various methods can be used for this derivative knowledge vector mining, such as derivative mining based on principal component analysis (PCA).

[0143] Let ($V_{f1}$) be an ($n\times m$) matrix ($n$ represents the number of data points, and $m$ represents the number of features). First, calculate the covariance matrix of ($V_{f1}$) ($C = \frac{1}{n - 1}V_{f1}^TV_{f1}$), and then solve the eigenvalues ($\lambda_i$) and eigenvectors ($u_i$) ($i = 1, 2, \cdots, m$) of the covariance matrix $C$. Sort the eigenvectors according to the magnitude of the eigenvalues, and select the first $k$ ($k < m$) eigenvectors to form the transformation matrix $U = [u_1, u_2, \cdots, u_k]$. Then the sensor monitoring derivative knowledge vector spectrum ($V_{d1}$) can be obtained through ($V_{d1}=V_{f1}U$). The meaning of this formula is to transform the original first sensor monitoring knowledge vector spectrum ($V_{f1}$) through the transformation matrix ($U$) composed of the main eigenvectors to obtain a vector spectrum ($V_{d1}$) with derivative knowledge. This transformation can highlight the main information in the data, reduce the data dimension while retaining key features.

[0144] 2) Integrate to obtain the second sensor monitoring knowledge vector spectrum: Integrate the sensor monitoring derivative knowledge vector spectrum ($V_{d1}$) with the first sensor monitoring knowledge vector spectrum ($V_{f1}$) at the same pyramid depth to obtain the second sensor monitoring knowledge vector spectrum ($V_{s1}$). The integration method can be weighted summation. Let ($V_{d1}=(v_{d11}, v_{d12}, \cdots, v_{d1m})$), ($V_{f1}=(v_{f11}, v_{f12}, \cdots, v_{f1m})$), and the elements of the integrated ($V_{s1}$) are ($v_{s1i}=w_1v_{d1i}+w_2v_{f1i}$) ($i = 1, 2, \cdots, m$), where $w_1$ and $w_2$ are weight coefficients. For example, $w_1 = 0.4$ and $w_2 = 0.6$. These weight coefficients can be set according to the importance of the data or experience. Through this integration method, the knowledge obtained from derivative mining is combined with the knowledge of the original first sensor monitoring knowledge vector spectrum to obtain a more comprehensive second sensor monitoring knowledge vector spectrum.

[0145] 3) Subsequent rounds of derivative knowledge vector mining: The obtained second sensor monitoring knowledge vector spectrum (V_{s1}) is used as the first sensor monitoring knowledge vector spectrum to be processed corresponding to the next round of derivative knowledge vector mining. Repeat the above derivative knowledge vector mining and integration process to gradually construct the second sensor monitoring knowledge vector spectrum with different pyramid depths. For example, in the second round of derivative knowledge vector mining, (V_{s1}) is used as the first sensor monitoring knowledge vector spectrum to be processed, and the second sensor monitoring knowledge vector spectrum (V_{s2}) of the second round is obtained according to the above PCA derivative mining and integration method, and so on.

[0146] Among them, there is a special definition for the second sensor monitoring knowledge vector spectrum of the second pyramid depth, which is the first sensor monitoring knowledge vector spectrum with the largest feature recognition. The feature recognition can be measured by calculating the correlation, discreteness and other indicators between the various features in the vector spectrum. For example, the information gain (InformationGain) of the feature can be calculated, assuming (X) is a feature vector, (Y) is a classification label (here it can be a classification related to the state of the processing equipment, such as normal state or fault state), information gain (IG (X, Y) = H (Y) - H (Y|X)), where (H (Y)) is the entropy of (Y), and (H (Y|X)) is the conditional entropy of (Y) under the condition of known (X). By calculating the information gain of the features in each first sensor monitoring knowledge vector spectrum, the one with the largest information gain is selected as the second sensor monitoring knowledge vector spectrum of the second pyramid depth.

[0147] For the second sensor monitoring knowledge vector spectrum of the remaining pyramid depth, it is obtained by mining the derivative knowledge vectors of the second sensor monitoring knowledge vector spectrum of the annotated pyramid depth of the upper and lower levels and integrating the first sensor monitoring knowledge vector spectrum of the same pyramid depth. For example, for the second sensor monitoring knowledge vector spectrum of the third pyramid depth (V_{s3}), it is obtained by mining the derivative knowledge vectors of the annotated second sensor monitoring knowledge vector spectrum of the second pyramid depth (V_{s2}) (such as using the above-mentioned PCA method) and integrating the first sensor monitoring knowledge vector spectrum of the third pyramid depth (V_{f3}) (using a weighted summation method similar to the above).

[0148] Next, pooled knowledge vector mining is performed based on the second sensor monitoring knowledge vector spectra of several pyramid depths to obtain target sensor monitoring knowledge vector spectra of several pyramid depths.

[0149] 1) The first round of pooling knowledge vector mining

[0150] Determination of the second sensor monitoring knowledge vector spectrum to be processed: The second sensor monitoring knowledge vector spectrum obtained by the last round of derivative knowledge vector mining is the second sensor monitoring knowledge vector spectrum to be processed in the first round of pooled knowledge vector mining. For example, after several rounds of derivative knowledge vector mining, the last second sensor monitoring knowledge vector spectrum obtained is (V_{sn}), and this (V_{sn}) becomes the second sensor monitoring knowledge vector spectrum to be processed in the first round of pooled knowledge vector mining.

