A feature bag priority migration method for reaction heat behavior cross-system discrimination

Abstract

The present application relates to the technical field of thermal behavior identification, and particularly relates to a feature package priority migration method for reaction thermal behavior cross-system identification, comprising: obtaining a plurality of identification samples; a feature extraction module performing multi-dimensional feature extraction on reaction thermal behavior of any identification sample and obtaining identification accuracy of any feature; a feature screening module calculating average identification accuracy of any feature to classify the feature into a feature package and determining priority of the feature package; a central control module adjusting a preset identification accuracy standard at a second feature quantity level by counting the number of screened features and determining a feature package quantity level and adjusting an identification accuracy interval when the feature package quantity does not meet the standard; and selecting a best feature package combination according to the priority of the feature package for identification of specific samples, and migrating the method when identifying a new reaction system thermal behavior sample. The present application can fully exert the high classification accuracy advantage of high-priority feature packages.

Description

Technical Field

[0001] This invention relates to the field of thermal behavior discrimination technology, and in particular to a feature packet priority transfer method for cross-system discrimination of reaction thermal behavior. Background Technology

[0002] In identifying the intrinsically safe (IS), non-initiated (NI), thermal runaway (TR), and rapid and safe state (QFS) thermal behaviors of isothermal semi-batch reaction systems, current research mainly focuses on single reaction systems, emphasizing the establishment of connections (i.e., criteria) between single characteristics and thermal behaviors, lacking in-depth research on the generalization of thermal behavior characteristics across systems. It is noteworthy that different criteria often yield different results for the same reaction system, possibly because: 1) different criteria use different information (characteristics) to represent the reaction system; 2) existing criteria mostly employ single characteristics, which may miss important information and fail to fully characterize various thermal behaviors. Furthermore, the thermal behaviors of different reaction systems are influenced by numerous factors such as reaction thermodynamic parameters and operating conditions, exhibiting cross-system differences, which significantly limits the application scope of the criteria. Therefore, establishing universal cross-system criteria with excellent identification performance is crucial for identifying the thermal behavior of reaction systems.

[0003] The above analysis shows that the key to thermal behavior identification lies in feature extraction and recognition, which essentially falls under the category of pattern recognition. The successful applications of pattern recognition methods in recent years in fields such as speech recognition, image processing, and fault detection have provided new ideas for the characterization and identification of thermal behavior in reaction systems, and also offered new avenues for the study of cross-system thermal behavior criteria.

[0004] This invention designs a pattern classification method for cross-system discrimination of thermal behavior in isothermal semi-batch reactions. The purpose is to discard redundant features with poor accuracy based on the role played by each individual feature, group features with similar recognition accuracy into a feature package, set feature package priorities based on the recognition accuracy of different feature packages, and adaptively select the best feature package combination, thereby achieving high-precision cross-system discrimination of thermal behavior of reaction systems. Summary of the Invention

[0005] To address this, the present invention provides a feature packet priority transfer method for cross-system discrimination of reaction thermal behavior, which overcomes the problem that thermal behavior discrimination in the prior art is often limited to a single reaction system and the accuracy is low due to the single thermal behavior discrimination feature.

[0006] To achieve the above objectives, the present invention provides a feature packet priority transfer method for cross-system discrimination of reaction thermal behavior, comprising:

[0007] Step S1: Obtain several identification samples. The feature extraction module performs multi-dimensional feature extraction of reaction thermal behavior on any of the identification samples and obtains the identification accuracy of any feature.

[0008] Step S2: The feature filtering module calculates the average recognition accuracy of any of the features, and filters the features based on the comparison results between the average recognition accuracy and the preset recognition accuracy standard.

[0009] Step S3: The feature filtering module determines the recognition accuracy boundary based on the recognition accuracy interval, classifies the filtered features into feature packages based on the recognition accuracy boundary, and determines the priority of the feature packages.

[0010] Step S4: The central control module counts the number of filtered features and adjusts the preset recognition accuracy standard at the second feature quantity level. The feature filtering module then re-filters the features based on the adjusted preset recognition accuracy standard.

[0011] Step S5: The central control module calculates the number of feature packets and determines the feature packet quantity level. If the feature packet quantity does not meet the standard, the adjustment method for the recognition accuracy interval is determined according to the feature packet quantity level.

[0012] Step S6: For the identification of specific samples, the best feature package combination is selected according to the priority of the feature package. When identifying samples of thermal behavior of new reaction systems, the method is transferred.

[0013] Further, in step S2, the feature filtering module calculates the average recognition accuracy P of any one of the features, and sets...

[0014]

[0015] Among them, F i Let be the accuracy of any feature in the i-th identified sample, where i ranges from 1 to m, and m is the total number of identified samples;

[0016] The feature filtering module has a preset recognition accuracy standard P0. The feature filtering module compares the average recognition accuracy P of a single feature with the preset recognition accuracy standard P0 to determine whether the recognition accuracy of a single feature meets the standard.

[0017] If the average recognition accuracy of a single feature is at the first accuracy level, the feature filtering module determines that the recognition accuracy of the feature meets the standard.

[0018] If the average recognition accuracy of a single feature is at the second accuracy level, the feature filtering module determines that the recognition accuracy of the feature does not meet the standard and removes the feature.

[0019] The first accuracy level satisfies P≥P0, and the second accuracy level satisfies P<P0.

[0020] Further, in step S3, the feature filtering module categorizes features at the first accuracy level based on the average recognition accuracy. The feature filtering module has several recognition accuracy boundaries, including a first recognition accuracy boundary P1, a second recognition accuracy boundary P2, ..., the nth recognition accuracy boundary P... n Where, P1 < P2 < ... < P n The feature selection module compares the average recognition accuracy P of a single feature at the first accuracy level with the recognition accuracy boundary, where...

[0021] If the result is the first accuracy comparison, the feature filtering module classifies the feature into the first feature package and determines the priority of the first feature package as the first priority.

[0022] If the result is the second accuracy comparison, the feature filtering module classifies the feature into the second feature package and determines the priority of the second feature package as the second priority.

[0023] If the accuracy comparison result is n, the feature filtering module classifies the feature into the nth feature package and determines the priority of the nth feature package as the nth priority.

[0024] If it is the (n+1)th accuracy comparison result, the feature filtering module classifies the feature into the (n+1)th feature package and determines the priority of the (n+1)th feature package as the (n+1)th priority.

[0025] The first accuracy comparison result satisfies P ≥ P n The second accuracy comparison result satisfies P. n-1 ≤P<P n The accuracy comparison result of the nth time satisfies P1≤P<P2, and the accuracy comparison result of the (n+1)th time satisfies P0≤P<P1. The priority level is first priority > second priority > nth priority > n+1th priority.

