Emergency rescue strategy generation method and system based on large language model

Abstract

The present invention discloses a method and system for generating an emergency rescue disposal strategy based on a large language model, and relates to the technical field of data verification. The present invention periodically collects periodic training data from an authorized and authenticated data source through a pre-processing unit, and a docking protection unit generates a corresponding data verification sequence based on the periodic training data. An emergency rescue management and control platform generates a data verification sequence for verification based on the transmitted periodic training data using the same generation method. The verified periodic training data is received by comparing the data verification sequence and input into an emergency rescue processing model for model training. In this way, only the data that has passed the verification is guaranteed to participate in the training of the final emergency rescue processing model, thereby providing reliable data guarantee for model training.

Description

Technical Field

[0001] The present invention relates to the technical field of data verification, and in particular to a method and system for generating an emergency rescue disposal strategy based on a large language model. Background Art

[0002] In today's society, various emergencies such as fires, earthquakes, floods, traffic accidents, etc. occur frequently. These emergencies often pose a huge threat to people's lives and property. Therefore, efficient and accurate emergency rescue and disposal strategies are crucial. Rapidly generating scientific and reasonable emergency rescue and disposal strategies can greatly improve rescue efficiency and reduce disaster losses.

[0003] At present, the generation of emergency rescue strategies relies heavily on various models. However, the training of these models requires a huge amount of training data. The existing technology mainly collects data from trusted data sources for training, but there are many problems in the data transmission process:

[0004] First, there is a risk that data may be intercepted during transmission. Once the training data is intercepted, it may be maliciously tampered with or used for improper purposes, which will seriously affect the training accuracy of the model. For example, in emergency rescue scenarios, if key data about a specific disaster type is intercepted and modified, it may cause deviations in the generated emergency rescue disposal strategy and fail to accurately respond to the actual situation.

[0005] Secondly, the data verification methods such as MD5 values ​​used in the existing technology are obviously not accurate enough. Although MD5 values ​​can be used to verify data integrity to a certain extent, they may be cracked and forged. In the face of complex network environments and potential malicious attacks, MD5 values ​​are difficult to ensure the authenticity and integrity of data, and they are relatively conventional, so they cannot provide reliable data protection for model training.

[0006] In order to solve the above problems, the present invention proposes a solution. Summary of the invention

[0007] The purpose of the present invention is to provide a method and system for generating emergency rescue strategies based on a large language model, in order to solve the problems raised in the above background technology;

[0008] The purpose of the present invention can be achieved through the following technical solutions:

[0009] The method for generating emergency rescue and disposal strategies based on a large language model includes the following steps:

[0010] Step 1: The pre-processing unit collects the periodic training data of the current training period and transmits it to the docking protection unit;

[0011] Step 2: The docking protection unit receives the periodic training data and performs binary conversion on it to obtain pre-processed binary data of the current training cycle;

[0012] Step 3: Generate a data verification sequence of the current training cycle according to a preset generation rule for the preprocessed binary data, and transmit the periodic training data of the current training cycle and its data verification sequence to a receiving verification unit;

[0013] Step 4: After receiving the periodic training data of the current training cycle and its data verification sequence, the receiving and verification unit generates a data verification sequence for verification of the current training cycle according to the same generation rules as step 3, and compares it with the received data verification sequence for consistency. If the comparison passes, the periodic training data of the current training cycle is transmitted to the strategy generation unit.

[0014] Furthermore, the generation rule of the data verification sequence for the current training cycle in step 3 is as follows:

[0015] S11: from left to right, taking every 4 characters in the preprocessed binary data as a group of target strings to obtain a plurality of groups of target strings, and marking the target strings as A1, A2, ..., Aa, from left to right, according to the position of each group of target strings in the preprocessed binary data, where a≥1;

[0016] S12: Referring to the four-digit binary numbers of numbers 0 to 15, the four-digit binary numbers of numbers 0, 1, ..., 15 are marked as B0, B1, ..., B15 in sequence;

[0017] S13: firstly, the target string A1 is self-determined. If the target string is not 1111, the target string A1 is first selected as the identification object, and a number of target strings are extracted from the target strings A2, A3, ..., Aa according to the preset selection conditions as ascending strings of the target string A1;

[0018] If the target string A1 is 1111, then the target strings A2, A3, ..., Aa are self-determined in the order of the target strings A2, A3, ..., Aa, until one of the target strings is not 1111, the target string is selected as the identification object, and several ascending strings of the target string are obtained;

