Intelligent management and control method for port business contract rate

Abstract

The application relates to the technical field of port business management, and particularly discloses an intelligent management and control method for port business contract rates, which comprises the following steps: a CRNN model is used to process a paper contract scanning image, and a contract text is recognized; a named entity recognition model combined with Bi-LSTM and CRF is used to extract rate rules from the contract text, and a rate mapping function is used to automatically generate rates in a business billing system; a rate difference calculation formula is used to compare the generated rates with existing rates in the business billing system, and a difference report is generated; a cosine similarity algorithm is used to compare the rate rule differences in new and old contract texts, a difference level is divided, and a difference comparison table is generated. The intelligent management and control method for port business contract rates solves the problems of low efficiency, high error rate and lagging management and control of manual operation, and improves the efficiency and accuracy of contract processing, rate extraction and generation and comparison analysis.

Description

Technical Field

[0001] This invention relates to the field of port business management technology, and in particular to an intelligent management and control method for port business contract rates. Background Technology

[0002] In the port business operation system, contract rate management is a core link that is directly related to port revenue accounting and customer cooperation pricing strategies. Its management quality directly affects the economic benefits of port enterprises and the stability of customer cooperation.

[0003] First, port business contracts are still mainly in paper form. Staff need to manually enter the text information of the paper contracts word by word into the computer system. In this process, a contract with multiple pages of clauses and complex rate rules often requires 1-2 staff members to spend 2-3 hours to complete the entry. This is not only inefficient, but also prone to errors in text information entry due to staff fatigue and negligence. If these errors are not discovered in time, they will directly affect the accuracy of revenue accounting.

[0004] Secondly, in the process of extracting rate rules and configuring them into the business billing system, staff need to filter out rate-related clauses from the entered text, such as the rate standards corresponding to different types of goods, different container types, and different loading and unloading methods, and then manually configure the rate data according to the parameter format of the billing system. Because port contract rate rules often have multiple dimensions, such as differentiating rates by cargo weight range, transportation distance, and loading and unloading efficiency requirements, manual extraction can easily miss some dimensions of rate rules, resulting in incomplete rate data configured into the system.

[0005] Furthermore, when it is necessary to verify the consistency between the newly configured rates and the existing historical rates in the system, manual comparison requires checking the values ​​of each rate dimension one by one. For contracts containing more than 10 rate dimensions, manual comparison takes more than 1 hour and is difficult to fully cover all dimensions, easily overlooking differences. When comparing the rate rules of new and old contracts, staff need to read the rate-related content of both contracts clause by clause in order to locate the differences. For contracts with many clauses, the comparison can take 3-4 hours, which is seriously lagging behind the efficient needs of port business operations. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent management and control method for port business contract rates. Through a three-stage process of text recognition, rate extraction and generation, and two-dimensional comparison, it solves the problems of low efficiency, high error rate and lagging management and control in current manual operations.

[0007] To achieve the above objectives, the present invention provides an intelligent management and control method for port commercial contract rates, comprising the following steps:

[0008] S1. The CRNN model is used to process the scanned image of the paper contract and identify the contract text;

[0009] S2. A named entity recognition model based on the combination of Bi-LSTM and CRF extracts rate rules from contract text and automatically generates rates in the business billing system through a rate mapping function.

[0010] S3. Using the rate difference calculation formula, compare the generated rate with the existing rate in the business billing system to generate a difference report;

[0011] S4. Use the cosine similarity algorithm to compare the differences in rate rules between the old and new contract texts, classify the difference levels, and generate a difference comparison table.

[0012] Preferably, in S1, the CRNN model includes a convolutional neural network (CNN) and a recurrent neural network (RNN). The CNN part uses the ReLU activation function to extract feature information of the text region from the scanned image of the paper contract, and filters out interfering elements in the image, including stamps, paper wrinkles, and blurred handwriting, through multi-layer convolution operations.

[0013] The RNN part uses an LSTM structure, which stores the feature information of preceding characters through memory units and forgets irrelevant information. It performs temporal modeling on the text feature sequence output by the CNN and predicts the character sequence.

