An intelligent control method and system for a vacuum system based on artificial intelligence

Abstract

The application discloses an intelligent control method and system for a vacuum system based on artificial intelligence, and relates to the technical field of automatic control.The method comprises the following steps: collecting a multi-source operation data sequence in real time to form a multi-dimensional time sequence data set; inputting the multi-dimensional time sequence data set into a vacuum load prediction model for prediction to obtain a corresponding predicted vacuum degree vector; taking the minimization of vacuum degree fluctuation, energy consumption and cross residues as a target, constructing a fitness function, and based on the predicted vacuum degree vector, optimizing in a feasible solution space by using an improved dragonfly algorithm to generate an optimal control decision of the vacuum system; and issuing the optimal control decision to an execution bottom layer of the vacuum system, and coupling a microsecond-level adaptive PID method to perform residue compensation according to real-time vacuum state data collected in real time to complete physical execution and closed loop of the intelligent control of the vacuum system.The application solves the problems of vacuum control lag, high energy consumption and cross residues in the prior art.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, and in particular to an intelligent control method and system for a vacuum system based on artificial intelligence. Background Technology

[0002] Liquid chromatography-triple quadrupole mass spectrometry (LC-MS / MS) is a core analytical instrument in fields such as environmental monitoring, clinical testing, and food safety. The vacuum system is fundamental to the normal operation of the mass spectrometer, and its stability directly determines ion transmission efficiency and signal-to-noise ratio.

[0003] In existing vacuum system control technologies, stepped control based on preset tables or traditional single-loop proportional-integral-derivative (PID) control methods are commonly used. However, in actual operation of liquid chromatography-triple quadrupole mass spectrometry (LC-MS / MS), gradient elution in the liquid phase causes drastic and complex changes in the gas load entering the ion source; simultaneously, mass spectrometry scanning modes (such as polarity switching and multi-reaction monitoring channel switching) can cause transient abrupt changes in the vacuum environment. Traditional control methods have the following significant drawbacks:

[0004] 1) Severe lag: Traditional PID relies on post-event feedback. There is a delay of milliseconds or even seconds from the vacuum degree deviating from the set value to the adjustment action taking effect. It cannot cope with the microsecond-level gas load change when the polarity switches, resulting in drastic fluctuations in vacuum degree.

[0005] 2) Energy consumption versus lifespan contradiction: In order to maintain the vacuum level under extreme working conditions, molecular pumps and backing pumps are usually operated at full load, resulting in huge energy waste and wear and tear on mechanical parts.

[0006] 3) Cross-contamination problem: If the flow guide valve is not controlled accurately when the solvent gradient changes or the scanning mode is switched, the residual gas in the chamber may desorb or backflow, resulting in serious cross-residue and reducing the accuracy of trace analysis.

[0007] Therefore, there is an urgent need for an intelligent control method for vacuum systems that can combine front-end chromatographic and mass spectrometric states for feedforward prediction and has microsecond-level high dynamic response capability. Summary of the Invention

[0008] This invention provides an intelligent control method and system for vacuum systems based on artificial intelligence, which solves the problems of vacuum control lag, high energy consumption and easy generation of cross residues in the prior art.

[0009] In a first aspect, embodiments of the present invention provide an intelligent control method for a vacuum system based on artificial intelligence, the method comprising:

[0010] Real-time acquisition of multi-source operational data sequences, including the vacuum system, from the liquid chromatography-triple quadrupole mass spectrometer constitutes a corresponding multidimensional time-series dataset.

[0011] The multidimensional time series dataset is input into the vacuum load prediction model built based on deep learning for prediction, and the corresponding predicted vacuum degree vector is obtained.

[0012] With the goal of minimizing vacuum fluctuations, energy consumption, and cross-residuals, a fitness function is constructed. Based on the predicted vacuum vector, an improved dragonfly algorithm is used to find the optimal solution space and generate the optimal control decision for the vacuum system.

[0013] The optimal control decision is sent down to the execution layer of the vacuum system, and residual compensation is performed by coupling a microsecond-level adaptive PID method based on the real-time vacuum state data, thus completing the physical execution and closed loop of the intelligent control of the vacuum system.

[0014] The technical solution provided in this application has at least the following beneficial effects:

[0015] By constructing an STS-MHA-LSTM vacuum load prediction model, the solvent changes (spatial characteristic flow) of liquid chromatography and the scanning action (temporal dynamic flow) of mass spectrometry are decoupled and fused for prediction, achieving feedforward prediction of vacuum change trends. Combined with microsecond-level timer interrupts of field-programmable gate arrays and adaptive PID with phase determination, a perfect closed loop of "predictive feedforward + microsecond-level residual feedback" is achieved, completely eliminating vacuum fluctuations caused by polarity switching and solving the control lag problem. A fitness function incorporating vacuum fluctuations, energy consumption, and cross-residue is constructed, enabling control decisions to ensure vacuum while considering energy consumption and residue suppression. Precise truncation and smooth transition of gas flow paths are achieved at the physical execution layer, significantly reducing gas desorption backflow on the inner wall of the cavity. In the Dragonfly algorithm improvement, a PSO mechanism is introduced, directly using the cognitive and social acceleration terms of PSO as position. Offset injection avoids step size oscillations and divergences caused by traditional dual velocity vectors. Combined with adaptive adjustment of the convergence factor, it greatly improves the convergence speed and robustness of global optimization. By improving the dragonfly algorithm, it finds the lowest energy consumption action combination that meets the vacuum requirements, avoiding the "one-size-fits-all" high-speed full-load operation in traditional methods, significantly reducing energy consumption and extending equipment life. For the mechanical delay of the guide valve, the feedforward buffer register is preloaded using the polarity switching timestamp, realizing "adjustment before movement". For residual compensation, a nonlinear adaptive PID based on the error phase judgment mechanism is proposed. The integral is locked in the abrupt phase and a nonlinear damping surge factor is introduced, so that the damping force is proportionally amplified as the absolute value of the error increases, effectively suppressing overshoot and oscillation caused by polarity switching. The steady-state phase is restored by the fuzzy PID to eliminate drift, realizing microsecond-level precise locking of vacuum under all operating conditions.