[0026] Furthermore, the method for determining the recognition accuracy boundary is as follows: the feature filtering module includes a recognition accuracy interval r0 and a maximum recognition accuracy P. max The feature filtering module uses a preset recognition accuracy standard P0 as the baseline, and sets a recognition accuracy boundary every time a recognition accuracy interval r0 is passed.

[0027] The first recognition accuracy boundary is P1 = P0 + r0;

[0028] The second recognition accuracy boundary is P2 = P0 + 2r0;

[0029] The nth recognition accuracy boundary P n =P0+nr0;

[0030] Among them, P n <P max .

[0031] Furthermore, in step S4, the central control module compares the number of filtered features M with the preset feature number standard M0 to determine whether the number of filtered features meets the standard.

[0032] If the number of filtered features is at the first feature number level, the central control module determines that the number of filtered features meets the standard;

[0033] If the number of filtered features is at the second feature number level, the central control module determines that the number of filtered features does not meet the standard and needs to adjust the preset recognition accuracy standard.

[0034] The first feature quantity level satisfies M≥M0, and the second feature quantity level satisfies M<M0.

[0035] Furthermore, the central control module is equipped with an adjustment method for the preset recognition accuracy standard under the second feature quantity level;

[0036] The first method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the first preset recognition accuracy standard.

[0037] The second method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the second preset recognition accuracy standard.

[0038] The third method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the third preset recognition accuracy standard.

[0039] The feature filtering module re-filters the features according to the adjusted preset recognition accuracy standard;

[0040] First preset recognition accuracy standard < Second preset recognition accuracy standard < Third preset recognition accuracy standard < Preset recognition accuracy standard.

[0041] Furthermore, in step S5, the central control module calculates the preset recognition accuracy standard P. j With maximum recognition accuracy P max The number Q of recognition accuracy intervals r0 contained between them is set.

[0042]

[0043] The central control module calculates the number of feature packets R.

[0044] If Q is a positive integer, then set R = Q;

[0045] If Q is not a positive integer, then set R to be the smallest positive integer greater than Q;

[0046] Wherein, at the first feature quantity level, a preset recognition accuracy standard P is set. j j = 0;

[0047] At the second feature quantity level, a preset recognition accuracy standard P is set. j j = a, b, c.

[0048] Furthermore, the central control module is equipped with a minimum feature packet number threshold R. min and the maximum feature packet number threshold R max The central control module will compare the number of feature packets R with R... min and R max A comparison is performed to determine whether the number of feature packets meets the standard;

[0049] If the number of feature packets is at the first feature packet level, the central control module determines that the number of feature packets is too large and does not meet the standard, and the recognition accuracy interval needs to be increased.

[0050] If the number of feature packets is at the second level, the central control module determines that the number of feature packets meets the standard;

[0051] If the number of feature packets is at the third feature packet level, the central control module determines that the number of feature packets is too small and does not meet the standard, and the recognition accuracy interval needs to be reduced.

[0052] Wherein, the first feature packet quantity level satisfies R > R max The second feature packet quantity level satisfies R min ≤R≤R max The number level of the third feature packet satisfies R < R min .

[0053] Furthermore, the central control module is equipped with a method to increase the recognition accuracy interval under the first feature packet quantity level;

[0054] The first method to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the first increased recognition accuracy interval;

[0055] The second way to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the second increased recognition accuracy interval;

[0056] The third way to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the third increased recognition accuracy interval;

[0057] Among them, the first increase in recognition accuracy interval > the second increase in recognition accuracy interval > the third increase in recognition accuracy interval > the recognition accuracy interval.

[0058] Furthermore, the central control module is equipped with a method to reduce the recognition accuracy interval at the third feature packet quantity level;

[0059] The first way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the first reduced recognition accuracy interval;

[0060] The second way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the second reduced recognition accuracy interval.

[0061] The third way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the third reduced recognition accuracy interval;

[0062] Among them, the first interval for reducing recognition accuracy is less than the second interval for reducing recognition accuracy, which is less than the third interval for reducing recognition accuracy.

[0063] This invention also protects a feature packet priority transfer method for cross-system discrimination of reaction thermal behavior, the steps of which are:

[0064] Obtain the curves of state variables changing with the reaction under different thermal behavior conditions, and extract multi-dimensional features based on the curves of state variable changes;

[0065] The average recognition accuracy of each feature in the multi-dimensional features is calculated based on the total number of recognized samples.

[0066] Set a preset recognition accuracy standard P0 and a recognition accuracy interval r0, and calculate the maximum recognition accuracy P for each feature. max Calculate the preset recognition accuracy standard P0 and the maximum recognition accuracy P max The number Q of recognition accuracy intervals r0 between them, and the preset recognition accuracy standard P0 and the maximum recognition accuracy P based on the recognition accuracy intervals r0. max The values ​​within the range are divided into the first recognition accuracy boundary P1, the second recognition accuracy boundary P2, ..., the nth recognition accuracy boundary P n Where, P1 < P2 < ... < P n ;

[0067] The average recognition accuracy P of a single feature that is greater than the preset recognition accuracy standard P0 is compared with each recognition accuracy boundary. The features are sorted from high to low according to the average recognition accuracy of the single feature. Features with average recognition accuracy within the same recognition accuracy range are grouped into a feature package. A recognition accuracy interval is set between two adjacent feature packages. A feature package priority structure is established according to the recognition accuracy, with feature packages with higher recognition accuracy having higher priority.

[0068] Multi-dimensional feature extraction was performed on the thermal behavior samples of the new reaction system. Feature packages with different priorities were obtained according to the feature package priority structure, and the best feature package combination was selected for identification.

[0069] Compared with existing technologies, the advantages of this invention lie in that it groups features with similar average recognition accuracy into a feature package and establishes a feature package priority structure based on average recognition accuracy. Based on feature package priority, the high classification accuracy advantage of high-priority feature packages can be fully utilized.

[0070] Furthermore, research has found that increasing feature dimensions can effectively improve the accuracy of thermal behavior recognition. This invention evaluates the effectiveness of a single feature by calculating its average recognition accuracy across several samples, and discards redundant features with poor accuracy by setting a preset recognition accuracy standard P0. Through this technical solution, the accuracy of thermal behavior recognition is improved on the one hand, and the efficiency of thermal behavior recognition is improved by eliminating redundant features on the other.

[0071] Furthermore, this invention sorts features from highest to lowest based on their average recognition accuracy, and then groups features with similar average recognition accuracy into one feature bag. Features with significantly different average recognition accuracy are placed in different feature bags. This invention sets a recognition accuracy interval between adjacent feature bags to divide them into feature bags. A feature bag priority structure is established according to recognition accuracy. Based on the feature bag priority, the high classification accuracy advantage of high-priority feature bags can be fully utilized.

