A high-purity hydrogen fluoride condensing parameter optimization method

Abstract

The application relates to the field of fluorine chemical technology, and discloses a high-purity hydrogen fluoride condensation parameter optimization method. The method receives real-time monitoring data of a condensation process, utilizes a physical parameter analysis model, combines temperature parameters, pressure parameters and historical operation records to analyze condensation optimization intentions, and generates a parameter optimization vector; a condensation state index graph is constructed according to distributed storage operation state data, a node represents a condensation state point, an edge represents a physical conversion relationship between state points, and the graph structure and index efficiency are updated in real time based on an adaptive adjustment mechanism; according to the parameter optimization vector, a multi-level mapping calculation method is used to dynamically locate and optimize the adjustment domain related to the optimization intention in the condensation state index graph; and in the adjustment domain, a multi-dimensional fusion matching algorithm is used to screen a candidate parameter set meeting the optimization intention and to sort and output. The method improves the pertinence and adaptability of condensation parameter optimization, and can effectively adapt to the high-purity hydrogen fluoride condensation demand under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of fluorochemical technology, specifically to a method for optimizing the condensation parameters of high-purity hydrogen fluoride. Background Technology

[0002] High-purity hydrogen fluoride is a key basic material for high-tech industries such as electronics and photovoltaic energy, and its purity directly affects the performance and quality of downstream products. In the production process of high-purity hydrogen fluoride, the condensation stage is the core process for achieving gas-liquid separation and removing impurities. Precise control of condensation parameters can effectively improve product purity and yield. However, the condensation process is dynamically affected by multiple factors such as the composition of the raw gas, ambient temperature, and system pressure, making parameter optimization a challenging process.

[0003] The adjustment of high-purity hydrogen fluoride condensation parameters often relies on operator experience or simple PID control, which is insufficient to handle complex and variable operating conditions. Traditional methods underutilize real-time monitoring data of the condensation process, often adjusting parameters based solely on a single temperature or pressure indicator, lacking in-depth analysis of the relationships between multiple parameters. This leads to a disconnect between optimization intentions and actual operating requirements. Furthermore, the operational status data of the condensation system is scattered across different devices or modules, with inconsistent data formats, making it difficult to form a complete status representation and hindering operators from quickly grasping the overall system operating status.

[0004] The description of the condensation state often relies on static models, which cannot reflect the physical transformation relationships between state points in real time. When the feed gas flow rate, composition, or environmental conditions fluctuate, the static model struggles to dynamically update the state characterization, causing parameter adjustments to lag behind changes in operating conditions. Furthermore, during parameter optimization, the determination of the adjustment domain lacks scientific calculation methods and often depends on empirically defined ranges, easily leading to over- or under-adjustment, affecting condensation efficiency. Simultaneously, parameter selection often employs single-dimensional matching methods, ignoring the coupling effects between multiple parameters such as temperature, pressure, and flow rate, making it difficult to select the truly optimal parameter combination suitable for the current operating conditions. These problems result in poor stability of the high-purity hydrogen fluoride condensation process, large fluctuations in product purity, and an inability to meet the stringent requirements of high-end industries for high-purity hydrogen fluoride.

[0005] With the rapid development of high-tech industries, the purity requirements for high-purity hydrogen fluoride are constantly increasing. Traditional methods for optimizing condensation parameters are no longer sufficient to meet the high precision and stability demands of industrial production. Real-time sensing, dynamic modeling, and precise parameter adjustment of the condensation process have become key issues in improving the production quality of high-purity hydrogen fluoride. Summary of the Invention

[0006] The purpose of this invention is to provide a method for optimizing the condensation parameters of high-purity hydrogen fluoride to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a method for optimizing the condensation parameters of high-purity hydrogen fluoride, the method comprising:

[0008] Receive real-time monitoring data of the condensation process, use a physical parameter analysis model, analyze the condensation optimization intention based on the temperature parameters, pressure parameters and historical operation records in the real-time monitoring data, and generate a parameter optimization vector based on the condensation optimization intention;

[0009] For the distributed storage of operational status data in the condensation system, a condensation status index graph is constructed. The condensation status index graph uses nodes to represent condensation status points and edges to represent the physical transformation relationships between status points. The graph structure and index efficiency are updated in real time based on an adaptive adjustment mechanism.

[0010] Based on the parameter optimization vector, the adjustment domain related to the optimization intention is dynamically located in the condensation state index diagram using a multi-level mapping calculation method.

[0011] Within the adjustment domain, a multi-dimensional fusion matching algorithm is used to filter and sort a set of candidate parameters that meet the optimization intent.

[0012] Preferably, the real-time monitoring data of the receiving condensation process is analyzed using a physical parameter analysis model. Based on the temperature parameters, pressure parameters, and historical operation records in the real-time monitoring data, the condensation optimization intention is analyzed, and a parameter optimization vector is generated in combination with the condensation optimization intention, including:

[0013] Real-time monitoring data is mapped to a fixed-dimensional feature space through physical modeling to generate temperature feature embeddings and pressure feature embeddings.

[0014] Acquire historical operation records, analyze historical temperature fluctuations, historical pressure changes and operation adjustment behaviors, construct operation profiles, and transform operation profile data into operation preference embeddings;

[0015] Access the condensation knowledge base, extract physical constraint features related to real-time monitoring data, and generate constraint feature embeddings;

[0016] A weighted fusion-based algorithm is used to integrate temperature feature embedding, operation preference embedding, and constraint feature embedding to perform condensation optimization intent parsing. Each parsed intent is assigned an optimization weight, and a parameter optimization vector is generated based on the optimization weight. The parameter optimization vector covers the current optimization intent and reflects the system operation characteristics.

[0017] Preferably, for the distributed storage of operational state data in the condensation system, a condensation state index graph is constructed. The condensation state index graph uses nodes to represent condensation state points and edges to represent the physical transformation relationships between state points. The graph structure and index efficiency are updated in real time based on an adaptive adjustment mechanism, including:

[0018] Signal processing techniques are used to extract state point features from the operation state data. Each condensation state point is used as a node in the condensation state index graph structure, and a corresponding state point feature is assigned to each node.

[0019] The edges in the condensation state index graph structure are created based on the physical transformation relationship between state points, and the initial strength of each edge is set to obtain the condensation state index graph. The strength is initialized by comparing the transformation similarity between two state points.

[0020] Based on the adaptive adjustment mechanism, the changes in state point parameters and their transformation relationships are tracked to ensure the real-time performance of the condensation state index graph. When a new state point is added or a state point is modified, an incremental adjustment strategy is used to update only the affected nodes and edges, and the strength of the edges is dynamically adjusted according to the operation records.

[0021] Deploy a real-time condensation state index graph in a distributed system.

[0022] Preferably, the step of dynamically locating the adjustment domain related to the optimization intention in the condensation state index diagram using a multi-level mapping calculation based on the parameter optimization vector includes:

[0023] The condensation state index graph is pre-trained using a feature mapping algorithm to generate the state embedding of each node in the graph.