[0016] In one optional implementation, the multi-source operational data sequence includes a liquid chromatography data sequence from a liquid chromatography system, a mass spectrometry scanning data sequence from a mass spectrometry scanning system, and a vacuum status data sequence from a vacuum system.

[0017] The data at each moment in the liquid chromatography data sequence includes the high-pressure pump flow rate, the organic phase ratio, and the column oven temperature.

[0018] The data at each moment in the mass spectrometry scan data sequence includes the target m / z range span, scan speed, polarity reversal trigger signal, and dwell time.

[0019] The data at each moment in the vacuum state data sequence includes the measured vacuum level of the ion source chamber, the measured vacuum level of the analyzer chamber, the current of the frequency converter pre-pump, and the speed of the molecular pump.

[0020] In one optional implementation, the vacuum load prediction model is constructed based on the STS-MHA-LSTM algorithm. The vacuum load prediction model includes an input layer, a spatial flow processing branch constructed based on the GRU algorithm, a temporal flow processing branch constructed based on the LSTM algorithm, a multi-head self-attention layer constructed based on the Attention mechanism, a fully connected layer, and an output layer.

[0021] In one alternative implementation, a multi-dimensional time-series dataset is input into a vacuum load prediction model built based on deep learning for prediction, resulting in the corresponding predicted vacuum level, including:

[0022] The multidimensional time series dataset is input into the input layer of the vacuum load prediction model built based on deep learning, and the multidimensional time series dataset is decoupled according to physical meaning, and divided into spatial feature flow and temporal dynamic flow.

[0023] The spatial state hidden layer vector of the spatial feature flow is extracted using the spatial flow processing branch of the vacuum load prediction model, and the temporal dynamic hidden layer vector of the temporal dynamic flow is extracted using the temporal flow processing branch of the vacuum load prediction model.

[0024] The spatial state hidden layer vector and the temporal dynamic hidden layer vector are concatenated, and the resulting concatenated vector is input into the multi-head self-attention layer. Based on the attention weight matrix calculated by the multi-head attention mechanism, the concatenated vector is weighted and fused to obtain a high-order feature matrix.

[0025] The high-order feature matrix is ​​flattened, passed through a fully connected layer and an activation function, and then the output layer is used to output the predicted vacuum vector for the future period.

[0026] In one alternative implementation, the spatial feature stream is composed of the high-pressure pump flow rate, the organic phase ratio, the column oven temperature, the target m / z range span, and the polarity conversion trigger signal at each moment in the multidimensional time-series dataset.

[0027] The time dynamic stream is composed of the scan rate, the dwell time, the measured vacuum level of the ion source chamber, the measured vacuum level of the analyzer chamber, the variable frequency forepump current, and the molecular pump speed at each moment in the multidimensional time series dataset.

[0028] In one alternative implementation, a fitness function is constructed with the objectives of minimizing vacuum fluctuations, energy consumption, and cross-residue. Based on the predicted vacuum vector, an improved dragonfly algorithm is used to search for optimal solutions within the feasible solution space, generating the optimal control decision for the vacuum system, including:

[0029] With the goal of minimizing vacuum fluctuations, energy consumption, and cross-residue, a fitness function is constructed, and the control decisions of the vacuum system are encoded as position vectors of individuals in the improved Dragonfly algorithm. The control decisions include the frequency of the backing pump, the opening degree of the flow guide valve, and the target rotational speed of the molecular pump.

[0030] The chaotic sequence is generated using the Logistic mapping and then mapped to the feasible solution space of individuals in the improved Dragonfly algorithm, resulting in an initial population including several initial individuals and the initial velocity of each initial individual.

[0031] Based on the fitness function and the initial population, a convergence factor and PSO mechanism are introduced, and an improved dragonfly algorithm is used to iteratively optimize and solve the problem, outputting the optimal control decision for the vacuum system.

[0032] In one alternative implementation, the fitness function is formulated as follows:

[0033]

[0034] In the formula, Let be the fitness value of the alternative control decision for individual X; Let be the predicted vacuum degree in the predicted vacuum degree vector at time k, for the l-th future time step; The target vacuum level; is the measured vacuum level of the analyzer chamber at time k in the multidimensional time series dataset; k is the time series indicator; l is the time step indicator for future periods; m is the total time steps for future periods; For individual X, the target speed and maximum rated speed of the molecular pump are the alternative control decisions. For individual X, the frequency of the upstream pump and the maximum rated frequency are the alternative control decisions. The opening degree of the diversion valve in the alternative control decision for individual X; This is the desorption attenuation coefficient; The residence time at time k in the multidimensional time series dataset; Energy consumption weighting balancing factor; The contribution coefficient for the coordinated air extraction of the backing pump; is the fitness weighting coefficient; e is the exponent base.

[0035] In one alternative implementation, based on the fitness function and the initial population, a convergence factor and a PSO mechanism are introduced. An improved dragonfly algorithm is used for iterative optimization to solve the problem and output the optimal control decision for the vacuum system, including:

[0036] Using the fitness function, we obtain the fitness value of each individual in the initial population or the population updated in the last iteration. Based on the fitness value, we determine the optimal position of each individual and the global optimal individual in the population.