[0072] Furthermore, appropriate feature dimensions can effectively improve the accuracy of thermal behavior recognition. This invention uses a preset feature quantity standard to judge whether the number of features meets the standard, and lowers the preset recognition accuracy standard when the number of features does not meet the standard in order to appropriately increase the number of selected features, thereby further improving the accuracy of thermal behavior recognition.

[0073] Furthermore, when lowering the preset recognition accuracy standard to appropriately increase the number of selected features, this invention calculates the difference between the number of selected features and the preset feature quantity standard. Based on the value of this difference, different adjustment coefficients are used to adjust the preset recognition accuracy standard. When the number of selected features is significantly lower, the preset recognition accuracy standard is lowered more significantly; conversely, when the number of selected features is less lower, the preset recognition accuracy standard is lowered less significantly. This achieves an appropriate increase in the number of selected features. By setting recognition accuracy standard adjustment coefficients to adjust the preset recognition accuracy standard in stages and limiting the value of these coefficients, this invention avoids insufficient selected features or redundant features among the selected features, further improving the recognition accuracy of thermal behavior.

[0074] Furthermore, an appropriate number of feature packets can effectively improve the accuracy of thermal behavior recognition. This invention achieves this by setting a minimum feature packet number threshold R. min and the maximum feature packet number threshold R max To determine whether the number of feature packets meets the standard, and when the number of feature packets does not meet the standard, to determine the adjustment method for the recognition accuracy interval, so as to control the number of feature packets within a reasonable range, thereby improving the recognition accuracy and recognition efficiency of thermal behavior.

[0075] Furthermore, when adjusting the recognition accuracy interval, the present invention, based on the experience of adjusting the preset recognition accuracy standard, also adjusts the recognition accuracy interval to the corresponding value by setting the recognition accuracy interval adjustment coefficient to increase or decrease the number of feature packets, so as to control the number of feature packets within a reasonable range, thereby improving the recognition accuracy and recognition efficiency of thermal behavior.

[0076] Furthermore, this invention introduces the migration method into the cross-system thermal behavior discrimination of isothermal semi-batch reaction systems, thereby improving the classification accuracy. Attached Figure Description

[0077] Figure 1 This is a flowchart of a feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to an embodiment of the present invention;

[0078] Figure 2 This is a schematic diagram of the feature packet priority migration method according to an embodiment of the present invention;

[0079] Figure 3 This is a flowchart of the automatic feature extraction process according to an embodiment of the present invention;

[0080] Figure 4(a) shows the ξ during the isothermal semi-intermittent homogeneous reaction process. B Curve showing change over time;

[0081] Figure 4(b) shows the ξ during the isothermal semi-intermittent homogeneous reaction process. ac Curve showing change over time;

[0082] Figure 4(c) shows the curve of τ changing with time during the isothermal semi-intermittent homogeneous reaction process;

[0083] Figure 5(a) shows the intrinsically safe thermal behavior of the isothermal semi-batch homogeneous reaction under the condition of ξ. B Curve showing change over time;

[0084] Figure 5(b) shows the intrinsically safe thermal behavior of the isothermal semi-batch homogeneous reaction under the condition of ξ. ac Curve showing change over time;

[0085] Figure 5(c) shows the curve of τ versus time for an isothermal semi-intermittent homogeneous reaction under intrinsically safe thermal behavior. Detailed Implementation

[0086] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0087] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0088] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0089] Please see Figure 1 and Figure 2 As shown, Figure 1 This is a flowchart of a feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the feature packet priority migration method according to an embodiment of the present invention; the feature packet priority migration method for cross-system discrimination of reaction thermal behavior according to the present invention includes:

[0090] Step S1: Obtain several identification samples. The feature extraction module performs multi-dimensional feature extraction of reaction thermal behavior on any of the identification samples and obtains the identification accuracy of any feature.

[0091] Step S2: The feature filtering module calculates the average recognition accuracy of any of the features, and filters the features based on the comparison results between the average recognition accuracy and the preset recognition accuracy standard.

[0092] Step S3: The feature filtering module determines the recognition accuracy boundary based on the recognition accuracy interval, classifies the filtered features into feature packages based on the recognition accuracy boundary, and determines the priority of the feature packages.

[0093] Step S4: The central control module counts the number of filtered features and adjusts the preset recognition accuracy standard at the second feature quantity level. The feature filtering module then re-filters the features based on the adjusted preset recognition accuracy standard.

[0094] Step S5: The central control module calculates the number of feature packets and determines the feature packet quantity level. If the feature packet quantity does not meet the standard, the adjustment method for the recognition accuracy interval is determined according to the feature packet quantity level.

[0095] Step S6: For the identification of specific samples, the best feature package combination is selected according to the priority of the feature package. When identifying samples of thermal behavior of new reaction systems, the method is transferred.

[0096] When identifying thermal behavior samples of new reaction systems, the above method is transferred to automatically select the optimal feature package combination for thermal behavior identification. This invention introduces the transfer method into the cross-system thermal behavior discrimination of isothermal semi-batch reaction systems, improving classification accuracy.

[0097] This invention groups features with similar average recognition accuracy into a feature pack, and establishes a feature pack priority structure based on average recognition accuracy. Based on this feature pack priority, the higher classification accuracy of high-priority feature packs can be fully utilized.

[0098] In step S1, the feature extraction method includes:

[0099] Artificial feature extraction: Figures 4 and 5 show the state variables (such as dimensionless conversion rate ξ) under different thermal behaviors of the isothermal semi-batch homogeneous reaction system. B Dimensionless cumulative degree ξ ac The curves showing the change of ξ (dimensional temperature τ, etc.) with the reaction are shown in Figures 4 and 5. Among them, Figure 4(a) shows the change of ξ during the isothermal semi-batch homogeneous reaction. B The curve of ξ changing with time, Figure 4(b) shows the ξ during the isothermal semi-intermittent homogeneous reaction process. acThe curves showing the change of τ over time are shown in Figure 4(c). In Figures 4(a), 4(b), and 4(c), the dotted arrows indicate the direction of jacket temperature increase. The dashed, solid, and dotted lines represent three different thermal behaviors: dashed line represents NI (non-initiated), solid line represents TR (thermal runaway), and dotted line represents QFS (rapid and safe state). The figure shows the change of τ over time. j ξ is the dimensionless jacket temperature, and θ is the dimensionless time. Figure 5(a) shows the intrinsically safe thermal behavior of the isothermal semi-batch homogeneous reaction under the condition of ξ. B The curves showing the change over time are shown in Figure 5(b), which illustrates the intrinsically safe thermal behavior of the isothermal semi-intermittent homogeneous reaction. ac The curves showing the change of τ over time are shown in Figure 5(c), which illustrates the change of τ over time in an isothermal semi-intermittent homogeneous reaction under intrinsically safe thermal behavior. In Figures 5(a), 5(b), and 5(c), the dotted arrows indicate the direction of jacket temperature increase. The figures also show the changes in τ over time. j Here, θ represents the dimensionless jacket temperature, and θ represents the dimensionless time. For different thermal behaviors, multiple features are extracted from the state variable change curves, including the maximum curvature of the conversion rate during the rising phase, the slope of the conversion rate at the initial moment, the maximum slope of the conversion rate-time curve, the maximum value of the conversion rate, the maximum value of the cumulative degree, and the maximum curvature of the temperature-time curve during the concave phase. The types and number of multi-dimensional features extracted for different reaction thermal behaviors can be the same, but their values ​​may differ.

