Engineering enterprise purchase intelligent management method, device and equipment and storage medium

Abstract

The invention discloses an engineering enterprise purchase intelligent management method and device, equipment and a storage medium. The method comprises the steps of obtaining a purchase demand text and a supplier information text of an engineering project; performing feature extraction on the purchase demand text to obtain a material demand vector; analyzing the supplier information text to obtain a supplier capability vector; inputting the material demand vector into a demand prediction model, and outputting a demand prediction tensor of a next stage; inputting a supplier capability vector and the demand prediction tensor into a cost calculation network to generate initial cost tensors for different suppliers; and taking an initial cost tensor and the supplier capability vector as input parameters, solving a constraint optimization problem taking cost minimization and supply stability maximization as objective functions, and outputting an optimal supplier allocation strategy related to procurement of the engineering project. According to the invention, cost minimization and supply stability maximization can be realized at the same time in supplier distribution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to an engineering enterprise procurement intelligent management method, device, equipment and storage medium. BACKGROUND

[0002] The current engineering enterprise procurement management method is usually implemented according to the following process: first, the project manager sorts the procurement requirements of the engineering project into a structured table (such as Excel or BOM list), and exports the historical quotations of the suppliers from the supplier management system; then, based on the procurement data and the historical quotations of the suppliers and using a machine learning model, the cost of each supplier is estimated; finally, the suppliers are sorted, and the supplier with the highest comprehensive score is selected as the procurement object.

[0003] The technical problem of the existing engineering enterprise procurement management method is that, due to the dependence on static existing procurement data and local cost optimization, it is difficult to cope with the dynamic demand changes of the engineering project, and thus it is difficult to achieve a collaborative decision that takes into account both cost and supply stability in complex and variable engineering scenarios, so it cannot simultaneously achieve cost minimization and supply stability maximization in supplier allocation. SUMMARY

[0004] The engineering enterprise procurement intelligent management method, device, equipment and storage medium provided by the embodiments of the present application can simultaneously achieve cost minimization and supply stability maximization in supplier allocation.

[0005] An embodiment of the present application provides an engineering enterprise procurement intelligent management method, comprising: obtaining a procurement requirement text and a supplier information text of an engineering project; performing feature extraction on the procurement requirement text to obtain a material requirement vector; performing analysis on the supplier information text to obtain a supplier capability vector; inputting the material requirement vector into a demand prediction model to output a demand prediction tensor of the next stage; inputting the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers; taking the initial cost tensor and the supplier capability vector as input parameters, and solving a constraint optimization problem with a cost minimization and a supply stability maximization as objective functions, and outputting an optimal supplier allocation strategy related to the procurement of the engineering project.

[0006] As an improvement of the above-mentioned solution, the feature extraction on the procurement requirement text to obtain a material requirement vector comprises the following sub-steps: Performing field-based segmentation on the procurement demand text based on an engineering material dictionary to obtain a segmented sequence with engineering field semantics; Extracting material specification entities, material quantity entities, and demand urgency entities from the segmented sequence using a named entity recognition model; Mapping the material specification entities, material quantity entities, and demand urgency entities to a unified vector space respectively, and performing vector splicing to generate a material demand vector.

[0007] As an improvement of the above-mentioned scheme, the analysis of the supplier information text to obtain a supplier capability vector includes the following sub-steps: Structurally extracting the supplier's historical delivery rate, quality pass rate, and capacity utilization rate data from the supplier information text; Comparing and calculating the extracted historical delivery rate, quality pass rate, and capacity utilization rate data with industry benchmark data to obtain relative performance indicators; Vector fusing the relative performance indicators with the supplier's certification level code to generate the supplier capability vector.

[0008] As an improvement of the above-mentioned scheme, the input of the material demand vector into the demand prediction model to output a next-stage demand prediction tensor includes the following sub-steps: Inputting the material demand vector into the demand prediction model of an encoder-decoder structure composed of a gated recurrent unit to obtain a hidden state sequence; Applying an attention mechanism to the hidden state sequence to calculate weighted time-series context features; Inputting the weighted time-series context features into a fully connected output layer to generate initial demand predictions for multiple future time slices; Stacking the initial demand predictions for multiple future time slices along the time dimension to form the next-stage demand prediction tensor.

