Intelligent supervision method, system and device for medicine collection and storage medium

Abstract

The invention discloses an intelligent supervision method, system and device for medicine collection and a storage medium. The method comprises the steps that a multivariable time sequence and target data serve as an input variable and an output variable of a preset prediction model respectively; preprocessing the multivariable time sequence to obtain a plurality of patch sequences; performing feature extraction on all patch sequences by adopting a multi-layer channel Mamba model to obtain target features; inputting the target feature into a linear layer of a preset prediction model to obtain a predicted medicine demand quantity; obtaining a supplier evaluation information set of a plurality of medicine suppliers, and scoring the medicine suppliers to obtain scoring data of each medicine supplier; and obtaining medicine collection suggestions according to all the score data. According to the invention, efficient, intelligent and comprehensive medicine collection and supervision are realized, the data processing efficiency and the real-time performance are high in the whole process, the efficiency of medicine collection and supervision and the scientificity of decision making are improved, and the stability and the safety of medicine supply are ensured.

Description

Technical Field

[0001] The present invention relates to the fields of medical treatment and artificial intelligence, and in particular to a method, system, device and storage medium for intelligent supervision of centralized procurement of medicines. Background Art

[0002] The supervision of centralized procurement of medicines is of great significance in reducing drug prices, improving drug quality, optimizing resource allocation, standardizing market order and improving the management level of medical institutions. Therefore, the research efforts on the supervision of centralized procurement of medicines are gradually increasing at this stage.

[0003] The existing centralized pharmaceutical procurement supervision system mainly relies on manual operations and traditional information technology, including modules such as procurement plan formulation, supplier selection, procurement execution and payment settlement. However, due to slow data processing speed, low accuracy, inability to monitor in real time and lack of intelligent analysis capabilities, these systems are difficult to meet the increasingly complex and huge regulatory needs. Therefore, the existing centralized pharmaceutical procurement supervision technology has the following problems:

[0004] 1. Low data processing efficiency: Manual data entry and processing are slow and error-prone, making it difficult to cope with massive amounts of data;

[0005] 2. Poor real-time performance: It is difficult to achieve real-time monitoring of supplier supply conditions, drug inventory status, etc., resulting in delayed supervision;

[0006] 3. Low level of intelligence: lack of data analysis and forecasting capabilities, unable to predict changes in purchasing demand and market fluctuations in advance, affecting purchasing decisions;

[0007] 4. Many regulatory loopholes: Incomplete supplier qualification review and procurement demand forecasting lead to drug quality problems and improper procurement. For example, a medical institution purchased a batch of substandard drugs due to lax supplier qualification review, which affected the treatment effect of patients; due to inaccurate forecasting, the institution experienced drug shortages many times and was forced to make emergency purchases, which increased costs.

[0008] These problems reflect the shortcomings of existing technologies, and therefore an efficient, intelligent and comprehensive solution is urgently needed. Summary of the invention

[0009] In view of this, the present invention provides a method, system, device and storage medium for intelligent supervision of centralized medicine procurement to solve the problems of low data processing efficiency, poor real-time performance, low intelligence and many supervision loopholes in the existing centralized medicine procurement supervision technology.

[0010] The present invention provides a method for intelligent supervision of centralized procurement of medicines, the method comprising:

[0011] Providing a multivariate time series and target data, using the multivariate time series as an input variable of a preset prediction model, and using the target data as an output variable of the preset prediction model; wherein the multivariate time series is composed of historical pharmaceutical procurement data, and the target data is composed of predicted pharmaceutical demand;

[0012] Preprocessing the multivariate time series by sequentially using channel mixing, instance normalization, and data segmentation to obtain a plurality of patch sequences;

[0013] Using a multi-layer channel Mamba model to extract features from all the patch sequences to obtain target features;

[0014] Inputting the target feature into the linear layer of the preset prediction model to obtain the predicted medicine demand;

[0015] Obtaining supplier evaluation information sets of multiple pharmaceutical suppliers, scoring the corresponding pharmaceutical suppliers according to each of the supplier evaluation information sets, and obtaining scoring data of each pharmaceutical supplier;

[0016] Based on all the scoring data, a pharmaceutical procurement recommendation corresponding to the predicted pharmaceutical demand is obtained.

[0017] Optionally, the method of sequentially adopting channel mixing, instance normalization and data segmentation to preprocess the multivariate time series to obtain multiple patch sequences includes:

[0018] Determine a channel mixing ratio and a channel random arrangement array, and perform channel mixing on the multivariate time series according to the channel mixing ratio and the channel random arrangement data to obtain a mixed sample sequence;

[0019] Performing instance normalization operation on the mixed sample sequence to obtain a normalized time series;

[0020] The normalized time series is patch segmented according to preset patch parameters to obtain a plurality of patch sequences.

[0021] Optionally, the expression of the mixed sample sequence is specifically:

[0022] X′=X+λ·X[:perm];

[0023] Wherein, X and X′ are the multivariate time series and the mixed sample sequence respectively; λ is the channel mixing ratio, specifically, it obeys the normal distribution with mean 0 and standard deviation σ; perm is the channel random permutation array, and X[:perm] means that the multivariate time series is randomly permuted according to the channel random permutation array;

[0024] The expression of the normalized time series is specifically:

[0025] X′ norm =InstanceNorm(X′);

[0026] Among them, X′ norm is the normalized time series, InstanceNorm(·) is the instance normalization operation function;

[0027] The expression of the patch sequence is specifically:

[0028] X v =Patching(X′ norm :v,P,S);

[0029] Among them, X v is the patch sequence, Patching(·) is the patch segmentation function, X′ norm :v is the normalized time series X′ norm ; P and S are the preset patch parameters, P is the patch length, and S is the step length between adjacent patches.

[0030] Optionally, the multi-layer channel Mamba model includes multi-layer channel Mamba blocks, and each layer of the channel Mamba blocks includes a block Mamba module and a channel attention module;

[0031] The multi-layer channel Mamba model is used to extract features from all the patch sequences to obtain target features, including:

[0032] Input each of the patch sequences into the block Mamba module in the channel Mamba block of the first layer to extract features, so as to obtain a key feature sequence corresponding to each of the patch sequences;

[0033] Input each of the key sequence sequences into the channel attention module in the channel Mamba block of the first layer for feature processing, so as to obtain a first layer output feature sequence corresponding to each of the patch sequences;

[0034] Input each of the first layer output feature sequences into the block Mamba module and the channel attention module in the channel Mamba block of the second layer in sequence, and obtain the second layer output feature sequence corresponding to each of the patch sequences in the same way;

[0035] Similarly, the target feature is obtained by iteratively processing the channel Mamba blocks of each layer in the multi-layer channel Mamba model.