[0024] The parameter optimization vector is mapped to the state embedding space using a depth alignment model;

[0025] Calculate the matching degree between the mapped parameter optimization vector and all state embeddings, and select a specified number of nodes with the highest matching degree from the condensation state index graph as the initial adjustment nodes;

[0026] Based on the neighboring nodes of each initial adjustment node, expand outward to the first level of neighboring nodes, determine whether the matching degree of the neighboring nodes reaches a specific threshold. If it does not reach the threshold, stop the expansion. If it does reach the threshold, add it to the candidate node set and repeat the expansion process to continue exploring the next level of neighboring nodes until the preset level limit is reached or the cumulative number of nodes is met.

[0027] The set of all candidate nodes obtained through multi-level mapping expansion is determined as the regulation domain.

[0028] Preferably, within the adjustment domain, the step of using a multi-dimensional fusion matching algorithm to filter and sort a set of candidate parameters that conform to the optimization intention includes:

[0029] For each condensation state point within the regulation domain that may be related to the optimization intention, extract multi-dimensional operational features;

[0030] A multi-dimensional fusion algorithm is used to fuse operational features from different dimensions to generate fused operational features for each state point.

[0031] Calculate the matching degree between the fusion operation features and the parameter optimization vector of each state point, and filter state points with matching degrees higher than a preset threshold as candidate parameter sets;

[0032] The candidate parameter set is sorted according to matching degree, operation preference, and parameter priority.

[0033] Preferably, the method further includes:

[0034] Obtain the historical operation sequence of the high-purity hydrogen fluoride condensation process, and divide the process into stable operation segments and variable operation segments based on the parameter change trends in the historical operation sequence;

[0035] Based on the differences in parameter values ​​between the stable operation segment and the variable operation segment, the types of parameters to be optimized are selected.

[0036] For each parameter type to be optimized, based on the data similarity of the operation points in the stable operation segment and the variable operation segment, combined with time characteristics and parameter adjustment records, the optimization capability index of each operation point in the stable operation segment is obtained.

[0037] Based on the parameter adjustment records and parameter reproduction data of each operation point in the changing operation segment, and combined with the optimization capability index, the core node probability of each operation point in the stable operation segment is obtained.

[0038] Based on the probability selection of the core nodes, each operation point in the stable operation segment is selected as the core node. Combining the operation data of the same parameter type in the changing operation segment, an optimized chain structure is constructed, and an optimization scheme for different parameters is determined based on the optimized chain structure.

[0039] Preferably, under each type of parameter to be optimized, based on the data similarity of the operation point in the stable operation segment and the changing operation segment, combined with time characteristics and parameter adjustment records, the optimization capability index of each operation point in the stable operation segment is obtained, including:

[0040] Under any parameter type to be optimized, all points containing operational data in the stable operation segment are taken as candidate operation points, and all points containing operational data in the variable operation segment are taken as adjustment points.

[0041] Record any candidate operation point as the selected candidate point, and any adjustment point as the selected adjustment point;

[0042] Based on the parameter adjustment time and adjustment duration of the selected candidate point and the selected adjustment point, the delay influence intensity of the selected candidate point relative to the selected adjustment point is obtained;

[0043] Calculate the similarity between the data of each adjustment behavior of the selected candidate point in the stable operation segment and the data of each adjustment behavior of the selected control point in the variable operation segment, and obtain the similarity feature factor corresponding to the selected control point under each adjustment behavior;

[0044] Using the intensity of the delay effect as a weight, the similarity feature factors corresponding to the selected adjustment point under each adjustment behavior are weighted and averaged to obtain the parameter optimization force of the selected candidate point relative to the selected adjustment point. The mean of the parameter optimization force of the selected candidate point relative to all adjustment points is used as the optimization capability index of the selected candidate point.

[0045] Preferably, the step of obtaining the core node probability of each operation point within the stable operation segment based on the parameter adjustment records and parameter reproduction data of each operation point within the changing operation segment, combined with the optimization capability index, includes:

[0046] Based on the adjustment duration and decision interval of each adjustment behavior of the selected candidate points, and combined with the number of parameter repetitions, the decision-making power index of the selected candidate points is obtained.

[0047] The product of the optimization capability index and the decision-making capability index of the selected candidate point is normalized to generate the core node probability of the selected candidate point.

[0048] Preferably, the step of selecting each operation point within a stable operation segment as a core node based on the probability of the core node, and constructing an optimized chain structure by combining operation data of the same parameter type in the changing operation segment, includes:

[0049] The candidate operation points with a core node probability greater than a preset probability threshold are respectively used as the core nodes of each chain in the optimized chain structure;

[0050] Obtain the decision-making power index for each adjustment point, and multiply the delay impact intensity of each adjustment point and the core node with the decision-making power index of the adjustment point as the following indicator of each adjustment point relative to the core node.

[0051] For any core node, construct a chain structure in descending order of the following index of each adjustment point relative to that core node, with nodes at the same level in the chain structure having the same following index.

[0052] The chain structure of all core nodes constitutes an optimized chain structure.

[0053] Preferably, determining the optimization scheme for different parameters based on the optimization chain structure includes:

[0054] In optimizing the chain structure, the layer containing the first shared child node of different core nodes is taken as the first target layer; after the first target layer, the layer with the most child nodes is taken as the second target layer.

[0055] A preset first optimization strategy is used for operation points between the core node and the first target layer, a preset second optimization strategy is used for operation points between the first target layer and the second target layer, and a preset third optimization strategy is used for operation points between the second target layer and the bottom layer.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] This method for optimizing high-purity hydrogen fluoride condensation parameters provides a more systematic and precise technical approach through multi-stage synergy. At the data utilization level, the method receives real-time monitoring data of the condensation process and, combined with a physical parameter analysis model, deeply integrates real-time parameters such as temperature and pressure with historical operation records. This allows for a more comprehensive capture of operating conditions, making the analysis of optimization intentions more aligned with actual production needs. This data-driven approach to generating optimization intentions avoids the subjectivity of traditional experience-based adjustments, making the generation of parameter optimization vectors more targeted and effectively adaptable to condensation requirements under different raw material compositions and flow rate fluctuations.

[0058] At the state representation level, a condensation state index graph is constructed based on the distributed storage of operational state data in the condensation system. This graph visually presents condensation state points and their transition relationships through nodes and edges, overcoming the limitations of traditional static models in state description. Furthermore, the introduction of an adaptive adjustment mechanism allows the index graph to update its structure and index efficiency in real time as operating conditions change, ensuring the timeliness and accuracy of state representation. This enables operators or control systems to monitor the dynamic changes in the condensation system in real time, clearly identify key nodes in state transitions, and provide a clear state reference framework for subsequent parameter adjustments.

[0059] At the regulation domain localization level, a multi-level mapping calculation method based on parameter optimization vectors is used to dynamically locate the regulation domain, breaking through the traditional extensive mode of empirically defining the regulation range. Multi-level mapping calculation, through layer-by-layer focusing, can accurately delineate the regulation range relevant to the current optimization intention from a complex state space, reducing the search area for invalid parameters and improving the efficiency of parameter optimization. Simultaneously, the dynamic localization method can adjust the regulation domain boundary in real time according to changes in operating conditions, ensuring that the regulation range always remains consistent with actual needs and avoiding regulation failure due to fluctuations in operating conditions.