[0019] S14: after a target string is selected as the identification object according to step S13 and a plurality of ascending strings of the identification object are obtained, the identification object and the plurality of ascending strings selected in S13 are removed from the target strings A1, A2, ..., Aa;

[0020] S15: selecting identification objects for all target strings remaining after being eliminated in S14 again according to steps S13 to S14, obtaining a number of ascending strings for the selected identification objects, and eliminating each identification object from all remaining target strings after a number of ascending strings are obtained after an identification object is selected;

[0021] When one of the remaining target strings is selected as the identification object and the number of ascending strings obtained is 0, the identification object selection for all the remaining target strings is stopped;

[0022] S16: re-mark all target strings selected as identification objects as identification strings, and mark all identification strings as G1, G2, ..., Gg in the order of selection, 1≤g <a;

[0023] S17: Calculate and obtain the verification feature value of the recognition string G1 according to the preset complex calculation rule;

[0024] S18: Calculate and obtain verification feature values ​​of identification strings G2, G3, ..., Gg in sequence;

[0025] S19: In the order of target strings A1, A2, ..., Aa, all remaining target strings that are not marked as recognition strings or selected as ascending strings are re-marked as segment strings and re-labeled as R1, R2, ..., Rr, 1≤r <a;

[0026] S110: Take the segment string R1 as an identification string, take all other segment strings as all ascending strings of the identification string, and calculate and obtain the verification feature value of the segment string R1 according to step S17;

[0027] S111: According to the sequence of the recognition strings G1, G2, ..., Gg, and the fragment string R1, the corresponding verification feature values ​​are concatenated to obtain a data verification sequence of the current training cycle.

[0028] The emergency rescue strategy generation system based on the large language model includes:

[0029] A pre-processing unit, which periodically collects periodic training data from an authorized data source and transmits the periodic training data to a docking protection unit;

[0030] The docking protection unit receives the periodic training data of a training cycle each time, performs binary conversion on the periodic training data to obtain pre-processed base data of the corresponding training cycle, and then generates a data verification sequence of the corresponding training cycle based on the pre-processed base data according to a generation rule;

[0031] The docking protection unit generates a data verification sequence for each training cycle and transmits it together with the periodic training data of the corresponding training cycle to the emergency rescue management platform;

[0032] The emergency rescue management platform will perform consistency check on each periodic training data received during a training cycle. If the consistency check passes, the corresponding periodic training data will be input into the emergency rescue processing model for model training.

[0033] Beneficial effects of the present invention:

[0034] (1) The present invention periodically collects periodic training data from an authorized and authenticated data source through a pre-processing unit, and a corresponding data verification sequence is generated by a docking protection unit based on the periodic training data. The emergency rescue management and control platform uses the same generation method to generate a data verification sequence for verification based on the transmitted periodic training data. The verified periodic training data is received by comparing the data verification sequence and input into the emergency rescue processing model for model training. In this way, it is ensured that only the data that has passed the verification participates in the training of the final emergency rescue processing model, providing reliable data guarantee for model training;

[0035] (2) In the process of generating the data verification sequence, the present invention is based on the characteristics of binary. By comparing the ascending features of binary numbers in the periodic training data, several identification strings and their ascending strings are extracted therefrom. For each identification string, a corresponding ascending matrix is ​​constructed in combination with its corresponding ascending strings, and the verification eigenvalues ​​of the identification string and its ascending string are obtained based on the ascending matrix. The data features extracted in this way are more concealed and secure. At the same time, the ascending string rules used to calculate the verification eigenvalues ​​have high stability and reliability. This unique data verification method can effectively ensure the integrity and accuracy of the training data during the generation of emergency rescue disposal strategies. Even in a complex network environment, the risk of data being intercepted or tampered with can be greatly reduced. By continuously optimizing the construction of the ascending matrix and the calculation method of the verification eigenvalue, the efficiency and accuracy of data verification can be further improved. The accuracy of the data ensures the accuracy of the model training, and the accuracy of the model further ensures that the generated strategy is more scientific, reasonable, and effective, and can better deal with various emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention will be further described below in conjunction with the accompanying drawings.