[0151] Pooling knowledge vector mining process: pooling knowledge vector mining is performed on the second sensor monitoring knowledge vector spectrum (V_{sn}) to obtain the sensor monitoring pooling knowledge vector spectrum. The pooling operation can adopt the maximum pooling (Max-pooling) or average pooling (Average-pooling) method.

[0152] Taking maximum pooling as an example, suppose (V_{sn}) is a (ptimesq) matrix and divide it into several small areas (for example, (2times2) small areas). For each small area (begin{bmatrix}v_{ij}&v_{i(j+1)}v_{(i+1)j}&v_{(i+1)(j+1)}end{bmatrix}), the result after maximum pooling is (max(v_{ij},v_{i(j+1)},v_{(i+1)j},v_{(i+1)(j+1)})). In this way, the entire (V_{sn}) is mined for pooled knowledge vectors to obtain the sensor monitoring pooled knowledge vector spectrum (V_{p1}). The purpose of maximum pooling is to reduce the dimension of the data while retaining the main features of the data and extract the local maximum features in the data.

[0153] 2) Integration to obtain target sensor monitoring knowledge vector spectrum

[0154] The target sensor monitoring knowledge vector spectrum is obtained by integrating the sensor monitoring pooling knowledge vector spectrum (V_{p1}) with the second sensor monitoring knowledge vector spectrum (V_{sn}) of the same pyramid depth. This integration process includes the following steps.

[0155] Interaction to obtain the first sensor monitoring interaction knowledge vector spectrum: First, interact (V_{p1}) with (V_{sn}) to obtain the first sensor monitoring interaction knowledge vector spectrum (V_{i1}). The interaction method can be element multiplication. Let (V_{p1}=(v_{p11}, v_{p12}, cdots, v_{p1m})), (V_{sn}=(v_{sn1}, v_{sn2}, cdots, v_{snm})), then the element (v_{i1i}=v_{p1i}timesv_{sni}) ((i=1, 2, cdots, m)) of (V_{i1}). This interaction method can fuse the pooled features with the features of the original second sensor monitoring knowledge vector spectrum to highlight the common information of the two.

[0156] Determine the attention confidence coefficient and conduct the second interaction: Determine the attention confidence coefficient corresponding to each attention indicator according to the first sensor monitoring interaction knowledge vector spectrum (V_{i1}). The attention confidence coefficient can be determined by calculating the stability of the feature, the correlation with other features, and other factors. For example, for a feature (v_{i1j}) in (V_{i1}), calculate its Pearson Correlation Coefficient (r_{j, k}) ((k=1, 2, cdots, m, kneqj)) with other features, and then determine its attention confidence coefficient (c_j) according to the sum of the absolute values ​​of the correlation coefficients, such as (c_j=sum_{k=1, kneqj}^{m}|r_{j, k}|).

[0157] According to each attention confidence coefficient, each attention index of the second sensor monitoring knowledge vector spectrum (V_{sn}) is interacted to obtain the second sensor monitoring interaction knowledge vector spectrum (V_{i2}). Assume that the attention index vector of (V_{sn}) is (A_{sn}=(a_{sn1}, a_{sn2}, cdots, a_{snm})), then the element (v_{i2i}=c_ia_{sni}) of (V_{i2}) ((i=1, 2, cdots, m)). This interaction based on the attention confidence coefficient can adjust the second sensor monitoring knowledge vector spectrum according to the importance of the feature.

[0158] Determine the target sensor monitoring knowledge vector spectrum: determine the processing environment sensor monitoring knowledge vector spectrum (V_{e}) corresponding to the second sensor monitoring interaction knowledge vector spectrum (V_{i2}). The processing environment sensor monitoring knowledge vector spectrum can be obtained by performing similar knowledge vector mining on data collected by processing environment related sensors (such as temperature, humidity sensors, etc.).

[0159] According to the processing environment sensor monitoring knowledge vector spectrum (V_{e}) and the second sensor monitoring interaction knowledge vector spectrum (V_{i2}), the target sensor monitoring knowledge vector spectrum (V_{t1}) is determined. For example, a weighted summation method can be used, assuming (V_{e}=(v_{e1}, v_{e2}, cdots, v_{em})), then the element (v_{t1i}=w_3v_{i2i}+w_4v_{ei}) of (V_{t1}) ((i=1, 2, cdots, m)), where (w_3) and (w_4) are weight coefficients, for example (w_3=0.7), (w_4=0.3). This target sensor monitoring knowledge vector spectrum (V_{t1}) combines the knowledge vector spectrum of the processing equipment itself and the knowledge vector spectrum of the processing environment, and can more comprehensively reflect the status of the furniture board processing equipment.