[0009] As an improvement of the above-mentioned scheme, the input of the supplier capability vector and the demand prediction tensor into the cost calculation network to generate an initial cost tensor for different suppliers includes the following sub-steps: Expanding and replicating the supplier capability vector to align its dimensions with the demand prediction tensor, and performing tensor splicing to form a joint input tensor; Inputting the joint input tensor into the cost calculation network containing convolutional layers and fully connected layers to extract deep cost correlation features; Simultaneously outputting the initial cost tensor for different suppliers and different materials through the regression output layer of the cost calculation network.

[0010] As an improvement of the above scheme, the initial cost tensor and the supplier capability vector are taken as input parameters, and a constrained optimization problem with a cost minimization and a supply stability maximization as objective functions is solved, and an optimal supplier allocation strategy related to the procurement of the engineering project is output, including the following sub-steps: According to the initial cost tensor, a first objective function with a procurement total cost minimization as an objective is constructed; According to the historical delivery rate and the capacity utilization rate in the supplier capability vector, a second objective function with a supply stability maximization as an objective is constructed; The first objective function and the second objective function are combined into a single comprehensive objective function by weighted summation, and non-negativity constraints of supplier allocation proportions and demand complete allocation constraints are introduced to form the constrained optimization problem; The constrained optimization problem is solved using a linear programming algorithm, and the procurement allocation proportion for each supplier is output as the optimal supplier allocation strategy.

[0011] Another embodiment of the present application provides an engineering enterprise procurement intelligent management device, comprising: An acquisition module is configured to acquire procurement demand text and supplier information text of an engineering project; A feature extraction module is configured to perform feature extraction on the procurement demand text to obtain a material demand vector; An analysis module is configured to analyze the supplier information text to obtain a supplier capability vector; A prediction module is configured to input the material demand vector into a demand prediction model to output a demand prediction tensor of a next stage; A generation module is configured to input the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers; A solving module is configured to take the initial cost tensor and the supplier capability vector as input parameters, and solve a constrained optimization problem with a cost minimization and a supply stability maximization as objective functions, and output an optimal supplier allocation strategy related to the procurement of the engineering project.

[0012] As an improvement of the above scheme, the feature extraction module is specifically configured to: Perform domain word segmentation processing based on an engineering material dictionary on the procurement demand text to obtain a word segmentation sequence with engineering domain semantics; Extract material specification entities, material quantity entities, and demand urgency entities from the word segmentation sequence using a named entity recognition model; The material specification entity, the material quantity entity and the demand urgency entity are respectively mapped to a unified vector space, and vector splicing is performed to generate a material demand vector.

[0013] Another embodiment of the present application provides an engineering enterprise procurement intelligent management device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the engineering enterprise procurement intelligent management method of the above-mentioned embodiments of the application when executing the computer program.

[0014] Another embodiment of the present application provides a storage medium, the computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the engineering enterprise procurement intelligent management method of the above-mentioned embodiments of the application when the computer program runs.

[0015] Compared with the prior art, the embodiments of the present application have the following beneficial effects: First, the procurement demand text and the supplier information text of the engineering project are obtained, the material demand vector which can accurately represent the material demand is obtained by feature extraction on the procurement demand text, and the supplier capability vector which can comprehensively reflect the supplier capability is obtained by analyzing the supplier information text; then the material demand vector is input into the demand prediction model to output the demand prediction tensor of the next stage, and the demand prediction is dynamically reflected by the model prediction instead of relying on the existing static data; then the supplier capability vector and the demand prediction tensor are input into the cost calculation network to generate the initial cost tensor for different suppliers, and the cost is calculated by combining the supplier capability and the dynamic demand, avoiding the limitation of local cost estimation in the prior art; finally, the initial cost tensor and the supplier capability vector are taken as input parameters to solve the constraint optimization problem with the cost minimization and the supplier stability maximization as the objective function, and the optimal supplier allocation strategy is finally output by double objective optimization instead of single sorting selection. Therefore, the embodiments of the present application can cope with the dynamic demand change of the engineering project, realize the collaborative decision considering the cost and the supply stability, and thus realize the cost minimization and the supply stability maximization in the supplier allocation. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a flowchart of an engineering enterprise procurement intelligent management method provided by an embodiment of the present application; Figure 2 is a structural schematic diagram of an engineering enterprise procurement intelligent management device provided by an embodiment of the present application; Figure 3 is a structural schematic diagram of an engineering enterprise procurement intelligent management device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] Clearly, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present application.