[0060] At the parameter selection level, a multi-dimensional fusion matching algorithm is used to filter and rank the candidate parameter set, overcoming the limitations of traditional single-dimensional parameter selection. This algorithm comprehensively considers the matching degree between multiple dimensions such as temperature, pressure, and condensation rate and the optimization intent. Through multi-factor weight allocation and collaborative calculation, it ensures that the selected candidate parameter set fully covers the optimization requirements. The ranked output provides operators or automatic control systems with a clear priority for parameter selection, facilitating the rapid determination of the optimal adjustment scheme and reducing the cost of parameter trial and error. Attached Figure Description

[0061] Figure 1 This is a timeline diagram of the high-purity hydrogen fluoride condensation parameter optimization method described in this invention;

[0062] Figure 2 A flowchart for constructing and updating the condensation state index map;

[0063] Figure 3 A flowchart for dynamic positioning of the adjustment domain;

[0064] Figure 4 This is a flowchart illustrating the parameter optimization scheme determined based on the optimized chain structure. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Please see Figure 1 This invention provides a method for optimizing the condensation parameters of high-purity hydrogen fluoride. The method includes: achieving precise optimization of condensation parameters through dynamic analysis of real-time monitoring data and historical operation records, combined with a physical parameter analytical model and a condensation state index diagram. The specific process includes:

[0067] The system receives real-time monitoring data of the condensation process, including temperature and pressure parameters. Combined with historical operation records, it analyzes the condensation optimization intent using a physical parameter analysis model, generating a parameter optimization vector. For the distributed storage of operation status data, a condensation state index graph is constructed, where nodes represent condensation state points and edges represent the physical transformation relationships between state points. The graph structure and index efficiency are updated in real-time based on an adaptive adjustment mechanism. Based on the parameter optimization vector, the system dynamically locates the adjustment domain in the condensation state index graph through multi-level mapping calculations. Within the adjustment domain, a multi-dimensional fusion matching algorithm is used to filter and sort the candidate parameter set, outputting the optimal parameter adjustment scheme.

[0068] Example 1: See Figure 2 Temperature and pressure parameters in real-time monitoring data are transformed into a fixed-dimensional feature representation space using physical modeling techniques. The dynamic changes of temperature parameters over consecutive time stamps are mapped to a time-series feature vector, which includes fluctuation frequency, gradient trend, and steady-state offset features. Pressure parameters are decomposed into spatial distribution features using a pressure conduction model, forming a multi-dimensional tensor structure. During feature embedding, the original data within the time window is sampled in segments, and the feature contribution value of each segment is calculated based on the thermodynamic transfer function. These are ultimately integrated into temperature and pressure feature embedding vectors. These embedding vectors have dimensional consistency, facilitating subsequent fusion processing.

[0069] The analysis of historical operation records encompasses two dimensions: operational behavior patterns and system response characteristics. Historical temperature fluctuation data is divided into low-fluctuation, medium-fluctuation, and high-fluctuation zones based on standard deviation, recording the frequency and magnitude of manual intervention by operators within each zone. Historical pressure change data is categorized into gradual, steep, and abrupt change phases based on the slope of change, and the valve opening adjustment sequence corresponding to each phase is statistically analyzed. Operational adjustment behaviors are analyzed by constructing a behavior matrix based on action type (e.g., increasing opening, adjusting cooling rate), action duration, and system recovery time after adjustment. These three types of data are encoded into operational preference embedding vectors, with behavioral patterns encoded using weighted frequency coding and system responses encoded using differential coding.

[0070] The condensation knowledge base stores physical constraint rules, including data such as phase transition critical points, saturated vapor pressure curves, and material heat transfer limits. Based on the combination of real-time temperature and pressure parameters, matching constraint entries are retrieved from the knowledge base. If the real-time temperature is within the gas-liquid mixed phase range, the critical heat flux limit at the corresponding pressure is associated; if the real-time pressure is close to the equipment design threshold, a safety margin parameter is associated. The extracted physical constraint features are quantized using boundary values ​​to generate constraint feature embedding vectors, with the vector dimensions aligned with the temperature and pressure feature embeddings.

[0071] Temperature feature embedding vectors, operation preference embedding vectors, and constraint feature embedding vectors are integrated through a three-layer weighted fusion mechanism. The first layer of fusion uses linear superposition, assigning weight coefficients α based on the confidence level of temperature parameters in real-time data, β based on the time proximity of operation records, and γ based on the enforcement level of constraints. The second layer of fusion processes the superposition results using a non-linear activation function to generate an intent feature map. The third layer of fusion performs cluster analysis on the intent feature map to identify core optimization target clusters, with each cluster corresponding to a condensation optimization intent. Each intent is assigned an optimization weight based on cluster density, ultimately generating a parameter optimization vector covering multi-objective optimization scenarios. Elements of this vector include parameters such as the target temperature deviation threshold, pressure regulation sensitivity coefficient, and dynamic balance priority.

[0072] For the operational status data of distributed storage, multi-channel signal processing technology is used to extract status point features. The operational status data includes parameters such as condenser tube wall temperature distribution, cooling medium flow rate, and condensate level height collected by equipment sensors. Each condensation status point is defined as a combination of all parameter values ​​at a specific timestamp. The spectral features of each parameter are extracted using wavelet transform, and then principal component analysis is used to reduce the dimensionality to a fixed-length status point feature vector. The condensation status index map uses the status point feature vector as node attributes, and the node identifier uses a spatiotemporal hybrid encoding (timestamp + equipment partition number).

[0073] The physical transition relationships between state points are modeled based on the principle of thermodynamic continuity. If the time interval between two state points is within the equipment response delay range, and the parameter changes conform to the mass conservation equation of the condensation process, then a directed edge connection is established. The initial strength of the edge is calculated based on the transition similarity: for parameters of the same class, cosine similarity is used; for parameters of different classes, the Jacobian matrix partial derivative is used to fit the similarity. The initial weight of the edge is set to the reciprocal of the similarity value to ensure that highly similar transition paths have low weight strength.

[0074] The adaptive adjustment mechanism includes a dynamic update strategy and a distributed collaborative strategy. When new state point data is added, existing nodes within its adjacent time range are retrieved, and potential connecting edges are identified through a transformation relationship verification algorithm. Nodes affected by the new edge trigger local topology reconstruction, updating only the feature vectors of the affected node and its third-order neighbor nodes. When state point parameters are modified, a backpropagation mechanism is used: first, the feature vector of the node is corrected, and then propagation is performed along outgoing and incoming edges to neighboring nodes, iterating until the parameter change is below the propagation threshold. Edge strength adjustment is based on frequent path analysis in the operation records; the edge weights of high-frequency transformation paths decay according to a logarithmic function, while the edge weights of low-frequency transformation paths are superimposed with random perturbations. The updated graph structure is broadcast to each computing node through a distributed message queue, and the version number is bound to the transaction ID of the operation record.

[0075] The condensation state index graph is deployed in a distributed graph database environment, employing a sharded storage mechanism. State nodes are horizontally sharded according to time series, and transformation edges are vertically sharded according to device physical location. Graph traversal requests are distributed to the corresponding shards through a routing layer, and query results are merged using a near real-time strategy. Graph structure update operations follow a two-phase commit protocol to ensure consistency across shard transactions. Index efficiency optimization is achieved through a two-layer cache: a local cache stores the adjacency relationships of hot nodes, and a global cache stores the subgraph structure of frequently queried paths. The cache invalidation mechanism is linked to adaptive adjustment events to ensure data timeliness.