[0037] Based on the individual optimal position and the global optimal individual, a convergence factor and PSO mechanism are introduced to perform behavioral simulations of separation, alignment, cohesion, food attraction, and enemy avoidance on the initial population or the population updated in the previous iteration, so as to update the individual positions and obtain the updated population.

[0038] Using the fitness function, calculate the fitness of each updated individual in the updated population, and update the updated individual with the best fitness value as the global best individual;

[0039] The position of the population is repeatedly updated. When the number of iterations reaches the maximum number of iterations or the fitness value of the globally optimal individual meets the requirements, the iterative optimization of the population is terminated, and the position of the globally optimal individual is output.

[0040] By decoding the spatial position of the globally optimal individual, the optimal control decision of the vacuum system can be obtained.

[0041] In one optional implementation, the optimal control decision is sent down to the execution layer of the vacuum system, and residual compensation is performed using a microsecond-level adaptive PID method based on real-time vacuum state data, thereby completing the physical execution and closed-loop of intelligent control of the vacuum system, including:

[0042] The optimal control decision is directly fed forward to the vacuum system, and the optimal back pump frequency, optimal guide valve opening and optimal molecular pump target speed in the optimal control decision are extracted as the dominant feedforward quantities.

[0043] The optimal backing pump frequency and the optimal molecular pump target speed are directly sent to the bottom-level driver of the vacuum system via a field-programmable gate array as the reference operating point.

[0044] By combining the optimal flow guide valve opening with the timestamp of the real-time polarity conversion trigger signal in the real-time mass spectrometry scanning data, the optimal flow guide valve opening with the timestamp is preloaded into the buffer register of the flow guide valve driver of the vacuum system within a preset period before the polarity switching action occurs, thus offsetting the mechanical execution delay.

[0045] During the control cycle, based on the high-frequency hardware timer interrupt of the field programmable gate array, the real-time vacuum status data of the vacuum system is continuously collected, and the real-time measured vacuum degree of the analyzer chamber is extracted to calculate the residual that still exists after the feedforward command is issued.

[0046] Based on the residual, a nonlinear adaptive PID method based on the error phase determination mechanism is used to calculate the microsecond-level compensation voltage, which is then cascaded and superimposed with the optimal flow guide valve opening to obtain the absolute analog control quantity of the flow guide valve sent to the vacuum system, thus completing the physical execution and closed loop of the intelligent control of the vacuum system.

[0047] Secondly, embodiments of the present invention provide an intelligent control system for a vacuum system based on artificial intelligence, used to implement an intelligent control method for a vacuum system, the system comprising:

[0048] The data acquisition unit is used to acquire multi-source operational data sequences, including the vacuum system, from the liquid chromatography-triple tandem quadrupole mass spectrometer in real time, forming a corresponding multi-dimensional time-series dataset.

[0049] The vacuum load prediction unit is used to input a multi-dimensional time series dataset into a vacuum load prediction model built based on deep learning for prediction, and obtain the corresponding predicted vacuum degree vector.

[0050] The control decision optimization unit is used to construct a fitness function with the goal of minimizing vacuum fluctuation, energy consumption and cross-residue. Based on the predicted vacuum vector, it uses an improved dragonfly algorithm to optimize in the feasible solution space and generate the optimal control decision for the vacuum system.

[0051] The decision-making physical execution unit is used to send the optimal control decision to the execution layer of the vacuum system, and to perform residual compensation by coupling a microsecond-level adaptive PID method based on the real-time vacuum state data collected in real time, thereby completing the physical execution and closed loop of the intelligent control of the vacuum system.

[0052] A third aspect of this invention provides an electronic device, which includes:

[0053] At least one processor; and a memory communicatively connected to the at least one processor; wherein,

[0054] The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention;

[0057] Figure 2 This is a flowchart illustrating the steps of an intelligent control method for a vacuum system based on artificial intelligence, provided in an embodiment of the present invention.

[0058] Figure 3 This is a schematic diagram of the functional units of an intelligent control system for a vacuum system based on artificial intelligence, provided in an embodiment of the present invention. Detailed Implementation

[0059] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0060] The present invention will be further described below with reference to the accompanying drawings.

[0061] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0062] like Figure 1As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0063] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0064] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program for an artificial intelligence-based intelligent control system for vacuum systems.

[0065] exist Figure 1 In the illustrated electronic device, the network interface 1004 is mainly used for intelligent control of the vacuum system based on artificial intelligence with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device, and the electronic device calls the electronic program of the intelligent control system of the vacuum system based on artificial intelligence stored in the memory 1005 through the processor 1001, and executes the intelligent control method of the vacuum system based on artificial intelligence provided in the embodiment of the present invention.

[0066] Reference Figure 2 The present invention provides an intelligent control method for a vacuum system based on artificial intelligence, the method comprising:

[0067] S201: Real-time acquisition of multi-source operational data sequences, including the vacuum system, from the liquid chromatography-triple quadrupole mass spectrometer to form a corresponding multi-dimensional time-series dataset;

[0068] S202: Input the multi-dimensional time series dataset into the vacuum load prediction model built based on deep learning for prediction, and obtain the corresponding predicted vacuum degree vector;

[0069] S203: With the goal of minimizing vacuum fluctuations, energy consumption, and cross-residuals, a fitness function is constructed. Based on the predicted vacuum vector, an improved dragonfly algorithm is used to find the optimal solution space and generate the optimal control decision for the vacuum system.