[0018] Referring to Figure 1 is a flowchart of an engineering enterprise procurement intelligent management method provided by an embodiment of the present application. The engineering enterprise procurement intelligent management method comprises the following steps. S10, obtaining a procurement demand text and a supplier information text of an engineering project; S11, performing feature extraction on the procurement demand text to obtain a material demand vector; S12, performing analysis on the supplier information text to obtain a supplier capability vector; S13, inputting the material demand vector into a demand prediction model to output a demand prediction tensor of a next stage; S14, inputting the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers; S15, taking the initial cost tensor and the supplier capability vector as input parameters, and solving a constraint optimization problem with a cost minimization and a supplier stability maximization as objective functions to output an optimal supplier allocation strategy related to procurement of the engineering project.

[0019] Compared with the prior art, the embodiments of the present application have the following beneficial effects: The procurement demand text and the supplier information text of an engineering project are acquired first, the material demand vector accurately representing the material demand is obtained by feature extraction on the procurement demand text, and the supplier capability vector comprehensively reflecting the supplier capability is obtained by analyzing the supplier information text; then the material demand vector is input into a demand prediction model to output a demand prediction tensor of the next stage, the demand prediction model is used for prediction instead of relying on existing static data, and the engineering demand change is dynamically reflected; then the supplier capability vector and the demand prediction tensor are input into a cost calculation network to generate an initial cost tensor for different suppliers, the cost is calculated in combination with the supplier capability and the dynamic demand, and the limitation of local cost estimation in the prior art is avoided; finally, the initial cost tensor and the supplier capability vector are taken as input parameters, a constraint optimization problem with a target function of minimizing the cost and maximizing the supply stability is solved, and the optimal supplier allocation strategy is finally output by double-target optimization instead of single sorting selection. Therefore, the embodiment of the present application can cope with the dynamic demand change of the engineering project, realize the collaborative decision considering the cost and the supply stability, and thus realize the cost minimization and the supply stability maximization simultaneously in the supplier allocation.

[0020] In the embodiment, for the procurement demand text, the construction plan document, the material application form, the design change notice and the progress plan attachment stored in the project management system are automatically grabbed through the interface connected with the project management system, and the supplementary demand description uploaded by the project management personnel through the system (such as temporary material adjustment demand) is received; for the supplier information text, on one hand, the qualification certificate scan of the settled supplier is called from the enterprise internal supplier management system, the annual supply agreement and the past performance evaluation report, and on the other hand, the latest product quotation, the capacity proof file and the quality detection report are obtained through the supplier self-service submission platform. After acquisition, the two types of texts are preliminarily standardized, the scan type text is converted into editable text through OCR technology, the texts in different formats (Word, Excel and PDF) are uniformly converted into TXT format, the redundant format symbols and invalid annotations are removed, and finally the procurement demand text and the supplier information text to be processed subsequently are formed.

[0021] As an improvement of the above embodiment, the feature extraction on the procurement demand text to obtain the material demand vector comprises the following sub-steps: performing domain segmentation processing on the procurement demand text based on an engineering material dictionary to obtain a segmentation sequence with engineering domain semantics; extracting a material specification entity, a material quantity entity and a demand urgency entity from the segmentation sequence by using a named entity recognition model; mapping the material specification entity, the material quantity entity and the demand urgency entity to a unified vector space respectively, and performing vector splicing to generate a material demand vector.