[0076] This implementation achieves closed-loop optimization of condensation parameters: real-time monitoring data drives the generation of parameter optimization vectors, a distributed index graph provides dynamic control paths, and an adaptive mechanism ensures execution reliability under complex operating conditions. The physical parameter analytical model integrates thermodynamic laws and operational experience, while the state index graph quantitatively represents the evolution of the system's operating state.

[0077] Example 2: See Figure 3 The node state embedding generation process of the condensation state index graph is pre-trained using a feature mapping algorithm. This algorithm employs an unsupervised learning mechanism to handle the topological relationships and node attributes in the graph. The original state point feature vector of each node is input into the graph neural network model and undergoes three convolutional operations to capture neighborhood information. The first convolutional layer aggregates the feature mean of direct neighbor nodes, the second convolutional layer introduces edge weight parameters for weighted aggregation, and the third convolutional layer fuses the node's own features with neighborhood features. The output vector is processed by dimensionality reduction to form a fixed-dimensional state embedding vector, which simultaneously encodes the node attributes and its structural position in the global graph. During pre-training, a negative sampling strategy is used to optimize the embedding space, making nodes with close physical transformation relationships closer together in the embedding space.

[0078] The parameter optimization vector is transformed into the state embedding space using a deep alignment model. This model comprises a dual-channel structure with a bidirectional long short-term memory network and an attention mechanism. The long short-term memory network resolves the temporal dependencies of the parameter optimization vector, recombining elements such as temperature deviation threshold and pressure regulation sensitivity coefficient according to their temporal correlation. The attention mechanism identifies the core and auxiliary parameters in the optimization intent and assigns differentiated encoding weights. The dual-channel outputs are matrix-concatenated in a fusion layer to generate an optimization intent vector with the same dimension as the state embedding space. This mapping process preserves the multi-objective characteristics of the original optimization vector, making the regulation intent present as a multi-focal distribution in the embedding space.

[0079] The matching degree calculation employs an improved similarity metric function. This function first calculates the cosine similarity between the optimized intent vector and the state embedding vectors of all nodes as the base matching value. A dynamic decay factor is then applied, adjusting the weights based on the node's freshness: the matching value of nodes added within the last three months is multiplied by a coefficient of 1.2, while the matching value of historical nodes decays linearly with time. Finally, a graph path constraint factor is introduced; when two nodes have a high-intensity transition edge, their matching values ​​generate a synergistic gain. After calculation, all nodes are sorted by their final matching values, and the top K nodes are selected as initial adjustment nodes. The value of K is dynamically adjusted based on the real-time system load, ranging from 5 to 15.

[0080] Neighborhood expansion employs a tiered threshold triggering mechanism. Starting from each initial adjustment node, its first-order neighbor node set is retrieved. For each neighbor node, real-time matching degree recalculation is performed: based on the initial matching degree and edge weight conversion coefficients, a temporary matching value for the current expansion level is generated. The preset dynamic threshold T is determined by the average matching degree of the initial adjustment node and the system operational complexity; the formula is a multivariate function, but the specific equation is not elaborated. When a neighbor node's temporary matching value ≥ T, it is included in the candidate set, and a second-order neighbor expansion for that node is initiated; otherwise, the branch path is terminated. The upper limit of the expansion depth is set to a third-order neighborhood, and a global candidate set capacity threshold is also set. When the total number of candidate nodes reaches the capacity threshold, the expansion process is immediately terminated to prevent system resource overload.

[0081] The boundary definition of the adjustment domain employs spatial clustering post-processing techniques. The candidate node set obtained from multi-level expansion is used to delineate the region boundaries using a density-based clustering algorithm. The core parameters of the algorithm include the neighborhood radius ε and the minimum number of contained points minPts, whose values ​​are automatically configured based on the physical distribution density of the devices. After clustering, several sub-adjustment domains are generated, with nodes within each sub-domain satisfying the connectivity requirements of the transformation path. For large-scale sub-domains spanning multiple devices, a secondary partitioning is implemented to ensure that a single sub-domain covers no more than three physical devices. The final output adjustment domain data format is a graph-structured subset, carrying all node features and connectivity relationships.

[0082] The generation of the candidate parameter set involves multi-dimensional feature extraction and fusion calculation. Four-dimensional operational features are extracted for each condensation state point within the regulation domain: temporal features include parameter duration and frequency of change cycles; spatial features include device location topology coding and sensor distribution density; operational features include frequency of manual intervention and proportion of automatic adjustment; and physical features include heat transfer efficiency and phase change stability indicators. These multi-dimensional features are independently normalized through parallel processing channels to eliminate dimensional differences. The multi-dimensional fusion algorithm employs a cross-attention mechanism: temporal and spatial features generate feature pair A, and operational and physical features generate feature pair B. A and B are then weighted by bidirectional attention, ultimately outputting a fused operational feature vector with a dimension one-quarter that of the original features.

[0083] The matching degree screening employs a triple verification mechanism. It fuses the operational feature vector and parameter optimization vector for similarity calculation, generating a basic matching degree based on a hybrid metric model of Manhattan distance and cosine similarity. The first verification compares the actual optimization performance of the state point under similar historical conditions; if the performance rating is below C, the matching degree is reduced by 30%. The second verification checks the current system load status, increasing the screening weight of stability indicators when the equipment is near full load. The third verification applies a real-time risk assessment matrix, imposing a matching degree decay factor on operational parameters that may trigger chain reactions. Candidate state points with preset thresholds are sorted in descending order of their original matching degree to form a primary candidate queue.

[0084] The final ranking output incorporates a multi-criteria decision-making model. A comprehensive evaluation matrix is ​​generated for each state point in the initial candidate queue. This matrix contains three dimensions: a technical dimension (matching degree and parameter adjustment sensitivity); an experience dimension (operator preference rating and number of historical successes); and a system dimension (implementation complexity and energy conversion efficiency). The weights of each criterion are dynamically configured using a weight vector generated during the optimization intent analysis phase. The ranking process employs an ideal solution approximation method: a positive ideal solution for the technical dimension is defined as the highest matching degree, a positive ideal solution for the experience dimension as the highest preference rating, and a negative ideal solution for the system dimension as the highest implementation complexity. The Euclidean distance between each candidate point and the ideal solution is calculated, and a final ranking list is generated based on comprehensive proximity. The top five are output as recommended parameter adjustment schemes.

[0085] The system architecture employs a pipelined parallel design. The matching computation layer is deployed on a GPU-accelerated cluster to handle intensive vector operations; the extended retrieval layer uses a distributed graph database to achieve near real-time neighborhood traversal; and the decision ranking layer runs on an in-memory computing engine, ensuring a ranking response latency of less than 50 milliseconds. All layers are connected via a high-speed data bus, and the data format uses a binary protocol to improve transmission efficiency. The entire processing flow establishes a breakpoint resumption mechanism, allowing execution to resume from the most recent verification point if any stage is interrupted.