[0100] Feature self-extraction: In terms of feature extraction of reaction thermal behavior, the inventors have conducted a thorough literature review and have not found any research reports on feature self-extraction. Feature self-extraction can generate certain features that are difficult to discover manually, especially some implicit features whose physical meaning is not clear but are effective for identification.

[0101] Therefore, this invention introduces autoencoders (deep learning) into the study of self-extraction of thermal behavior features, and proposes a cascaded autoencoder, such as... Figure 3 As shown, an autoencoder consists of an encoder and a decoder, both essentially neural network layers. The encoder is used for feature extraction, while the decoder uses the extracted features to reconstruct the (input) signal to verify the effectiveness of the feature extraction. Compared to a single autoencoder, cascaded autoencoders can achieve deep dimensionality reduction of large datasets in complex reaction systems, which is beneficial for removing redundant information and thus helps improve the accuracy of identifying complex reaction thermal behaviors.

[0102] Based on the two feature extraction approaches mentioned above, this invention uses both manually extracted and automatically extracted features as inputs to the feature package priority transfer method for cross-system discrimination of reaction thermal behavior, so as to give full play to the intuitiveness of manually extracted features and the convenience of automatically extracted features.

[0103] Feature mining research, as a crucial foundation of pattern recognition, directly impacts the reliability of thermal behavior representation. Previous research by the inventors revealed that increasing feature dimensions can effectively improve the accuracy of thermal behavior recognition. To comprehensively and accurately represent different thermal behaviors, this invention includes both manual and automatic thermal behavior representation. 1) To effectively represent various thermal behaviors, a systematic analysis of the state parameter changes in the reaction system, combined with thermal behavior category attributes, is proposed to mine effective combinations of linear and nonlinear features that characterize various thermal behaviors; this process is the manual feature extraction process. 2) Based on the systematic analysis of manually extracted features, automatic feature extraction is performed, fully comparing manually extracted and automatically extracted features to optimize feature combinations and extraction methods to adapt to complex reaction scenarios.

[0104] In step S6, for the identification of a specific sample, the optimal feature package combination is selected based on the feature package priority structure. Initially, the target number of feature packages can be set to obtain the characteristics of the sample under the reactive thermal behavior to be identified. The feature package priority structure of the sample's characteristics under the reactive thermal behavior to be identified is then determined according to the feature package priority structure. The feature packages are calculated sequentially from high to low priority to determine whether they meet the requirements of the current reactive thermal behavior. If the target number of feature packages is reached during the calculation, the feature packages that meet the requirements at that time are selected for identification, and subsequent feature packages are not evaluated, thus determining the optimal feature package combination. If no feature package meets the requirements of the current reactive thermal behavior after traversing all feature packages, the two feature packages with the highest priority are selected for identification, thus determining the optimal feature package combination.

[0105] This invention employs the k-nearest neighbor (k-NN) method to calculate the Euclidean distance between each feature packet of the sample to be identified and the training samples. If the calculated Euclidean distance is less than a set value, the feature packet is considered valid; otherwise, it is invalid.

[0106] For example, the specific implementation of whether the current reaction thermal behavior requirement is met is as follows: calculate the minimum Euclidean distance between the feature packs of the sample and all training samples. If the minimum Euclidean distance is less than a set value, the feature pack is considered valid; otherwise, it is invalid. If it is still invalid after traversing all feature packs, the feature packs with the first two priorities are determined as the best feature pack combination.

[0107] For example, the present invention can also sort all Euclidean distances from smallest to largest according to the Euclidean distance between the calculated sample and all training samples, count the top 10 training samples' reaction-thermal behavior categories, and use the reaction-thermal behavior category with the largest number of training samples as the reaction-thermal behavior category of the new sample.

[0108] Based on the established feature package priorities, the optimal feature package combination suitable for the thermal behavior of a new reaction system is automatically selected when identifying samples of the new reaction system's thermal behavior. This method allows for the selection of the most suitable feature information for the thermal behavior samples of a new reaction system while eliminating redundant information, thereby improving the accuracy and efficiency of cross-system thermal behavior identification. The above classification process uses homogeneous reactions as an example, but is not limited to the identification of homogeneous reaction thermal behavior.

[0109] This invention classifies data samples of a single thermal behavior reaction system without involving loops, with the basic statement execution count being constant and the computational complexity being O(1), making it very suitable for online processing.

[0110] Specifically, in step S2, the feature filtering module calculates the average recognition accuracy P of any of the features, and sets...

[0111]

[0112] Among them, F i Let be the accuracy of any feature in the i-th identified sample, where i ranges from 1 to m, and m is the total number of identified samples;

[0113] The feature filtering module has a preset recognition accuracy standard P0. The feature filtering module compares the average recognition accuracy P of a single feature with the preset recognition accuracy standard P0 to determine whether the recognition accuracy of a single feature meets the standard.

[0114] If the average recognition accuracy of a single feature is at the first accuracy level, the feature filtering module determines that the average recognition accuracy of the feature meets the standard.

[0115] If the average recognition accuracy of a single feature is at the second accuracy level, the feature filtering module determines that the average recognition accuracy of the feature does not meet the standard and removes the feature.

[0116] The first accuracy level satisfies P≥P0, and the second accuracy level satisfies P<P0.

[0117] In this embodiment, the preset recognition accuracy standard P0 is set to 60% < P0 < 80%, and P0 = 62% is preferred in this embodiment.

[0118] Research has found that increasing feature dimensions can effectively improve the accuracy of thermal behavior recognition. This invention evaluates the effectiveness of a single feature by calculating its average accuracy across several recognition samples, and discards redundant features with poor accuracy by setting a preset recognition accuracy standard P0. Through this technical solution, the accuracy of thermal behavior recognition is improved on the one hand, and the efficiency of thermal behavior recognition is improved by eliminating redundant features on the other.