[0022] In the embodiment, the professional terms in the engineering context are effectively recognized by performing the field-based segmentation processing on the procurement demand text based on the engineering material dictionary, so as to avoid the semantic distortion caused by general segmentation; then the material specification entity, the material quantity entity and the demand urgency entity are extracted from the segmented sequence by using the named entity recognition model, so as to realize the structured extraction of the key procurement elements in the unstructured text; then the three types of extracted entities are respectively mapped to a unified vector space and are spliced, so as to improve the dimension integrity and semantic consistency of the material demand representation; finally, the generated material demand vector directly reflects the specification attribute, the quantity and the time sensitivity of the material, so as to finally improve the accuracy of subsequent demand prediction and supplier matching. Therefore, the embodiment realizes the semantic, structured and computable representation of the engineering procurement demand text by combining the field dictionary guided segmentation, the entity recognition and the vector splicing.

[0023] In the embodiment, first, the procurement demand text is subjected to the field-based segmentation processing based on the engineering material dictionary, so as to obtain the segmented sequence with the engineering field semantics. The engineering material dictionary is pre-constructed and contains the material category terms (such as “steel bar”, “concrete” and “cable”) commonly used in the engineering construction field, the specification description terms (such as “HRB400E”, “C30” and “YJV22-4×120”) and the demand correlation terms (such as “emergency procurement” and “quarterly demand”), and the dictionary library will be updated regularly according to the newly added material types in the industry. When processing, the procurement demand text is pre-processed to remove meaningless punctuation marks and redundant spaces, and the text is uniformly converted to lowercase format; then the improved forward maximum matching segmentation algorithm is adopted, which preferentially matches the field terms in the engineering material dictionary on the basis of the general segmentation logic, so as to avoid the segmentation of “HRB400E steel bar” into “HRB400”, “E”, “steel bar” and other fragments without complete field semantics; finally, the originally continuous procurement demand text (such as “Project Q3 needs to urgently procure 50 tons of HRB400E steel bar”) is segmented into the segmented sequence of “project”, “Q3”, “need”, “urgent procurement”, “HRB400E steel bar” and “50 tons”, so as to ensure that each segmented unit carries clear engineering procurement semantics.

[0024] Next, the material specification entity, the material quantity entity and the demand urgency entity are extracted from the word sequence by using a named entity recognition model. The named entity recognition model is obtained by fine-tuning a BERT pre-training model based on the input of the above word sequence and the position encoding of each word, and the output of the entity label corresponding to each word. In the model fine-tuning stage, the engineering enterprise historical procurement demand text annotation dataset (annotation categories include "material specification", "material quantity", "demand urgency" and "others") is used for training, and the attention mechanism is optimized for the characteristics of the engineering field. The words containing specification parameters (such as "HRB400E steel bar"), the words describing the quantity (such as "50 tons") and the words describing the urgency (such as "urgent procurement") are allocated higher attention weights, and the association recognition logic of numerical units and specification terms is increased to avoid missing the unit information by identifying only "50 tons" as a quantity. When processing, the word sequence is input into the model, and the model will judge the entity category one by one, and finally extract the material specification entity "HRB400E steel bar", the material quantity entity "50 tons" and the demand urgency entity "urgent procurement" from the above example word sequence.

[0025] Finally, the material specification entity, the material quantity entity and the demand urgency entity are respectively mapped to a unified vector space and vector spliced to generate a material demand vector. In the specific mapping process, an improved vector embedding strategy is used for different entities: for the material specification entity, a Word2Vec algorithm incorporating domain semantic weights is used to map it to a vector S with a dimension of d1, where each element of S corresponds to the eigenvalue of the specification term in the engineering material semantic space; for the material quantity entity, the numerical part is first normalized (mapped to the [0, 1] interval), and then combined with the semantic embedding of the quantity unit to form a vector Q with a dimension of d2; for the demand urgency entity, a semantic intensity coefficient matrix is constructed based on the pre-set urgency level (such as "urgent", "routine" and "low priority") to map the entity to a vector U with a dimension of d3, where the element value of U is positively correlated with the urgency level. The dimension of the unified vector space is set to d1+d2+d3, and the vector splicing is realized by the following formula: V = [S; Q; U] Where V is the final generated material demand vector, ";" represents the column stacking of the vector; S is the mapping vector of the material specification entity, d1 is 256; Q is the mapping vector of the material quantity entity, d2 is 64; U is the mapping vector of the demand urgency entity, d3 is 32. By this splicing method, the characteristics of the three entities are completely integrated into a single vector, ensuring that the subsequent demand prediction model can obtain comprehensive procurement demand information based on the vector.