[0070] S204: The optimal control decision is sent down to the execution layer of the vacuum system, and residual compensation is performed by coupling a microsecond-level adaptive PID method based on the real-time vacuum state data collected in real time, thereby completing the physical execution and closed loop of the intelligent control of the vacuum system.

[0071] The technical solution provided in this application has at least the following beneficial effects:

[0072] By constructing a vacuum load prediction model using a spatio-temporal synchronous multi-head attention long short-term memory (STS-MHA-LSTM) network, the solvent changes in liquid chromatography (spatial characteristic flow) and the scanning actions in mass spectrometry (temporal dynamic flow) are decoupled and fused for prediction. This achieves feedforward prediction of vacuum degree change trends. Combined with microsecond-level timer interrupts from a field-programmable gate array and adaptive PID with phase determination, a perfect closed loop of "predictive feedforward + microsecond-level residual feedback" is achieved, completely eliminating vacuum degree fluctuations caused by polarity switching and solving the control lag problem. A fitness function incorporating vacuum degree fluctuations, energy consumption, and cross-residue is constructed, enabling control decisions to ensure vacuum degree while considering energy consumption and residue suppression. Precise truncation and smooth transition of gas flow paths are achieved at the physical execution layer, significantly reducing gas desorption backflow on the inner wall of the cavity. In the Dragonfly algorithm improvement, particle swarm optimization is introduced. The Optimization (PSO) mechanism directly injects the cognitive and social acceleration terms of PSO as position offsets, avoiding step size oscillations and divergences caused by traditional dual velocity vectors. Combined with adaptive adjustment of the convergence factor, it greatly improves the convergence speed and robustness of global optimization. By improving the dragonfly algorithm, it finds the lowest energy consumption action combination that meets the vacuum requirements, avoiding the "one-size-fits-all" high-speed full-load operation in traditional methods, significantly reducing energy consumption and extending equipment life. For the mechanical delay of the guide valve, the feedforward buffer register is preloaded using the polarity switching timestamp, realizing "adjustment before movement". For residual compensation, a nonlinear adaptive PID based on the error phase judgment mechanism is proposed. The integral is locked in the abrupt phase and a nonlinear damping surge factor is introduced, so that the damping force is proportionally amplified as the absolute value of the error increases, effectively suppressing overshoot and oscillation caused by polarity switching. The steady-state phase is restored by the fuzzy PID to eliminate drift, realizing microsecond-level precise locking of vacuum under all operating conditions.

[0073] In one optional implementation, the multi-source operational data sequence includes a liquid chromatography data sequence from a liquid chromatography system, a mass spectrometry scanning data sequence from a mass spectrometry scanning system, and a vacuum status data sequence from a vacuum system.

[0074] The data at each moment in the liquid chromatography data sequence includes the high-pressure pump flow rate, the organic phase ratio, and the column oven temperature.

[0075] The data at each moment in the mass spectrometry scan data sequence includes the target m / z range span, scan speed, polarity reversal trigger signal, and dwell time.

[0076] The data at each moment in the vacuum state data sequence includes the measured vacuum level of the ion source chamber, the measured vacuum level of the analyzer chamber, the current of the frequency converter pre-pump, and the speed of the molecular pump.

[0077] In this embodiment, liquid chromatography (LC) domain data acquisition: the high-pressure pump flow rate, organic phase ratio, and column oven temperature are read at a fixed period (e.g., 100ms) via the LC system's communication bus; these three parameters directly determine the liquid flow rate and volatility entering the ion source, and are the decisive factors for the "source gas load" of the vacuum system.

[0078] Mass spectrometry scanning domain data acquisition: Capture the real-time status of the mass spectrometry scanning controller, including the target m / z range, scanning speed, polarity reversal trigger signal, and residence time. Polarity reversal causes the high-voltage power supply to reverse instantaneously, triggering a drastic change in the discharge mode within the ion source and a pulsed release of desorbed gas. The m / z range and residence time determine the configuration of the RF and DC components of the quadrupole, affecting the ion residue level in the analyzer chamber. These constitute the "internal excitation gas load".

[0079] Vacuum system domain data acquisition: The measured vacuum levels of the ion source chamber and analyzer chamber are acquired in real time by a Pirani / cold cathode composite vacuum gauge installed on the vacuum chamber; the current of the frequency converter back pump (reflecting the mechanical pumping load) and the speed of the molecular pump (reflecting the high vacuum pumping capability) are read through the driver bus, which constitute the "system response status".

[0080] The three domain data mentioned above are aligned and interpolated based on a unified high-precision timestamp (such as GPS time synchronization or internal system clock synchronization). After removing outliers, a multidimensional time series dataset with strong physical coupling characteristics is constructed.

[0081] In one optional implementation, the vacuum load prediction model is constructed based on the STS-MHA-LSTM algorithm. The vacuum load prediction model includes an input layer, a spatial flow processing branch constructed based on the Gated Recurrent Unit (GRU) algorithm, a temporal flow processing branch constructed based on the Long Short-Term Memory (LSTM) algorithm, a multi-head self-attention layer constructed based on the Attention mechanism, a fully connected layer, and an output layer.

[0082] In one alternative implementation, a multi-dimensional time-series dataset is input into a vacuum load prediction model built based on deep learning for prediction, resulting in the corresponding predicted vacuum level, including:

[0083] S2021: Input the multidimensional time series dataset into the input layer of the vacuum load prediction model built based on deep learning, and decouple the multidimensional time series dataset according to physical meaning, dividing it into spatial feature flow and temporal dynamic flow;

[0084] S2022: Since the spatial feature flow is state-driven, it does not require extremely long time memory and is expected to respond quickly to sudden changes. The spatial flow processing branch of the vacuum load prediction model is used to extract the spatial state hidden layer vector of the spatial feature flow. Since the vacuum degree change has thermodynamic and hydrodynamic inertia and is affected by the cumulative influence of historical scanning strategies, it requires long-term memory. The time flow processing branch of the vacuum load prediction model is used to extract the time dynamic hidden layer vector of the time dynamic flow.