[0014] Preferably, in S2, the forward LSTM and backward LSTM in the Bi-LSTM are used to collaboratively capture the semantic information of the text context, and the CRF layer is used to perform logical constraint correction on the output entity label sequence.

[0015] Preferably, in S2, the rate mapping function expression includes the following two forms:

[0016] In a linear adjustment scenario, structured rate data R sys It is calculated using the following formula:

[0017]

[0018] Among them, R ext This represents the original rate rules extracted from the contract text, stored in multidimensional data format, including service type, box type, billing unit, and rate value; α is the dynamic adjustment coefficient for the contract rate; p is the risk compensation coefficient; β is the basic contract rate;

[0019] In nonlinear adjustment scenarios, structured rate data R' sys It is calculated using the following formula:

[0020]

[0021] Where δ represents the initial offset of the external rate data under nonlinear conditions; γ is the scaling factor; and ∈ is the final offset of the nonlinear adjustment.

[0022] Preferably, in S3, the difference D for each dimension γ i Make a separate determination, D i The calculation formula is as follows:

[0023]

[0024] Where i is the i-th dimension of the rate calculation; R sys,i R represents the calculated rate value for the i-th dimension; e,i This represents the existing rate value for the i-th dimension stored in the existing business contract;

[0025] Set the difference threshold for each dimension to 5%. If there exists a difference in any dimension D... i When the percentage is >5%, a difference report is automatically generated; when all D... i When the rate is ≤5%, the rate verification process is completed directly without generating a difference report; the generated difference report includes the difference dimension name, R sys,i Numerical value, R e,i Numerical values ​​and single-dimensional differences

[0026] Preferably, S4 includes the following steps:

[0027] S41. Extract the rate rule text from the old and new contracts. The rate rule text extracted from the new contract will be denoted as Text. new The rate rule text extracted from the old contract is denoted as Text. old The two texts are preprocessed, including removing meaningless characters and extracting feature words;

[0028] S42. The TF-IDF algorithm is used to calculate the weight of each feature word in the corresponding text. Based on the feature word weights, the cosine similarity algorithm is used to calculate the similarity Sim between the two texts. The calculation formula is as follows:

[0029]

[0030] Where m represents the number of feature words extracted from the text of the new and old contract rate rules, and the feature words include the name of the goods, the container type, the rate value, and the billing unit; w new,j Let w be the term frequency-inverse document frequency (TF-IDF) weight of the j-th feature word in the new contract. old,j The term frequency-inverse document frequency (TF-IDF) weights of the j-th feature word in the old contract.

[0031] Preferably, in S4, three levels of difference are determined based on the cosine similarity Sim value:

[0032] If Sim≥0.8 is a low difference level, the core content of the new and old contract rate rules is the same, and the differences only exist in non-critical information;

[0033] If 0.5≤Sim<0.8 is considered a medium difference level, it indicates that there are some core differences in the fee rate rules between the old and new contracts;

[0034] If Sim < 0.5, it is considered a high difference level, indicating a significant difference in the core content of the fee rate rules between the old and new contracts. A difference comparison table is generated for each difference level, and the old contract terms, new contract terms, and difference types are marked.

[0035] Therefore, the present invention employs the above-mentioned intelligent management method for port business contract rates, and the beneficial effects are as follows:

[0036] (1) This invention uses a CRNN model to automatically recognize paper contract texts, replacing manual input, so that the work that previously required a lot of time can be completed quickly, while avoiding manual input errors and ensuring the accuracy of subsequent revenue accounting.

[0037] (2) This invention uses a model combining Bi-LSTM and CRF to efficiently extract rate rules from contract texts and then automatically generate the rates required by the business billing system using a rate mapping function. This solves the problem of dimension omissions that are easy to occur when extracting data manually, making the rate data more complete.

[0038] (3) This invention uses the rate difference calculation formula and cosine similarity algorithm to quickly compare the newly generated rate with the existing rate and the rate rules of the new and old contracts. Compared with manual verification, it can greatly reduce the time consumption and fully cover all dimensions, and generate difference reports and comparison tables in a timely manner to meet the high-efficiency needs of port business operations.