[0037] Figure 1 is a flow chart of the method of the present invention;

[0038] Figure 2 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] like Figure 1 , Figure 2 As shown, a method and system for generating emergency rescue disposal strategies based on a large language model include a data collection terminal and an emergency rescue management and control platform;

[0041] The data collection end is used to periodically provide training data to the emergency rescue management and control platform through an authorized and authenticated data source, wherein the data source includes but is not limited to the official website of the emergency management department, local news websites, etc., and the authorization and authentication refers to the identity and qualification authentication through the emergency rescue management and control platform. The data collection end includes a pre-processing unit and a docking protection unit;

[0042] At each training cycle, the pre-processing unit collects all the training data stored in the training cycle to obtain the periodic training data of the corresponding training cycle, and transmits it to the docking protection unit. The training data refers to the detailed text description of each emergency event. These descriptions should cover all aspects of the event, including the event type (such as fire, earthquake, flood, etc.), the location and time of occurrence, the personnel and property involved, the development trend of the event, etc.

[0043] The docking protection unit performs binary conversion on the periodic training data received for each training cycle to obtain pre-processed binary data of the corresponding training cycle, and then generates a data verification sequence of the corresponding training cycle according to a preset generation rule;

[0044] After receiving the periodic training data of the current training cycle, the docking protection unit performs binary conversion to obtain the preprocessed binary data of the current training cycle, and then generates the data verification sequence of the current training cycle according to the preset generation rule. The docking protection rule is as follows:

[0045] S11: from left to right, taking every 4 characters in the preprocessed binary data as a group of target strings to obtain a plurality of groups of target strings, and marking the target strings as A1, A2, ..., Aa, from left to right, according to the position of each group of target strings in the preprocessed binary data, where a≥1;

[0046] S12: Referring to the four-digit binary numbers of numbers 0 to 15, the four-digit binary numbers of numbers 0, 1, ..., 15 are marked as B0, B1, ..., B15 in sequence;

[0047] S13: First, perform self-determination on the target string A1:

[0048] If the target string is not 1111, first select the target string A1 as the identification object, and extract several target strings from the target strings A2, A3,..., Aa as the ascending strings of the target string A1 according to the preset selection conditions. The selection conditions are as follows:

[0049] SS11: Obtain the subscript marked by the four-bit binary number consistent with the identification object, and use the subscript as the screening index C1 of the identification object;

[0050] SS12: Select the target string A2 based on the identification object: Obtain the subscript D1 marked by the four-bit binary number consistent with the target string A2, compare the sizes of D1 and C1. If C1 < D1, then temporarily select the target string A2 as the candidate string of the identification object; otherwise, do nothing;

[0051] SS13: Select the target strings A3, A4,..., Aa based on the identification object in turn according to SS12;

[0052] SS14: Mark all the candidate strings temporarily selected as the identification object as E1, E2,..., Ee in the order of being temporarily selected as the candidate string of the identification object, where e ≥ 1;

[0053] SS15: Select the first f candidate strings from the candidate strings E1, E2,..., Ee as the ascending strings of the identification object in the order from left to right, and mark the f ascending strings of the identification object as F1, F2,..., Ff in the order of the candidate strings E1, E2,..., Ee;

[0054] At this time, the ascending strings F1, F2,..., Ff of the identification object satisfy the following conditions: the subscripts of the ascending strings F1, F2,..., Ff increase in turn, and the subscript of the ascending string Ff is greater than or equal to the subscript of the ascending string Ff+1. Here, the subscript of the ascending string F1 refers to the subscript marked by the four-bit binary number consistent with the ascending string F1, and the subscripts of the remaining strings are类推;

[0055] If the target string A1 is 1111, perform self-determination on the target strings A2, A3,..., Aa in turn according to the order of the target strings A2, A3,..., Aa until a certain target string is not 1111, select the target string as the identification object, and obtain several ascending strings of the target string according to the steps from SS1 to SS5;

[0056] S14: after a target string is selected as the identification object according to step S13 and a plurality of ascending strings of the identification object are obtained, the identification object and the plurality of ascending strings selected in S13 are removed from the target strings A1, A2, ..., Aa;

[0057] S15: selecting identification objects for all target strings remaining after being eliminated in S14 again according to steps S13 to S14, obtaining a number of ascending strings for the selected identification objects, and eliminating each identification object from all remaining target strings after a number of ascending strings are obtained after an identification object is selected;

[0058] When one of the remaining target strings is selected as the identification object and the number of ascending strings obtained is 0, the identification object selection for all the remaining target strings is stopped;

[0059] S16: re-mark all target strings selected as identification objects as identification strings, and mark all identification strings as G1, G2, ..., Gg in the order of selection, 1≤g <a;

