Self-adaptive control system and method of plasma generator

Abstract

The invention relates to the technical field of plasma generators, and particularly discloses a self-adaptive control system and method of a plasma generator, and the method comprises the steps: carrying out the feature extraction of a voltage time sequence data set, a current time sequence data set and a plasma density time sequence data set through employing a data analysis and coding technology based on deep learning, meanwhile, time sequence feature extraction is carried out on curves of voltage, current and plasma density which correspond to each control mode and change along with time; therefore, the optimal control mode of the target plasma generator is intelligently screened out according to the fast scanning semantic query representation between the multivariable time sequence implicit cooperation feature of the plasma generator and the multivariable time sequence implicit coding feature of each control mode. By means of the mode, the future parameter change trend can be predicted more accurately, the optimal control mode can be intelligently screened out, and compared with a traditional threshold value judgment method, early warning can be carried out in advance before the parameters are close to but do not exceed the safety range.

Description

Technical Field

[0001] The present application relates to the technical field of plasma generators, and more specifically, to an adaptive control system and method for a plasma generator. Background Art

[0002] Plasma generators are widely used in many fields such as industry, scientific research and medicine, such as semiconductor manufacturing, material processing, environmental purification, etc. However, plasma generation is a highly complex and dynamic process involving the simultaneous changes of multiple physical quantities (such as voltage, current, plasma density). In order to ensure that the plasma generator operates efficiently and stably in different application scenarios, it needs to be precisely controlled.

[0003] Patent CN117148710A discloses an adaptive plasma monitoring method and system. Specifically, multiple control modes are first set for the plasma generator to be monitored to form curves of voltage, current and plasma density over time in each mode. Then, the operating constraint range is set and monitored in real time. If the parameter exceeds the range, an abnormal command is issued. After receiving the command, the system selects the optimal control mode and switches to the mode within a preset time.

[0004] In this patent, based on the fact that the parameters at the current moment and in the next 2T period exceed the operating constraint range pre-set by the operator, an abnormal command is issued and the optimal control mode is screened. However, although the patent can predict future operations, it relies on the operating constraint range (i.e., threshold) pre-set by the operator. This method is relatively static and difficult to adapt to dynamically changing working environments or unforeseen situations. If the working conditions change, the preset threshold may no longer apply, resulting in the system being unable to accurately determine whether an abnormality has occurred. In addition, this method only issues an abnormal command and adjusts the control mode after detecting that the parameters exceed the preset threshold, which may cause a certain response lag, especially in rapidly changing or time-sensitive application scenarios. The best adjustment time may be missed, affecting the stability and performance of the system.

[0005] Therefore, an optimized adaptive control scheme for a plasma generator is desired. Summary of the invention

[0006] The present application provides an adaptive control system and method for a plasma generator, which can more accurately predict future parameter change trends and intelligently screen out the optimal control mode. Compared with the traditional threshold judgment method, it can provide early warning before the parameters are close to but have not exceeded the safety range, thereby ensuring that the equipment is always in the optimal working state.

[0007] In a first aspect, a method for adaptively controlling a plasma generator is provided, comprising:

[0008] Setting a corresponding set of control modes for a target plasma generator;

[0009] Forming a curve of voltage, current and plasma density changing with time corresponding to each control mode in the set of control modes;

[0010] monitoring the voltage, current and plasma density of the target plasma generator in real time to obtain a voltage time series data set, a current time series data set and a plasma density time series data set;

[0011] Based on the voltage time series data set, the current time series data set and the plasma density time series data set, the optimal control mode of the target plasma generator is screened, including: performing time series arrangement and sequence coding on the voltage time series data set, the current time series data set and the plasma density time series data set to obtain multivariate time series implicit synergistic features of the plasma generator; performing time series feature extraction on the curves of voltage, current and plasma density corresponding to each control mode in the set of control modes over time to obtain a set of multivariate time series implicit coding features of the control mode; performing fast scanning semantic query coding on the set of multivariate time series implicit synergistic features of the plasma generator and the multivariate time series implicit coding features of the control mode to obtain current state-control mode prototype query response semantic coding features; based on the current state-control mode prototype query response semantic coding features, the optimal control mode of the target plasma generator is obtained;

[0012] The optimal control mode is obtained, and the control mode of the target plasma generator is switched.

[0013] In a second aspect, an adaptive control system for a plasma generator is provided, comprising:

[0014] A control mode set setting module, used to set a corresponding control mode set for a target plasma generator;

[0015] A time variation curve forming module, used for forming a curve of voltage, current and plasma density varying with time corresponding to each control mode in the set of control modes;

[0016] A time series data set acquisition module, used for real-time monitoring of the voltage, current and plasma density of the target plasma generator to obtain a voltage time series data set, a current time series data set and a plasma density time series data set;

[0017] A control mode screening module, for screening the optimal control mode of the target plasma generator based on the voltage time series data set, the current time series data set and the plasma density time series data set, comprising: a multivariable time series implicit coordination coding unit, for performing time series arrangement and sequence coding on the voltage time series data set, the current time series data set and the plasma density time series data set to obtain a multivariable time series implicit coordination feature of the plasma generator; a time series feature extraction unit, for performing time series feature extraction on a curve of voltage, current and plasma density varying with time corresponding to each control mode in the set of control modes to obtain a set of multivariable time series implicit coding features of the control mode; a fast scanning semantic query coding unit, for performing fast scanning semantic query coding on a set of multivariable time series implicit coordination features of the plasma generator and multivariable time series implicit coding features of the control mode to obtain a current state-control mode prototype query response semantic coding feature; an optimal control mode determination unit, for obtaining the optimal control mode of the target plasma generator based on the current state-control mode prototype query response semantic coding feature;

[0018] The control mode switching module is used to obtain the optimal control mode and switch the control mode of the target plasma generator.

[0019] The present application provides an adaptive control system and method for a plasma generator, which uses data analysis and coding technology based on deep learning to perform time series arrangement and multivariate time series implicit feature extraction on voltage time series data sets, current time series data sets and plasma density time series data sets, and extracts time series features of the curves of voltage, current and plasma density corresponding to each control mode over time, so as to intelligently screen out the optimal control mode of the target plasma generator based on the fast scanning semantic query representation between the multivariate time series implicit synergistic features of the plasma generator and the multivariate time series implicit coding features of each control mode. In this way, the future parameter change trend can be predicted more accurately, and the optimal control mode can be intelligently screened out. Compared with the traditional threshold judgment method, it can give an early warning before the parameter is close but has not exceeded the safety range, thereby ensuring that the equipment is always in the optimal working state. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0021] Figure 1 It is a schematic flow chart of the adaptive control method of the plasma generator according to an embodiment of the present application.