[0039] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0040] Figure 1 This is an overall flowchart of an embodiment of an intelligent management and control method for port business contract rates according to the present invention;

[0041] Figure 2 This is a detailed data interaction flowchart of an embodiment of an intelligent management and control method for port business contract rates according to the present invention. Detailed Implementation

[0042] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0043] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0044] like Figure 1 , Figure 2 As shown, an intelligent management method for port business contract rates is characterized by the following steps:

[0045] S1. When digitizing paper contracts, it is necessary to extract text from the scanned image first. This invention uses a CRNN model to process the scanned image of the paper contract in order to identify the contract text.

[0046] The CRNN model consists of two parts: a convolutional neural network (CNN) and a recurrent neural network (RNN). The CNN part uses the ReLU activation function to extract feature information of the text region from the two-dimensional scanned image of the paper contract. It also filters out interfering elements in the image, such as stamps, paper wrinkles, and blurred handwriting, through multi-layer convolution operations.

[0047] Specifically, the CNN model's structure consists of multiple layers of alternating convolutional and pooling layers. First, a preprocessed scanned image of the contract is input into the network. Preprocessing operations include converting the image to grayscale and resizing it appropriately. Then, the first convolutional layer uses a 5×5 kernel, which slides across the image to capture basic local features, including text edges and simple textures. After each convolutional operation, the ReLU activation function is applied, preserving the positive values ​​of the features while suppressing negative values ​​to zero. This not only enhances the model's ability to represent complex features but also speeds up training and avoids the vanishing gradient problem that can occur in deep networks.

[0048] Pooling layers are connected after convolutional layers. Max pooling is usually used to compress the feature maps obtained from convolution. This reduces the size and amount of data of the feature maps while retaining the most critical feature information. This helps to reduce the complexity of subsequent calculations and can also prevent the model from overfitting the training data to some extent.

[0049] As the network deepens, subsequent convolutional layers capture more complex and abstract features on top of the basic features extracted earlier, such as the overall outline and structure of text characters. Through these multi-layered convolution and pooling operations, CNNs can gradually filter out interfering elements in the image. For stamps, because their color and texture features differ significantly from text, they are gradually weakened in multi-layer processing; irregular textures caused by paper wrinkles are smoothed during pooling; and for characters with blurred handwriting, their key stroke structure features are enhanced in deep convolutions, thereby reducing the impact of blurring and ultimately outputting feature information that accurately represents the text region.

[0050] The RNN part adopts an LSTM structure to process sequence data, capture and utilize the temporal relationship between characters. Preceding character feature information is stored in memory units, and irrelevant information is forgotten, so as to perform temporal modeling on the text feature sequence output by the CNN and predict the character sequence.

[0051] The LSTM structure is the core of the RNN part. It consists of a plurality of serially connected memory units, and each memory unit is connected to each other through information flow to form a chain for processing sequence data. Each memory unit has three important control mechanisms, which are respectively responsible for the input, retention and output of information. When the text feature sequence extracted by the CNN enters the LSTM in order, the first mechanism determines which of the current features are worth retaining. For example, the character features of key clauses in a contract are allowed to enter the unit for storage; the second mechanism actively cleans up unnecessary information in the unit. For example, when processing a new clause, the irrelevant modifier information stored previously is gradually forgotten to avoid occupying space and affecting subsequent judgments; the third mechanism determines which information in the current unit can be transmitted to the next unit, and the filtered effective information is transmitted sequentially along the unit chain, so that the entire sequence processing remains coherent.

[0052] When processing contract text, the feature sequence enters these memory units in reading order. The character information processed by preceding units is selectively retained, and when subsequent units process new characters, relevant information retained previously can be called. For example, after "party a" is identified, this information is appropriately retained, and when "the" is encountered subsequently, the complete expression "party a" can be accurately identified in combination with the preceding memory. Through this structure, LSTM can sort out the sequence relationship between characters, even for long contract sentences, it can still remember the key context associations, and finally predict a complete and coherent contract text based on these associations.