[0060] S17: Calculate and obtain the verification feature value of the recognition string G1 according to the preset recalculation rule. The recalculation rule is as follows:

[0061] S171: Obtain all ascending strings of the identification string G1, and mark all the ascending strings as H1, H2, ..., Hh in sequence according to the order of the selected ascending strings, 1≤h <a;

[0062] S172: Root h. If the positive square root of h is not a positive integer, construct an ascending matrix M1 of the identification string G1 according to a preset filling construction rule. The filling construction rule is as follows:

[0063] SS21: Select a number from numbers 1 to h, the square of which is closest to h. For example, there is a number n among numbers 0 to h, the square of which is greater than h, and the square of n-1 is less than h. The selected number is marked as the matrix index J1 of the recognition string G1;

[0064] SS22: Utilizing Formulas Calculate and obtain the fill index K1 of the identification string G1;

[0065] SS23: Remove duplicates from the ascending strings H1, H2, ..., Hh, and re-mark the remaining ascending strings as I1, I2, ..., Ii, 1≤i≤h;

[0066] SS24: Obtain the total number of ascending strings that are consistent with the ascending string I1 from the ascending strings H1, H2, ..., Hh, and mark the total number as the frequency of the ascending string I1;

[0067] SS25: Calculate and obtain the frequencies of the ascending strings I2, I3, ..., Ii in sequence according to SS24 and select the frequency with the smallest value as the supplementary string of the identification string G1. If there are multiple ascending strings with the smallest frequency, select the first ascending string with the smallest frequency in the order of ascending strings H1, H2, ..., Hh as the supplementary string L1 of the identification string G1.

[0068] SS26: Construct an ascending matrix of the identification string G1 in the order of ascending strings H1, H2, ..., Hh, combined with the filling index of the identification string G1 and the filling string , where the ascending matrix M1 is a matrix with J1 rows and J1 columns, and K1 L1 are filled after Hh in the last row of the matrix;

[0069] If the positive square root of h is a positive integer, the positive square root of h is marked as the matrix index J1 of the identification string G1, and the ascending matrix of the identification string G1 is constructed in the order of the ascending strings H1, H2, ..., Hh combined with the matrix index J1 of the identification string G1 , where the ascending matrix M1 is a matrix with J1 rows and J1 columns;

[0070] S173: Calculate and obtain the eigenvalue N1 of the ascending matrix M1 using the equation |M1-N1×Q1|=0, where N1 is the identity matrix of the ascending matrix M1, the number of rows and columns in the identity matrix N1 is the same as the number of rows and columns in the ascending matrix M1, and the main diagonal elements in the identity matrix N1 are all 1, and the other elements are all 0;

[0071] S174: Obtain the absolute value of the characteristic value N1, and round up the absolute value to obtain the verification characteristic value of the recognition string G1;

[0072] S18: Calculate and obtain verification feature values ​​of identification strings G2, G3, ..., Gg in sequence;

[0073] S19: In the order of target strings A1, A2, ..., Aa, all remaining target strings that are not marked as recognition strings or selected as ascending strings are re-marked as segment strings and re-labeled as R1, R2, ..., Rr, 1≤r <a;

[0074] S110: Take the segment string R1 as an identification string, take all other segment strings as all ascending strings of the identification string, and calculate and obtain the verification feature value of the segment string R1 according to step S17;

[0075] S111: according to the sequence of the recognition strings G1, G2, ..., Gg, and the fragment string R1, the corresponding verification feature values ​​are concatenated to obtain a data verification sequence of the current training cycle;

[0076] The docking protection unit transmits the periodic training data of the current training period and its data verification sequence to the emergency rescue management and control platform;

[0077] An emergency rescue management and control platform, used for emergency response, the emergency rescue management and control platform includes a receiving and verification unit and a strategy generation unit;

[0078] After receiving the transmitted periodic training data of the current training cycle and its data verification sequence, the emergency rescue control platform first performs binary conversion on the periodic training data of the current training cycle to obtain pre-processed binary data of the current training cycle, and then generates a data verification sequence for verification of the current training cycle based on the pre-processed binary data according to steps S11 to S111;

[0079] Perform consistency comparison between the generated data verification sequence and the received data verification sequence, and if the comparison passes, transmit the periodic training data of the current training cycle to the strategy generation unit;

[0080] The strategy generation unit pre-stores the emergency rescue processing model, the initial emergency rescue processing model

[0081] The model parameters are selected and set by the model trainer, and after the strategy generation unit receives the transmitted periodic training data of the current training period, it inputs it into the emergency rescue processing model to train the emergency rescue processing model;

[0082] In the description of the specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0083] The above contents are merely examples and explanations of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the invention or exceed the scope defined by the claims, they shall all fall within the protection scope of the present invention.