[0036] Optionally, scoring the corresponding pharmaceutical supplier according to each supplier evaluation information to obtain scoring data of each pharmaceutical supplier includes:

[0037] Build a large language model for sentiment classification;

[0038] Using the sentiment classification large language model, sentiment classification is performed on the supplier evaluation information set corresponding to each of the pharmaceutical suppliers to obtain a sentiment classification result set corresponding to each of the pharmaceutical suppliers;

[0039] According to each of the sentiment classification result sets, the corresponding medical suppliers are scored to obtain the scoring data corresponding to each of the medical suppliers.

[0040] Optionally, constructing a large language model for sentiment classification includes:

[0041] Determine a sentiment classification task prefix, and construct a sentiment classification task instruction according to the sentiment classification task prefix;

[0042] When constructing the sentiment classification large language model for sentiment classification, the main task definition between input data and output data; wherein the input data includes the supplier evaluation information in the supplier evaluation information set, and the output data includes the sentiment classification results in the sentiment classification result set;

[0043] Providing multiple emotion classification examples, each of which includes corresponding input data and output data;

[0044] Model training is performed according to the sentiment classification task instruction, the main task definition and all the sentiment classification examples to obtain the sentiment classification large language model.

[0045] Optionally, before performing model training according to the sentiment classification task instruction, the main task definition and all the sentiment classification examples, the method further includes:

[0046] Use data enhancement methods to dynamically generate training samples;

[0047] Correspondingly, the model training is performed according to the sentiment classification task instruction, the main task definition and all the sentiment classification examples to obtain the sentiment classification large language model, including:

[0048] Based on the sentiment classification task instruction, the main task definition and all the sentiment classification examples, the training samples are used as training sets for model training to obtain the sentiment classification large language model.

[0049] Optionally, the sentiment classification result set corresponding to each of the medical suppliers includes at least one sentiment classification result; the sentiment classification result includes positive sentiment, negative sentiment, neutral sentiment and no sentiment output;

[0050] The step of scoring the corresponding pharmaceutical supplier according to each of the sentiment classification result sets to obtain the scoring data corresponding to each pharmaceutical supplier includes:

[0051] Select the sentiment classification result set of any one of the medical suppliers, traverse the type of each sentiment classification result in the selected sentiment classification result set, and when the sentiment classification result is the positive sentiment, add one point to the current score data of the corresponding medical supplier; when the sentiment classification result is the negative sentiment, reduce one point to the current score data of the corresponding medical supplier; when the sentiment classification result is the neutral sentiment or the no sentiment output, keep the current score data of the corresponding medical supplier unchanged;

[0052] The sentiment classification result set of each of the pharmaceutical suppliers is traversed, and the scoring data corresponding to each of the pharmaceutical suppliers is obtained in the same manner.

[0053] Optionally, obtaining a medicine procurement recommendation corresponding to the predicted medicine demand based on all the scoring data includes:

[0054] The pharmaceutical supplier corresponding to the maximum value among all the scoring data is determined as the target supplier corresponding to the predicted pharmaceutical demand.

[0055] In addition, the present invention also provides a medicine centralized procurement intelligent supervision system, which is applied to the aforementioned medicine centralized procurement intelligent supervision method, and the system includes:

[0056] A data acquisition module, used to provide a multivariate time series and target data, using the multivariate time series as an input variable of a preset prediction model, and using the target data as an output variable of the preset prediction model; wherein the multivariate time series is composed of historical pharmaceutical procurement data, and the target data is composed of predicted pharmaceutical demand;

[0057] A data processing module, used for preprocessing the multivariate time series by sequentially adopting channel mixing, instance normalization and data segmentation methods to obtain a plurality of patch sequences;

[0058] A feature extraction module, used for extracting features from all the patch sequences using a multi-layer channel Mamba model to obtain target features;

[0059] A demand prediction module, used for inputting the target feature into the linear layer of the preset prediction model to obtain the predicted medicine demand;

[0060] A supplier scoring module, used to obtain supplier evaluation information sets of multiple pharmaceutical suppliers, and score the corresponding pharmaceutical suppliers according to each supplier evaluation information set to obtain scoring data of each pharmaceutical supplier;

[0061] The centralized procurement suggestion module is used to obtain the centralized procurement suggestion for medicines corresponding to the predicted medicine demand based on all the scoring data.

[0062] In addition, the present invention also provides an intelligent supervision device for centralized procurement of medicines, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the method steps in the aforementioned intelligent supervision method for centralized procurement of medicines when running.

[0063] In addition, the present invention also provides a computer storage medium, which includes: at least one instruction, which implements the method steps in the aforementioned intelligent supervision method for centralized pharmaceutical procurement when the instruction is executed by a computer.

[0064] The beneficial effects of the present invention are as follows: since the multivariate time series not only contains the time series data of a single variable, but also incorporates the information of other related variables, providing a multivariate time series composed of historical medicine centralized procurement data as an input variable of a preset prediction model can not only provide rich information dimensions and capture the complex relationship between variables, but also improve the prediction accuracy of subsequent models, and facilitate enhancing the timeliness and effectiveness of medicine centralized procurement supervision; by sequentially adopting the methods of channel mixing, instance normalization and data segmentation to pre-process the multivariate time series, the artificial intelligence algorithm can be used to greatly improve the data processing speed and accuracy, and reduce the time and error rate of manual operation; the multi-layer channel Mamba model is adopted The model is used for feature extraction, which can be applied to complex multivariate time series and has better large-scale data capabilities, thereby ensuring the real-time, accuracy and reliability of the target features obtained; the target features are input into the linear layer of the preset prediction model, and predictions are made based on the target features, which can predict the pharmaceutical procurement demand and market fluctuations in advance, optimize procurement decisions, avoid drug shortages or excessive procurement, and improve the scientificity and rationality of procurement plans; then, the supplier evaluation information set of each pharmaceutical supplier is used to score the supplier, and the scoring data obtained is used to provide pharmaceutical centralized procurement recommendations, which can comprehensively evaluate and manage the supplier's qualifications, credit and supply history, and ensure the supplier's reliability and drug quality;

[0065] The intelligent supervision method, system, device and storage medium for centralized medicine procurement of the present invention realizes efficient, intelligent and comprehensive centralized medicine procurement supervision. The data processing efficiency is high and the real-time performance is high in the whole process, which improves the efficiency of centralized medicine procurement supervision and the scientific nature of decision-making, and ensures the stability and safety of drug supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0067] Figure 1 A flow chart of a method for intelligent supervision of centralized procurement of medicines in Embodiment 1 of the present invention is shown;

[0068] Figure 2 A flowchart of preprocessing a multivariate time series in the first embodiment of the present invention is shown;

[0069] Figure 3 A flowchart of feature extraction for all patch sequences in Embodiment 1 of the present invention is shown;

[0070] Figure 4 A flowchart showing how to score each pharmaceutical supplier in the first embodiment of the present invention is shown.