[0022] Figure 2 This is a schematic diagram of data flow for selecting the optimal control mode of the target plasma generator based on the voltage time series data set, the current time series data set and the plasma density time series data set in the adaptive control method of the plasma generator of the embodiment of the present application.

[0023] Figure 3 A schematic flowchart of screening the optimal control mode of the target plasma generator based on the voltage time series data set, the current time series data set and the plasma density time series data set in the adaptive control method of the plasma generator of the embodiment of the present application.

[0024] Figure 4 In the adaptive control method of the plasma generator of the embodiment of the present application, the voltage time series data set, the current time series data set and the plasma density time series data set are arranged in time series and sequence encoded to obtain a schematic flow chart of the multivariable time series implicit synergy characteristics of the plasma generator.

[0025] Figure 5 In the adaptive control method of the plasma generator of the embodiment of the present application, a schematic flowchart is provided for extracting the time-varying curves of the voltage, current and plasma density corresponding to each control mode in the set of control modes to obtain a set of multivariate time-series implicit coding features of the control modes.

[0026] Figure 6 In the adaptive control method of the plasma generator of an embodiment of the present application, a schematic flowchart of quickly scanning the semantic query encoding of the set of multivariable timing implicit collaborative features of the plasma generator and the multivariable timing implicit coding features of the control mode to obtain the current state-control mode prototype query response semantic coding features.

[0027] Figure 7 It is a schematic block diagram of an adaptive control system of a plasma generator according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0029] Based on this, the present application proposes an adaptive control method for a plasma generator, such as Figure 1As shown, the adaptive control method of the plasma generator includes: S1, setting a set of corresponding control modes for the target plasma generator; S2, forming a curve of voltage, current and plasma density corresponding to each control mode in the set of control modes over time; S3, real-time monitoring of the voltage, current and plasma density of the target plasma generator to obtain a voltage time series data set, a current time series data set and a plasma density time series data set; S4, based on the voltage time series data set, the current time series data set and the plasma density time series data set, screening the optimal control mode of the target plasma generator; S5, obtaining the optimal control mode, and switching the control mode of the target plasma generator.

[0030] Exemplarily, in step S1, a set of corresponding control modes is set for the target plasma generator. It should be understood that, referring to the existing patent CN117148710A, setting a set of corresponding control modes for the target plasma generator means defining a set of possible control modes for the plasma generator of the target to be monitored. This step first requires identifying which parameters can be adjusted in real time during operation, and these adjustable parameters generally include voltage, current, and plasma density. Once it is determined which variables are controllable, it is necessary to set the specific value range or adjustment rules of these variables to ensure that all subsequent operations are carried out within a safe and effective range. Next, based on the above-mentioned different variable value combinations, a series of different control modes are formed. Each control mode represents a set of operating parameter configurations under specific conditions, such as a combination of a certain voltage level, current intensity, and plasma density. By creating a control mode set covering a variety of possibilities, a diverse selection basis can be provided for subsequent steps, so that its working state can be flexibly adjusted under different conditions to ensure optimal performance. In addition, such a setting also provides the necessary flexibility for automatic control, enabling it to dynamically select the most appropriate control mode based on real-time monitoring data.

[0031] Exemplarily, in step S2, a curve of voltage, current and plasma density corresponding to each control mode in the set of control modes is formed over time. It should be understood that in step S2, for each defined control mode, the plasma generator is started within a set operation monitoring cycle (e.g., 2T), and the change of voltage, current and plasma density over time in the cycle is recorded. These data will be stored as an online monitoring data group. The specific process involves the following aspects: First, for each control mode, the plasma generator is started under its specific operating conditions, and the change of voltage, current and plasma density is continuously monitored in a complete operation monitoring cycle. This means that from the start of the plasma generator to the end of this period, the data points of these three key parameters will be continuously collected to ensure that the complete time series information is captured. Then, in order to accurately describe the change trend of these parameters over time, three different change curve functions will be set for each control mode, corresponding to voltage, current and plasma density respectively. Each change curve function is intended to fit the actual observed data through a mathematical model to obtain the optimal coefficient set. For example, for voltage, there will be a first change curve function for calculating the voltage value A(t) at each moment t. This function depends on a series of voltage coefficients {K_i}, and the optimal voltage coefficient group is obtained by minimizing the error of the voltage calculation value As(t) at the corresponding time point in the online monitoring data group. The same method is also applied to current and plasma density, and the second change curve function and the third change curve function are respectively established to obtain the optimal current coefficient group {K_j} and plasma density coefficient group {K_z}. These coefficient sets represent the parameter combination that best reflects the law of change of actual physical quantities under a given control mode. For details, please refer to the existing patent CN117148710A. Finally, all these change curve functions and their corresponding optimal coefficient sets will be saved as characteristic representations of voltage, current and plasma density changes over time under different control modes. These curves not only reflect the dynamic behavior under each control mode, but also provide important reference basis for subsequent steps, so that the current state can be quickly evaluated during real-time operation, and the most suitable control mode can be selected or potential problems can be warned in advance.

[0032] Exemplarily, in step S3, the voltage, current and plasma density of the target plasma generator are monitored in real time to obtain a voltage time series data set, a current time series data set and a plasma density time series data set. It should be understood that the continuous monitoring of the operating plasma generator mainly focuses on three key parameters: voltage, current and plasma density. The changes in these parameters directly reflect the working state and performance of the plasma generator. Through real-time monitoring, a series of data points that change over time can be collected to form a voltage time series data set, a current time series data set and a plasma density time series data set. Specifically: through real-time monitoring, the above three physical quantities can be measured multiple times in a short time, and the accuracy and timeliness of the data can be ensured. This is particularly important for capturing rapidly changing processes, such as transient behavior or abnormal fluctuations during plasma generation. Secondly, for the data obtained at each time point, they not only represent the current state, but also provide a basis for subsequent analysis. For example, within a time period of 2T (where T is the set operation monitoring period), the specific values ​​of voltage, current and plasma density will be recorded, and these values ​​constitute a sample point within the time period. As time goes by, such sample points continue to accumulate and eventually form a complete time series data set. Furthermore, these time series data sets are not just simple historical records, they can also be used to evaluate whether the plasma generator is working as expected. By analyzing these data sets, any trends or patterns that deviate from the normal range can be identified, thereby providing early warning of potential problems. In addition, this data is also one of the important bases for screening the optimal control mode, because different control modes will show different characteristics on these parameters.