[0085] S2023: The spatial state hidden layer vector and the temporal dynamic hidden layer vector are concatenated, and the resulting concatenated vector is input into the multi-head self-attention layer. The attention mechanism assigns weights by calculating the similarity between the query and the key, enabling the model to adaptively focus on the key spatiotemporal feature combinations that have the greatest impact on the future vacuum degree and suppress redundant features. Based on the attention weight matrix calculated by the multi-head attention mechanism, the concatenated vector is weighted and fused to obtain a high-order feature matrix.

[0086] S2024: Flatten the high-order feature matrix, pass it through a fully connected layer and activation function, and use the output layer to output the predicted vacuum vector for future periods.

[0087] In one alternative implementation, the spatial characteristic flow consists of the high-pressure pump flow rate, the organic phase ratio, the column oven temperature, the target m / z range span, and the polarity conversion trigger signal at each moment in the multidimensional time series dataset. These data determine the "spatial distribution intensity" and "source flux" of the gas load at the current moment, and are state-type driving variables.

[0088] The time dynamic flow consists of the scanning speed, the dwell time, the measured vacuum level of the ion source chamber, the measured vacuum level of the analyzer chamber, the variable frequency pre-pump current, and the molecular pump speed at each moment in the multi-dimensional time series dataset. These data have strong time-series recursion and inertial characteristics, which determine the "trend" of the dynamic evolution of the vacuum system and belong to dynamic response variables.

[0089] In one alternative implementation, a fitness function is constructed with the objectives of minimizing vacuum fluctuations, energy consumption, and cross-residue. Based on the predicted vacuum vector, an improved dragonfly algorithm is used to search for optimal solutions within the feasible solution space, generating the optimal control decision for the vacuum system, including:

[0090] S2031: To minimize vacuum fluctuations, energy consumption, and cross-residual effects, a fitness function is constructed, and the control decisions of the vacuum system are encoded as position vectors of individuals in the improved Dragonfly algorithm. The control decisions include the frequency of the backing pump, the opening degree of the flow guide valve, and the target rotational speed of the molecular pump.

[0091] S2032: Use Logistic mapping to generate chaotic sequences, and map the chaotic sequences to the feasible solution space of individuals in the improved Dragonfly algorithm to obtain an initial population including several initial individuals and the initial velocity of each initial individual.

[0092] S2033: Based on the fitness function and the initial population, a convergence factor and PSO mechanism are introduced, and an improved dragonfly algorithm is used to perform iterative optimization to solve the problem and output the optimal control decision of the vacuum system.

[0093] In one alternative implementation, the fitness function is formulated as follows:

[0094]

[0095] In the formula, Let be the fitness value of the alternative control decision for individual X; Let be the predicted vacuum degree in the predicted vacuum degree vector at time k, for the l-th future time step; The target vacuum level; is the measured vacuum level of the analyzer chamber at time k in the multidimensional time series dataset; k is the time series indicator; l is the time step indicator for future periods; m is the total time steps for future periods; For individual X, the target speed and maximum rated speed of the molecular pump are the alternative control decisions. For individual X, the frequency of the upstream pump and the maximum rated frequency are the alternative control decisions. The opening degree of the diversion valve in the alternative control decision for individual X; This is the desorption attenuation coefficient; The residence time at time k in the multidimensional time series dataset; Energy consumption weighting balancing factor; The contribution coefficient for the coordinated air extraction of the backing pump; is the fitness weighting coefficient; e is the exponent base.

[0096] In one alternative implementation, based on the fitness function and the initial population, a convergence factor and a PSO mechanism are introduced. An improved dragonfly algorithm is used for iterative optimization to solve the problem and output the optimal control decision for the vacuum system, including:

[0097] S20331: Using the fitness function, obtain the fitness value of each individual in the initial population or the population updated in the last iteration, and determine the optimal position of each individual and the global optimal individual in the population based on the fitness value;

[0098] S20332: Based on the individual optimal position and the global optimal individual, a convergence factor and PSO mechanism are introduced to perform behavioral simulations of separation, alignment, cohesion, food attraction, and enemy avoidance on the initial population or the population updated in the previous iteration, so as to update the individual positions and obtain the updated population.

[0099] The formula is:

[0100]

[0101] In the formula, The value for the separation behavior in the t-th iteration; Let i and j be the current individual position and the positions of neighboring individuals in the t-th iteration; i and j are individual indicators; N is the number of neighboring individuals; to avoid collisions between individuals and neighboring individuals, enhance population diversity, and prevent the dragonflies from dispersing and causing the solution to become overly concentrated;

[0102]

[0103] In the formula, The alignment behavior value for the t-th iteration; Let be the velocity of the neighboring individuals in the t-th iteration; keep the individual velocity consistent with that of the neighboring individuals to enhance group coordination and simulate the coordinated flight behavior of dragonfly groups during pollination.

[0104]

[0105] In the formula, is the cohesive behavior value for the t-th iteration; it causes individuals to move towards the center of neighboring individuals, enhancing group aggregation and simulating the behavior of dragonflies gathering towards the center of flower clusters during pollination.