[0071] Figure 5 A structural diagram of an intelligent supervision system for centralized pharmaceutical procurement in Embodiment 2 of the present invention is shown. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0073] Embodiment 1

[0074] This embodiment provides a method for intelligent supervision of centralized medical procurement, such as Figure 1 As shown, the method includes:

[0075] S1: providing a multivariate time series and target data, using the multivariate time series as an input variable of a preset prediction model, and using the target data as an output variable of the preset prediction model; wherein the multivariate time series is composed of historical pharmaceutical procurement data, and the target data is composed of predicted pharmaceutical demand;

[0076] S2: preprocessing the multivariate time series by sequentially using channel mixing, instance normalization, and data segmentation methods to obtain a plurality of patch sequences;

[0077] S3: using a multi-layer channel Mamba model to extract features from all the patch sequences to obtain target features;

[0078] S4: inputting the target feature into the linear layer of the preset prediction model to obtain the predicted medicine demand;

[0079] S5: obtaining supplier evaluation information sets of multiple pharmaceutical suppliers, and scoring the corresponding pharmaceutical suppliers according to each supplier evaluation information set to obtain scoring data of each pharmaceutical supplier;

[0080] S6: Based on all the scoring data, obtain a medicine procurement recommendation corresponding to the predicted medicine demand.

[0081] In this embodiment, since the multivariate time series not only contains the time series data of a single variable, but also incorporates the information of other related variables, providing a multivariate time series composed of historical pharmaceutical procurement data as an input variable of a preset prediction model can not only provide rich information dimensions and capture the complex relationship between variables, but also improve the prediction accuracy of subsequent models, which is convenient for enhancing the timeliness and effectiveness of pharmaceutical procurement supervision; by sequentially adopting the methods of channel mixing, instance normalization and data segmentation to pre-process the multivariate time series, artificial intelligence algorithms can be used to greatly improve the data processing speed and accuracy, and reduce the time and error rate of manual operations; using the multi-layer channel Mamba model Feature extraction can be applied to complex multivariate time series and has better large-scale data capabilities, thereby ensuring the real-time, accuracy and reliability of the target features obtained; the target features are input into the linear layer of the preset prediction model, and predictions are made based on the target features. This can predict medical procurement needs and market fluctuations in advance, optimize procurement decisions, avoid drug shortages or over-procurement, and improve the scientificity and rationality of procurement plans; then, each medical supplier is scored through the supplier evaluation information set, and medical procurement recommendations are provided based on the scoring data obtained. This can comprehensively evaluate and manage the supplier's qualifications, credit and supply history, and ensure the supplier's reliability and drug quality.

[0082] The intelligent supervision method for centralized medicine procurement in this embodiment realizes efficient, intelligent and comprehensive centralized medicine procurement supervision. The data processing efficiency is high and the real-time performance is high in the whole process, which improves the efficiency of centralized medicine procurement supervision and the scientific nature of decision-making, and ensures the stability and safety of drug supply.

[0083] Each step of the method for intelligent supervision of centralized pharmaceutical procurement in this embodiment is described in detail below.

[0084] Preferably, in this embodiment S1, the historical medicine procurement data constituting the multivariate time series includes the daily sales volume, weather information and seasonal information of at least one medicine in a specified historical time period, and these data are cleaned and aligned to ensure that the sales volume, weather information and seasonal information are aligned in time (i.e., there are corresponding sales volume, weather information and seasonal information data at each time point), and then a suitable model (such as a linear regression model, a time series analysis model, etc.) is selected to fit the corresponding multivariate time series, which is usually a two-dimensional array or a three-dimensional array. The target data consisting of the predicted medicine demand refers to the demand for the corresponding medicine in the historical medicine procurement data in the specified future time period.

[0085] Using the above-mentioned multivariate time series as the input variable of the preset prediction model can not only provide rich information dimensions and capture the complex relationship between variables, but also improve the prediction accuracy and adapt to complex systems, thereby ensuring the accuracy and reliability of the preset prediction model; using the target data as the output variable of the preset prediction model facilitates the subsequent accurate prediction of the predicted medical demand.

[0086] In the above-mentioned historical pharmaceutical procurement data, the daily sales volume of medicines in a specified time period can be obtained through the pharmaceutical sales system, financial statements or sales records. Weather information (including temperature, humidity, rainfall, etc.) can be obtained from meteorological departments, weather forecast websites and other channels. Seasonal information can be directly generated through calendars or timestamps (for the convenience of subsequent processing, seasonal information can be encoded as numerical or categorical variables, for example, spring is encoded as 1, summer is encoded as 2, and so on).

[0087] Specifically, in this embodiment S1, the specific expressions of the multivariate time series and the target data are respectively X∈R L×V and Y∈R T×V , where X and Y are the multivariate time series and target data, respectively, V is the number of channels, L and T are the sequence length of the multivariate time series and the sequence length of the target data, respectively.

[0088] Preferably, if Figure 2 As shown, this embodiment S2 includes:

[0089] S21: determining a channel mixing ratio and a channel random arrangement array, and performing channel mixing on the multivariate time series according to the channel mixing ratio and the channel random arrangement data to obtain a mixed sample sequence;

[0090] S22: performing an instance normalization operation on the mixed sample sequence to obtain a normalized time series;

[0091] S23: performing patch segmentation on the normalized time series according to preset patch parameters to obtain a plurality of patch sequences.