[0033] Exemplarily, in step S4, based on the voltage time series data set, the current time series data set and the plasma density time series data set, the optimal control mode of the target plasma generator is screened. It should be understood that in screening the optimal control mode of the target plasma generator based on the voltage time series data set, the current time series data set and the plasma density time series data set, the technical concept of the present application is to use data analysis and coding technology based on deep learning to perform time series arrangement and multivariate time series implicit feature extraction on the voltage time series data set, the current time series data set and the plasma density time series data set, and at the same time extract the time series feature of the curves of the voltage, current and plasma density corresponding to each control mode, so as to intelligently screen out the optimal control mode of the target plasma generator according to the fast scanning semantic query representation between the multivariate time series implicit synergistic features of the plasma generator and the multivariate time series implicit coding features of each control mode. In this way, the future parameter change trend can be predicted more accurately, and the optimal control mode can be intelligently screened out. Compared with the traditional threshold judgment method, it can give an early warning before the parameter is close but has not exceeded the safety range, thereby ensuring that the device is always in the optimal working state.

[0034] like Figure 2 and Figure 3 As shown, based on the voltage time series data set, the current time series data set and the plasma density time series data set, the optimal control mode of the target plasma generator is screened, including: S41, time-series arrangement and sequence encoding of the voltage time series data set, the current time series data set and the plasma density time series data set to obtain multi-variable time series implicit synergistic features of the plasma generator; S42, time-series feature extraction of the curves of voltage, current and plasma density corresponding to each control mode in the set of control modes over time to obtain a set of multi-variable time series implicit coding features of the control mode; S43, fast scanning semantic query coding of the set of multi-variable time series implicit synergistic features of the plasma generator and the multi-variable time series implicit coding features of the control mode to obtain current state-control mode prototype query response semantic coding features; S44, based on the current state-control mode prototype query response semantic coding features, the optimal control mode of the target plasma generator is obtained.

[0035] Exemplarily, in step S41, the voltage time series data set, the current time series data set and the plasma density time series data set are arranged in time series and sequence encoded to obtain the multivariate time series implicit synergistic characteristics of the plasma generator. It should be understood that the generation of plasma is a highly dynamic process involving the simultaneous changes of multiple physical quantities (such as voltage, current and plasma density). These parameters not only have significant dynamic characteristics in the time dimension, but also have complex correlations and influence relationships on the time scale. By arranging and sequence encoding these time series data sets, the dynamic behavior between these parameters and their change patterns over time can be more comprehensively understood and captured.

[0036] In one embodiment, Figure 4 As shown, the voltage time series data set, the current time series data set and the plasma density time series data set are arranged in time series and sequence encoded to obtain the multivariate time series implicit synergy feature of the plasma generator, including: S411, arranging the voltage time series data set, the current time series data set and the plasma density time series data set into a plasma generator multivariate time series synergy matrix according to the time dimension and the variable sample dimension; S412, inputting the plasma generator multivariate time series synergy matrix into a sequence encoder based on the LSTM-RNN hybrid model to obtain the plasma generator multivariate time series implicit synergy feature vector as the plasma generator multivariate time series implicit synergy feature.

[0037] Exemplarily, in step S411, the voltage time series data set, the current time series data set and the plasma density time series data set are arranged into a plasma generator multivariate time series synergy matrix according to the time dimension and the variable sample dimension. It should be understood that, considering that the generation of plasma is a highly dynamic process, it involves the simultaneous change of multiple physical quantities (voltage, current and plasma density), that is, the voltage time series data set, the current time series data set and the plasma density time series data set have dynamic characteristics of time series in the time dimension, and there are mutual correlations and influence relationships on the time scale between the various parameters. Based on this, in order to more comprehensively understand and capture the dynamic behavior between these parameters, in the technical solution of the present application, the voltage time series data set, the current time series data set and the plasma density time series data set are arranged into a plasma generator multivariate time series synergy matrix according to the time dimension and the variable sample dimension, so as to provide data support for better capturing the interaction between different variables and their change patterns over time in the future.

[0038] Exemplarily, in step S412, the plasma generator multivariable timing coordination matrix is ​​input into a sequence encoder based on a LSTM-RNN hybrid model to obtain a plasma generator multivariable timing implicit coordination feature vector as the plasma generator multivariable timing implicit coordination feature. It should be understood that in order to further explore and refine the timing change trend and the mutual influence relationship between various parameters in the plasma generator multivariable timing coordination matrix, in the technical solution of the present application, the plasma generator multivariable timing coordination matrix is ​​input into a sequence encoder based on a LSTM-RNN hybrid model to obtain a plasma generator multivariable timing implicit coordination feature vector. It should be understood that RNN is a neural network for processing sequence data, which is suitable for natural language processing, speech recognition, time series prediction and other fields. The characteristic of RNN is that it has a "memory" function, that is, it can use previous information in the sequence to affect the current output. LSTM is a special RNN variant designed to overcome the long-term dependency problem in traditional RNN. LSTM controls the flow of information by introducing a gating mechanism (input gate, forget gate, and output gate), allowing the network to selectively remember or forget certain information, and can perform better when processing longer time series, and can effectively capture dependencies within a long time span. Therefore, this application uses a LSTM-RNN hybrid model to capture long-term trends and periodic characteristics in the plasma generation process by using the LSTM part to handle complex dynamic changes over a long time span. And use the RNN part to supplement the processing of rapid changes or local patterns in the short term to capture finer time series details, such as instantaneous fluctuations or abnormal changes in a short period of time. In this way, the model's ability to understand and capture features on different time scales is improved, thereby providing more precise support for intelligent control systems.