[0053] S2, after extracting the contract text, it is also necessary to identify the rate rules therefrom and convert them into rates available to the commercial billing system. The present invention extracts rate rules from contract texts based on a named entity recognition model combining bidirectional long short-term memory network (Bi-LSTM) and conditional random field (CRF), and automatically generates rates in the commercial billing system through a rate mapping function.

[0054] The Bi-LSTM and CRF combined named entity recognition model first converts words in the text into vector form, with each word represented by a 300-dimensional vector. This transforms textual information into numerical information that the model can process. The model's bidirectional long short-term memory network layer, comprising 128 hidden units, is responsible for in-depth processing of this vector information. In Bi-LSTM, the forward LSTM and backward LSTM are used collaboratively to capture the semantic information of the text's context. The forward LSTM analyzes the text word by word from left to right, capturing the influence of preceding words on the current word; the backward LSTM processes from right to left, focusing on the semantic connections brought by subsequent words. Through this bidirectional processing approach, the model can more comprehensively understand the meaning of each word in its context and accurately identify rate-related entity information in the text.

[0055] After processing by a bidirectional long short-term memory network, a conditional random field (CRF) layer is used to perform logical constraint correction on the output entity label sequence, extracting rate rules more accurately from the contract text. Then, a rate mapping function automatically converts these rules into rates that can be directly used in the business billing system, achieving a smooth conversion from text information to system data.

[0056] The rate mapping function expression of this invention includes the following two forms:

[0057] In a linear adjustment scenario, the system can identify structured rate data R. sys It is calculated using the following formula:

[0058]

[0059] Among them, R ext This represents the original rate rules extracted from the contract text, stored in multidimensional data format, including information such as service type, container type, billing unit, and rate value, i.e., R. ext = [S,C,U,V], where S represents the service type string (e.g., loading and unloading service, warehousing service), C represents the container type, U is the billing unit, and V represents the rate value; where the service type is the specific service name, such as loading and unloading service, warehousing service, etc.; the container type specifies the type of container for loading the goods; the billing unit specifies the unit standard for calculating the fee; and the rate value is the basic fee standard.

[0060] The calculation formula also needs to consider the following adjustment factors:

[0061] The contract rate dynamic adjustment coefficient α is used to adjust the base rate based on market supply and demand fluctuations, seasonal factors, etc.; the risk compensation coefficient p is used to adjust the contract rate with a risk premium, taking into account factors such as port operation risk and exchange rate fluctuation risk; the base contract rate β is the initial contract rate without considering the above adjustment factors. By combining these factors in a linear relationship, the linearly adjusted structured rate data can be obtained.

[0062] In nonlinear adjustment scenarios, structured rate data R' sys It is calculated using the following formula:

[0063]

[0064] Where δ represents the initial offset of external rate data under nonlinear conditions, used to correct rate fluctuations caused by sudden market events or special business needs under nonlinear conditions; γ is the scaling factor, which can be adjusted to change the magnitude of rate adjustment and achieve refined control; ∈ is the final offset of nonlinear adjustment, used to balance the impact of other unforeseen factors on rates; by integrating these parameters into the calculation according to nonlinear relationships, the structured rate data after nonlinear adjustment can be obtained.

[0065] S3. After the rate is generated, the generated rate is compared with the existing rate in the business billing system using the rate difference calculation formula, and finally a corresponding difference report is generated.

[0066] The calculation of the rate variance D involves multiple rate dimensions, including different fee categories such as loading and unloading fees, storage fees, and port charges, each of which is calculated as an independent dimension. That is, the variance D for each dimension i... i Make a separate determination, D i The calculation formula is as follows:

[0067]

[0068] Where i is the i-th dimension of the rate calculation; R sys,i R represents the calculated rate value for the i-th dimension; e,i This represents the existing rate value for the i-th dimension stored in the existing business contract.

[0069] The system pre-sets a difference threshold of 5% for each dimension. If there is a difference in any dimension D... i When the percentage is >5%, the system automatically generates a difference report; when all D i If the difference is ≤5%, it is considered acceptable, and the rate verification process will be completed directly without generating a difference report; the generated difference report includes the difference dimension name, R sys,i Numerical value, R e,iValues and single-dimensional differences The specific situation of rate differences can be clearly understood.