[0160] 3) Pooled knowledge vector mining in subsequent rounds

[0161] The obtained target sensor monitoring knowledge vector spectrum (V_{t1}) is used as the second sensor monitoring knowledge vector spectrum to be processed corresponding to the next round of pooling knowledge vector mining. Repeat the above pooling knowledge vector mining and integration process to gradually construct target sensor monitoring knowledge vector spectra of different pyramid depths. For example, in the second round of pooling knowledge vector mining, (V_{t1}) is used as the second sensor monitoring knowledge vector spectrum to be processed, and the second round of target sensor monitoring knowledge vector spectrum (V_{t2}) is obtained according to the above pooling and integration method, and so on. Among them, the target sensor monitoring knowledge vector spectrum of the first pyramid depth is the second sensor monitoring knowledge vector spectrum of the first pyramid depth, and the target sensor monitoring knowledge vector spectrum of the remaining pyramid depth is obtained by pooling knowledge vector mining on the target sensor monitoring knowledge vector spectrum of the unlabeled pyramid depth of the upper and lower levels and integrating the second sensor monitoring knowledge vector spectrum of the same pyramid depth.

[0162] With this design, the knowledge information in the data can be deeply mined through the combined operation of derived knowledge vector mining and pooled knowledge vector mining. In derived knowledge vector mining, PCA and other methods are used to process knowledge vector spectra of different rounds, such as integrating knowledge vector spectra from different sources by weighted summation, which improves the comprehensiveness of knowledge. The special definition of the depth of the second pyramid helps to focus on the information with the highest feature recognition and improve the pertinence of data mining. In the process of pooled knowledge vector mining, the pooling operation effectively reduces the data dimension. At the same time, different knowledge vector spectra are integrated through steps such as interaction and determination of attention confidence coefficients, and combined with the processing environment sensor monitoring knowledge vector spectrum, the final target sensor monitoring knowledge vector spectrum can integrate the information of the equipment itself and the processing environment, providing a more accurate and comprehensive basis for the monitoring, control and decision-making of furniture board processing equipment.

[0163] In an optional embodiment, the furniture panel processing equipment control method is implemented based on a target intelligent decision-making processing algorithm, and the target intelligent decision-making processing algorithm is debugged according to the following steps: obtaining a number of first sensor monitoring training data; the first sensor monitoring training data includes a first processing task training label but does not include a second processing task training label; obtaining a number of second sensor monitoring training data; the second sensor monitoring training data includes a second processing task training label but does not include the first processing task training label; debugging the initial intelligent decision processing algorithm at least once based on the first sensor monitoring training data and the second sensor monitoring training data, until the first debugging error and the second debugging error meet the set debugging standard requirements, thereby obtaining the target intelligent decision processing algorithm.

[0164] Among them, the debugging process includes: inputting a number of the first sensor monitoring training data and a number of the second sensor monitoring training data into the initial intelligent decision-making processing algorithm to obtain the first equipment control training characteristics of each corresponding first processing task and the second equipment control training characteristics of each second processing task; determining the first debugging error based on each of the first equipment control training characteristics and the corresponding training labels of each first processing task; determining the second debugging error based on each of the second equipment control training characteristics and the corresponding training labels of each second processing task; optimizing the algorithm weights of the initial intelligent decision-making processing algorithm based on the first debugging error and the second debugging error, and using the initial intelligent decision-making processing algorithm after optimizing the algorithm weights as the initial intelligent decision-making processing algorithm before the next round of debugging processing.

[0165] Under a preferred design idea, the linked processing task is a processing task whose process priority is less than the set process priority; the obtaining of a number of first sensor monitoring training data includes: obtaining a number of initial sensor monitoring training data; wherein the initial sensor monitoring training data includes a first processing task training label and a second processing task training label; and filtering at least part of the data set of each initial sensor monitoring training data according to the set process priority to obtain a number of corresponding first sensor monitoring training data.

[0166] It can be understood that the above debugging process of the target intelligent decision-making processing algorithm includes the following two branches.

[0167] Branch 1: Debugging basis of target intelligent decision-making processing algorithm: data acquisition

[0168] 1) Acquisition of first sensor monitoring training data

[0169] Initial sensor monitoring training data: First, obtain a number of initial sensor monitoring training data, which are the basic data source for the entire debugging process. The initial sensor monitoring training data contains rich information, including the first processing task training label and the second processing task training label. For example, the initial sensor monitoring training data can be collected from a large number of furniture board processing historical records. For example, these historical records contain 1,000 sets of data points, each of which contains various sensor monitoring data during the processing, such as temperature, pressure, vibration and other data, as well as the corresponding processing task-related labels (first processing task training label and second processing task training label).

[0170] Screening to obtain the first sensor monitoring training data: Screen at least part of the data set of each initial sensor monitoring training data according to the set process priority to obtain a number of corresponding first sensor monitoring training data. Setting the process priority is an indicator used to distinguish the importance or processing order of different processing tasks. For example, the process priority is set to 0.6 (for example, the value range is 0 to 1). If the process priority of a processing task in a certain initial sensor monitoring training data is less than this set value, it will be screened out as the first sensor monitoring training data. For example, after screening, 600 groups of first sensor monitoring training data are obtained, which only include the first processing task training label and do not include the second processing task training label. The purpose of this step is to separate the data related to a specific processing task (the first processing task) so that the intelligent decision-making processing algorithm can be debugged for the first processing task later.