[0039] Exemplarily, in step S42, the curves of the voltage, current and plasma density corresponding to each control mode in the set of control modes are subjected to time series feature extraction to obtain a set of multivariable time series implicit coding features of the control mode. It should be understood that, considering that each control mode also contains the parameter time series variation information of the voltage, current and plasma density under specific circumstances, therefore, in order to more accurately understand and analyze the time series characteristics of each parameter under each control mode, so as to quickly identify the control mode that best suits the current conditions, in the technical solution of the present application, the curves of the voltage, current and plasma density corresponding to each control mode in the set of control modes are subjected to time series feature extraction to obtain a set of multivariable time series implicit coding feature vectors of the control mode.

[0040] In particular, in a specific embodiment of the present application, first, the curves of the voltage, current and plasma density corresponding to each control mode in the set of control modes are arranged according to the time dimension to obtain a sequence of multiple control mode multivariate time series feature vectors. In other words, the curves of the parameters corresponding to each control mode over time are essentially one-dimensional time series data. Specifically, if the measured values ​​of voltage, current or plasma density at different time points are arranged in order, a sequence of vectors can be formed, each element in each vector represents a measured value at a time point, and the sequence of the entire vector describes the change of the physical quantity over time. Afterwards, in order to effectively capture the dynamic behavior between the timings under each control mode, the sequence of each control mode multivariate time series feature vector is input into the control mode timing feature extractor based on the gated cyclic unit to capture and extract the implicit correlation characteristics of the timings between different times in each control mode, and obtain a set of control mode multivariate time series implicit coding feature vectors as the set of control mode multivariate time series implicit coding features.

[0041] In one embodiment, Figure 5 As shown, the time-varying curves of voltage, current and plasma density corresponding to each control mode in the set of control modes are subjected to timing feature extraction to obtain a set of control mode multivariate timing implicit coding features, including: S421, arranging the time-varying curves of voltage, current and plasma density corresponding to each control mode in the set of control modes according to the time dimension to obtain a sequence of multiple control mode multivariate timing feature vectors; S422, inputting the sequence of each control mode multivariate timing feature vector in the sequence of the multiple control mode multivariate timing feature vectors into a control mode timing feature extractor based on a gated cyclic unit to obtain a set of control mode multivariate timing implicit coding feature vectors as the set of control mode multivariate timing implicit coding features.

[0042] Exemplarily, in step S43, the set of the plasma generator multivariable time series implicit synergistic features and the control mode multivariable time series implicit coding features are quickly scanned and semantically queried to obtain the current state-control mode prototype query response semantic coding features. It should be understood that, considering that the plasma generator multivariable time series implicit synergistic features represent the real-time state characteristics of the plasma generator under the current operating conditions. The set of control mode multivariable time series implicit coding features represents the comprehensive dynamic characteristics of a specific control mode over a period of time. Therefore, in order to extract the semantic information and features that best match the current plasma generator multivariable time series implicit synergistic features from the set of control mode multivariable time series implicit coding features, so as to screen out the optimal control mode, in the technical solution of the present application, the plasma generator multivariable time series implicit synergistic features and the control mode multivariable time series implicit coding features are quickly scanned and semantically queried to obtain the current state-control mode prototype query response semantic coding features.

[0043] In one embodiment, Figure 6 As shown, a set of the plasma generator multivariable timing implicit coordination feature and the control mode multivariable timing implicit coding feature are quickly scanned for semantic query coding to obtain a current state-control mode prototype query response semantic coding feature, including: S431, based on the plasma generator multivariable timing implicit coordination feature vector, a subset of the control mode multivariable timing implicit coding feature vectors that are quickly matched is determined from the set of the control mode multivariable timing implicit coding feature vectors; S432, a subset of the plasma generator multivariable timing implicit coordination feature vector and the control mode multivariable timing implicit coding feature vector are cross-domain query response coded to obtain a current state-control mode prototype query response semantic coding vector as the current state-control mode prototype query response semantic coding feature.

[0044] In one embodiment, in step S431, based on the plasma generator multivariable timing implicit synergistic feature vector, a subset of the control mode multivariable timing implicit coding feature vectors that are quickly matched is determined from the set of the control mode multivariable timing implicit coding feature vectors, including: calculating the mutual information between the plasma generator multivariable timing implicit synergistic feature vector and each control mode multivariable timing implicit coding feature vector in the set of the control mode multivariable timing implicit coding feature vectors to obtain a set of plasma generator-control mode multivariable fast semantic matching factors; Identify the first plasma generator-control mode multivariable semantic maximum and the second plasma generator-control mode multivariable semantic maximum from the set of generator-control mode multivariable fast semantic matching factors; based on the positions of the first plasma generator-control mode multivariable semantic maximum and the second plasma generator-control mode multivariable semantic maximum in the set of plasma generator-control mode multivariable fast semantic matching factors, determine a subset of the control mode multivariable temporal implicit coding feature vectors that are quickly matched from the set of control mode multivariable temporal implicit coding feature vectors. Specifically, the process can be expressed as follows:

[0045] V={v 21 ,v 22 ,...,v 2i ,...,v 2n}

[0046] x 1 =sigmoid(v 1 )

[0047] x 2i =sigmoid(v 2i )

[0048] MI i =H(x 1 )+H(x 2i )-H(x 1 ,x 2i )

[0049]

[0050] {k sub}={v 2arg max1 ,...,v 2argmax2}

[0051] Where V is the set of multivariable time series implicit coding feature vectors of the control mode, v 21 , v 22 , v 2i and v 2nare the first, second, i-th and n-th control mode multivariable time series implicit coding feature vectors in the set of control mode multivariable time series implicit coding feature vectors, respectively, v 1 is the implicit synergistic eigenvector of the plasma generator multivariate time series, sigmoid(·) is the sigmoid function, x 1 is the plasma generator multivariate time series transformation feature vector, x 2i Yes 2i The corresponding control mode multivariable time series transformation feature vector, H(x 1 ) represents x 1 The information entropy, H(x 2i ) represents x 2i The information entropy, H(x 1 ,x 2i ) represents x 1 and x 2i The joint entropy between i Represents x 1 and x 2i Plasma generator-control mode multivariable fast semantic matching factor between, is a set of plasma generator-control mode multivariable fast semantic matching factors, n is the number of plasma generator-control mode multivariable fast semantic matching factors in the set of plasma generator-control mode multivariable fast semantic matching factors, max(·) is the maximum value in the set, MI max1 is the first plasma generator-control mode multivariable semantic maximum, MI max2 is the second plasma generator-control mode multivariate semantic maximum value, arg max(·) is the position of the maximum value in the set, arg max1 is the first determined fast matching position, argmax2 is the second determined fast matching position, v 2arg max1 The multivariate temporal implicit coding feature vector of the control mode that first determines the fast matching position, v 2arg max2 The multivariate temporal implicit coding feature vector of the control mode for the second fast matching position is represented by {k sub} is a subset of the control mode multivariate time series implicit encoding feature vectors.