[0086] Example 3: See Figure 4 The historical operation sequence analysis of the high-purity hydrogen fluoride condensation process employs a time-series segmentation algorithm, dividing continuous operation records into stable and variable operation segments. The segmentation process is based on a parameter fluctuation detection mechanism, constructing sliding time windows for core parameters such as temperature and pressure, with a window length of 1.5 times the system response period. The parameter change rate within each window is calculated using discrete differentiation. When the mean differential value of three consecutive windows exceeds a set threshold, it is marked as the starting point of the change, and so on, until the differential value of subsequent windows falls below the threshold, marking the ending point of the change. A stable operation segment is defined as a smooth interval between variable segments, with its parameter standard deviation not exceeding 3% of the equipment's nominal value. The operation segment division results are stored in the form of time interval labels, accompanied by statistical characteristics of the parameters within each segment.

[0087] The selection of parameter types to be optimized is based on the parameter difference analysis between the stable and variable sections. Two indices are calculated for each process parameter: the difference degree (D) reflects the magnitude of numerical change, and the sensitivity (S) characterizes the system response intensity. The difference degree (D) is defined as the relative deviation between the mean of the stable section and the extreme value of the variable section, and the sensitivity (S) is measured by the rate of change in condensation efficiency caused by parameter changes. Parameter selection conditions must meet the following:

[0088]

[0089] in: To optimize the potential value of the parameters, and As a weighted average of difference and sensitivity, and The normalization coefficient is... This is the filtering threshold. The saturation characteristics of the contribution of the function to the difference degree. The function enhances the discriminative power of sensitivity. It satisfies... The parameter types are entered into the set to be optimized. Their physical meaning is that they are process parameters that have both significant operational differences and are sensitive to system performance.

[0090] The optimization capability index of operation points within the stable operating segment is calculated using a cross-time period comparison method. After selecting the parameter type to be optimized, each time point with an operation record within the stable segment is designated as a candidate operation point, and operation points of the same type within the variable segment are designated as adjustment points. Candidate operation points With adjustment point The correlation strength is determined by the time delay factor. and data similarity The decision is made jointly. The time delay factor is calculated by applying an exponential decay process to the time difference between the two operations, reflecting the time decay effect of historical operations. Data similarity is calculated by aligning the operation trajectories of the two points using a dynamic time warping algorithm, and includes two components: parameter value similarity and operation action similarity.

[0091] Optimize capability indicators The generation process is divided into three stages: first, calculation For a single adjustment point Local optimization force This reflects the contribution of a point in a specific historical adjustment; then, the effects of all adjustment points are aggregated to obtain the global optimization force. This represents the average performance of the point in historical optimizations; finally, a stability correction coefficient is introduced. This coefficient is based on Calculation of the duration of the stable phase and the range of parameter fluctuations. Final optimization capability index. The larger the value, the more likely the operation point is to become a core node for optimization.

[0092] The calculation, integration, and optimization capabilities of core node probabilities, along with decision-making behavior characteristics. Decision-making power indicators. Reflecting the operation point The quality of historical decisions comprises three components: adjustment duration. Reflecting the degree of prudence in decision-making, and the interval between adjacent decisions. The ratio reflects the decision frequency and the number of times the parameter is repeated. This reflects the repeatability of the decision. These three components are combined through a product to form... Its mathematical characteristics emphasize long-term, low-frequency, and reproducible decision-making patterns. Core node probability. Obtained through normalization:

[0093]

[0094] Where: the denominator is all the candidate operation points. The sum of the products ensures the normalization of the probability values. The physical meaning of this formula is: it considers both the performance of the operation point in historical optimization (…). ), and also consider the reliability of its decision-making behavior ( This helps avoid misjudgments caused by relying solely on data similarity.

[0095] The optimized chain structure is constructed using a hierarchical expansion algorithm. First, a selection process is employed. Greater than the threshold The operation points are used as the core node set, and the threshold is... The value is adaptively adjusted based on system complexity, typically taking the upper quartile of the probability distribution. For each core node... Calculate all adjustment points Follow-up indicators This indicator is determined by the product of the intensity of the delayed impact and the decision-making power indicator, i.e. The following index reflects the strength of the historical following relationship between the adjustment point and the core node; the larger the value, the more likely the adjustment point is to be influenced by the core node.

[0096] The hierarchical division of the chain structure is based on the numerical distribution of the follower index. The values ​​are divided into five intervals based on the 0.2 quantile, corresponding to a five-layer chain structure: the first layer contains only the core node itself; the second layer includes... The adjustment point, among which and The fourth and fifth quantiles; and so on up to the fifth level. The adjustment point. Nodes within the same level have considerable following strength, and a radial structure from the core to the edge is formed between different levels. The final generated optimized chain structure is a directed acyclic graph, with edges pointing from higher levels to lower levels, and edge weights corresponding to the following index values.

[0097] The parameter optimization scheme is generated based on the hierarchical characteristics of the chain structure. A precise control strategy is adopted for nodes from the core to the third layer, with parameter adjustments controlled within ±5% of the nominal value. A range-based control strategy is adopted for nodes from the third to the fifth layer, allowing parameter fluctuations within ±15%. Nodes outside the fifth layer maintain parameter stability. The core idea of ​​this allocation method is that nodes closer to the core have a significant impact on the system and require precise control, while edge nodes can be more relaxed to reduce operational complexity. The control parameters for each layer are obtained by weighted averaging of the historical best values ​​of the nodes at that layer; the weights are the follow-up index values.

[0098] The system uses a time-series database to store historical operation records, with operation segment markers appended as metadata. The optimization capability index calculation module is deployed on the stream processing engine, receiving newly generated operation data in real time and updating index values. Core node probabilities are recalculated every 24 hours, dynamically adjusting and optimizing the chain structure. A visual interface displays the parameter distribution ranges at each level and presents the strength of the following relationships between nodes in the form of a heatmap. Operators can adjust threshold parameters using interactive tools, and the system automatically records manual interventions for subsequent model optimization.

[0099] The operational segment analysis mechanism established in this implementation deconstructs historical operational patterns from a time perspective and identifies key optimization nodes through quantitative indicators. The optimization chain structure organizes discrete operational points into an organic whole, preserving the decision-making authority of core nodes while also considering the operational flexibility of peripheral nodes. Probabilistic modeling avoids boundary effects caused by rigid divisions, and a hierarchical strategy achieves differentiated parameter control precision. The entire scheme provides a structured optimization decision-making framework while maintaining the continuity of the physical process.

[0100] Example 4: In the data comparison and analysis between stable and variable operating segments, the calculation of the optimization capability index of operating points requires processing multi-dimensional historical operating data. Taking temperature parameter optimization in an actual operating cycle as an example, five typical operating points (T1-T5) in the stable operating segment and three adjustment points (Q1-Q3) in the variable operating segment are selected for illustration. The time distribution characteristics of the operating points and parameter adjustment records are shown in Table 1.

[0101] Table 1: Time distribution characteristics of operation points and parameter adjustment records.