[0171] 2) Acquisition of Second Sensor Monitoring Training Data

[0172] Corresponding to the acquisition of the first sensor monitoring training data, a number of second sensor monitoring training data are acquired. These data include the second processing task training label but do not include the first processing task training label. For example, from the remaining 400 sets of initial sensor monitoring training data (1000-600), after further screening or labeling, 300 sets of second sensor monitoring training data are obtained. These data are specifically used to debug the intelligent decision processing algorithm for the second processing task.

[0173] Branch 2: Debugging process

[0174] 1) Input training data into the initial intelligent decision-making processing algorithm

[0175] Input a number of first sensor monitoring training data and a number of second sensor monitoring training data into the initial intelligent decision processing algorithm. The initial intelligent decision processing algorithm can be an algorithm based on a neural network, such as a multi-layer perceptron (MLP). For example, this MLP has an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined according to the number of features of the input sensor monitoring training data. For example, after the first sensor monitoring training data and the second sensor monitoring training data are preprocessed, each data point has 20 features, then the input layer has 20 nodes. The hidden layer can be set to 2 layers, with 10 and 5 nodes in each layer respectively. The number of nodes in the output layer is determined according to the number of equipment control training features to be obtained, for example, 3 nodes (corresponding to different equipment control parameters).

[0176] When the first sensor monitoring training data is input, the first equipment control training features of the corresponding first processing tasks are obtained through MLP calculation. Similarly, when the second sensor monitoring training data is input, the second equipment control training features of the second processing tasks are obtained. For example, for the first processing task, the first equipment control training features obtained may be the predicted values ​​of parameters such as the speed of the cutting equipment and the pressure of the drilling equipment; for the second processing task, the second equipment control training features obtained may be the predicted values ​​of parameters such as the rotation speed of the polishing equipment and the grinding pressure.

[0177] 2) Determine the first debugging error and the second debugging error

[0178] Determination of the first debugging error: Determine the first debugging error based on each first device control training feature and each corresponding first processing task training label. Here, the mean square error (MSE) can be used as the error measurement indicator. Suppose the first device control training feature is (C_{1i}=(c_{1i1}, c_{1i2}, cdots, c_{1in}))((i=1, 2, cdots, m), (m) is the number of first sensor monitoring training data, (n) is the dimension of the first device control training feature), and the corresponding first processing task training label is (L_{1i}=(l_{1i1}, l_{1i2}, cdots, l_{1in})). Then the first debugging error can be: (E_1=frac{1}{m}sum_{i=1}^{m}sum_{j=1}^{n}(c_{1ij}-l_{1ij})^2). The meaning of this formula is to calculate the sum of squared errors between each first device control training feature and the corresponding training label, and then average them to get the first debugging error. For example, if (m=600), (n=3), the value of the first debugging error is obtained by calculating the sum of squared errors of each data point and averaging them.

[0179] Determination of the second debugging error: Similarly, the second debugging error is determined based on each second device control training feature and each corresponding second processing task training label. Assume that the second device control training feature is (C_{2i}=(c_{2i1}, c_{2i2}, cdots, c_{2in})), and the corresponding second processing task training label is (L_{2i}=(l_{2i1}, l_{2i2}, cdots, l_{2in})). Then the above second debugging error can be exemplified as: (E_2=frac{1}{k}sum_{i=1}^{k}sum_{j=1}^{n}(c_{2ij}-l_{2ij})^2) ((k) is the number of second sensor monitoring training data). For example, if (k=300), (n=3), the value of the second debugging error is calculated according to the above formula.

[0180] 3) Optimize the algorithm weights of the initial intelligent decision-making processing algorithm

[0181] The algorithm weights of the initial intelligent decision-making processing algorithm are optimized based on the first debugging error and the second debugging error. In a neural network (such as MLP), the algorithm weights determine the mapping relationship from input data to output data. Taking the backpropagation algorithm as an example, based on the calculated first debugging error (E_1) and second debugging error (E_2), starting from the output layer, the error is backpropagated to adjust the connection weights between the layers.

[0182] Let (w_{ij}) be the connection weight from the (i)th layer to the (j)th layer in the neural network, and the learning rate be (eta) (e.g. (eta=0.01)). According to the formula of the back propagation algorithm (Deltaw_{ij}=-etafrac{partialE}{partialw_{ij}}) (where (E=E_1+E_2)), calculate the update amount of the weight (Deltaw_{ij}), and then update the weight to (w_{ij}=w_{ij}+Deltaw_{ij}). By continuously adjusting the weight, the first debugging error and the second debugging error are gradually reduced.

[0183] The initial intelligent decision-making processing algorithm after optimizing the algorithm weights is used as the initial intelligent decision-making processing algorithm before the next round of debugging processing, and the above debugging process is repeated until the first debugging error and the second debugging error meet the set debugging standard requirements. The set debugging standard requirements can be a specific error threshold, such as (E_1leq0.05) and (E_2leq0.05). When this requirement is met, the target intelligent decision-making processing algorithm is obtained, which can accurately obtain the equipment control decision characteristics of the first processing task and the second processing task based on the input sensor monitoring data.