[0052] In one embodiment, identifying a first plasma generator-control mode multivariable semantic maximum and a second plasma generator-control mode multivariable semantic maximum from the set of plasma generator-control mode multivariable fast semantic matching factors includes: extracting the maximum plasma generator-control mode multivariable fast semantic matching factor from the set of plasma generator-control mode multivariable fast semantic matching factors as the first plasma generator-control mode multivariable semantic maximum; dividing the set of plasma generator-control mode multivariable fast semantic matching factors by the first plasma generator-control mode multivariable semantic maximum by position to obtain a set of plasma generator-control mode multivariable fast semantic matching coefficients; extracting the maximum plasma generator-control mode multivariable fast semantic matching coefficient from the set of plasma generator-control mode multivariable fast semantic matching coefficients as the second plasma generator-control mode multivariable semantic maximum.

[0053] That is, the real-time operating state of the current plasma generator (i.e., the multivariate time series implicit synergistic feature vector) is first compared with the historical data under various predefined control modes (i.e., the control mode multivariate time series implicit encoding feature vector). Mutual information is a statistical measure used to measure the amount of information shared between two random variables. In this scenario, it is used to evaluate the similarity between the time series characteristics of voltage, current, and plasma density under the current operating state and the corresponding characteristics under each control mode. For example, in the semiconductor manufacturing process, if the voltage fluctuation of the current plasma generator is large, and a specific control mode also shows a similar voltage fluctuation pattern in the past records, the mutual information value between the two states is high. This means that the control mode may be more suitable for dealing with problems arising under the current conditions. In this way, the system can generate a series of "plasma generator-control mode multivariate fast semantic matching factors", which reflect the correlation or matching degree between different control modes and the current state. Next, the system needs to find the largest values ​​from all the calculated matching factors, and the first maximum value and the second maximum value are particularly pointed out here. Considering that plasma generation is a complex process, there may be multiple different control modes that can better adapt to the current conditions. Therefore, retaining the second highest matching factor can help further analysis and ensure that other potential best options are not missed. Finally, according to the position of the previously selected maximum matching factor, the system locates the corresponding control mode and forms a subset of control modes that are most likely to become the best choice.

[0054] In one embodiment, in step S432, a subset of the plasma generator multivariable timing implicit synergy feature vector and the control mode multivariable timing implicit coding feature vector is subjected to cross-domain query response coding to obtain a current state-control mode prototype query response semantic coding vector, including: performing a linear transformation on the plasma generator multivariable timing implicit synergy feature vector to obtain a plasma generator multivariable timing implicit synergy query vector and a plasma generator multivariable timing implicit synergy value vector, and using a subset of the control mode multivariable timing implicit coding feature vector as a subset of the key vector, performing converter-based query coding on the query vector, the value vector and the subset of the key vector to obtain the current state-control mode prototype query response semantic coding vector. Specifically, the process can be expressed by the formula:

[0055] v q =v 1 W q +b q

[0056] v v =v 1 W v +b v

[0057]

[0058] Among them, W q and b q are the query embedding matrix and the query bias vector, respectively, v q is the plasma generator multivariate time series implicit collaborative query vector, W v and b v are the value embedding matrix and value bias vector, respectively, v v is the plasma generator multivariate time series implicit synergy value vector, v 2j T Yes 2j The transposed vector of v 2j The length of , softmax(·) is a normalized exponential function, is the matrix multiplication, v r is the current state-control mode prototype query response semantic encoding vector.

[0059] That is, firstly, the plasma generator multivariate time series implicit synergistic feature vector is linearly transformed to generate the plasma generator multivariate time series implicit synergistic query vector and value vector. This linear transformation can be understood as a mapping mechanism that simplifies complex time series data while retaining important dynamic information. The query vector represents the characteristics of the voltage, current and plasma density changing with time under the current plasma generator operating state, which is used to "query" or find the most similar control mode; while the value vector contains the specific numerical information of these characteristics, providing a substantial reference basis in the subsequent matching process. For example, in the semiconductor manufacturing process, if the operating parameters of the current plasma generator show a certain time series pattern, then the query vector reflects the structural characteristics of this pattern, and the value vector records the specific parameter values. These two vectors together constitute a complete perspective describing the current operating state. Next, the system selects a part of the pre-defined control mode multivariate time series implicit encoding feature vector set as a subset of the key vector. The key vector here refers to the feature representation of those control modes that have been identified as highly relevant to the current operating state. This step is chosen to narrow the search scope and focus on the control modes that are most likely to be the best options, thereby improving efficiency and ensuring accuracy. After the previous mutual information calculation, several control modes that best match the current operating state have been found. Now, the system extracts the feature vectors corresponding to these modes to form a smaller but very representative subset for detailed comparison in the next step. Finally, the system uses transformer-based query encoding technology, combined with query vectors, value vectors, and key vector subsets, to obtain a comprehensive current state-control mode prototype query response semantic encoding vector.

[0060] In summary, by calculating the mutual information between the implicit synergistic features of the plasma generator multivariable timing and the implicit coding features of the control mode multivariable timing, the module can quickly evaluate the semantic similarity between different feature vectors. As a statistic, mutual information can quantify the degree of information sharing between two feature vectors, so as to efficiently screen out the most relevant candidate vectors. Identifying the first maximum value and the second maximum value from the mutual information sequence further narrows the search range and concentrates on processing the most relevant feature vectors. This method not only improves the processing speed, but also reduces the waste of computing resources. Through mutual information calculation and maximum value recognition, the fast scanning semantic query coding can flexibly select the feature vector subsequence that is most relevant to the query vector.