[0106]

[0107] In the formula, Let be the food attraction behavior value for the t-th iteration; Let t be the location of the food source in the t-th iteration; guide individuals to move towards the food source (optimal solution), simulating the behavior of dragonflies attracted by pollen, and accelerate the global search;

[0108]

[0109] In the formula, Let be the enemy avoidance behavior value in the t-th iteration; Let t be the enemy's position in the t-th iteration, i.e., the individual with the worst fitness value; move the individual away from the enemy (inferior solution) to avoid getting trapped in local optima, simulate the behavior of a dragonfly avoiding dangerous areas, and enhance the robustness of global exploration;

[0110]

[0111] In the formula, This represents the position change value obtained using the traditional dragonfly algorithm at iteration t+1,t. As global search weight;

[0112]

[0113]

[0114] In the formula, Let be the optimal position of the i-th individual in the t-th iteration; Accelerate one's own cognition; The acceleration coefficient of social cognition; A random number in the interval [0,1]. It is the globally optimal individual in the t-th iteration; Let be the convergence factor for the t-th iteration; Let i be the i-th updated individual in the population during the (t+1)-th iteration. For the i-th updated individual in the t-th iteration, in the initial iteration, t represents the initial individual; t is an indicator of the iteration number. These are the maximum and minimum values ​​of the convergence factor; The maximum number of iterations is given. The traditional Dragonfly algorithm has already defined a core position change value (calculated from behaviors such as separation, alignment, and aggregation). If the independent velocity vector of the standard PSO is forcibly introduced at this time, it will result in two "vectors" driving the movement of individuals within the same iteration cycle. This will cause serious step size oscillation and divergence in mathematical optimization. Therefore, this embodiment adopts the "step size injection method".

[0115] S20333: Using the fitness function, calculate the fitness of each updated individual in the updated population, and update the updated individual with the best fitness value as the global best individual;

[0116] S20334: Repeatedly update the position of the population. When the number of iterations reaches the maximum number of iterations or the fitness value of the globally optimal individual meets the requirements, terminate the iterative optimization of the population and output the position of the globally optimal individual.

[0117] S20335: Decode the spatial position of the globally optimal individual to obtain the optimal control decision of the vacuum system.

[0118] In one optional implementation, the optimal control decision is sent down to the execution layer of the vacuum system, and residual compensation is performed using a microsecond-level adaptive PID method based on real-time vacuum state data, thereby completing the physical execution and closed-loop of intelligent control of the vacuum system, including:

[0119] S2041: The optimal control decision is directly sent to the vacuum system as a feedforward quantity, and the optimal backing pump frequency, optimal guide valve opening and optimal molecular pump target speed in the optimal control decision are extracted as the dominant feedforward quantities.

[0120] S2042: The optimal backing pump frequency and the optimal molecular pump target speed are directly sent to the bottom driver of the vacuum system via a field programmable gate array as the reference operating point.

[0121] S2043: Combine the optimal flow guide valve opening with the timestamp of the real-time polarity conversion trigger signal in the real-time mass spectrometry scan data, and preload the timestamped optimal flow guide valve opening into the buffer register of the flow guide valve driver of the vacuum system within a preset period before the polarity switching action occurs, so as to offset the mechanical execution delay.

[0122] S2044: During the control cycle, based on the high-frequency hardware timer interrupt of the field-programmable gate array, the real-time vacuum status data of the vacuum system is continuously acquired, and the real-time measured vacuum degree of the analyzer chamber is extracted. The residual remaining after the feedforward command is issued is calculated using the following formula:

[0123]

[0124] In the formula, For microsecond-level discrete sampling time The residual; For microsecond-level discrete sampling time The real-time measured vacuum level of the analyzer chamber; The target vacuum level;

[0125] S2045: Based on the residual, a nonlinear adaptive PID method based on the error phase determination mechanism is used to calculate the microsecond-level compensation voltage, and it is cascaded and superimposed with the optimal flow guide valve opening to obtain the absolute analog control quantity of the flow guide valve sent to the vacuum system, thus completing the physical execution and closed loop of the intelligent control of the vacuum system.

[0126] In this embodiment, the phase determination is as follows: Based on the residual, the rate of change of the residual is calculated in real time. If the rate of change of the residual exceeds the abrupt change threshold, it is determined that the current phase is in a polarity switching abrupt change phase; otherwise, it is determined to be a steady-state tracking phase.

[0127] Coupling compensation under abrupt phase changes: Under abrupt phase changes, a nonlinear adaptive PID method based on an error phase determination mechanism is used to forcibly lock out the integral term, retaining only the proportional and nonlinear derivative terms, to calculate the microsecond-level compensation voltage. The formula is as follows:

[0128]

[0129] In the formula, For microsecond-level compensation voltage, at microsecond-level discrete sampling times The signal is applied directly to the flow guide valve driver via a digital-to-analog converter to instantly change the valve plate angle in order to compensate for the gas load residuals that the feedforward quantity fails to accurately cover. This is the residual from the previous microsecond-level discrete sampling time. It has a high-frequency sampling period in the microsecond range; These are the proportional gain coefficient and the fundamental differential gain coefficient; It is a nonlinear damping surge factor, a constant greater than 0 (obtained by calibration of the fluid impedance characteristics of the vacuum chamber, preferably in the range of 5~15), which makes the magnitude of the damping force no longer constant, but automatically amplified proportionally as the absolute value of the error increases.

[0130] Coupling compensation under steady-state phase: Under steady-state tracking phase, conventional fuzzy adaptive PID is restored to eliminate the tiny drifts that the feedforward prediction fails to cover with a very small step size.