[0184] It can be seen that by screening training data according to different processing task labels, the intelligent decision-making processing algorithm can be debugged for specific processing tasks (the first processing task and the second processing task). The use of mean square error to measure the debugging error can intuitively reflect the difference between the equipment control training characteristics and the training labels. The back propagation algorithm is used to optimize the algorithm weights according to the debugging error to continuously improve the accuracy of the algorithm. Repeat the debugging process until the set debugging requirements are met to ensure that the target intelligent decision-making processing algorithm can make effective control decisions on furniture board processing equipment while meeting the accuracy requirements, thereby improving processing efficiency and quality.

[0185] Under another preferred design idea, a number of first sensor monitoring training data are input into an initial intelligent decision-making processing algorithm to obtain each first device control training feature, including: for each first sensor monitoring training data, performing multiple rounds of knowledge vector mining processing on the first sensor monitoring training data through the initial intelligent decision-making processing algorithm to obtain a number of first sensor monitoring training knowledge vector spectra with a pyramid depth; performing derivative knowledge vector mining based on a number of first sensor monitoring training knowledge vector spectra to obtain a number of second sensor monitoring training knowledge vector spectra with a pyramid depth; determining the first device control training feature based on the second sensor monitoring training knowledge vector spectrum with a target pyramid depth.

[0186] The above design idea first processes multiple rounds of knowledge vector mining for each first sensor monitoring training data.

[0187] 1) Purpose and principle of multi-round knowledge vector mining

[0188] In the process of inputting a plurality of first sensor monitoring training data into the initial intelligent decision processing algorithm to obtain each first device control training feature, multiple rounds of knowledge vector mining processing are performed for each first sensor monitoring training data. The purpose of such multiple rounds of processing is to gradually extract a knowledge vector spectrum with more depth and semantic information from the original first sensor monitoring training data, so that the first device control training feature can be determined more accurately later.

[0189] Its principle is based on the hierarchical abstraction of data. Each round of mining processing will further explore the hidden information in the data based on the previous one, which is similar to building a pyramid structure, gradually building up higher-level knowledge representations from the bottom-level raw data.

[0190] 2) Example of multi-round knowledge vector mining process

[0191] For example, the first sensor monitoring training data is a vector containing multiple sensor data, such as temperature sensor data (T), pressure sensor data (P) and vibration sensor data (V). In the first round of knowledge vector mining, the original data are firstly subjected to separate feature extraction.

[0192] For the temperature data (T=(t_1, t_2, cdots, t_n)), calculate its mean (mu_T=frac{1}{n}sum_{i=1}^{n}t_i), variance (sigma_T^2=frac{1}{n-1}sum_{i=1}^{n}(t_i-mu_T)^2), maximum value (T_{max}=max(t_1, t_2, cdots, t_n)) and minimum value (T_{min}=min(t_1, t_2, cdots, t_n)). Similarly, similar calculations are performed on the pressure data (P) and vibration data (V) to obtain their respective features such as mean, variance, maximum and minimum values.

[0193] These features are combined to form the first sensor monitoring training knowledge vector spectrum of the first pyramid depth. For example, the vector spectrum of the first pyramid depth (V_1=(mu_T, sigma_T^2, T_{max}, T_{min}, mu_P, sigma_P^2, P_{max}, P_{min}, mu_V, sigma_V^2, V_{max}, V_{min})).

[0194] In the second round of knowledge vector mining, the relevant features in the vector spectrum (V_1) of the first pyramid depth are fused. For example, the features related to temperature and pressure can be fused by weighted summation. Let the temperature feature vector be (T_f=(mu_T, sigma_T^2)), the pressure feature vector be (P_f=(mu_P, sigma_P^2)), and the elements of the fused vector (F) (f_i=w_1t_i+w_2p_i) (where (w_1=0.6), (w_2=0.4)). In this way, the first sensor monitoring training knowledge vector spectrum of the second pyramid depth is constructed. As the number of rounds increases, more relevant low-level vectors are continuously fused, and more complex fusion algorithms are used, such as the principal component analysis (PCA) algorithm. For example, at the third pyramid depth, multiple vectors of the second pyramid depth are subjected to PCA analysis. After PCA processing, the vector originally composed of multiple features is converted into a vector composed of several main component features, thereby obtaining the first sensor monitoring training knowledge vector spectrum of the third pyramid depth.

[0195] Secondly, derived knowledge vector mining is performed based on several first sensor monitoring training knowledge vector spectra.

[0196] 1) Methods and significance of derived knowledge vector mining

[0197] After obtaining the first sensor monitoring training knowledge vector spectrum of several pyramid depths, the derived knowledge vector mining is carried out. The purpose of derived knowledge vector mining is to further mine more valuable information based on the existing knowledge vector spectrum, and to generate new knowledge representation forms through the transformation and combination of data, so as to more comprehensively reflect the relationship between data and equipment control training characteristics.

[0198] For example, a derivative knowledge vector mining method based on wavelet transform can be used. Wavelet transform can decompose signals into components of different scales and frequencies, which is very effective for mining local features and trends in data.

[0199] 2) Specific operations of derived knowledge vector mining

[0200] Suppose the first sensor monitoring training knowledge vector spectrum is (V_{k1}, V_{k2}, cdots, V_{kn}) ((n) is the number of pyramid depths). For each (V_{ki}), perform a wavelet transform. Taking discrete wavelet transform (DWT) as an example, suppose the wavelet function is (psi(t)) and the scaling function is (varphi(t)), and perform a discrete wavelet transform on each element (v_{kij}) in (V_{ki}) ((j=1, 2, cdots, m), (m) is the dimension of the vector spectrum).