[0102]

[0103] The calculation of the delay effect intensity considers the time decay effect and operational correlation. Taking the stable segment operation point T2 and the variable segment adjustment point Q1 as an example, the time interval between their recording times is 1 hour and 45 minutes. The time decay coefficient is set to decay by 15% per hour, then the delay effect intensity of T2 on Q1 is 0.85^(1.75)≈0.72. When calculating the data similarity, the temperature change trajectories of T2 and Q1 in the 30 minutes before adjustment are first aligned. The waveform matching degree is obtained by using the dynamic time warping algorithm, which is 0.68. Then, combined with the similarity score of the adjustment actions of the two, which is 0.75, the final data similarity is (0.68+0.75) / 2=0.715. Therefore, the local optimization force of T2 on Q1 is L=0.72×0.715≈0.515.

[0104] The calculation of global optimization power requires aggregating the influence of all adjustment points. Continuing with T2 as an example, its local optimization power relative to Q2 and Q3 is calculated to be 0.61 and 0.42, respectively. Assuming the weight distribution of the three adjustment points (Q1-Q3) is 0.4, 0.35, and 0.25, then the global optimization power of T2 is G = 0.4 × 0.515 + 0.35 × 0.61 + 0.25 × 0.42 ≈ 0.526. The stability correction coefficient ξ is determined based on the duration characteristics of the stable segment in which T2 is located. This segment lasts for 3.2 hours with a temperature fluctuation range of ±0.8℃. After normalization, ξ = 0.85. Finally, the optimization capability index of T2 is C = 0.526 × 0.85 ≈ 0.447.

[0105] The decision-making power index reflects the decision-making quality characteristics of the operation point. Taking T2 as an example, the ratio of its adjustment duration of 15 minutes to the adjacent decision interval of 3.2 hours is 0.078, and the square root of the parameter recurrence count of 5 times is 2.236. Therefore, R = (15 / 192) × 2.236 ≈ 0.175. The calculation of the core node probability requires summing the C·R products of all candidate operation points as the denominator. Assuming the sum of the C·R products of the five stable operation points is 0.382, then the core node probability of T2 is P = 0.447 × 0.175 / 0.382 ≈ 0.205.

[0106] The follow-through index of adjustment points is calculated based on their historical association with the core node. Once T2 is identified as the core node, the follow-through of each adjustment point to it needs to be calculated. Taking Q1 as an example, its time delay intensity with T2 is 0.72. Q1's own decision-making power index is calculated to be 0.12 based on its adjustment duration of 25 minutes and the system average interval. Therefore, the follow-through index F = 0.72 × 0.12 = 0.086. Similarly, the follow-through indices of Q2 and Q3 to T2 are calculated to be 0.054 and 0.038, respectively. These values ​​will be used for hierarchical division during subsequent optimization chain construction.

[0107] The operation feature analysis module employs a sliding window mechanism to process real-time data streams. During feature extraction at each operation point, the system automatically configures the start and end times of the analysis window, with the window length dynamically adjusted based on the parameter type: a 30-minute window for temperature parameters and a 45-minute window for pressure parameters. The raw sampled data within the window is smoothed before statistical features such as mean, range, and rate of change are extracted. Operational behavior features record metadata such as the triggering method (automatic / manual), executing device, and completion status of the adjustment action. This feature data is stored in a distributed feature library for subsequent real-time queries.

[0108] The dynamic update mechanism for optimization capability indicators adopts an event-driven model. When the system detects a new operation record, it first determines the type of operation segment it belongs to. If it is a changing segment operation, the indicators of the relevant stable segment operation points are recalculated; if it is a stable segment operation, the characteristic data of that point is updated. The indicator recalculation process uses an incremental calculation strategy, updating only the affected nodes rather than performing a full calculation. The system maintains an operation point association graph, recording the potential influence relationships between points, for quickly locating the range of nodes that need to be updated.

[0109] The threshold setting for core node probabilities employs an adaptive algorithm. The system periodically (every 6 hours) statistically analyzes the probability distribution characteristics of all candidate operation points and calculates their quartile values. The initial threshold is set to the third quartile. When the system detects changes in operating conditions (such as equipment switching or raw material batch replacement), it automatically lowers the threshold by 5 percentage points to expand the range of core nodes. Threshold adjustment records are stored in conjunction with the operation log for subsequent analysis of the effectiveness of optimization decisions.

[0110] The visual monitoring interface displays the optimization chain structure in the form of a heatmap. Core nodes are shown as red circles, with their radius proportional to their probability. Adjustment points are displayed as blue rectangles of varying shades based on their follow-up index values, with directed arrows indicating the following relationship. Operators can view the historical evolution of the optimization chain using the timeline slider or click on any node to view its detailed characteristic data. The right panel of the interface displays the current optimal parameter adjustment suggestions in real time, including the target value range, priority adjustment equipment, and estimated impact duration.

[0111] The parameter optimization scheme generation module employs a tiered recommendation strategy. For parameters directly related to core nodes, the system provides adjustment suggestions accurate to 0.1℃; for adjustment points with strong tracking (tracking index > 0.05), the recommended adjustment range is controlled within ±2℃; for other nodes, only monitoring suggestions are provided. Each recommended scheme is accompanied by an impact assessment report, listing the parameters of adjacent processes that may be affected and their expected direction of change. After the operator confirms the execution, the system automatically records the deviation data between the actual adjustment value and the predicted value, which is used to optimize subsequent recommendation algorithms.

[0112] The historical data analysis module supports multi-dimensional operation pattern mining. The system provides optimization capability index analysis from multiple perspectives, including time (shift / day / week), equipment (reactor number), and raw material batch. Operators can compare the distribution differences of core nodes under different operating conditions or query the historical optimization trajectory of specific parameters. The analysis results can be exported as structured reports, including indicator trend charts and correlation network diagrams for each operation point, providing data references for process improvement.

[0113] The anomaly detection subsystem monitors the operational status of the optimization chain in real time. An anomaly alarm is triggered when the actual parameters of a core node deviate from the recommended values ​​for more than 15 minutes, or when the follow-up indicator suddenly drops by more than 50%. The alarm information includes the anomaly type, the scope of impact, and handling recommendations, and is sent to relevant personnel via audible and visual signals and mobile push notifications. The system automatically activates the anomaly handling contingency plan, including freezing relevant parameters and preparing backup equipment.

[0114] Example 5: The construction of the optimized chain structure begins with the identification of core nodes. The system iterates through the core node probability values ​​of all candidate operation points within the stable operation segment, and a preset probability threshold is dynamically adjusted based on real-time operating conditions. When the probability value of an operation point exceeds this threshold, the system automatically marks it as a core node and assigns it a unique identifier. The capacity limit of the core node set does not exceed 30% of the total number of operation points to avoid making the optimized chain structure too complex. The radiation range of each core node is automatically defined through historical correlation analysis, including adjustment points that have physical connections or operational logic correlations with it.

[0115] The calculation of the follow-up index of the adjustment point relative to the core node combines the time decay effect and decision quality assessment. The time decay factor is calculated based on the time interval between the adjustment point's operation and the core node; the longer the interval, the smaller the decay coefficient. The decision quality factor is calculated based on the historical success rate of the adjustment point's decisions and the frequency of parameter recurrence. The follow-up index is generated using a two-factor product model, and the result reflects the timeliness and reliability of the adjustment point's response to the core node's instructions. The index value is standardized to a range of 0 to 1 for easy subsequent hierarchical division.