[0061] Exemplarily, in step S44, based on the semantic coding features of the current state-control mode prototype query response, the optimal control mode of the target plasma generator is obtained. In one embodiment, based on the semantic coding features of the current state-control mode prototype query response, the optimal control mode of the target plasma generator is obtained, including: inputting the semantic coding vector of the current state-control mode prototype query response into the optimal control mode filter based on the classifier to obtain the optimal control mode of the target plasma generator. That is, the semantic coding features of the current state-control mode prototype query response obtained by fast scanning and querying the set of the multivariable temporal implicit synergistic feature vector of the plasma generator and the multivariable temporal implicit coding feature vector of the control mode are classified and processed, so as to intelligently screen out the optimal control mode of the target plasma generator. In this way, the future parameter change trend can be predicted more accurately, and the optimal control mode can be intelligently screened out. Compared with the traditional threshold judgment method, it can give an early warning before the parameter is close but has not exceeded the safety range, thereby ensuring that the equipment is always in the optimal working state.

[0062] In one embodiment, the classifier-based optimal control mode filter is a Softmax classifier, which works by converting the score corresponding to each control mode into a probability distribution. For a given input feature (i.e., the current state-control mode prototype query response semantic encoding vector), the model calculates the probability value of each control mode. These probability values ​​add up to 1, and the control mode corresponding to the highest probability is considered to be the best choice. The Softmax classifier is a neural network model including a fully connected layer and a softmax activation function. The last layer of this model should have the same number of output units as the control modes, and each unit represents the score of a specific control mode. When a new current state-control mode prototype query response semantic encoding vector arrives, it will be immediately sent to the softmax classifier for prediction. The model outputs the probability distribution of each possible control mode, and the one with the highest probability is selected as the final decision.

[0063] In a preferred embodiment, inputting the current state-control mode prototype query response semantic encoding vector into a classifier-based optimal control mode filter to obtain the optimal control mode of the target plasma generator includes:

[0064] The feature set of the current state-control mode prototype query response semantic encoding vector is subjected to feature clustering to obtain a feature set within the current state-control mode prototype query response semantic encoding class and a feature set outside the current state-control mode prototype query response semantic encoding class, namely:

[0065]

[0066] in, represents the current state-control mode prototype query response semantic encoding class feature set, f 1i represents the i-th feature value of the feature set within the current state-control mode prototype query response semantic encoding class, f 2j Represents the jth feature value of the out-of-class feature set of the current state-control mode prototype query response semantic encoding;

[0067] The ratio of the number of feature values ​​in the feature set of the semantic coding class of the current state-control mode prototype query response to the number of feature values ​​in the feature set of the semantic coding vector of the current state-control mode prototype query response is calculated, that is:

[0068]

[0069] Wherein, k represents the number of feature values ​​in the feature set of the current state-control mode prototype query response semantic encoding class, n 1 represents the number of feature values ​​of the feature set of the current state-control mode prototype query response semantic encoding vector, and λ represents the ratio;

[0070] Calculate the ratio of the λth power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the current state-control mode prototype query response semantic encoding class to the λth power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the current state-control mode prototype query response semantic encoding vector to obtain the first current state-control mode prototype query response semantic encoding weight, that is:

[0071]

[0072] Among them, f p represents each feature value in the feature set of the current state-control mode prototype query response semantic encoding vector, w 1 represents the semantic encoding weight of the first current state-control mode prototype query response;

[0073] Calculate the ratio of the λ / 2 power of the sum of squares of all eigenvalues ​​in the feature set of the current state-control mode prototype query response semantic encoding class to the λ / 2 power of the sum of squares of all eigenvalues ​​in the feature set of the current state-control mode prototype query response semantic encoding vector to obtain the second current state-control mode prototype query response semantic encoding weight, that is:

[0074]

[0075] Among them, w 2Represents the second current state-control mode prototype query response semantic encoding weight;

[0076] For each feature value in the feature set within the semantic encoding class of the current state-control mode prototype query response, calculate the product of the feature value and the second current state-control mode prototype query response semantic encoding weight and add the first current state-control mode prototype query response semantic encoding weight to obtain the optimized feature value, that is:

[0077] f′ 1i =w 2 ×f 1i +w 1

[0078] Among them, f' 1i represents the i-th optimized feature value of the feature set within the semantic encoding class based on the current state-control mode prototype query response;

[0079] For each feature value in the feature set outside the semantic encoding class of the current state-control mode prototype query response, the product of the feature value and the weight of the semantic encoding of the first current state-control mode prototype query response is calculated to obtain the optimized feature value, that is:

[0080] f 2j '=w 1 ×f 2j

[0081] Among them, f 2j ' represents the jth optimized feature value of the out-of-class feature set based on the current state-control mode prototype query response semantic encoding;

[0082] The optimized current state-control mode prototype query response semantic coding vector based on the optimized feature values ​​of the feature set within the current state-control mode prototype query response semantic coding class and the feature set outside the current state-control mode prototype query response semantic coding class is input into the classifier-based optimal control mode filter to obtain the optimal control mode of the target plasma generator.

[0083] Here, when the set of the plasma generator multivariate timing implicit collaborative feature vector and the set of the control mode multivariate timing implicit coding feature vector respectively represent the plasma generator multivariate timing collaborative feature and the plasma generator multivariate timing implicit coding feature under different control modes, when performing semantic query encoding based on the fast scanning mechanism, the differences in fast scanning semantic queryability caused by the differences in the timing distribution characteristics of the source data are taken into account, and it is expected to improve the semantically consistent aggregation expression effect of the current state-control mode prototype query response semantic coding vector obtained by response semantic encoding.

[0084] Therefore, while the current state-control mode prototype query response semantic coding vector is feature clustered, the interaction description of the key feature information of the current state-control mode prototype query response semantic coding vector in the clustering process is carried out, and the geometric equivariant topology of the feature is constructed by low-rank harmonic modulation of the clustering features of the current state-control mode prototype query response semantic coding vector and the equivariance of the feature as a whole, so as to obtain the translation and rotation symmetry of the schematic distribution of the clustering features of the current state-control mode prototype query response semantic coding vector relative to the feature as a whole, thereby introducing geometric message passing into the feature expression of the current state-control mode prototype query response semantic coding vector and realizing the cluster mapping symmetry of the current state-control mode prototype query response semantic coding vector through irreducible low-rank coefficient manipulation, thereby improving the consistency of the clustering-based feature representation of the current state-control mode prototype query response semantic coding vector, thereby improving the accuracy of the optimal control mode of the target plasma generator obtained by inputting the current state-control mode prototype query response semantic coding vector into the classifier-based optimal control mode filter.