[0084] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for generating emergency rescue strategies based on a large language model, characterized in that: The following steps are involved: Step 1: The pre-processing unit collects the periodic training data of the current training period and transmits it to the docking protection unit; Step 2: The docking protection unit receives the periodic training data and performs binary conversion on it to obtain pre-processed binary data of the current training cycle; Step 3: Generate a data verification sequence of the current training cycle according to a preset generation rule for the preprocessed binary data, and transmit the periodic training data of the current training cycle and its data verification sequence to a receiving verification unit; Step 4: After receiving the periodic training data of the current training cycle and its data verification sequence, the receiving verification unit generates a data verification sequence for verification of the current training cycle according to the same generation rule as step 3, and compares the data verification sequence with the received data verification sequence for consistency. If the comparison is passed, the periodic training data of the current training cycle is transmitted to the strategy generation unit, and the strategy generation unit pre-stores the emergency rescue processing model. After receiving the transmitted periodic training data of the current training cycle, the strategy generation unit inputs the data into the emergency rescue processing model to train the emergency rescue processing model; The generation rules for the data verification sequence of the current training cycle in step 3 are as follows: S11: from left to right, taking every 4 characters in the preprocessed binary data as a group of target strings to obtain a plurality of groups of target strings, and marking the target strings as A1, A2, ..., Aa from left to right according to the position of each group of target strings in the preprocessed binary data, where a≥2; S12: Referring to the four-digit binary numbers of numbers 0 to 15, the four-digit binary numbers of numbers 0, 1, ..., 15 are marked as B0, B1, ..., B15 in sequence; S13: firstly, the target string A1 is self-determined. If the target string is not 1111, the target string A1 is first selected as the identification object, and a number of target strings are extracted from the target strings A2, A3, ..., Aa according to the preset selection conditions as ascending strings of the target string A1; If the target string A1 is 1111, then the target strings A2, A3, ..., Aa are self-determined in the order of the target strings A2, A3, ..., Aa, until one of the target strings is not 1111, the target string is selected as the identification object, and several ascending strings of the target string are obtained; S14: after a target string is selected as the identification object according to step S13 and a plurality of ascending strings of the identification object are obtained, the identification object and the plurality of ascending strings selected in S13 are removed from the target strings A1, A2, ..., Aa; S15: selecting identification objects for all target strings remaining after being eliminated in S14 again according to steps S13 to S14, obtaining a number of ascending strings for the selected identification objects, and eliminating each identification object from all remaining target strings after a number of ascending strings are obtained after an identification object is selected; When the number of ascending strings obtained after selecting one from all the remaining target strings as the identification object is 0, stop selecting the identification object from all the remaining target strings at this time; S16: Re-label all the target strings selected as the identification object as identification strings, and in the order of selection, label all the identification strings as G1, G2,..., Gg in sequence, where 1 ≤ g < a; S17: Calculate and obtain the verification feature value of the identification string G1 according to the preset re-calculation rule; S18: Calculate and obtain the verification feature values of the identification strings G2, G3,..., Gg in sequence; S19: In the order of the target strings A1, A2,..., Aa, re-label all the remaining target strings that are not labeled as identification strings or not selected as ascending strings as segment strings, and re-label them as R1, R2,..., Rr, where 1 ≤ r < a; S110: Take the segment string R1 as an identification string, and calculate and obtain the verification feature value of the segment string R1 according to the steps of S17 for all the ascending strings of the remaining segment strings as the ascending strings of the identification string; S111: Concatenate the corresponding verification feature values in the order of the identification strings G1, G2,..., Gg, and the segment string R1 to obtain the data verification sequence of the current training cycle.