[0119] Specifically, in step S3, the feature filtering module categorizes features at the first accuracy level based on the average recognition accuracy. The feature filtering module has several recognition accuracy boundaries, including a first recognition accuracy boundary P1, a second recognition accuracy boundary P2, ..., the nth recognition accuracy boundary P... n Where, P1 < P2 < ... < P n The feature selection module compares the average recognition accuracy P of a single feature at the first accuracy level with the recognition accuracy boundary, where...

[0120] If the result is the first accuracy comparison, the feature filtering module classifies the feature into the first feature package and determines the priority of the first feature package as the first priority.

[0121] If the result is the second accuracy comparison, the feature filtering module classifies the feature into the second feature package and determines the priority of the second feature package as the second priority.

[0122] If the accuracy comparison result is n, the feature filtering module classifies the feature into the nth feature package and determines the priority of the nth feature package as the nth priority.

[0123] If it is the (n+1)th accuracy comparison result, the feature filtering module classifies the feature into the (n+1)th feature package and determines the priority of the (n+1)th feature package as the (n+1)th priority.

[0124] Wherein, the first accuracy comparison result satisfies P≥P n The second accuracy comparison result satisfies P. n-1 ≤P<P n The accuracy comparison result of the nth accuracy satisfies P1≤P<P2, and the accuracy comparison result of the (n+1)th accuracy satisfies P0≤P<P1. The priority level is first priority > second priority > nth priority > n+1th priority; different priorities correspond to different recognition accuracy ranges.

[0125] Specifically, the method for determining the accuracy boundary is as follows:

[0126] The feature selection module includes an accuracy interval r0 and a maximum accuracy P. max The feature selection module uses a preset recognition accuracy standard P0 as a baseline, and sets a recognition accuracy boundary every time a recognition accuracy interval r0 is passed. Therefore,

[0127] The first recognition accuracy boundary is P1 = P0 + r0;

[0128] The second recognition accuracy boundary is P2 = P0 + 2r0;

[0129] The nth recognition accuracy boundary P n =P0+nr0;

[0130] Among them, P n <P max .

[0131] In this embodiment, the recognition accuracy interval r0 is set to 2% < r0 < 7%, and preferably r0 = 4%. The maximum recognition accuracy P max Set as the maximum recognition accuracy for any feature.

[0132] This invention sorts features from highest to lowest based on their average recognition accuracy, then groups features with similar average recognition accuracy into one feature bag. Features with significantly different average recognition accuracy are placed in different feature bags. This invention uses an accuracy interval between adjacent feature bags to divide them. It is worth noting that although different feature bags may correspond to different recognition accuracies, they have a certain degree of information complementarity for hot behavior recognition. A feature bag priority structure is established according to recognition accuracy. Based on the feature bag priority, the high classification accuracy advantage of high-priority feature bags can be fully utilized.

[0133] Specifically, in step S4, the central control module compares the number of filtered features M with a preset feature quantity standard M0 to determine whether the number of filtered features meets the standard.

[0134] If the number of filtered features is at the first feature number level, the central control module determines that the number of filtered features meets the standard;

[0135] If the number of filtered features is at the second feature number level, the central control module determines that the number of filtered features does not meet the standard and needs to adjust the preset recognition accuracy standard.

[0136] The first feature quantity level satisfies M≥M0, and the second feature quantity level satisfies M<M0.

[0137] In this embodiment, the preset feature quantity standard M0 can be determined based on the total number of feature dimensions of the identified samples. In this embodiment, the preset feature quantity standard M0 is limited to not less than 1 / 2 of the total number of feature dimensions.

[0138] Specifically, the central control module is equipped with a mode for adjusting the preset recognition accuracy standard at a second feature quantity level, wherein...

[0139] The first method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the first preset recognition accuracy standard.

[0140] The second method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the second preset recognition accuracy standard.

[0141] The third method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the third preset recognition accuracy standard.

[0142] The feature filtering module re-filters the features according to the adjusted preset recognition accuracy standard;

[0143] The first preset recognition accuracy standard < the second preset recognition accuracy standard < the third preset recognition accuracy standard < the preset recognition accuracy standard, and the setting should be made according to the actual working conditions.

[0144] This invention provides a preferred embodiment in which a preset recognition accuracy standard is adjusted using a recognition accuracy standard adjustment coefficient. The specific implementation method is as follows:

[0145] The central control module calculates the difference ΔM between the filtered feature quantity M and the preset feature quantity standard M0 at the second feature quantity level, setting ΔM = M0 - M. The central control module includes a first preset feature quantity difference ΔM1 and a second preset feature quantity difference ΔM2, where ΔM1 < ΔM2. The central control module compares the feature quantity difference ΔM with ΔM1 and ΔM2 respectively to determine the adjustment method for the preset recognition accuracy standard.

[0146] The first method for adjusting the recognition accuracy standard is to use a first recognition accuracy standard adjustment coefficient e1 to adjust the preset recognition accuracy standard to the first preset recognition accuracy standard P. a Set P a =P0×e1;

[0147] The second method for adjusting the recognition accuracy standard is to use a second recognition accuracy standard adjustment coefficient e2 to adjust the preset recognition accuracy standard to the second preset recognition accuracy standard P. b Set P b =P0×e2;

[0148] The third recognition accuracy standard adjustment method is to use a third recognition accuracy standard adjustment coefficient e3 to adjust the preset recognition accuracy standard to the third preset recognition accuracy standard P. c Set P c =P0×e3;

[0149] The feature filtering module re-filters the features according to the adjusted preset recognition accuracy standard;

[0150] Wherein, the first recognition accuracy standard adjustment method satisfies ΔM≥ΔM2, the second recognition accuracy standard adjustment method satisfies ΔM1≤ΔM<ΔM2, the third recognition accuracy standard adjustment method satisfies ΔM<ΔM1, 0.8<e1<e2<e3<1, and P0 is the preset recognition accuracy standard.

[0151] In this embodiment, 0.8 < e1 < 0.85 < e2 < 0.9 < e3 < 1 is set. In this embodiment, e1 = 0.83, e2 = 0.88, e3 = 0.95, the first preset feature quantity difference ΔM1 = 0.1 × M0, and the second preset feature quantity difference ΔM2 = 0.2 × M0.

[0152] Appropriate feature dimensions can effectively improve the accuracy of thermal behavior recognition. This invention uses a preset feature quantity standard to judge whether the number of features meets the standard, and lowers the preset recognition accuracy standard when the number of features does not meet the standard to appropriately increase the number of selected features, thereby further improving the accuracy of thermal behavior recognition.