[0092] In the channel mixing process, the channel mixing ratio and channel random arrangement data are first determined to facilitate channel mixing of multivariate time series as needed, which can not only promote the information interaction between different variables (channels) and enhance the model's understanding of the overall characteristics of multivariate time series data, but also help capture the potential relationship and dependence between variables and improve the model's prediction ability; it can also effectively alleviate information redundancy and improve data processing efficiency; the mixed sample sequence generated by channel mixing is a virtual sample, which can increase data diversity. After channel mixing, instance normalization is performed. By normalizing each sample (or instance), the data distribution between different samples tends to be consistent, which helps stabilize the training process and reduce the model's sensitivity to initial parameters and training data; it can also better handle data with different distributions and improve generalization ability; the normalized time series obtained after instance normalization helps the model find the optimal solution faster, thereby accelerating the convergence speed of the training process. By performing patch segmentation on the normalized time series according to preset patch parameters, the normalized time series of a large data set can be divided into multiple small batches for processing, thereby reducing the complexity of a single calculation. Small batch processing can help the model update parameters faster during training. At the same time, by using techniques such as gradient accumulation, the training efficiency can be further improved without sacrificing model performance. In addition, it can also prevent the model from overfitting and further improve the generalization ability of the model.

[0093] Specifically, in S21, the expression of the mixed sample sequence is:

[0094] X′=X+λ·X[:perm];

[0095] Wherein, X and X′ are the multivariate time series and the mixed sample sequence respectively; λ is the channel mixing ratio, specifically, it obeys the normal distribution with mean 0 and standard deviation σ; perm is the channel random permutation array, and X[:perm] means that the multivariate time series is randomly arranged according to the channel random permutation array.

[0096] Specifically, in S22, the expression of the normalized time series is:

[0097] X′ norm =InstanceNorm(X′);

[0098] Among them, X′ norm is the normalized time series, and InstanceNorm(·) is the instance normalization operation function.

[0099] Specifically, in S23, the expression of the patch sequence is specifically:

[0100] X v =Patching(X′ norm :v,P,S);

[0101] Among them, X v is the patch sequence, Patching(·) is the patch segmentation function, X′ norm :v is the normalized time series X′ norm ; P and S are the preset patch parameters, P is the patch length, and S is the step length between adjacent patches.

[0102] Preferably, in this embodiment S3, the multi-layer channel Mamba model includes multi-layer channel Mamba blocks, and each layer of the channel Mamba blocks includes a block Mamba module and a channel attention module;

[0103] like Figure 3 As shown, S3 includes:

[0104] S31: inputting each of the patch sequences into the block Mamba module in the channel Mamba block of the first layer to perform feature extraction, so as to obtain a key feature sequence corresponding to each of the patch sequences;

[0105] S32: input each of the key sequence sequences into the channel attention module in the channel Mamba block of the first layer for feature processing, so as to obtain a first layer output feature sequence corresponding to each of the patch sequences;

[0106] S33: inputting each of the first layer output feature sequences into the block Mamba module and the channel attention module in the channel Mamba block of the second layer in sequence, and obtaining the second layer output feature sequence corresponding to each of the patch sequences in the same manner;

[0107] S34: Similarly, the channel Mamba blocks of each layer in the multi-layer channel Mamba model are used to perform iterative processing to obtain the target feature.

[0108] In this embodiment, the multi-layer channel Mamba model refers to a deep learning model constructed by stacking multiple channel Mamba blocks (or channel Mamba modules) based on the Mamba model framework, wherein the channel Mamba block is a new architecture that combines the characteristics of the structured state space model (SSM) and modern activation functions, and is intended to improve the efficiency and performance of sequence modeling tasks. Through the multi-layer channel Mamba model stacked with channel Mamba blocks, it is possible to utilize the linear scalability of SSM to achieve efficient calculations on the sequence length, avoiding the common secondary calculation complexity problems in the Transformer model; it also alleviates the modeling constraints of the convolutional neural network through the global receptive field and dynamic weighting mechanism, and provides advanced modeling capabilities similar to Transformer; the number of channel Mamba blocks can be flexibly adjusted according to specific task requirements, and the depth and complexity of the model can be changed by increasing or decreasing the number of channel Mamba blocks.

[0109] In the processing of each layer of channel Mamba blocks in this embodiment, the block Mamba module is first used for processing, and the feature representation of each patch sequence can be output, which is convenient for subsequent channel attention processing; the channel attention module is then used for feature processing, so that important feature channels can be enhanced and unimportant channels can be suppressed, and then the features after channel attention weighted processing can be output to obtain feature information that is more useful for the task; through the iterative processing of multiple layers of channel Mamba blocks, the deep features of the time series data can be gradually extracted, more accurate target features can be obtained, and strong support can be provided for subsequent prediction tasks.

[0110] Specifically, for the patch sequence X in S23 v , the key feature sequence obtained after feature extraction in the first-layer block Mamba module is Z1=M(X v ), the first-layer output feature sequence obtained by feature processing in the first-layer channel attention module is Z′1=A(Z1); where M(·) and A(·) represent the block Mamba module feature extraction function and the channel attention feature processing function respectively. For the l-th layer channel Mamba block, its output l-th layer output feature sequence is Z′ l , for the l+1th layer channel Mamba block, its output l+1th layer output feature sequence is Z l+1 ′, when the number of layers of the channel Mamba block is k, the target feature output by S3 in this embodiment is Z k ′.

[0111] Specifically, in this embodiment S4, the output target feature is Z k ′ is input into the linear layer of the preset prediction model for target prediction. The prediction process is:

[0112] Y P =W proj Flatten(Z k ′);

[0113] Among them, Y P is the model prediction result, i.e. the predicted demand for medicine; W proj is the linear layer weight matrix, and Flatten(·) represents the operation function expanded into a one-dimensional vector.

[0114] In this embodiment S5, pharmaceutical suppliers may use web crawlers to collect information from drug purchasing websites, health forums, or customer feedback emails, and their corresponding supplier evaluation information sets obtain text data from user comments, feedback forms, social media, or other relevant data sources.

[0115] Preferably, if Figure 4 As shown, in this embodiment S5, the corresponding pharmaceutical supplier is scored according to each supplier evaluation information, and the scoring data of each pharmaceutical supplier is obtained, including:

[0116] S51: Building a large language model for sentiment classification;

[0117] S52: using the sentiment classification large language model to perform sentiment classification on the supplier evaluation information set corresponding to each of the pharmaceutical suppliers, to obtain a sentiment classification result set corresponding to each of the pharmaceutical suppliers;

[0118] S53: Score the corresponding pharmaceutical supplier according to each of the sentiment classification result sets, and obtain the score data corresponding to each pharmaceutical supplier.