[0085] Exemplarily, in step S5, the optimal control mode is obtained, and the control mode of the target plasma generator is switched. It should be understood that once the optimal control mode is determined, the system needs to perform the actual control mode switch. This means adjusting the operating parameters of the plasma generator to the values ​​specified by the newly selected mode. For example, if the optimal mode requires a higher voltage level or a different current intensity, the control system will change these settings accordingly. This process should be completed as quickly and smoothly as possible to reduce the impact on the plasma generation process and ensure the stability and safety of the system during the switch.

[0086] In summary, the adaptive control method of the plasma generator according to the embodiment of the present application is explained, which uses data analysis and coding technology based on deep learning to perform time series arrangement and multivariate time series implicit feature extraction on the voltage time series data set, the current time series data set and the plasma density time series data set, and at the same time extracts the time series features of the curves of the voltage, current and plasma density corresponding to each control mode, so as to intelligently screen out the optimal control mode of the target plasma generator according to the fast scanning semantic query representation between the multivariate time series implicit synergistic features of the plasma generator and the multivariate time series implicit coding features of each control mode. In this way, it is possible to more accurately predict the future parameter change trend and intelligently screen out the optimal control mode. Compared with the traditional threshold judgment method, it can give an early warning before the parameter is close but has not exceeded the safety range, thereby ensuring that the equipment is always in the optimal working state.

[0087] Figure 7Schematic block diagram of the adaptive control system of the plasma generator according to the embodiment of the present application. Figure 7 As shown, the adaptive control system 100 of the plasma generator includes: a control mode set setting module 110, which is used to set a corresponding control mode set for the target plasma generator; a time change curve forming module 120, which is used to form a curve of voltage, current and plasma density corresponding to each control mode in the control mode set over time; a time series data set acquisition module 130, which is used to monitor the voltage, current and plasma density of the target plasma generator in real time to obtain a voltage time series data set, a current time series data set and a plasma density time series data set; a control mode screening module 140, which is used to screen the optimal control mode of the target plasma generator based on the voltage time series data set, the current time series data set and the plasma density time series data set; a control mode switching module 150, which is used to obtain the optimal control mode and switch the control mode of the target plasma generator.

[0088] In one embodiment, the control mode screening module includes: a multivariable timing implicit coordination coding unit, which is used to perform timing arrangement and sequence coding on the voltage timing data set, the current timing data set and the plasma density timing data set to obtain the multivariable timing implicit coordination characteristics of the plasma generator; a timing feature extraction unit, which is used to perform timing feature extraction on the curves of voltage, current and plasma density changing with time corresponding to each control mode in the set of control modes to obtain a set of control mode multivariable timing implicit coding characteristics; a fast scanning semantic query coding unit, which is used to perform fast scanning semantic query coding on the set of the plasma generator multivariable timing implicit coordination characteristics and the control mode multivariable timing implicit coding characteristics to obtain the current state-control mode prototype query response semantic coding characteristics; an optimal control mode determination unit, which is used to obtain the optimal control mode of the target plasma generator based on the current state-control mode prototype query response semantic coding characteristics.

[0089] In one embodiment, the multivariate timing implicit synergy encoding unit is used to: arrange the voltage timing data set, the current timing data set, and the plasma density timing data set into a plasma generator multivariate timing synergy matrix according to the time dimension and the variable sample dimension; input the plasma generator multivariate timing synergy matrix into a sequence encoder based on the LSTM-RNN hybrid model to obtain a plasma generator multivariate timing implicit synergy feature vector as the plasma generator multivariate timing implicit synergy feature.

[0090] Here, those skilled in the art will appreciate that the specific operations of the various modules and units in the adaptive control system of the plasma generator described above have been described in detail in the referenced description. Figures 1 to 6 The adaptive control method of the plasma generator has been described in detail, and therefore, its repeated description will be omitted.

[0091] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0092] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0094] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0095] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An adaptive control method for a plasma generator, characterized in that: include: Setting a corresponding set of control modes for a target plasma generator; Forming a curve of voltage, current and plasma density changing with time corresponding to each control mode in the set of control modes; monitoring the voltage, current and plasma density of the target plasma generator in real time to obtain a voltage time series data set, a current time series data set and a plasma density time series data set; Based on the voltage time series data set, the current time series data set and the plasma density time series data set, the optimal control mode of the target plasma generator is screened, including: performing time series arrangement and sequence coding on the voltage time series data set, the current time series data set and the plasma density time series data set to obtain multivariate time series implicit synergistic features of the plasma generator; performing time series feature extraction on the curves of voltage, current and plasma density corresponding to each control mode in the set of control modes over time to obtain a set of multivariate time series implicit coding features of the control mode; performing fast scanning semantic query coding on the set of multivariate time series implicit synergistic features of the plasma generator and the multivariate time series implicit coding features of the control mode to obtain current state-control mode prototype query response semantic coding features; based on the current state-control mode prototype query response semantic coding features, the optimal control mode of the target plasma generator is obtained; The optimal control mode is obtained, and the control mode of the target plasma generator is switched.

2. The adaptive control method of a plasma generator according to claim 1, characterized in that: The voltage time series data set, the current time series data set and the plasma density time series data set are time-series arranged and sequence-coded to obtain multivariate time series implicit synergistic features of the plasma generator, including: Arranging the voltage time series data set, the current time series data set and the plasma density time series data set into a plasma generator multivariate time series coordination matrix according to the time dimension and the variable sample dimension; The plasma generator multivariate time series synergy matrix is ​​input into a sequence encoder based on an LSTM-RNN hybrid model to obtain a plasma generator multivariate time series implicit synergy feature vector as the plasma generator multivariate time series implicit synergy feature.