2. The method for generating emergency rescue strategies based on a large language model according to claim 1, characterized in that: The selection conditions in S13 are as follows: SS11: Obtain the subscript marked by the four-bit binary number consistent with the identification object, and use the subscript as the screening index C1 of the identification object; SS12: Select the target string A2 based on the identification object: Obtain the subscript D1 marked by the four-bit binary number consistent with the target string A2, compare the sizes of D1 and C1, if C1 < D1, then temporarily select the target string A2 as the candidate string of the identification object, otherwise, do nothing; SS13: Select the target strings A3, A4,..., Aa based on the identification object in sequence according to SS12; SS14: In the order of the candidate strings temporarily selected as the identification object, label all the candidate strings temporarily selected as the identification object as E1, E2,..., Ee in sequence, where e ≥ 1; SS15: Select the first f candidate strings from the candidate strings E1, E2,..., Ee as the ascending strings of the identification object in the order from left to right, and label the f ascending strings of the identification object as F1, F2,..., Ff in the order of the candidate strings E1, E2,..., Ee; At this time, the ascending strings F1, F2,..., Ff of the identification object satisfy the following conditions: the subscripts of the ascending strings F1, F2,..., Ff increase in sequence, and the subscript of the ascending string Ff is greater than or equal to the subscript of the ascending string Ff + 1.

3. The method for generating emergency rescue strategies based on a large language model according to claim 1, characterized in that: The re-calculation rule for calculating and obtaining the verification feature value of the identification string G1 is as follows: S171: Obtain all the ascending strings of the identification string G1, and in the order of selection of the ascending strings, label all the ascending strings as H1, H2,..., Hh in sequence, where 1 ≤ h < a; S172: Execute the square root of h. If the positive square root of h is not a positive integer, construct an ascending matrix M1 of the identification string G1 according to a preset filling construction rule; If the positive square root of h is a positive integer, the positive square root of h is marked as the matrix index J1 of the identification string G1, and the ascending matrix of the identification string G1 is constructed in the order of the ascending strings H1, H2, ..., Hh combined with the matrix index J1 of the identification string G1 , where the ascending matrix M1 is a matrix with J1 rows and J1 columns; S173: Calculate and obtain the eigenvalues ​​of the ascending matrix M1 using the equation |M1-N1×Q1|=0, where N1 is the identity matrix of the ascending matrix M1, the number of rows and columns in the identity matrix N1 is the same as the number of rows and columns in the ascending matrix M1, and the main diagonal elements in the identity matrix N1 are all 1, and the other elements are all 0; S174: Obtain the absolute value of the characteristic value, and round up the absolute value to obtain the verification characteristic value of the recognition string G1.

4. The method for generating emergency rescue strategies based on a large language model according to claim 3 is characterized in that: In S172, the filling construction rule of the ascending matrix M1 of the identification string G1 is as follows: SS21: Select a number from 1 to h. If there is a number n among 0 to h, the square of n is greater than h and the square of n-1 is less than h, mark the selected number as the matrix index J1 of the recognition string G1. SS22: Utilizing Formulas Calculate and obtain the fill index K1 of the identification string G1; SS23: Remove duplicates from the ascending strings H1, H2, ..., Hh, and re-mark the remaining ascending strings as I1, I2, ..., Ii, 1≤i≤h; SS24: Obtain the total number of ascending strings that are consistent with the ascending string I1 from the ascending strings H1, H2, ..., Hh, and mark the total number as the frequency of the ascending string I1; SS25: Calculate and obtain the frequencies of the ascending strings I2, I3, ..., Ii in sequence according to SS24 and select the frequency with the smallest value as the supplementary string of the identification string G1. If there are multiple ascending strings with the smallest frequency, select the first ascending string with the smallest frequency in the order of ascending strings H1, H2, ..., Hh as the supplementary string L1 of the identification string G1. SS26: Construct an ascending matrix of the identification string G1 in the order of ascending strings H1, H2, ..., Hh, combined with the filling index of the identification string G1 and the filling string , where the ascending matrix M1 is a matrix with J1 rows and J1 columns, and K1 L1s are filled after Hh in the last row of the matrix.

5. An emergency rescue strategy generation system based on a large language model, the system being used to implement the method of claim 1, characterized in that: include: A pre-processing unit, which periodically collects periodic training data from an authorized data source and transmits the periodic training data to a docking protection unit; The docking protection unit receives the periodic training data of a training cycle each time, performs binary conversion on the periodic training data to obtain pre-processed base data of the corresponding training cycle, and then generates a data verification sequence of the corresponding training cycle based on the pre-processed base data according to a generation rule; The docking protection unit generates a data verification sequence for each training cycle and transmits it together with the periodic training data of the corresponding training cycle to the emergency rescue management platform; The emergency rescue management platform will perform consistency check on each periodic training data received during a training cycle. If the consistency check passes, the corresponding periodic training data will be input into the emergency rescue processing model for model training.

Images