[0119] Using the sentiment classification language model to classify the supplier evaluation information can make full use of big data to comprehensively evaluate and manage the supplier's qualifications, credit and supply history, ensure the supplier's reliability and drug quality, and reduce the drug quality and supply risks caused by supplier problems. It is then convenient to predict medical demand based on the above steps, provide optimized procurement suggestions, help decision makers formulate reasonable procurement plans, reduce operating costs in the process of medical procurement, avoid waste of resources, and improve the allocation efficiency and benefits of medical resources.

[0120] Furthermore, the present embodiment S51 includes:

[0121] S511: Determine a sentiment classification task prefix, and construct a sentiment classification task instruction according to the sentiment classification task prefix;

[0122] S512: when constructing the sentiment classification large language model for sentiment classification, a main task definition between input data and output data; wherein the input data includes the supplier evaluation information in the supplier evaluation information set, and the output data includes the sentiment classification result in the sentiment classification result set;

[0123] S513: providing a plurality of emotion classification examples, each of the emotion classification examples including corresponding input data and output data;

[0124] S514: Perform model training according to the sentiment classification task instruction, the main task definition and all the sentiment classification examples to obtain the sentiment classification large language model.

[0125] Through the above steps, it is easier to better guide the model to solve the problem of sentiment classification and obtain a large sentiment classification language model with higher classification accuracy.

[0126] Specifically, in S511 of this embodiment, a NER prefix is ​​selected as a sentiment classification task prefix, for example:

[0127] Definition: Given the following context, output the relevant entities. Explain the answer step by step.

[0128] Context: The service from this drug supplier was very prompt, but the quality of the drugs could be improved.

[0129] In this embodiment S512, the main task definition is constructed as follows:

[0130] If the input data contains positive emotions, the output data is "positive"; if the input data contains negative emotions, the output data is "negative"; if the input data contains neutral emotions, the output data should be "neutral"; if the input data does not contain emotions, the output data should be "none".

[0131] In this embodiment S513, the sentiment classification example is as follows:

[0132] Positive Sentiment Example 1:

[0133] Input data: I am very satisfied with the service of this supplier. The aspect is service.

[0134] Output data: positive.

[0135] Positive Sentiment Example 2:

[0136] Input data: The quality of the medicines provided by this company is very good. Aspect is quality.

[0137] Output data: positive.

[0138] Negative Sentiment Example 1:

[0139] Input data: Their delivery time is very slow. Aspect is the delivery time.

[0140] Output data: negative.

[0141] Negative Sentiment Example 2:

[0142] Input data: The packaging of the drug does not meet the standards. The aspect is packaging.

[0143] Output data: negative.

[0144] Neutral Sentiment Example 1:

[0145] Input data: Their prices are relatively stable. Aspect is price.

[0146] Output data: neutral.

[0147] Neutral Sentiment Example 2:

[0148] Input data: The location of the supplier is relatively convenient. Aspect is location.

[0149] Output data: neutral.

[0150] Based on the above sentiment classification task instructions, main task definition and all sentiment classification examples, the model is trained to obtain a sentiment classification language model. The final inference example generated by the model is as follows:

[0151] Definition: Given the following context, output the relevant entities. Explain the answer step by step.

[0152] Context: The service from this drug supplier was very prompt, but the quality of the drugs could be improved.

[0153] Next solve the following tasks:

[0154] If the input data contains positive emotions, the output data is "positive"; if the input data contains negative emotions, the output data is "negative"; if the input data contains neutral emotions, the output data should be "neutral"; if the input data does not contain emotions, the output data should be "none".

[0155] Positive Sentiment Example 1:

[0156] Input data: I am very satisfied with the service of this supplier. The aspect is service.

[0157] Output data: positive.

[0158] Positive Sentiment Example 2:

[0159] Input data: The quality of the medicines provided by this company is very good. Aspect is quality.

[0160] Output data: positive.

[0161] Negative Sentiment Example 1:

[0162] Input data: Their delivery time is very slow. Aspect is the delivery time.

[0163] Output data: negative.

[0164] Negative Sentiment Example 2:

[0165] Input data: The packaging of the drug does not meet the standards. The aspect is packaging.

[0166] Output data: negative.

[0167] Neutral Sentiment Example 1:

[0168] Input data: Their prices are relatively stable. Aspect is price.

[0169] Output data: neutral.

[0170] Neutral Sentiment Example 2:

[0171] Input data: The location of the supplier is relatively convenient. Aspect is location.

[0172] Output data: neutral.

[0173] Based on the above sentiment classification language model, when new data is input, the corresponding sentiment classification result can be output. For example, when the input data is "The service of this drug supplier is very fast, but the quality of the drugs needs to be improved. The aspect is service", the sentiment classification result corresponding to the input data can be output as "positive".

[0174] Preferably, before S514, the method further includes:

[0175] Use data enhancement methods to dynamically generate training samples;

[0176] Correspondingly, S514 includes:

[0177] Based on the sentiment classification task instruction, the main task definition and all the sentiment classification examples, the training samples are used as training sets for model training to obtain the sentiment classification large language model.

[0178] By dynamically generating training samples and continuously generating new and unique training samples, ensuring that each iteration is exposed to different data combinations, the diversity of the training set during model training can be improved, thereby improving the generalization ability of the model and effectively improving the robustness of the model in dealing with data changes and noise.

[0179] Specifically, the process of dynamically generating training samples is as follows:

[0180] (1) Extract variables: First, extract relevant variables from the existing training data by prompting the sentiment classification large language model. For example, for a question: "A warehouse has 3 boxes of apples and 4 boxes of oranges. Each box of apples weighs 2 kg, and each box of oranges weighs 5 kg. How many kilograms of fruit are there in total?", extract the four variables {num_apples}, {num_oranges}, {weight_apple}, and {weight_orange}.

[0181] (2) Generate delegitimization problem: Generate a delegitimization problem containing placeholder variables. The form of the delegitimization problem is:

[0182] "A warehouse has {num_apples} boxes of apples and {num_oranges} boxes of oranges. Each box of apples weighs {weight_apple} kilograms, and each box of oranges weighs {weight_orange} kilograms. How many kilograms of fruit are there in total?".