[0153] This invention, when lowering the preset recognition accuracy standard to appropriately increase the number of selected features, calculates the difference between the number of selected features and the preset feature quantity standard. Based on the value of this difference, different adjustment coefficients are used to adjust the preset recognition accuracy standard. When the number of selected features is significantly lower, the preset recognition accuracy standard is lowered more significantly; conversely, when the number of selected features is less lower, the preset recognition accuracy standard is lowered less significantly. This achieves an appropriate increase in the number of selected features. By setting recognition accuracy standard adjustment coefficients to adjust the preset recognition accuracy standard in stages and limiting the value of these coefficients, this invention avoids insufficient selected features or redundant features among the selected features, further improving the recognition accuracy of thermal behavior.

[0154] Specifically, in step S5, the central control module determines the adjustment method for the recognition accuracy interval based on the feature packet quantity level as follows: calculate the generalized preset recognition accuracy standard P. j With maximum recognition accuracy P max The number Q of recognition accuracy intervals r0 contained between them is set.

[0155]

[0156] The central control module calculates the number of feature packets R.

[0157] If Q is a positive integer, then set R = Q;

[0158] If Q is not a positive integer, then set R to be the smallest positive integer greater than Q;

[0159] Among them, at the first feature quantity level, the generalized preset recognition accuracy standard P j j = 0;

[0160] At the second feature quantity level, the generalized preset recognition accuracy standard P j j = a, b, c. The subscripts a, b, c represent different preset recognition accuracy standards, with the first preset recognition accuracy standard P. a <Second preset recognition accuracy standard P> b <Third preset recognition accuracy standard P> c <Preset recognition accuracy standard P0.

[0161] Specifically, the central control module has a minimum feature packet number threshold R. min and the maximum feature packet number threshold R max The central control module will compare the number of feature packets R with R... min and R max A comparison is performed to determine whether the number of feature packets meets the standard.

[0162] If the number of feature packets is at the first feature packet level, the central control module determines that the number of feature packets is too large and does not meet the standard, and the recognition accuracy interval needs to be increased.

[0163] If the number of feature packets is at the second level, the central control module determines that the number of feature packets meets the standard;

[0164] If the number of feature packets is at the third feature packet level, the central control module determines that the number of feature packets is too small and does not meet the standard, and the recognition accuracy interval needs to be reduced.

[0165] Wherein, the first feature packet quantity level satisfies R > R max The second feature packet quantity level satisfies R min ≤R≤R max The number level of the third feature packet satisfies R < R min .

[0166] During the identification of specific samples, it was found that when the number of feature packets is greater than 7, the accuracy rate of sample identification can reach over 95%. However, when the number of feature packets is greater than 15, although the accuracy rate can reach over 96%, the identification efficiency will significantly decrease. Considering both accuracy and identification efficiency, this embodiment preferably uses R. min =7,R max =15.

[0167] Specifically, the central control module is configured to increase the recognition accuracy interval at a first feature packet quantity level, wherein...

[0168] The first method to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the first increased recognition accuracy interval;

[0169] The second way to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the second increased recognition accuracy interval;

[0170] The third way to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the third increased recognition accuracy interval;

[0171] Among them, the first increase in recognition accuracy interval > the second increase in recognition accuracy interval > the third increase in recognition accuracy interval > the recognition accuracy interval.

[0172] This invention provides a preferred embodiment in which the recognition accuracy interval is increased by using an adjustment coefficient to increase the recognition accuracy interval. The specific implementation method is as follows:

[0173] The central control module calculates the number of feature packets R and the critical value R between the first feature packet number level and the maximum feature packet number level. max The difference ΔR is too large due to the large number of feature packets. d Set ΔR d =RR max The central control module has a first preset feature packet quantity difference ΔR value. d1 The difference ΔR between the number of the second preset feature packets and the number of the second preset feature packets is too large. d2 ΔR d1 <ΔR d2 The central control module will determine the difference ΔR between the excessive number of feature packets. d respectively with ΔR d1 and ΔR d2 The comparison is performed to determine how to increase the recognition accuracy interval, whereby...

[0174] The first method to increase the recognition accuracy interval is to adjust the recognition accuracy interval to the first increased recognition accuracy interval r by selecting the first recognition accuracy interval increase adjustment coefficient δ1. d1 , set r d1 =r0×δ1;

[0175] The second method to increase the recognition accuracy interval is to adjust the recognition accuracy interval to the second increased recognition accuracy interval r by selecting the second recognition accuracy interval increase adjustment coefficient δ2. d2 , set r d2 =r0×δ2;

[0176] The third method to increase the recognition accuracy interval is to adjust the recognition accuracy interval to the third increased recognition accuracy interval r by selecting the third recognition accuracy interval increase adjustment coefficient δ3. d3 , set rd3 =r0×δ3;

[0177] Wherein, the first method of increasing the recognition accuracy interval satisfies ΔR d ≥ΔR d2 The second method of increasing the recognition accuracy interval satisfies ΔR d1 ≤ΔR d <ΔR d2 The third method of increasing the recognition accuracy interval satisfies ΔR d <ΔR d1 , 1<δ3<δ2<δ1<1.2, r0 is the recognition accuracy interval.

[0178] In this embodiment, we set 1 < δ3 < 1.08 < δ2 < 1.15 < δ1 < 1.2. Preferably, δ3 = 1.05, δ2 = 1.1, δ1 = 1.18, and 1 < ΔR. d1 <4<ΔR d2 <7, In this embodiment, the difference ΔR between the number of the first preset feature packets and the actual number of packets is too large. d1 =3, the difference ΔR is too large due to the excessive number of the second preset feature packets. d2 =6.

[0179] Specifically, the central control module is equipped with a method to reduce the recognition accuracy interval at the third feature packet quantity level, wherein,

[0180] The first way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the first reduced recognition accuracy interval;

[0181] The second way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the second reduced recognition accuracy interval.

[0182] The third way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the third reduced recognition accuracy interval;

[0183] Among them, the first interval for reducing recognition accuracy is less than the second interval for reducing recognition accuracy, which is less than the third interval for reducing recognition accuracy.