[0070] S4, in order to compare the differences of rate rules between new and old contract texts, the present invention uses a cosine similarity algorithm to compare the differences of rate rules between new and old contract texts, divides difference levels and generates a difference comparison table, which includes the following steps:

[0071] S41, extracting rate rule texts from new and old contracts, and marking the rate rule text extracted from the new contract as Text new , marking the rate rule text extracted from the old contract as Text old , preprocessing the two texts respectively, and the preprocessing operations include removing meaningless characters and extracting feature words.

[0072] Meaningless characters include punctuation marks in the text, which mainly function for sentence breaking and do not involve specific rate content; there are also redundant spaces and line breaks, which are only typographical separation elements and have no substantial impact on semantics; in addition, common function words such as "de", "le", "zai" in Chinese, and guiding words such as "genju", "anzhao" are also meaningless characters that can be removed in the extraction of core information of rate rules; there may also be special symbols, such as decorative symbols like asterisks and horizontal lines, or garbled codes and incomplete characters generated during text scanning, none of which contain effective rate rule information, and these will be cleaned up in the preprocessing stage to ensure the accuracy of subsequent feature word extraction.

[0073] Extracted feature words include cargo type names such as specific cargo types like coal, steel, and containers; container type specifications include specification information of loading containers such as 20-foot standard containers and 40-foot high containers; rate values are specific fee amounts or calculation proportions; billing units are tons, containers or days.

[0074] S42, adopting the TF-IDF algorithm to calculate the weight of each feature word in the corresponding (respective) text, and based on the weights of these feature words, adopting the cosine similarity algorithm to calculate the similarity Sim of the two texts, and the calculation formula is:

[0075]

[0076] Wherein, m is the number of feature words extracted from the rate rule texts of new and old contracts, and the feature words include core words such as cargo type name, container type specification, rate value, and billing unit; w new,j is the term frequency-inverse document frequency TF-IDF weight of the j-th feature word of the new contract, w old,j is the term frequency-inverse document frequency TF-IDF weight of the j-th feature word of the old contract.

[0077] Next, based on the cosine similarity Sim value, three levels of difference are defined:

[0078] If Sim≥0.8 is a low difference level, it indicates a high degree of similarity, meaning that the core content of the new and old contract rate rules is consistent, with differences only in non-critical information.

[0079] If 0.5≤Sim<0.8 is considered a medium difference level, it indicates that the similarity is at an intermediate level, corresponding to some core dimension differences between the old and new contract rate rules.

[0080] If Sim < 0.5, it indicates a high level of difference, meaning that the similarity is low and the core content of the fee rate rules in the old and new contracts is significantly different. For different levels of difference, the system generates a difference comparison table and marks the old contract terms, the new contract terms, and the specific types of differences.

[0081] Example 1

[0082] Taking a new "Container Loading and Unloading Contract of a Shipping Company" added in May 2024 at a certain coastal hub port (hereinafter referred to as "Port Y") as an example, the method of this invention is applied to complete the intelligent management of rates. The specific process is as follows:

[0083] S1. Using the CRNN model to recognize paper contract text: The input data is a scanned copy of the container loading and unloading contract, which has 12 pages in total, including a blue business stamp, partial paper creases on page 6, and handwritten modification annotations on page 10.

[0084] The model parameters are set as follows: the CNN part has 4 convolutional layers with 5×5 kernels, plus 3 max pooling layers, and the activation function is ReLU; the RNN part has 2 LSTM layers with 256 hidden units.

[0085] The processing results showed that the model achieved an accuracy rate of 97.8% in automatically identifying and filtering business chapters, a 96.3% accuracy rate in repairing text in wrinkled areas, and a 95.1% accuracy rate in recognizing handwritten annotations. The final output was a 18,000-word structured contract text, completed in 6 minutes and 15 seconds. Compared to the traditional method of manual data entry by two staff members taking 2.5 hours, this represents a 24-fold increase in efficiency, with no character recognition errors, thus avoiding the impact of manual data entry errors on subsequent rate processing.