3. The adaptive control method for a plasma generator according to claim 2, characterized in that: The time series feature extraction is performed on the curves of voltage, current and plasma density changing with time corresponding to each control mode in the set of control modes to obtain a set of multivariate time series implicit coding features of the control mode, including: Arranging the curves of voltage, current and plasma density varying with time corresponding to each control mode in the set of control modes according to the time dimension to obtain a sequence of multivariate time series feature vectors of multiple control modes; Each sequence of control mode multivariable timing feature vectors in the sequence of the multiple control mode multivariable timing feature vectors is input into a control mode timing feature extractor based on a gated cyclic unit to obtain a set of control mode multivariable timing implicit coding feature vectors as the set of control mode multivariable timing implicit coding features.

4. The adaptive control method for a plasma generator according to claim 3, characterized in that: The plasma generator multivariable time series implicit synergy feature and the control mode multivariable time series implicit coding feature are quickly scanned for semantic query coding to obtain the current state-control mode prototype query response semantic coding feature, including: Based on the plasma generator multivariable timing implicit synergistic feature vector, determining a subset of the control mode multivariable timing implicit coding feature vectors for fast matching from the set of the control mode multivariable timing implicit coding feature vectors; A subset of the plasma generator multivariate timing implicit collaborative feature vector and the control mode multivariate timing implicit coding feature vector is cross-domain query response coded to obtain a current state-control mode prototype query response semantic coding vector as the current state-control mode prototype query response semantic coding feature.

5. The adaptive control method for a plasma generator according to claim 4, characterized in that: Based on the plasma generator multivariable time series implicit synergistic feature vector, a subset of the control mode multivariable time series implicit coding feature vectors for fast matching is determined from the set of the control mode multivariable time series implicit coding feature vectors, including: Calculating the mutual information between the plasma generator multivariable time series implicit synergy feature vector and each control mode multivariable time series implicit coding feature vector in the set of the control mode multivariable time series implicit coding feature vector to obtain a set of plasma generator-control mode multivariable fast semantic matching factors; identifying a first plasma generator-control mode multivariate semantic maximum and a second plasma generator-control mode multivariate semantic maximum from the set of plasma generator-control mode multivariate fast semantic matching factors; Based on the positions of the first plasma generator-control mode multivariable semantic maximum and the second plasma generator-control mode multivariable semantic maximum in the set of plasma generator-control mode multivariable fast semantic matching factors, a subset of the control mode multivariable timing implicit coding feature vectors for fast matching is determined from the set of the control mode multivariable timing implicit coding feature vectors.

6. The adaptive control method for a plasma generator according to claim 5, characterized in that: Identifying a first plasma generator-control mode multivariate semantic maximum and a second plasma generator-control mode multivariate semantic maximum from the set of plasma generator-control mode multivariate fast semantic matching factors, comprising: Extracting the maximum plasma generator-control mode multivariable fast semantic matching factor from the set of plasma generator-control mode multivariable fast semantic matching factors as the first plasma generator-control mode multivariable semantic maximum value; Dividing the set of plasma generator-control mode multivariable fast semantic matching factors by the first plasma generator-control mode multivariable semantic maximum by position to obtain a set of plasma generator-control mode multivariable fast semantic matching coefficients; The maximum plasma generator-control mode multivariable fast semantic matching coefficient is extracted from the set of the plasma generator-control mode multivariable fast semantic matching coefficients as the second plasma generator-control mode multivariable semantic maximum value.

7. The adaptive control method for a plasma generator according to claim 6, characterized in that: A cross-domain query response encoding is performed on a subset of the plasma generator multivariable timing implicit coordination feature vector and the control mode multivariable timing implicit coding feature vector to obtain a current state-control mode prototype query response semantic coding vector, including: linearly transforming the plasma generator multivariable timing implicit coordination feature vector to obtain a plasma generator multivariable timing implicit coordination query vector and a plasma generator multivariable timing implicit coordination value vector, and using a subset of the control mode multivariable timing implicit coding feature vector as a subset of the key vector, and performing converter-based query encoding on the query vector, the value vector and the subset of the key vector to obtain the current state-control mode prototype query response semantic coding vector.

8. The adaptive control method for a plasma generator according to claim 7, characterized in that: Based on the semantic coding features of the current state-control mode prototype query response, the optimal control mode of the target plasma generator is obtained, including: inputting the current state-control mode prototype query response semantic coding vector into the classifier-based optimal control mode filter to obtain the optimal control mode of the target plasma generator.

9. An adaptive control system for a plasma generator, characterized in that: include: A control mode set setting module, used to set a corresponding control mode set for a target plasma generator; A time variation curve forming module, used for forming a curve of voltage, current and plasma density varying with time corresponding to each control mode in the set of control modes; A time series data set acquisition module, used for real-time monitoring of the voltage, current and plasma density of the target plasma generator to obtain a voltage time series data set, a current time series data set and a plasma density time series data set; A control mode screening module, for screening the optimal control mode of the target plasma generator based on the voltage time series data set, the current time series data set and the plasma density time series data set, comprising: a multivariable time series implicit coordination coding unit, for performing time series arrangement and sequence coding on the voltage time series data set, the current time series data set and the plasma density time series data set to obtain a multivariable time series implicit coordination feature of the plasma generator; a time series feature extraction unit, for performing time series feature extraction on a curve of voltage, current and plasma density varying with time corresponding to each control mode in the set of control modes to obtain a set of multivariable time series implicit coding features of the control mode; a fast scanning semantic query coding unit, for performing fast scanning semantic query coding on a set of multivariable time series implicit coordination features of the plasma generator and multivariable time series implicit coding features of the control mode to obtain a current state-control mode prototype query response semantic coding feature; an optimal control mode determination unit, for obtaining the optimal control mode of the target plasma generator based on the current state-control mode prototype query response semantic coding feature; The control mode switching module is used to obtain the optimal control mode and switch the control mode of the target plasma generator.

10. The adaptive control system of the plasma generator according to claim 9, characterized in that: The multivariate time series implicit collaborative coding unit is used to: Arranging the voltage time series data set, the current time series data set and the plasma density time series data set into a plasma generator multivariate time series coordination matrix according to the time dimension and the variable sample dimension; The plasma generator multivariate time series synergy matrix is ​​input into a sequence encoder based on an LSTM-RNN hybrid model to obtain a plasma generator multivariate time series implicit synergy feature vector as the plasma generator multivariate time series implicit synergy feature.

Images