[0026] As an improvement of the above embodiment, the parsing of the supplier information text to obtain a supplier capability vector comprises the following sub-steps: structurally extracting historical delivery rate, quality pass rate and capacity utilization rate data of the supplier from the supplier information text; comparing and calculating the extracted historical delivery rate, quality pass rate and capacity utilization rate data with industry benchmark data to obtain a relative performance index; vector fusing the relative performance index and the authentication level code of the supplier to generate the supplier capability vector.

[0027] In this embodiment, the historical delivery rate, quality pass rate and capacity utilization rate data of the supplier are first structurally extracted from the supplier information text, which can convert the unstructured or semi-structured supplier description into quantifiable indicators; then the extracted historical delivery rate, quality pass rate and capacity utilization rate data are compared and calculated with the industry benchmark data to realize the relative evaluation of the actual performance capability of the supplier, avoiding the industry difference deviation caused by relying only on absolute values; then the obtained relative performance index is vector fused with the authentication level code of the supplier, which can improve the dimension richness and discriminability of the supplier capability representation; finally, the generated supplier capability vector reflects the relative level and quality level of the supplier in delivery, quality and capacity, which finally improves the description accuracy of the supplier stability in subsequent cost modeling and optimization distribution.

[0028] Specifically, the historical delivery rate, quality pass rate, and capacity utilization rate of suppliers are first extracted from the supplier information text in a structured manner. This supplier information text includes annual supply reports, quality inspection reports, capacity registration documents, and past cooperation and performance records submitted by suppliers. The data presentation includes both structured statements such as "historical delivery rate: 95%" and unstructured descriptions such as "100 orders received in the past 12 months, 92 delivered on time." The extraction process employs an improved domain-adaptive information extraction model. This model, based on a traditional rule engine, incorporates semantic association rules from the engineering procurement domain—for example, pre-setting the mapping relationship between keywords such as "timely delivery" and "on-time performance" and historical delivery rates, the association logic between expressions such as "sampling inspection passed" and "quality met" and quality pass rates, and the correspondence rules between terms such as "actual capacity" and "maximum production load" and capacity utilization rates. During processing, the supplier information text is first converted into a standardized text format, and redundant comments and formatting symbols are removed. Then, the text is scanned sentence by sentence by the model to identify semantic fragments related to historical delivery rate, quality pass rate, and capacity utilization rate. Data conversion is performed on unstructured descriptions (such as calculating the historical delivery rate as the ratio of "92 batches delivered on time" to "100 batches of orders"). Finally, the specific data of the three core indicators are accurately extracted from various types of text to ensure that the data of each indicator can directly reflect the supplier's performance and production capacity.

[0029] Secondly, the extracted historical delivery rate, quality pass rate, and capacity utilization rate data are compared with industry benchmark data to obtain relative performance indicators. The industry benchmark data comes from the annual supplier capability white paper published by the engineering industry association, including the industry minimum, maximum, and average historical delivery rates for different material categories, the industry standard and excellent quality pass rates, and the industry normal and saturated capacity utilization ranges. To avoid indicator bias caused by a single comparison method, this embodiment uses an improved weighted relative value algorithm, the specific formula of which is as follows: , Representing the The relative performance indicators of each metric Corresponding historical delivery rate Corresponding quality pass rate Corresponding capacity utilization rate; To extract the supplier's information from the supplier information text Actual data for each indicator; The first in the industry benchmark data The minimum value of each indicator; The first in the industry benchmark data The maximum value of each indicator; For the first The weight coefficient of the index is determined based on the core demand of "timeliness priority, quality guarantee second, and capacity adaptation auxiliary" in engineering procurement. Through the formula calculation, the original index data of different dimensions can be converted into relative performance indicators of a unified dimension (value range [0, 1]), which directly reflects the relative competitiveness of the supplier in the industry.