[0116] The hierarchical structure is constructed using a quantile discretization strategy. All followability index values ​​are divided into five tiers based on the 0.2 quantile interval: the highest tier includes adjustment points for the top 20% of values, and the lowest tier includes adjustment points for the bottom 20%. Each core node independently generates a five-layer chain structure, with the first layer being the core node itself, and the subsequent four layers filled with adjustment points according to the strength of followability. Nodes at the same level follow an equivalence principle, meaning that nodes within the same level have approximately the same range of followability index values. The data storage for the chain structure uses a doubly linked list format, with pointers establishing predecessor-successor relationships between nodes.

[0117] The integrity verification mechanism of the chain structure is activated after the hierarchical construction is completed. The system checks whether each adjustment point is reasonably assigned to the radiation range of a core node. For operation points not covered by any chain, a compensation algorithm is activated. The compensation algorithm is based on the shortest path principle in graph theory, associating the operation point with the nearest neighbor core node chain. For adjustment points with multiple core node association conflicts, the system automatically assigns them to the core node chain with the highest followability index. The final optimized chain structure global view is a forest topology, with the root being the core node and the branches being adjustment points at different levels.

[0118] The hierarchical implementation of the optimization strategy adopts a three-stage segmentation model. The first adjustment point level shared by different core nodes is located in the optimization chain structure; this level is marked as the first critical layer of strategy segmentation. The system scans the number of child nodes in each layer after this critical layer, and marks the level where the peak number of child nodes is located as the second critical layer. In the hierarchical strategy binding mechanism of operation points, operation points from the core node to the first critical layer are bound to the first optimization strategy, which focuses on precise control characteristics. Operation points from the first critical layer to the second critical layer are bound to the second optimization strategy, emphasizing dynamic balance characteristics. Operation points from the second critical layer to the terminal node are bound to the third optimization strategy, emphasizing operational robustness.

[0119] The execution criteria for the first optimization strategy include parameter constraints, response speed, and accuracy requirements. The adjustment range of the target parameter is limited to a narrow fluctuation range, with each adjustment step not exceeding five percent of the base value. The control system checks parameter deviations every thirty seconds, triggering rapid correction within ten seconds of abnormal fluctuations. The temperature control accuracy for key process parameters is maintained at ±0.5 degrees Celsius, and the pressure control accuracy is maintained at ±0.02 MPa. Execution records include the timestamp of each adjustment, operator identity, and the deviation between the actual and target values.

[0120] The second optimization strategy establishes a range-based control framework. The target parameter is allowed to fluctuate within a 15% range above and below the baseline value, and the system dynamically adjusts the range width based on real-time load conditions. The control cycle is extended to a check every three minutes, and the response time is extended to thirty minutes. This layer of operation points supports batch coordinated adjustment, allowing up to eight related parameters to change simultaneously. The system automatically generates multiple parameter combination schemes for operators to choose from; the differences between the schemes mainly reflect the balance between energy consumption and quality control indicators.

[0121] The third optimization strategy implements the principle of minimal intervention. Parameters are maintained as they are unless absolutely necessary, and adjustments are only initiated when they exceed preset safety boundaries. The detection cycle is adjusted to fifteen minutes, and the response time is extended to two hours. The adjustment process adopts a step-by-step, gradual approach, with each adjustment not exceeding one-third of the total parameter range. The strategy automatically avoids peak production periods, prioritizing operation during equipment idle times. After each operation, the system monitors relevant process parameters for forty-eight hours, recording the time required to return to steady state.

[0122] The optimized solution's overall output uses a tree-structured navigation interface. Operators can expand the core node chain layer by layer to view the current parameter status and suggested adjustment directions for each operational point. The interface distinguishes three strategy zones with different colors: the core zone is marked in red, the balanced zone in yellow, and the robust zone in green. Clicking on any operational point brings up a details panel containing information such as real-time monitoring curves, historical adjustment records, and related equipment status. Adjustment decisions support cross-level collaborative operations, and the system automatically checks for strategy conflicts and provides a sequence of optimization suggestions.

[0123] A dynamic adjustment mechanism continuously tracks and optimizes the operational efficiency of the chain. Every six hours, the system calculates the policy compliance index for each level of operation. When the compliance of a certain level consistently falls below a set standard, a structural adjustment is triggered. The structural adjustment includes three dimensions: core node reset, level re-division, and policy parameter readjustment. Core node reset recalculates the probability distribution, level re-division adjusts the quantile cut-off points, and policy parameter readjustment optimizes the control parameter thresholds for each level. The entire process is executed asynchronously in the background, ensuring the continuity of foreground operations.

[0124] The fault-tolerant system establishes a multi-layered protection mechanism. When an abnormal parameter of a core node is detected, its management authority is automatically and temporarily transferred to an adjacent core node at the same layer. If an adjustment point goes offline, the system automatically freezes the automatic adjustment function of its chain and switches to manual monitoring mode. Cross-device parameter conflicts employ a priority arbitration mechanism, determining the processing order based on the position of the operation point in the process flow. All abnormal events generate independent analysis reports, indicating the root cause and processing trajectory.

[0125] The version control system stores the historical evolution record of the optimization chain. Each structural adjustment generates a new version snapshot, supporting structural backtracking and comparative analysis at any point in time. A version difference visualization function marks newly added core nodes, transfer operation points, and strategy change levels. Operators can load the optimization chain structure of historical versions to perform simulations and compare the expected performance differences of different versions under the same operating conditions. Version data is retained for thirty days, and important versions support permanent archiving.

[0126] This implementation constructs a hierarchical control system that allocates operation points of varying criticality through a hierarchical strategy. The chain structure maintains the control authority of core nodes while ensuring operational flexibility for edge nodes. A three-segment model balances the conflicting demands for precise control and operational robustness. A dynamic adjustment mechanism ensures the system continuously adapts to complex and ever-changing industrial environments. The entire solution organizes discrete process parameters into an organic whole, constructing a collaborative optimization framework based on physical correlations.

[0127] 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. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing condensation parameters of high-purity hydrogen fluoride, characterized in that, The method includes: Receive real-time monitoring data of the condensation process, use a physical parameter analysis model, analyze the condensation optimization intention based on the temperature parameters, pressure parameters and historical operation records in the real-time monitoring data, and generate a parameter optimization vector based on the condensation optimization intention; For the distributed storage of operational status data in a condensation system, a condensation status index graph is constructed. The condensation status index graph uses nodes to represent condensation status points and edges to represent the physical transformation relationships between these points. Signal processing techniques are used to extract condensation status point features from the operational status data. Each condensation status point is treated as a node in the condensation status index graph structure, and corresponding condensation status point features are assigned to each node. Edges are created in the condensation status index graph structure based on the physical transformation relationships between condensation status points, and an initial strength is set for each edge to obtain the condensation status index graph. The strength is initialized by comparing the transformation similarity between two condensation status points, and the graph structure and index efficiency are updated in real time based on an adaptive adjustment mechanism. Based on the parameter optimization vector, the adjustment domain related to the optimization intention is dynamically located in the condensation state index diagram using a multi-level mapping calculation method. Within the adjustment domain, a multi-dimensional fusion matching algorithm is used to filter and sort a set of candidate parameters that meet the optimization intent. The method further includes: Obtain the historical operation sequence of the high-purity hydrogen fluoride condensation process, and divide the process into stable operation segments and variable operation segments based on the parameter change trends in the historical operation sequence; Based on the differences in parameter values ​​between the stable operation segment and the variable operation segment, the types of parameters to be optimized are selected. For each parameter type to be optimized, based on the data similarity of the operation points in the stable operation segment and the variable operation segment, combined with time characteristics and parameter adjustment records, the optimization capability index of each operation point in the stable operation segment is obtained. Based on the parameter adjustment records and parameter reproduction data of each operation point in the changing operation segment, and combined with the optimization capability index, the core node probability of each operation point in the stable operation segment is obtained. Based on the probability selection of the core nodes, each operation point in the stable operation segment is selected as the core node. Combining the operation data of the same parameter type in the changing operation segment, an optimized chain structure is constructed, and an optimization scheme for different parameters is determined based on the optimized chain structure.

2. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 1, characterized in that, The real-time monitoring data of the receiving condensation process is used to analyze the condensation optimization intention based on the temperature parameters, pressure parameters, and historical operation records in the real-time monitoring data using a physical parameter analysis model. This analysis is then combined with the condensation optimization intention to generate a parameter optimization vector, including: Real-time monitoring data is mapped to a fixed-dimensional feature space through physical modeling to generate temperature feature embeddings and pressure feature embeddings. Acquire historical operation records, analyze historical temperature fluctuations, historical pressure changes and operation adjustment behaviors, construct operation profiles, and transform operation profile data into operation preference embeddings; Access the condensation knowledge base, extract physical constraint features related to real-time monitoring data, and generate constraint feature embeddings; A weighted fusion-based algorithm is used to integrate temperature feature embedding, operation preference embedding, and constraint feature embedding to perform condensation optimization intent parsing. Each parsed intent is assigned an optimization weight, and a parameter optimization vector is generated based on the optimization weight. The parameter optimization vector covers the current optimization intent and reflects the system operation characteristics.

3. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 2, characterized in that, The aforementioned method constructs a condensation state index graph for the distributed storage of operational state data in the condensation system. This index graph uses nodes to represent condensation state points and edges to represent the physical transformation relationships between these points. The graph structure and index efficiency are updated in real-time based on an adaptive adjustment mechanism, including: Based on the adaptive adjustment mechanism, the changes in the parameters of the condensation state points and their transformation relationships are tracked to ensure the real-time performance of the condensation state index graph. When a new condensation state point is added or modified, an incremental adjustment strategy is used to update only the affected nodes and edges, and the strength of the edges is dynamically adjusted according to the operation records. Deploy a real-time condensation state index graph in a distributed system.

4. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 3, characterized in that, The adjustment domain related to the optimization intention is dynamically located in the condensation state index map using a multi-level mapping calculation based on the optimized vector according to the parameters. include: The condensation state index graph is pre-trained using a feature mapping algorithm to generate the state embedding of each node in the graph. The parameter optimization vector is mapped to the state embedding space using a depth alignment model; Calculate the matching degree between the mapped parameter optimization vector and all state embeddings, and select a specified number of nodes with the highest matching degree from the condensation state index graph as the initial adjustment nodes; Based on the neighboring nodes of each initial adjustment node, expand outward to the first level of neighboring nodes, determine whether the matching degree of the neighboring nodes reaches a specific threshold. If it does not reach the threshold, stop the expansion. If it does reach the threshold, add it to the candidate node set and repeat the expansion process to continue exploring the next level of neighboring nodes until the preset level limit is reached or the cumulative number of nodes is met. The set of all candidate nodes obtained through multi-level mapping expansion is determined as the regulation domain.

5. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 4, characterized in that, Within the adjustment domain, a multi-dimensional fusion matching algorithm is used to filter and sort a set of candidate parameters that meet the optimization intent, including: For each condensation state point related to the optimization intention contained within the regulation domain, multi-dimensional operational features are extracted; A multi-dimensional fusion algorithm is used to fuse operational features from different dimensions to generate fused operational features for each condensation state point. Calculate the matching degree between the fusion operation feature and the parameter optimization vector of each condensation state point, and filter out condensation state points with a matching degree higher than a preset threshold as candidate parameter sets; The candidate parameter set is sorted according to matching degree, operation preference, and parameter priority.

6. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 1, characterized in that, Under each parameter type to be optimized, based on the data similarity of the operation point in the stable operation segment and the variable operation segment, combined with time characteristics and parameter adjustment records, the optimization capability index of each operation point in the stable operation segment is obtained, including: Under any parameter type to be optimized, all points containing operational data in the stable operation segment are taken as candidate operation points, and all points containing operational data in the variable operation segment are taken as adjustment points. Record any candidate operation point as the selected candidate point, and any adjustment point as the selected adjustment point; Based on the parameter adjustment time and adjustment duration of the selected candidate point and the selected adjustment point, the delay influence intensity of the selected candidate point relative to the selected adjustment point is obtained; Calculate the similarity between the data of each adjustment behavior of the selected candidate point in the stable operation segment and the data of each adjustment behavior of the selected control point in the variable operation segment, and obtain the similarity feature factor corresponding to the selected control point under each adjustment behavior; Using the intensity of the delay effect as a weight, the similarity feature factors corresponding to the selected adjustment point under each adjustment behavior are weighted and averaged to obtain the parameter optimization force of the selected candidate point relative to the selected adjustment point. The mean of the parameter optimization force of the selected candidate point relative to all adjustment points is used as the optimization capability index of the selected candidate point.

7. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 6, characterized in that, The process of obtaining the core node probability of each operation point within a stable operation segment by combining the parameter adjustment records and parameter reproduction data of each operation point within the changing operation segment with the optimization capability index includes: Based on the adjustment duration and decision interval of each adjustment behavior of the selected candidate points, and combined with the number of parameter repetitions, the decision-making power index of the selected candidate points is obtained. The product of the optimization capability index and the decision-making capability index of the selected candidate point is normalized to generate the core node probability of the selected candidate point.

8. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 7, characterized in that, The step of selecting each operation point within a stable operation segment as a core node based on the probability of the core node, and constructing an optimized chain structure by combining operation data of the same parameter type in the changing operation segment, includes: The candidate operation points with a core node probability greater than a preset probability threshold are respectively used as the core nodes of each chain in the optimized chain structure; Obtain the decision-making power index for each adjustment point, and multiply the delay impact intensity of each adjustment point and the core node with the decision-making power index of the adjustment point as the following indicator of each adjustment point relative to the core node. For any core node, construct a chain structure in descending order of the following index of each adjustment point relative to that core node, with nodes at the same level in the chain structure having the same following index size. The chain structure of all core nodes constitutes an optimized chain structure.

9. The method for optimizing high-purity hydrogen fluoride condensation parameters according to claim 8, characterized in that, The process of determining optimization schemes with different parameters based on the optimization chain structure includes: In optimizing the chain structure, the layer containing the first shared child node of different core nodes is taken as the first target layer; after the first target layer, the layer with the most child nodes is taken as the second target layer. A preset first optimization strategy is used for operation points between the core node and the first target layer, a preset second optimization strategy is used for operation points between the first target layer and the second target layer, and a preset third optimization strategy is used for operation points between the second target layer and the bottom layer.

Images