[0184] This invention provides a preferred embodiment in which the recognition accuracy interval is reduced by using a coefficient to decrease the recognition accuracy interval. The specific implementation is as follows:

[0185] The central control module calculates the number of feature packets R and the minimum feature packet threshold R at the third feature packet quantity level. min The difference ΔR is too small in the number of feature packets. x Set ΔR x =R min-R, the central control module has a first preset feature packet quantity too small difference ΔR value. x1 The difference ΔR between the number of the second preset feature packets and the number of the second preset feature packets is too small. x2 ΔR x1 <ΔR x2 The central control module will calculate the difference ΔR between the insufficient number of feature packets. x respectively with ΔR x1 and ΔR x2 A comparison is performed to determine how to reduce the recognition accuracy interval, where,

[0186] The first method to reduce the recognition accuracy interval is to adjust the recognition accuracy interval to the first reduced recognition accuracy interval r by selecting the first recognition accuracy interval reduction adjustment coefficient η1. x1 , set r x1 =r0×η1;

[0187] The second method to reduce the recognition accuracy interval is to adjust the recognition accuracy interval to the second reduced recognition accuracy interval r by selecting the second recognition accuracy interval reduction adjustment coefficient η2. x2 , set r x2 =r0×η2;

[0188] The third method to reduce the recognition accuracy interval is to adjust the recognition accuracy interval to the third reduction recognition accuracy interval r by selecting the third recognition accuracy interval reduction adjustment coefficient η3. x3 , set r x3 =r0×η3;

[0189] Wherein, the first method of reducing the recognition accuracy interval satisfies ΔR x ≥ΔR x2 The second method for reducing the recognition accuracy interval satisfies ΔR x1 ≤ΔR x <ΔR x2 The third method for reducing the recognition accuracy interval satisfies ΔR x <ΔR x1 , 0.8 < η1 < η2 < η3 < 1, r0 is the recognition accuracy interval.

[0190] In this embodiment, we set 0.8 < η1 < 0.88 < η2 < 0.95 < η3 < 1. Preferably, η1 = 0.83, η2 = 0.9, η3 = 0.97, and 2 < ΔR. x1 <4<ΔR x2 <6, In this embodiment, the difference ΔR between the number of the first preset feature packets and the actual number of packets is too small. x1 =3, the difference ΔR is too small for the second preset feature packet quantity. x2 =5.

[0191] An appropriate number of feature packets can effectively improve the accuracy of thermal behavior recognition. This invention addresses this by setting a minimum feature packet number threshold R. min and the maximum feature packet number threshold R max To determine whether the number of feature packets meets the standard, and when the number of feature packets does not meet the standard, to determine the adjustment method for the recognition accuracy interval, so as to control the number of feature packets within a reasonable range, thereby improving the recognition accuracy and recognition efficiency of thermal behavior.

[0192] When adjusting the recognition accuracy interval, this invention, based on experience in adjusting the preset recognition accuracy standard, also adjusts the recognition accuracy interval to the corresponding value by setting the recognition accuracy interval adjustment coefficient to increase or decrease the number of feature packets, so as to control the number of feature packets within a reasonable range, thereby improving the recognition accuracy and recognition efficiency of thermal behavior.

[0193] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0194] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A feature packet priority transfer method for cross-system discrimination of reaction thermal behavior, characterized in that, include: Step S1: Obtain several identification samples. The feature extraction module performs multi-dimensional feature extraction of reaction thermal behavior on any of the identification samples and obtains the identification accuracy of any feature. Step S2: The feature filtering module calculates the average recognition accuracy of any of the features, and filters the features based on the comparison results between the average recognition accuracy and the preset recognition accuracy standard. Step S3: The feature filtering module determines the recognition accuracy boundary based on the recognition accuracy interval, classifies the filtered features into feature packages based on the recognition accuracy boundary, and determines the priority of the feature packages. Step S4: The central control module counts the number of filtered features and adjusts the preset recognition accuracy standard at the second feature quantity level. The feature filtering module then re-filters the features based on the adjusted preset recognition accuracy standard. Step S5: The central control module calculates the number of feature packets and determines the feature packet quantity level. If the feature packet quantity does not meet the standard, the adjustment method for the recognition accuracy interval is determined according to the feature packet quantity level. Step S6: For the identification of specific samples, the best feature package combination is selected according to the priority of the feature package. When identifying thermal behavior samples of new reaction systems, the method is transferred. In step S6, for the identification of a specific sample, the optimal feature package combination is selected based on the feature package priority structure. The number of target feature packages is initially set to obtain the characteristics of the sample under the reaction thermal behavior to be identified. The feature package priority structure of the sample under the reaction thermal behavior to be identified is determined according to the feature package priority structure. The feature packages are then calculated sequentially from high to low priority to determine whether they meet the requirements of the current reaction thermal behavior. If the set target number of feature packages is reached during the calculation, the feature packages that meet the requirements at that time are selected for identification, and subsequent feature packages are not evaluated, thus determining the optimal feature package combination. If no feature package meets the requirements of the current reaction thermal behavior after traversing all feature packages, the two feature packages with the highest priority are selected for identification, thus determining the optimal feature package combination.

2. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 1, characterized in that, In step S2, the feature filtering module calculates the average recognition accuracy P of any of the features and sets... Among them, F i Let be the accuracy of any feature in the i-th identified sample, where i ranges from 1 to m, and m is the total number of identified samples; The feature filtering module has a preset recognition accuracy standard P0. The feature filtering module compares the average recognition accuracy P of a single feature with the preset recognition accuracy standard P0 to determine whether the average recognition accuracy of a single feature meets the standard. If the average recognition accuracy of a single feature is at the first accuracy level, the feature filtering module determines that the recognition accuracy of the feature meets the standard. If the average recognition accuracy of a single feature is at the second accuracy level, the feature filtering module determines that the recognition accuracy of the feature does not meet the standard and removes the feature. The first accuracy level satisfies P≥P0, and the second accuracy level satisfies P<P0.

3. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 2, characterized in that, In step S3, the feature filtering module categorizes features at the first accuracy level based on the average recognition accuracy. The feature filtering module has several recognition accuracy boundaries, including a first recognition accuracy boundary P1, a second recognition accuracy boundary P2, ..., the nth recognition accuracy boundary P... n Where, P1 < P2 < ... < P n The feature selection module compares the average recognition accuracy P of a single feature at the first accuracy level with the recognition accuracy boundary, where... If the result is the first accuracy comparison, the feature filtering module classifies the feature into the first feature package and determines the priority of the first feature package as the first priority. If the result is the second accuracy comparison, the feature filtering module classifies the feature into the second feature package and determines the priority of the second feature package as the second priority. If the accuracy comparison result is n, the feature filtering module classifies the feature into the nth feature package and determines the priority of the nth feature package as the nth priority. If it is the (n+1)th accuracy comparison result, the feature filtering module classifies the feature into the (n+1)th feature package and determines the priority of the (n+1)th feature package as the (n+1)th priority. The first accuracy comparison result satisfies P ≥ P n The second accuracy comparison result satisfies P. n-1 ≤P<P n The accuracy comparison result of the nth time satisfies P1≤P<P2, and the accuracy comparison result of the (n+1)th time satisfies P0≤P<P1. The priority level is first priority > second priority > nth priority > n+1th priority.

4. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 3, characterized in that, The method for determining the recognition accuracy boundary is as follows: the feature filtering module includes a recognition accuracy interval r0 and a maximum recognition accuracy P. max The feature filtering module uses a preset recognition accuracy standard P0 as the baseline, and sets a recognition accuracy boundary every time a recognition accuracy interval r0 is passed. The first recognition accuracy boundary is P1 = P0 + r0; The second recognition accuracy boundary is P2 = P0 + 2r0; The nth recognition accuracy boundary P n =P0+nr0; Among them, P n <P max .

5. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 4, characterized in that, In step S4, the central control module compares the number of filtered features M with the preset feature number standard M0 to determine whether the number of filtered features meets the standard. If the number of filtered features is at the first feature number level, the central control module determines that the number of filtered features meets the standard; If the number of filtered features is at the second feature number level, the central control module determines that the number of filtered features does not meet the standard and needs to adjust the preset recognition accuracy standard. The first feature quantity level satisfies M≥M0, and the second feature quantity level satisfies M<M0.

6. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 5, characterized in that, The central control module is equipped with a mode for adjusting the preset recognition accuracy standard under the second feature quantity level; The first method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the first preset recognition accuracy standard. The second method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the second preset recognition accuracy standard. The third method for adjusting the recognition accuracy standard is that the central control module adjusts the preset recognition accuracy standard to the third preset recognition accuracy standard. The feature filtering module re-filters the features according to the adjusted preset recognition accuracy standard; First preset recognition accuracy standard < Second preset recognition accuracy standard < Third preset recognition accuracy standard < Preset recognition accuracy standard.

7. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 6, characterized in that, In step S5, the central control module determines the adjustment method for the recognition accuracy interval based on the feature packet quantity level as follows: calculate the generalized preset recognition accuracy standard P. j With maximum recognition accuracy P max The number Q of recognition accuracy intervals r0 contained between them is set. The central control module calculates the number of feature packets R. If Q is a positive integer, then set R = Q; If Q is not a positive integer, then set R to be the smallest positive integer greater than Q; Among them, at the first feature quantity level, the generalized preset recognition accuracy standard P j j = 0; At the second feature quantity level, the generalized preset recognition accuracy standard P j j = a, b, c, where subscripts a, b, c represent different preset recognition accuracy standards, and the first preset recognition accuracy standard P a <Second preset recognition accuracy standard P> b <Third preset recognition accuracy standard P> c <Preset recognition accuracy standard P0.

8. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 7, characterized in that, The central control module has a minimum feature packet number threshold R. min and the maximum feature packet number threshold R max The central control module will compare the number of feature packets R with R... min and R max A comparison is performed to determine whether the number of feature packets meets the standard; If the number of feature packets is at the first feature packet level, the central control module determines that the number of feature packets is too large and does not meet the standard, and the recognition accuracy interval needs to be increased. If the number of feature packets is at the second level, the central control module determines that the number of feature packets meets the standard; If the number of feature packets is at the third feature packet level, the central control module determines that the number of feature packets is too small and does not meet the standard, and the recognition accuracy interval needs to be reduced. Wherein, the first feature packet quantity level satisfies R > R max The second feature packet quantity level satisfies R min ≤R≤R max The number level of the third feature packet satisfies R < R min ; The central control module is equipped with a method to increase the recognition accuracy interval under the first feature packet quantity level; The first method to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the first increased recognition accuracy interval; The second way to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the second increased recognition accuracy interval; The third way to increase the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the third increased recognition accuracy interval; Among them, the first increase in recognition accuracy interval > the second increase in recognition accuracy interval > the third increase in recognition accuracy interval > the recognition accuracy interval.

9. The feature packet priority transfer method for cross-system discrimination of reaction thermal behavior according to claim 8, characterized in that, The central control module is equipped with a method to reduce the recognition accuracy interval at the third feature packet quantity level; The first way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the first reduced recognition accuracy interval; The second way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the second reduced recognition accuracy interval. The third way to reduce the recognition accuracy interval is that the central control module adjusts the recognition accuracy interval to the third reduced recognition accuracy interval; Among them, the first interval for reducing recognition accuracy is less than the second interval for reducing recognition accuracy, which is less than the third interval for reducing recognition accuracy.

10. A feature packet priority transfer method for cross-system discrimination of reaction thermal behavior, characterized in that, The steps of the migration method are as follows: Obtain the curves of state variables changing with the reaction under different thermal behavior conditions, and extract multi-dimensional features based on the curves of state variable changes; The average recognition accuracy of each feature in the multi-dimensional features is calculated based on the total number of recognized samples. Set a preset recognition accuracy standard P0 and a recognition accuracy interval r0, and calculate the maximum recognition accuracy P for each feature. max Calculate the preset recognition accuracy standard P0 and the maximum recognition accuracy P max The number Q of recognition accuracy intervals r0 between them, and the preset recognition accuracy standard P0 and the maximum recognition accuracy P based on the recognition accuracy intervals r0. max The values ​​within the range are divided into the first recognition accuracy boundary P1, the second recognition accuracy boundary P2, ..., the nth recognition accuracy boundary P n Where, P1 < P2 < ... < P n ; The average recognition accuracy P of a single feature that is greater than the preset recognition accuracy standard P0 is compared with each recognition accuracy boundary. The features are sorted from high to low according to the average recognition accuracy of the single feature. Features with average recognition accuracy within the same recognition accuracy range are grouped into a feature package. A recognition accuracy interval is set between two adjacent feature packages. A feature package priority structure is established according to the recognition accuracy, with feature packages with higher recognition accuracy having higher priority. Multi-dimensional feature extraction is performed on samples of the thermal behavior of the new reaction system. Feature packages with different priorities are obtained according to the feature package priority structure, and the optimal feature package combination is selected for identification. Specifically, the number of target feature packages is set in the early stage, the features of the samples under the reaction thermal behavior to be identified are obtained, and the feature package priority structure of the features of the samples under the reaction thermal behavior to be identified is determined according to the feature package priority structure. The feature packages are calculated in descending order of priority to see if they meet the requirements of the current reaction thermal behavior. If the set target number of feature packages is reached during the calculation, the feature packages that meet the requirements at this time are selected for identification, and subsequent feature packages are not evaluated. The optimal feature package combination is then determined. If no feature package meets the requirements of the current reaction thermal behavior after traversing all feature packages, the two feature packages with the highest priority are selected for identification to determine the optimal feature package combination.

Images