[0183] (3) Define variable ranges: Define a value range for each variable so that random sampling can be performed during training. For example:

[0184] {'num_apples':'random.randint(1,50)',

[0185] 'num_oranges':'random.randint(1,50)',

[0186] 'weight_apple':'random.uniform(1,10)',

[0187] 'weight_orange':'random.uniform(1,10)'}.

[0188] (4) Generate solution functions: prompt the sentiment large language model to generate functions that solve these de-lexicalization problems. For example:

[0189] def solution(num_apples,num_oranges,weight_apple,weight_orange):

[0190] total_weight=num_apples*weight_apple+num_oranges*weight_orangereturn total_weight.

[0191] (5) Sampling new values: Sampling new variable values ​​from the defined value range to create new training samples. For example:

[0192] A warehouse has 25 boxes of apples and 30 boxes of oranges. Each box of apples weighs 3.5 kg, and each box of oranges weighs 4.2 kg. How many kilograms of fruit are there in total?

[0193] (6) Compute new answers: Use the generated function and the new sampled values ​​to compute new training labels. For example:

[0194] Ground Truth: 213.5 kg.

[0195] (7) Dynamically generate training samples: Continuously generate new and unique training samples according to the above process.

[0196] Specifically, the sentiment classification result set corresponding to each of the medical suppliers includes at least one sentiment classification result; the sentiment classification results include positive sentiment, negative sentiment, neutral sentiment and no sentiment output.

[0197] Since the supplier evaluation information set corresponding to each pharmaceutical supplier may include multiple supplier evaluation information, the corresponding sentiment classification results may also include multiple ones, and the sentiment classification result set may include at least one sentiment classification result.

[0198] Specifically, S53 of this embodiment includes:

[0199] S531: Select the sentiment classification result set of any one of the pharmaceutical suppliers, traverse the type of each sentiment classification result in the selected sentiment classification result set, and when the sentiment classification result is the positive sentiment, add one point to the current score data of the corresponding pharmaceutical supplier; when the sentiment classification result is the negative sentiment, reduce one point to the current score data of the corresponding pharmaceutical supplier; when the sentiment classification result is the neutral sentiment or the no sentiment output, keep the current score data of the corresponding pharmaceutical supplier unchanged;

[0200] S532: Traverse the sentiment classification result set of each of the pharmaceutical suppliers, and obtain the score data corresponding to each of the pharmaceutical suppliers in the same manner.

[0201] Through the above method, each pharmaceutical supplier can be scored simply and effectively, and the supplier's qualifications, credit and supply history can be comprehensively evaluated and managed to ensure the supplier's reliability and drug quality, and reduce drug quality and supply risks caused by supplier problems. Based on accurate demand forecasts, the system can provide optimized procurement suggestions, help decision makers formulate reasonable procurement plans, avoid resource waste, and improve the efficiency and benefits of pharmaceutical resource allocation.

[0202] Specifically, for any sentiment classification result of any pharmaceutical supplier, when the output sentiment classification result is positive, that is, the output data is "positive", the current score data of the pharmaceutical supplier is +1; when the output sentiment classification result is negative, that is, the output data is "negative", the current score data of the pharmaceutical supplier is -1; when the output sentiment classification result is neutral, that is, the output data is "neutral", the current score data of the pharmaceutical supplier remains unchanged; when the output sentiment classification result is no sentiment output, that is, the output data is "none", the current score data of the pharmaceutical supplier remains unchanged. Each sentiment classification result of the pharmaceutical supplier is scored according to this process, and the final accumulated score data is the final score data of the pharmaceutical supplier.

[0203] Preferably, this embodiment S6 includes:

[0204] The pharmaceutical supplier corresponding to the maximum value among all the scoring data is determined as the target supplier corresponding to the predicted pharmaceutical demand.

[0205] The pharmaceutical supplier with the highest score data representing the best comprehensive evaluation of its qualifications, credit and supply history can be identified as the target supplier for pharmaceutical procurement. Determining the target supplier based on the above method and conducting centralized procurement according to the predicted pharmaceutical demand can provide optimized procurement suggestions, help decision makers formulate reasonable procurement plans, avoid waste of resources, and improve the efficiency and effectiveness of the allocation of pharmaceutical resources.

[0206] The complete intelligent supervision method for centralized procurement of medicines described in this embodiment introduces artificial intelligence technology and big data analysis technology, realizes the intelligence, real-time and high efficiency of the centralized procurement supervision system of medicines, and has significant beneficial effects. First, by using artificial intelligence algorithms and big data technology, the data processing speed and accuracy are greatly improved, and the time and error rate of manual operation are reduced. The system can monitor the entire process of centralized procurement of medicines in real time, including the real-time update of the supplier's supply situation and the inventory status of medicines, timely discover and deal with problems, and enhance the timeliness and effectiveness of supervision. Through the demand forecasting module, using machine learning algorithms and big data analysis, the demand for medicine procurement and market fluctuations can be predicted in advance, procurement decisions can be optimized, drug shortages or excessive procurement can be avoided, and the scientificity and rationality of procurement plans can be improved. The supplier management module comprehensively evaluates and manages the qualifications, credit and supply history of suppliers to ensure the reliability of suppliers and the quality of medicines, and reduce the quality and supply risks of medicines caused by supplier problems. Based on accurate demand forecasting, the system can provide optimized procurement suggestions, help decision makers formulate reasonable procurement plans, avoid waste of resources, and improve the efficiency and benefits of the allocation of medical resources. Through intelligent supervision means, the system covers all aspects of centralized procurement of medicines, reduces blind spots and loopholes in supervision, and improves the overall supervision level and reliability. In addition, through efficient data processing, real-time monitoring and intelligent analysis and prediction, the system significantly reduces the operating costs in the process of centralized procurement of medicines and improves the overall economic benefits. In short, the present invention provides an efficient, intelligent and comprehensive solution for the supervision of centralized procurement of medicines, which not only improves the efficiency of supervision and the scientific nature of decision-making, but also ensures the stability and safety of drug supply, and has significant social and economic benefits.