[0201] According to the formula (W_{kij}(a,b)=frac{1}{sqrt{a}}int_{-infty}^{infty}v_{kij}(t)psi(frac{tb}{a})dt) (where (a) is the scale parameter and (b) is the translation parameter), the coefficients after wavelet transformation (W_{kij}) are obtained. These coefficients constitute the derived knowledge vector spectrum. Then, the vector spectra after wavelet transformation of different pyramid depths are combined to obtain the second sensor monitoring training knowledge vector spectra of several pyramid depths. For example, for the second sensor monitoring training knowledge vector spectrum (V_{s2}) of the second pyramid depth, it can be composed of part of the coefficients of the first sensor monitoring training knowledge vector spectrum of the first pyramid depth after wavelet transformation and part of the coefficients of the first sensor monitoring training knowledge vector spectrum of the second pyramid depth after wavelet transformation.

[0202] Finally, the first device control training feature is determined based on the second sensor monitoring training knowledge vector spectrum of the target pyramid depth

[0203] 1) Basis for selecting target pyramid depth: The selection of target pyramid depth is based on the analysis of the second sensor monitoring training knowledge vector spectrum of different pyramid depths. For example, it can be selected based on the contribution of the vector spectrum of each pyramid depth to the equipment control training features. This contribution can be measured by calculating the correlation between the vector spectrum and the known equipment control training features.

[0204] Assume that the second sensor monitoring training knowledge vector spectrum is (V_{s1}, V_{s2}, cdots, V_{sn}) ((n) is the number of pyramid depths), and the first device control training feature is (C=(c_1, c_2, cdots, c_m)). Calculate the Pearson Correlation Coefficient (r_{si}) of each (V_{si}) and (C). According to the size of (r_{si}), select the pyramid depth with the highest correlation as the target pyramid depth. For example, if (r_{s3}) is the largest of all correlation coefficients, then the target pyramid depth is the third pyramid depth.

[0205] 2) The process of determining the first device control training characteristics: Once the second sensor monitoring training knowledge vector spectrum of the target pyramid depth is determined, for example, the target pyramid depth is (k), and the corresponding second sensor monitoring training knowledge vector spectrum is (V_{sk}). A linear regression algorithm can be used to determine the first device control training characteristics. Let (V_{sk}=(v_{sk1}, v_{sk2}, cdots, v_{skp})) ((p) is the dimension of the vector spectrum), establish a linear regression model (c_j=beta_0+sum_{i=1}^{p}beta_iv_{ski}) ((j=1, 2, cdots, m)), where (beta_0, beta_1, cdots, beta_p) are regression coefficients.

[0206] These regression coefficients are estimated by the least squares method (LeastSquaresMethod). The goal of the least squares method is to minimize the sum of squared errors: (S=sum_{j=1}^{m}(c_j-(beta_0+sum_{i=1}^{p}beta_iv_{ski}))^2). Calculate the partial derivatives of (S) with respect to (beta_0, beta_1, cdots, beta_p) respectively, and set the partial derivatives to 0, solve the system of equations to obtain the regression coefficients (beta_0, beta_1, cdots, beta_p). Substitute the elements in (V_{sk}) into the linear regression model to obtain the first device control training features (C=(c_1, c_2, cdots, c_m)).

[0207] With this design, through multiple rounds of knowledge vector mining processing, different levels of information can be deeply mined from the original first sensor monitoring training data, and a pyramid-structured knowledge vector spectrum can be constructed to make the data representation more hierarchical and semantic. Derivative knowledge vector mining uses methods such as wavelet transform to further mine new knowledge representations, increasing the richness of data information. The first equipment control training features are determined based on the second sensor monitoring training knowledge vector spectrum of the target pyramid depth, the appropriate pyramid depth is selected through correlation analysis, and the linear regression algorithm is used to determine the features, which improves the accuracy of equipment control training feature determination, thereby helping to improve the accuracy and effectiveness of the entire furniture board processing equipment control.

[0208] It should be understood that the formula parameter content recorded in the above technical solution is in LaTeX format (an existing formula editing format), and technical personnel in this field can undoubtedly restore the corresponding original formula expression based on the formula parameter content recorded in the above technical solution.

[0209] Further, Figure 2 FIG. 2 is a schematic diagram of a control system 200 provided in an embodiment of the present invention. Figure 2 The control system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present invention. Figure 2 As shown, the control system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present invention. The memory 230 may be a separate device independent of the processor 210, or may be integrated in the processor 210. Optionally, as Figure 2 As shown, the control system 200 may also include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices. Optionally, the control system 200 may implement the corresponding processes corresponding to the storage engine or the components in the storage engine (such as the processing module) or the device deployed with the storage engine in each method of the embodiment of the present invention, which will not be described here for the sake of brevity. It should be understood that the processor of the embodiment of the present invention may be an integrated circuit chip with signal processing capabilities.

[0210] It is understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0211] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0212] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the embodiments of the present invention are not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the embodiments of the present invention, ordinary technicians in this field can also make many forms without departing from the purpose of the embodiments of the present invention and the scope of protection of the embodiments of the present invention, all of which are within the protection of the embodiments of the present invention.