[0131] This invention also provides an intelligent control system for a vacuum system based on artificial intelligence, referring to... Figure 3 The device may include the following units:

[0132] The data acquisition unit 301 is used to acquire multi-source operating data sequences, including the vacuum system, in the liquid chromatography-triple tandem quadrupole mass spectrometer in real time, forming a corresponding multi-dimensional time-series dataset.

[0133] The vacuum load prediction unit 302 is used to input a multi-dimensional time series dataset into a vacuum load prediction model built based on deep learning for prediction, and obtain the corresponding predicted vacuum degree vector.

[0134] The control decision optimization unit 303 is used to construct a fitness function with the goal of minimizing vacuum fluctuation, energy consumption and cross residue. Based on the predicted vacuum vector, it uses an improved dragonfly algorithm to optimize in the feasible solution space and generate the optimal control decision of the vacuum system.

[0135] The decision physical execution unit 304 is used to send the optimal control decision to the execution layer of the vacuum system, and perform residual compensation by coupling a microsecond-level adaptive PID method based on the real-time vacuum state data collected in real time, thereby completing the physical execution and closed loop of the intelligent control of the vacuum system.

[0136] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0137] Memory, used to store computer programs;

[0138] The processor, when executing the program stored in the memory, implements the artificial intelligence-based intelligent control method for vacuum systems of the present invention.

[0139] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0140] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or individual transistor logic devices, or discrete hardware components.

[0141] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the artificial intelligence-based intelligent control method for vacuum systems according to embodiments of the present invention.

[0142] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0144] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0146] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0147] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent control method for a vacuum system based on artificial intelligence, characterized in that, The method includes: Real-time acquisition of multi-source operational data sequences, including the vacuum system, from the liquid chromatography-triple quadrupole mass spectrometer constitutes a corresponding multidimensional time-series dataset. The multidimensional time series dataset is input into the vacuum load prediction model built based on deep learning for prediction, and the corresponding predicted vacuum degree vector is obtained. With the goal of minimizing vacuum fluctuations, energy consumption, and cross-residuals, a fitness function is constructed. Based on the predicted vacuum vector, an improved dragonfly algorithm is used to find the optimal solution space and generate the optimal control decision for the vacuum system. The optimal control decision is sent down to the execution layer of the vacuum system, and residual compensation is performed by coupling a microsecond-level adaptive PID method based on the real-time vacuum state data, thus completing the physical execution and closed loop of the intelligent control of the vacuum system.

2. The intelligent control method for a vacuum system based on artificial intelligence according to claim 1, characterized in that, The multi-source operational data sequence includes the liquid chromatography data sequence of the liquid chromatography system, the mass spectrometry scanning data sequence of the mass spectrometry scanning system, and the vacuum status data sequence of the vacuum system; The data at each moment in the liquid chromatography data sequence includes the high-pressure pump flow rate, the organic phase ratio, and the column oven temperature. The data at each moment in the mass spectrometry scan data sequence includes the target m / z range span, scan speed, polarity reversal trigger signal, and dwell time. The data at each moment in the vacuum state data sequence includes the measured vacuum level of the ion source chamber, the measured vacuum level of the analyzer chamber, the current of the frequency converter pre-pump, and the speed of the molecular pump.

3. The intelligent control method for a vacuum system based on artificial intelligence according to claim 2, characterized in that, The vacuum load prediction model is constructed based on the STS-MHA-LSTM algorithm. The vacuum load prediction model includes an input layer, a spatial flow processing branch constructed based on the GRU algorithm, a temporal flow processing branch constructed based on the LSTM algorithm, a multi-head self-attention layer constructed based on the Attention mechanism, a fully connected layer, and an output layer.

4. The intelligent control method for a vacuum system based on artificial intelligence according to claim 3, characterized in that, A multi-dimensional time-series dataset is input into a vacuum load prediction model built based on deep learning for prediction, resulting in the corresponding predicted vacuum level, including: The multidimensional time series dataset is input into the input layer of the vacuum load prediction model built based on deep learning, and the multidimensional time series dataset is decoupled according to physical meaning, and divided into spatial feature flow and temporal dynamic flow. The spatial state hidden layer vector of the spatial feature flow is extracted using the spatial flow processing branch of the vacuum load prediction model, and the temporal dynamic hidden layer vector of the temporal dynamic flow is extracted using the temporal flow processing branch of the vacuum load prediction model. The spatial state hidden layer vector and the temporal dynamic hidden layer vector are concatenated, and the resulting concatenated vector is input into the multi-head self-attention layer. Based on the attention weight matrix calculated by the multi-head attention mechanism, the concatenated vector is weighted and fused to obtain a high-order feature matrix. The high-order feature matrix is ​​flattened, passed through a fully connected layer and an activation function, and then the output layer is used to output the predicted vacuum vector for the future period.

5. The intelligent control method for a vacuum system based on artificial intelligence according to claim 4, characterized in that, The spatial feature stream is composed of the high-pressure pump flow rate, the organic phase ratio, the column oven temperature, the target m / z range span, and the polarity conversion trigger signal at each moment in the multidimensional time series dataset; The time dynamic stream is composed of the scan rate, the dwell time, the measured vacuum level of the ion source chamber, the measured vacuum level of the analyzer chamber, the variable frequency forepump current, and the molecular pump speed at each moment in the multidimensional time series dataset.