[0030] Finally, the relative performance indicators and the authentication level code of the supplier are vector fused to generate the supplier capability vector. The authentication level of the supplier is determined according to the engineering industry supplier qualification authentication standard and is divided into four levels of AAA, AA, A and B. The corresponding authentication level codes are represented by one-hot encoding, i.e. the AAA level code is [1, 0, 0, 0], the AA level code is [0, 1, 0, 0], the A level code is [0, 0, 1, 0], and the B level code is [0, 0, 0, 1]. The coding dimension is fixed at 4 dimensions to ensure that the features of different levels can be quantitatively compared. The vector fusion process adopts a feature balancing fusion strategy. First, the three relative performance indicators are combined into a 3-dimensional performance vector , and the authentication level code is taken as a 4-dimensional level vector (wherein - corresponding to the encoding values of AAA level to B level); and then the supplier capability vector is generated through the following fusion formula: is the final supplier capability vector, and the dimension is 7 dimensions; is the fusion weight of the performance vector, is the fusion weight of the level vector; "x" represents the element-level multiplication operation of the vector. Through this fusion method, the relative performance and qualification level features of the supplier are completely integrated into a single vector, ensuring that the subsequent cost calculation network and constraint optimization solving process can fully consider the comprehensive capability of the supplier based on the vector.

[0031] As an improvement of the above embodiment, the inputting the material demand vector into the demand prediction model to output a demand prediction tensor of the next stage includes the following sub-steps: inputting the material demand vector into the demand prediction model of the encoder-decoder structure composed of the gated recurrent unit to obtain a hidden state sequence; applying an attention mechanism to the hidden state sequence to calculate a weighted time sequence context feature; inputting the weighted time sequence context feature into a fully connected output layer to generate initial demand predictions of multiple future time slices; stacking the initial demand predictions of the multiple future time slices along the time dimension to form the demand prediction tensor of the next stage.

[0032] ​In this embodiment, the material demand vector is first input into the demand forecasting model, which is composed of an encoder-decoder structure with gated loop units. This effectively captures the dynamic evolution of engineering procurement demand over time. Then, an attention mechanism is applied to the hidden state sequence output by the model to achieve differentiated weighting of information at different historical time steps, highlighting contextual features that are more relevant to future demand. Next, the weighted temporal contextual features are input into the fully connected output layer to generate initial demand forecasts for multiple future time slices, which improves the responsiveness of the forecast results to fluctuations in material demand and changes in project schedule. Finally, the initial demand forecasts for each time slice are stacked along the time dimension to form the demand forecast tensor for the next stage, ultimately improving the coverage accuracy of multi-time period and multi-material demand in subsequent cost calculations and supplier allocation.

[0033] Specifically, in this embodiment, the material demand vector is first input into the demand prediction model, which is composed of an encoder-decoder structure of gated recurrent units (GRUs), to obtain the hidden state sequence. Both the encoder and decoder of the demand prediction model adopt a 3-layer GRU network structure, and a time-series processing step is added at the encoder input—the material demand vector is first arranged in time sequence according to the historical procurement cycle of the project (e.g., weekly, monthly) to form a material demand vector sequence of length T (each time step corresponds to a material demand vector for one cycle), ensuring that the data input to the model has a time dimension feature. Each layer of the encoder's GRU extracts temporal features from the input vector sequence. It dynamically adjusts the retention and forgetting ratio of historical information through reset and update gates. The reset gate focuses on filtering key features of the current cycle's demand, while the update gate focuses on integrating the demand correlation between historical cycles and the current cycle. The temporal features output by the encoder serve as the initial input to the decoder. The decoder gradually generates hidden states for future cycles through the GRU network, and finally outputs a sequence of hidden states containing the temporal features of each future cycle (the sequence length is consistent with the number of future time slices to be predicted). This sequence fully preserves the temporal correlation and dynamic change features of material demand.