[0086] S2. Rate Rule Extraction and Rate Generation: Rate rule extraction uses a named entity recognition model combining Bi-LSTM (256 hidden units) and CRF. The structured text output by S1 is labeled with entity combinations of "service type-cargo type-container type-billing unit-rate value". The core rate rules extracted include "container handling service-iron ore-40-foot high cube container-yuan / container-850 yuan" and "container handling service-copper ore-20-foot standard container-yuan / container-520 yuan", with no rate dimensions omitted.

[0087] The rate mapping is generated according to the Y-port billing system rules. Because this contract involves a "peak season premium for bulk commodities," it falls under a non-linear adjustment scenario. Therefore, a non-linear rate mapping function is called, and the parameter is assigned the value R. ext (Original rate, iron ore 40-foot high cube container loading and unloading) 850 yuan / container; γ is the scaling factor, in this implementation γ is the reciprocal of the peak season premium factor of 1.1; δ is the initial offset, port equipment loss compensation is 30 yuan; ∈ is the final offset, temporary manual scheduling cost is 20 yuan. R' is calculated according to the following formula. sys :

[0088]

[0089] The calculation result is R' sys = (850+30)×1.1+20=988 yuan / box, generating structured rate data that can be directly accessed by the Y Port billing system.

[0090] S3. Comparison of newly generated rates with existing rates in the system: The comparison is between the newly generated iron ore 40-foot high container loading and unloading rate (988 yuan / container) and the old rate of the shipping company in the Y port system (950 yuan / container).

[0091] The difference is calculated as the total number of dimensions for rate calculation, n=1, with only the iron ore 40-foot high-cube container loading and unloading rate, R. sys,i =988, R e,i =950, then:

[0092]

[0093] The result was determined to be a difference of 4% < threshold 5%, so the rate verification was completed directly without the need to generate a difference report, thus shortening the time of 1 hour spent on manual verification.

[0094] S4. Comparison of Old and New Contract Rate Rules: Text Preprocessing and Feature Extraction for Extracting the Text of the New Contract Rate Rules new In this embodiment, the fee rate text is "RMB 850 per 40-foot high cube container for iron ore loading and unloading, and RMB 520 per 20-foot standard container for copper ore loading and unloading" compared to the old contract fee rate rule text. old In this embodiment, the price is "Iron ore 40-foot high cube container loading and unloading: 820 yuan / container, copper ore 20-foot standard container loading and unloading: 500 yuan / container". After removing meaningless characters such as commas and spaces, the extracted feature words include iron ore, copper ore, 40-foot high cube container, 20-foot standard container, loading and unloading rate, 850 yuan / container, 520 yuan / container, 820 yuan / container, 500 yuan / container, for a total of m = 9 feature words.

[0095] The cosine similarity is calculated by calculating the weight of feature words through the TF-IDF algorithm, wherein core numerical values such as "850 yuan / case" and "820 yuan / case" have the highest weight, and substituting into the formula obtains: Sim=0.85.

[0096] The difference level and comparison table shows that when 0.7 < Sim < 0.9, it belongs to the medium difference level, that is, there are differences in core dimensions, only the rate values are adjusted, while there is no change in service type, cargo type and container type, and the following difference comparison table is generated:

[0097] Table 1 Difference comparison table for intelligent management and control of port business contract rates

[0098] Iron ore 40-foot high container loading and unloading rates 950 yuan / box 988 yuan / box Value increased Copper ore 20-foot container loading and unloading rates 510 yuan / box (old mapping) 592 yuan / box (New Mapping) Value increased

[0099] It can be seen from Table 1 that the terms of the new contract have increased to varying degrees in the container loading and unloading rates of iron ore and copper ore; this adjustment reflects the positive role of the intelligent management and control method for port business contract rates in dynamically adapting to market changes and optimizing cost structure, which helps to improve port operation revenue and market competitiveness, and provides a reference practical example for subsequent rate management.