6. The intelligent control method for a vacuum system based on artificial intelligence according to claim 5, characterized in that, With the objectives of minimizing vacuum fluctuations, energy consumption, and cross-residual effects, a fitness function is constructed. Based on the predicted vacuum vector, an improved dragonfly algorithm is used to find the optimal solution in the feasible solution space, generating the optimal control decision for the vacuum system, including: With the goal of minimizing vacuum fluctuations, energy consumption, and cross-residue, a fitness function is constructed, and the control decisions of the vacuum system are encoded as position vectors of individuals in the improved Dragonfly algorithm. The control decisions include the frequency of the backing pump, the opening degree of the flow guide valve, and the target rotational speed of the molecular pump. The chaotic sequence is generated using the Logistic mapping and then mapped to the feasible solution space of individuals in the improved Dragonfly algorithm, resulting in an initial population including several initial individuals and the initial velocity of each initial individual. Based on the fitness function and the initial population, a convergence factor and PSO mechanism are introduced, and an improved dragonfly algorithm is used to iteratively optimize and solve the problem, outputting the optimal control decision for the vacuum system.

7. The intelligent control method for a vacuum system based on artificial intelligence according to claim 6, characterized in that, The formula for the fitness function is: In the formula, For individuals X The fitness value of the corresponding alternative control decision; For a moment k In the predicted vacuum vector, the future first l Predicted vacuum degree at each time step; The target vacuum level; Central time of multidimensional time series data k The measured vacuum level of the analyzer chamber; k For timing indication; l This is an indicator of the time step for future cycles. m This represents the total time step for the future cycle; For individuals X The corresponding alternative control decisions include the target speed and maximum rated speed of the molecular pump; For individuals X The corresponding alternative control decision includes the backing pump frequency and maximum rated frequency. For individuals X The corresponding flow control valve opening in the alternative control decision; This is the desorption attenuation coefficient; Central time of multidimensional time series data k Duration of stay; Energy consumption weighting balancing factor; The contribution coefficient for the coordinated air extraction of the backing pump; For fitness weighting coefficients; e The base of the exponent.

8. The intelligent control method for a vacuum system based on artificial intelligence according to claim 7, characterized in that, Based on the fitness function and the initial population, a convergence factor and PSO mechanism are introduced. An improved dragonfly algorithm is used for iterative optimization to solve the problem and output the optimal control decision for the vacuum system, including: Using the fitness function, we obtain the fitness value of each individual in the initial population or the population updated in the last iteration. Based on the fitness value, we determine the optimal position of each individual and the global optimal individual in the population. Based on the individual optimal position and the global optimal individual, a convergence factor and PSO mechanism are introduced to perform behavioral simulations of separation, alignment, cohesion, food attraction, and enemy avoidance on the initial population or the population updated in the previous iteration, so as to update the individual positions and obtain the updated population. Using the fitness function, calculate the fitness of each updated individual in the updated population, and update the updated individual with the best fitness value as the global best individual; The position of the population is repeatedly updated. When the number of iterations reaches the maximum number of iterations or the fitness value of the globally optimal individual meets the requirements, the iterative optimization of the population is terminated, and the position of the globally optimal individual is output. By decoding the spatial position of the globally optimal individual, the optimal control decision of the vacuum system can be obtained.

9. The intelligent control method for a vacuum system based on artificial intelligence according to claim 8, characterized in that, The optimal control decision is sent to the execution layer of the vacuum system, and residual compensation is performed using a microsecond-level adaptive PID method based on the real-time vacuum state data. This completes the physical execution and closed-loop control of the vacuum system, including: The optimal control decision is directly fed forward to the vacuum system, and the optimal back pump frequency, optimal guide valve opening and optimal molecular pump target speed in the optimal control decision are extracted as the dominant feedforward quantities. The optimal backing pump frequency and the optimal molecular pump target speed are directly sent to the bottom-level driver of the vacuum system via a field-programmable gate array as the reference operating point. By combining the optimal flow guide valve opening with the timestamp of the real-time polarity conversion trigger signal in the real-time mass spectrometry scanning data, the optimal flow guide valve opening with the timestamp is preloaded into the buffer register of the flow guide valve driver of the vacuum system within a preset period before the polarity switching action occurs, thus offsetting the mechanical execution delay. During the control cycle, based on the high-frequency hardware timer interrupt of the field programmable gate array, the real-time vacuum status data of the vacuum system is continuously collected, and the real-time measured vacuum degree of the analyzer chamber is extracted to calculate the residual that still exists after the feedforward command is issued. Based on the residual, a nonlinear adaptive PID method based on the error phase determination mechanism is used to calculate the microsecond-level compensation voltage, which is then cascaded and superimposed with the optimal flow guide valve opening to obtain the absolute analog control quantity of the flow guide valve sent to the vacuum system, thus completing the physical execution and closed loop of the intelligent control of the vacuum system.

10. An intelligent control system for a vacuum system based on artificial intelligence, used to implement the intelligent control method for a vacuum system as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition unit is used to acquire multi-source operational data sequences, including the vacuum system, from the liquid chromatography-triple tandem quadrupole mass spectrometer in real time, forming a corresponding multi-dimensional time-series dataset. The vacuum load prediction unit is used to input a multi-dimensional time series dataset into a vacuum load prediction model built based on deep learning for prediction, and obtain the corresponding predicted vacuum degree vector. The control decision optimization unit is used to construct a fitness function with the goal of minimizing vacuum fluctuation, energy consumption and cross-residue. Based on the predicted vacuum vector, it uses an improved dragonfly algorithm to optimize in the feasible solution space and generate the optimal control decision for the vacuum system. The decision-making physical execution unit is used to send the optimal control decision to the execution layer of the vacuum system, and to perform residual compensation by coupling a microsecond-level adaptive PID method based on the real-time vacuum state data collected in real time, thereby completing the physical execution and closed loop of the intelligent control of the vacuum system.

Images