Claims

1. A control method for furniture board processing equipment, characterized in that: The method is applied to a control system, and the method includes: obtaining target sensor monitoring data collected by an IoT sensor installed on furniture board processing equipment; performing pyramid knowledge vector mining based on the target sensor monitoring data to obtain target sensor monitoring knowledge vector spectra of several pyramid depths; using a target sensor monitoring knowledge vector spectrum of a first pyramid depth among the several target sensor monitoring knowledge vector spectra as a first target sensor monitoring knowledge vector spectrum, obtaining a first equipment control decision feature of a first processing task based on the first target sensor monitoring knowledge vector spectrum, and obtaining a second equipment control decision feature of a second processing task based on target sensor monitoring knowledge vector spectra other than the first target sensor monitoring knowledge vector spectrum. policy feature; wherein, the first processing task includes a linkage processing task; in combination with the first equipment control decision feature and the second equipment control decision feature, the furniture board processing equipment is remotely controlled; if the first equipment control decision feature and the second equipment control decision feature meet the set equipment control decision conditions, the target sensor monitoring data, the first equipment control decision feature and the second equipment control decision feature are reported to the remote monitoring middle-end system, so that the remote monitoring middle-end system performs automatic production line scheduling based on the received plurality of target sensor monitoring data and the first equipment control decision feature and the second equipment control decision feature of each target sensor monitoring data.

2. The method according to claim 1, characterized in that The method of performing pyramid knowledge vector mining based on target sensor monitoring data to obtain target sensor monitoring knowledge vector spectra of several pyramid depths includes: performing multiple rounds of knowledge vector mining processing based on target sensor monitoring data to obtain first sensor monitoring knowledge vector spectra of several pyramid depths; performing pyramid knowledge vector mining based on first sensor monitoring knowledge vector spectra of several pyramid depths to obtain target sensor monitoring knowledge vector spectra of several pyramid depths; wherein each round of knowledge vector mining processing includes the following contents: performing pyramid knowledge vector mining based on first sensor monitoring data to obtain first implicit sensor monitoring knowledge vector spectrum; wherein the first sensor monitoring data is sensor monitoring data of attention index with a set proportion in sensor monitoring data to be processed corresponding to the current round of pyramid knowledge vector mining; the first pyramid ... The sensor monitoring data to be processed corresponding to the vector mining is the target sensor monitoring data; the first implicit sensor monitoring knowledge vector spectrum and the second implicit sensor monitoring knowledge vector spectrum are fused with attention indicators to obtain a third implicit sensor monitoring knowledge vector spectrum; the second implicit sensor monitoring knowledge vector spectrum is the sensor monitoring data other than the first sensor monitoring data in the sensor monitoring data to be processed; the third implicit sensor monitoring knowledge vector spectrum is subjected to pooled knowledge vector mining to obtain an original sensor monitoring knowledge vector spectrum; wherein the original sensor monitoring knowledge vector spectrum is the sensor monitoring data to be processed corresponding to the next round of pooled knowledge vector mining; the first sensor monitoring knowledge vector spectrum of the several pyramid depths includes the original sensor monitoring knowledge vector spectrum obtained by at least two rounds of knowledge vector mining in multiple rounds of knowledge vector mining.

3. The method according to claim 2, characterized in that There are differences in distribution labels of at least part of the attention indicators in the first sensor monitoring data based on which two consecutive rounds of pyramid knowledge vector mining are based and in the corresponding attention indicators in the sensor monitoring data to be processed.

4. The method according to claim 2, characterized in that: The method of performing pyramid knowledge vector mining based on the first sensor monitoring knowledge vector spectra of several pyramid depths to obtain target sensor monitoring knowledge vector spectra of several pyramid depths includes: performing derivative knowledge vector mining based on the first sensor monitoring knowledge vector spectra of several pyramid depths to obtain second sensor monitoring knowledge vector spectra of several pyramid depths; wherein the second sensor monitoring knowledge vector spectrum of the second pyramid depth is the first sensor monitoring knowledge vector spectrum with the largest feature recognition degree; the second sensor monitoring knowledge vector spectrum of the remaining pyramid depths is obtained by performing derivative knowledge vector mining on the second sensor monitoring knowledge vector spectra of the pyramid depths of the upper and lower levels and integrating the first sensor monitoring knowledge vector spectra of the same pyramid depth; performing pooled knowledge vector mining based on the second sensor monitoring knowledge vector spectra of several pyramid depths to obtain the target sensor monitoring knowledge vector spectra of several pyramid depths; wherein the target sensor monitoring knowledge vector spectrum of the first pyramid depth is the second sensor monitoring knowledge vector spectrum of the first pyramid depth; the target sensor monitoring knowledge vector spectrum of the remaining pyramid depths is obtained by performing pooled knowledge vector mining on the target sensor monitoring knowledge vector spectra of the unlabeled pyramid depths of the upper and lower levels and integrating the second sensor monitoring knowledge vector spectra of the same pyramid depth.