[0207] Embodiment 2

[0208] A medicine centralized procurement intelligent supervision system is applied to the medicine centralized procurement intelligent supervision method of embodiment 1, such as Figure 5 As shown, the system includes:

[0209] A data acquisition module, used to provide a multivariate time series and target data, using the multivariate time series as an input variable of a preset prediction model, and using the target data as an output variable of the preset prediction model; wherein the multivariate time series is composed of historical pharmaceutical procurement data, and the target data is composed of predicted pharmaceutical demand;

[0210] A data processing module, used for preprocessing the multivariate time series by sequentially adopting channel mixing, instance normalization and data segmentation methods to obtain a plurality of patch sequences;

[0211] A feature extraction module, used for extracting features from all the patch sequences using a multi-layer channel Mamba model to obtain target features;

[0212] A demand prediction module, used for inputting the target feature into the linear layer of the preset prediction model to obtain the predicted medicine demand;

[0213] A supplier scoring module, used to obtain supplier evaluation information sets of multiple pharmaceutical suppliers, and score the corresponding pharmaceutical suppliers according to each supplier evaluation information set to obtain scoring data of each pharmaceutical supplier;

[0214] The centralized procurement suggestion module is used to obtain the centralized procurement suggestion for medicines corresponding to the predicted medicine demand based on all the scoring data.

[0215] In this embodiment, since the multivariate time series not only contains the time series data of a single variable, but also incorporates the information of other related variables, the multivariate time series composed of historical medicine procurement data is provided as the input variable of the preset prediction model through the data acquisition module, which can not only provide rich information dimensions and capture the complex relationship between variables, but also improve the prediction accuracy of subsequent models, so as to enhance the timeliness and effectiveness of medicine procurement supervision; through the data processing module, the multivariate time series is pre-processed by using channel mixing, instance normalization and data segmentation methods in sequence, which can greatly improve the data processing speed and accuracy by using artificial intelligence algorithms, and reduce the time and error rate of manual operations; through the feature extraction module, the multi-layer channel Mamba model is used for It can extract features and is applicable to complex multivariate time series. It has better large-scale data capabilities, thereby ensuring the real-time, accuracy and reliability of the target features obtained. It inputs the target features into the linear layer of the preset prediction model, and uses the demand prediction module to make predictions based on the target features. It can predict medical procurement needs and market fluctuations in advance, optimize procurement decisions, avoid drug shortages or excessive procurement, and improve the scientificity and rationality of procurement plans. It then uses the supplier scoring module to score suppliers through the supplier evaluation information set of each medical supplier, and uses the centralized procurement suggestion module to provide medical centralized procurement suggestions based on the scoring data obtained. It can comprehensively evaluate and manage the supplier's qualifications, credit and supply history, and ensure the supplier's reliability and drug quality.

[0216] The intelligent supervision system for centralized pharmaceutical procurement of this embodiment realizes efficient, intelligent and comprehensive centralized pharmaceutical procurement supervision. The data processing efficiency is high and the real-time performance is high during the whole process, which improves the efficiency of centralized pharmaceutical procurement supervision and the scientific nature of decision-making, and ensures the stability and safety of drug supply.

[0217] The functions of each module in the intelligent supervision system for centralized medical procurement described in this embodiment are the same as the method steps of the intelligent supervision method for centralized medical procurement described in Example 1. Therefore, for details not included in this embodiment, please refer to Example 1 and Figures 1 to 4 The detailed description will not be repeated here.

[0218] Embodiment 3

[0219] A device for intelligent supervision of centralized medicine procurement includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed, the method steps in the method for intelligent supervision of centralized medicine procurement of embodiment 1 are implemented.

[0220] Through computer programs stored in memory and running on processors, efficient, intelligent and comprehensive medical procurement supervision is achieved. The data processing efficiency is high and real-time in the whole process, which improves the efficiency of medical procurement supervision and the scientific nature of decision-making, and ensures the stability and safety of drug supply.

[0221] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of a computer device, and uses various interfaces and lines to connect various parts of the entire computer device.

[0222] The memory can be used to store computer programs and / or models. The processor realizes various functions of the computer device by running or executing the computer programs and / or models stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, video data, etc.). In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0223] It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by a computer program. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0224] These computer programs may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0225] These computer programs can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0226] This embodiment also provides a computer storage medium, which includes: at least one instruction, which implements the method steps in the intelligent supervision method for centralized medical procurement of embodiment 1 when the instruction is executed by a computer.

[0227] By executing a computer storage medium containing at least one instruction, efficient, intelligent and comprehensive medical procurement supervision is achieved. The data processing efficiency is high and real-time in the whole process, which improves the efficiency of medical procurement supervision and the scientific nature of decision-making, and ensures the stability and safety of drug supply.

[0228] Similarly, for details not yet provided in this embodiment, please refer to Embodiment 1, Embodiment 2 and Figures 1 to 5 The detailed description will not be repeated here.

[0229] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for intelligent supervision of centralized medical procurement, characterized in that: The method comprises: Providing a multivariate time series and target data, using the multivariate time series as an input variable of a preset prediction model, and using the target data as an output variable of the preset prediction model; wherein the multivariate time series is composed of historical pharmaceutical procurement data, and the target data is composed of predicted pharmaceutical demand; Preprocessing the multivariate time series by sequentially using channel mixing, instance normalization, and data segmentation to obtain a plurality of patch sequences; Using a multi-layer channel Mamba model to extract features from all the patch sequences to obtain target features; Inputting the target feature into the linear layer of the preset prediction model to obtain the predicted medicine demand; Obtaining supplier evaluation information sets of multiple pharmaceutical suppliers, scoring the corresponding pharmaceutical suppliers according to each of the supplier evaluation information sets, and obtaining scoring data of each pharmaceutical supplier; Based on all the scoring data, a pharmaceutical procurement recommendation corresponding to the predicted pharmaceutical demand is obtained.

2. The intelligent supervision method for centralized medical procurement according to claim 1 is characterized in that: The method of channel mixing, instance normalization and data segmentation is sequentially used to preprocess the multivariate time series to obtain multiple patch sequences, including: Determine a channel mixing ratio and a channel random arrangement array, and perform channel mixing on the multivariate time series according to the channel mixing ratio and the channel random arrangement data to obtain a mixed sample sequence; Performing instance normalization operation on the mixed sample sequence to obtain a normalized time series; The normalized time series is patch segmented according to preset patch parameters to obtain a plurality of patch sequences.