[0100] Therefore, the present invention adopts the above-mentioned intelligent management and control method for port business contract rates, which replaces traditional manual operations through intelligent means, greatly improves the efficiency and accuracy of port business contract rate management and control, and provides strong support for the digital transformation of port business management.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solution of the present invention can still be modified or equivalently replaced, and these modifications or equivalent replacements cannot make the modified technical solution deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for intelligent management and control of port commercial contract rates, characterized in that, Includes the following steps: S1. The CRNN model is used to process the scanned image of the paper contract and identify the contract text; S2. A named entity recognition model based on the combination of Bi-LSTM and CRF extracts rate rules from contract text and automatically generates rates in the business billing system through a rate mapping function. S3. Using the rate difference calculation formula, compare the generated rate with the existing rate in the business billing system to generate a difference report; Specifically: For each dimension Difference Make a separate judgment. The calculation formula is as follows: in, The first calculation of the rate One dimension; Represents the calculated first Rate values ​​for each dimension; This indicates the first [item] stored in the existing business contract. Existing rate values ​​for each dimension; Set a difference threshold of 5% for each dimension. If any dimension has a difference threshold... Automatically generate a difference report when all At that time, the rate verification process is completed directly without the need to generate a difference report; the generated difference report includes the difference dimension name, numerical value Numerical values ​​and single-dimensional differences ; S4. Use the cosine similarity algorithm to compare the differences in rate rules between the old and new contract texts, classify the difference levels, and generate a difference comparison table. In S1, the CRNN model includes a convolutional neural network (CNN) and a recurrent neural network (RNN). The CNN part uses the ReLU activation function to extract feature information of the text region from the scanned image of the paper contract, and filters out interfering elements in the image, including stamps, paper wrinkles and blurred handwriting, through multi-layer convolution operations. The RNN part uses an LSTM structure, which stores the feature information of previous characters through memory units and forgets irrelevant information. It performs temporal modeling on the text feature sequence output by the CNN and predicts the character sequence. In S2, the forward LSTM and backward LSTM in Bi-LSTM are used to collaboratively capture the semantic information of the text context, and the CRF layer is used to perform logical constraint correction on the output entity label sequence. In S2, the rate mapping function expression includes the following two forms: In linear adjustment scenarios, structured rate data It is calculated using the following formula: ； in, This represents the original rate rules extracted from the contract text, stored in the form of multidimensional data, including service type, box type, billing unit, and rate value; This is the dynamic adjustment coefficient for the contract rate; This is the risk compensation coefficient; Based on the basic contract rate; In non-linear adjustment scenarios, structured rate data It is calculated using the following formula: ； in, This represents the initial offset applied to external rate data under nonlinear conditions; This is the scaling factor; This is the final offset for non-linear adjustment.

2. The intelligent management and control method for port business contract rates according to claim 1, characterized in that, S4 includes the following steps: S41. Extract the rate rule text from the old and new contracts. The rate rule text extracted from the new contract is denoted as... The rate rule text extracted from the old contract is denoted as The two texts are preprocessed, including removing meaningless characters and extracting feature words; S42. Calculate the weight of each feature word in the corresponding text using the TF-IDF algorithm. Based on the feature word weights, calculate the similarity between the two texts using the cosine similarity algorithm. The calculation formula is: ； in, The number of feature words extracted from the text of the rate rules in the old and new contracts. Feature words include cargo name, container type, rate value and billing unit. For the new contract The term frequency-inverse document frequency (TF-IDF) weights of each feature word. For the old contract, The term frequency-inverse document frequency (TF-IDF) weights of each feature word.

3. The intelligent management and control method for port business contract rates according to claim 1, characterized in that, In S4, based on cosine similarity The values ​​are divided into three levels of difference: like The low-difference level corresponds to the core content of the new and old contract rate rules being the same, with differences only in non-critical information. like The difference level corresponds to some core differences between the old and new contract rate rules. like The highest level of difference corresponds to significant differences in the core content of the fee rate rules between the old and new contracts; a difference comparison table is generated for each level of difference, and the old contract terms, the new contract terms, and the type of difference are marked.

Images