5. The method according to claim 4, characterized in that The method of performing derivative knowledge vector mining on the first sensor monitoring knowledge vector spectrum at several pyramid depths to obtain the second sensor monitoring knowledge vector spectrum at several pyramid depths includes: performing derivative knowledge vector mining on the first sensor monitoring knowledge vector spectrum to be processed to obtain a sensor monitoring derivative knowledge vector spectrum; wherein the first sensor monitoring knowledge vector spectrum to be processed based on the first round of derivative knowledge vector mining is the first sensor monitoring knowledge vector spectrum obtained by the last round of pooling knowledge vector mining; integrating the sensor monitoring derivative knowledge vector spectrum with the first sensor monitoring knowledge vector spectrum at the same pyramid depth to obtain a second sensor monitoring knowledge vector spectrum; and using the second sensor monitoring knowledge vector spectrum as the first sensor monitoring knowledge vector spectrum to be processed corresponding to the next round of derivative knowledge vector mining.

6. The method according to claim 4, characterized in that The method of performing pooled knowledge vector mining based on the second sensor monitoring knowledge vector spectrum of several pyramid depths to obtain the target sensor monitoring knowledge vector spectrum of several pyramid depths includes: performing pooled knowledge vector mining on the second sensor monitoring knowledge vector spectrum to be processed to obtain a sensor monitoring pooled knowledge vector spectrum; wherein the second sensor monitoring knowledge vector spectrum obtained by the last round of derivative knowledge vector mining is the second sensor monitoring knowledge vector spectrum to be processed in the first round of pooled knowledge vector mining; obtaining the target sensor monitoring knowledge vector spectrum by integrating the sensor monitoring pooled knowledge vector spectrum with the second sensor monitoring knowledge vector spectrum of the same pyramid depth; and using the target sensor monitoring knowledge vector spectrum as the second sensor monitoring knowledge vector spectrum to be processed corresponding to the next round of pooled knowledge vector mining.

7. The method according to claim 6, characterized in that The target sensor monitoring knowledge vector spectrum is obtained by integrating the sensor monitoring pooling knowledge vector spectrum with the second sensor monitoring knowledge vector spectrum of the same pyramid depth, including: interacting the sensor monitoring pooling knowledge vector spectrum with the second sensor monitoring knowledge vector spectrum of the same pyramid depth to obtain a first sensor monitoring interaction knowledge vector spectrum; determining the attention confidence coefficients corresponding to each attention indicator in the first sensor monitoring interaction knowledge vector spectrum according to the first sensor monitoring interaction knowledge vector spectrum, and interacting each attention indicator of the second sensor monitoring knowledge vector spectrum according to each attention confidence coefficient to obtain a second sensor monitoring interaction knowledge vector spectrum; determining the processing environment sensor monitoring knowledge vector spectrum corresponding to the second sensor monitoring interaction knowledge vector spectrum, and determining the target sensor monitoring knowledge vector spectrum according to the processing environment sensor monitoring knowledge vector spectrum and the second sensor monitoring interaction knowledge vector spectrum.

8. The method according to claim 1, characterized in that: The control method of furniture plate processing equipment is implemented according to a target intelligent decision-making processing algorithm, which is debugged according to the following steps: obtaining a number of first sensor monitoring training data; the first sensor monitoring training data includes a first processing task training label but does not include a second processing task training label; obtaining a number of second sensor monitoring training data; the second sensor monitoring training data includes a second processing task training label but does not include the first processing task training label; debugging the initial intelligent decision processing algorithm at least once according to the first sensor monitoring training data and the second sensor monitoring training data, until the first debugging error and the second debugging error meet the set debugging standard requirements, and the target intelligent decision processing algorithm is obtained; wherein the debugging process includes: inputting a number of the first sensor monitoring training data and a number of the second sensor monitoring training data into the initial intelligent decision processing algorithm to obtain the first equipment control training features of each corresponding first processing task and the second equipment control training features of each second processing task; determining the first debugging error according to each of the first equipment control training features and the corresponding first processing task training labels; determining the second debugging error according to each of the second equipment control training features and the corresponding second processing task training labels; The first debugging error and the second debugging error optimize the algorithm weight of the initial intelligent decision-making processing algorithm, and use the initial intelligent decision-making processing algorithm after optimizing the algorithm weight as the initial intelligent decision-making processing algorithm before the next round of debugging processing; the linkage processing task is a processing task whose process priority is less than the set process priority; the obtaining of a plurality of first sensor monitoring training data includes: obtaining a plurality of initial sensor monitoring training data; wherein the initial sensor monitoring training data includes a first processing task training label and a second processing task training label; screening at least part of the data set of each initial sensor monitoring training data according to the set process priority to obtain a plurality of corresponding first Sensor monitoring training data; inputting a plurality of first sensor monitoring training data into an initial intelligent decision processing algorithm to obtain each first device control training feature, including: for each first sensor monitoring training data, performing multiple rounds of knowledge vector mining processing on the first sensor monitoring training data through the initial intelligent decision processing algorithm to obtain a plurality of first sensor monitoring training knowledge vector spectra with a pyramid depth; performing derivative knowledge vector mining based on a plurality of first sensor monitoring training knowledge vector spectra to obtain a plurality of second sensor monitoring training knowledge vector spectra with a pyramid depth; determining the first device control training feature based on the second sensor monitoring training knowledge vector spectrum with a target pyramid depth.

9. A control system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 8.

Images