3. The intelligent supervision method for centralized medical procurement according to claim 2 is characterized in that: The expression of the mixed sample sequence is specifically: X′=X+λ·X[:perm]; Wherein, X and X′ are the multivariate time series and the mixed sample sequence respectively; λ is the channel mixing ratio, specifically, it obeys the normal distribution with mean 0 and standard deviation σ; perm is the channel random permutation array, and X[:perm] means that the multivariate time series is randomly permuted according to the channel random permutation array; The expression of the normalized time series is specifically: X′ norm =InstanceNorm(X′); Among them, X′ norm is the normalized time series, InstanceNorm(·) is the instance normalization operation function; The expression of the patch sequence is specifically: X v =Patching(X′ norm :v,P,S); Among them, X v is the patch sequence, Patching(·) is the patch segmentation function, X′ norm :v is the normalized time series X′ norm ; P and S are the preset patch parameters, P is the patch length, and S is the step length between adjacent patches.

4. The intelligent supervision method for centralized medical procurement according to claim 1 is characterized in that: The multi-layer channel Mamba model includes multi-layer channel Mamba blocks, and each layer of the channel Mamba block includes a block Mamba module and a channel attention module; The multi-layer channel Mamba model is used to extract features from all the patch sequences to obtain target features, including: Input each of the patch sequences into the block Mamba module in the channel Mamba block of the first layer to extract features, so as to obtain a key feature sequence corresponding to each of the patch sequences; Input each of the key sequence sequences into the channel attention module in the channel Mamba block of the first layer for feature processing, so as to obtain a first layer output feature sequence corresponding to each of the patch sequences; Input each of the first layer output feature sequences into the block Mamba module and the channel attention module in the channel Mamba block of the second layer in sequence, and obtain the second layer output feature sequence corresponding to each of the patch sequences in the same way; Similarly, the target feature is obtained by iteratively processing the channel Mamba blocks of each layer in the multi-layer channel Mamba model.

5. The intelligent supervision method for centralized medical procurement according to claim 1 is characterized in that: Scoring the corresponding pharmaceutical suppliers according to the evaluation information of each supplier to obtain scoring data of each pharmaceutical supplier includes: Build a large language model for sentiment classification; Using the sentiment classification large language model, sentiment classification is performed on the supplier evaluation information set corresponding to each of the pharmaceutical suppliers to obtain a sentiment classification result set corresponding to each of the pharmaceutical suppliers; According to each of the sentiment classification result sets, the corresponding medical suppliers are scored to obtain the scoring data corresponding to each of the medical suppliers.

6. The intelligent supervision method for centralized medical procurement according to claim 5 is characterized in that: The constructing of a large language model for sentiment classification includes: Determine a sentiment classification task prefix, and construct a sentiment classification task instruction according to the sentiment classification task prefix; When constructing the sentiment classification large language model for sentiment classification, the main task definition between input data and output data; wherein the input data includes the supplier evaluation information in the supplier evaluation information set, and the output data includes the sentiment classification results in the sentiment classification result set; Providing multiple emotion classification examples, each of which includes corresponding input data and output data; Model training is performed according to the sentiment classification task instruction, the main task definition and all the sentiment classification examples to obtain the sentiment classification large language model.

7. The intelligent supervision method for centralized medical procurement according to claim 6 is characterized in that: Before performing model training according to the sentiment classification task instruction, the main task definition and all the sentiment classification examples, the method further includes: Use data enhancement methods to dynamically generate training samples; Correspondingly, the model training is performed according to the sentiment classification task instruction, the main task definition and all the sentiment classification examples to obtain the sentiment classification large language model, including: Based on the sentiment classification task instruction, the main task definition and all the sentiment classification examples, the training samples are used as training sets for model training to obtain the sentiment classification large language model.

8. The intelligent supervision method for centralized medical procurement according to claim 5 is characterized in that: The sentiment classification result set corresponding to each of the medical suppliers includes at least one sentiment classification result; the sentiment classification results include positive sentiment, negative sentiment, neutral sentiment and no sentiment output; The step of scoring the corresponding pharmaceutical supplier according to each of the sentiment classification result sets to obtain the scoring data corresponding to each pharmaceutical supplier includes: Select the sentiment classification result set of any one of the medical suppliers, traverse the type of each sentiment classification result in the selected sentiment classification result set, and when the sentiment classification result is the positive sentiment, add one point to the current score data of the corresponding medical supplier; when the sentiment classification result is the negative sentiment, reduce one point to the current score data of the corresponding medical supplier; when the sentiment classification result is the neutral sentiment or the no sentiment output, keep the current score data of the corresponding medical supplier unchanged; The sentiment classification result set of each of the pharmaceutical suppliers is traversed, and the scoring data corresponding to each of the pharmaceutical suppliers is obtained in the same manner.

9. The intelligent supervision method for centralized medical procurement according to any one of claims 1 to 8, characterized in that: The method of obtaining a medicine procurement recommendation corresponding to the predicted medicine demand based on all the scoring data includes: The pharmaceutical supplier corresponding to the maximum value among all the scoring data is determined as the target supplier corresponding to the predicted pharmaceutical demand.

10. An intelligent supervision system for centralized medical procurement, characterized in that: Applied to the intelligent supervision method for centralized medical procurement as claimed in any one of claims 1 to 9, the system comprises: A data acquisition module, used to provide a multivariate time series and target data, using the multivariate time series as an input variable of a preset prediction model, and using the target data as an output variable of the preset prediction model; wherein the multivariate time series is composed of historical pharmaceutical procurement data, and the target data is composed of predicted pharmaceutical demand; A data processing module, used for preprocessing the multivariate time series by sequentially adopting channel mixing, instance normalization and data segmentation methods to obtain a plurality of patch sequences; A feature extraction module, used for extracting features from all the patch sequences using a multi-layer channel Mamba model to obtain target features; A demand prediction module, used for inputting the target feature into the linear layer of the preset prediction model to obtain the predicted medicine demand; A supplier scoring module, used to obtain supplier evaluation information sets of multiple pharmaceutical suppliers, and score the corresponding pharmaceutical suppliers according to each supplier evaluation information set to obtain scoring data of each pharmaceutical supplier; The centralized procurement suggestion module is used to obtain the centralized procurement suggestion for medicines corresponding to the predicted medicine demand based on all the scoring data.

11. An intelligent monitoring device for centralized medical procurement, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, and when the computer program is executed, the method steps in the method for intelligent supervision of centralized procurement of medicines as described in any one of claims 1 to 9 are implemented.

12. A computer storage medium, characterized in that: The computer storage medium includes: at least one instruction, which, when executed by a computer, implements the method steps in the intelligent supervision method for centralized pharmaceutical procurement as described in any one of claims 1 to 9.

Images