[0034] Secondly, an attention mechanism is applied to the hidden state sequence to calculate weighted temporal context features. This embodiment employs an improved temporal attention mechanism, which, unlike conventional attention which only focuses on feature correlation, incorporates a time decay factor to highlight the influence of recent hidden states on the prediction results. The specific weight calculation formula is as follows: , For the first Attention weights for each hidden state Indicates the time index of the hidden state in the sequence ( Corresponding to the earliest predicted time slice, (corresponding to the latest predicted time slice). For the first The relevance score between each hidden state and the current output of the decoder (calculated using cosine similarity); This is the time decay coefficient (with a value range of (0,1)), used to reduce the weight ratio of early hidden states; The total length of the hidden state sequence (i.e., the number of future time slices to be predicted). This is the final weighted temporal context feature; For the first There are several hidden state vectors. This formula considers both the correlation between each hidden state and the prediction target, and strengthens the influence of recent demand characteristics through a time decay factor, making the weighted temporal context features more consistent with the dynamic changes in engineering procurement needs.

[0035] Next, the weighted temporal context features are input into the fully connected output layer to generate initial demand forecasts for multiple future time slices. The fully connected output layer comprises two hidden layers and one output layer. The first hidden layer uses the ReLU activation function to transform the D-dimensional weighted temporal context features into higher-dimensional abstract features (2D dimension). The second hidden layer further optimizes the feature representation using the ReLU activation function. The output layer uses a linear activation function to map the abstract features to predicted values ​​that match the material demand dimension. The dimensionality of the output layer is consistent with the feature dimension of the material demand vector, ensuring that each predicted value corresponds to a core indicator of material demand (such as material quantity, supply priority related to demand urgency). Simultaneously, the number of neurons in the output layer is consistent with the number of future time slices to be predicted. Finally, the output contains initial demand forecasts for N time slices (each time slice corresponds to a complete set of material demand forecast indicators), achieving simultaneous forecasting of demand for multiple future cycles.

[0036] Finally, the initial demand forecasts for the multiple future time slices are stacked along the time dimension to form the demand forecast tensor for the next stage. Specifically, the stacking process uses time slices as the first dimension, material type as the second dimension (corresponding to different categories of materials in the project), and demand forecast indicators as the third dimension (such as quantity and supply priority). The initial demand forecast for each future time slice is treated as a time slice of the tensor, and these are stacked sequentially in chronological order to form a demand forecast tensor with dimensions of (number of future time slices × number of material types × demand indicator dimension). This tensor not only contains the demand forecast results for each future period but also clearly distinguishes the forecast information for different materials and indicators through its multi-dimensional structure, providing structured demand input data for the subsequent cost calculation network and ensuring the accuracy of supply and demand matching calculations.

[0037] As an improvement of the above embodiment, the inputting the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers includes the following sub-steps: The supplier capability vector is dimensionally replicated to be aligned with the demand prediction tensor, and is tensor-spliced to form a joint input tensor; The joint input tensor is input into the cost calculation network containing convolutional layers and fully connected layers to extract deep cost correlation features; The initial cost tensor for different suppliers and different materials is simultaneously output by the regression output layer of the cost calculation network.

[0038] In this embodiment, the supplier capability vector is dimensionally replicated to be aligned with the demand prediction tensor, and is tensor-spliced, which can realize the structural alignment of the supplier capability and the multi-period and multi-material demand in a unified tensor space; then the joint input tensor formed is input into the cost calculation network containing convolutional layers and fully connected layers, which realizes the automatic modeling of the high-order interaction between the supplier capability and the dynamic demand, and captures the cost difference patterns of different suppliers under different materials and time slices; then the initial cost tensor for different suppliers and different materials is simultaneously output by the regression output layer of the cost calculation network, which improves the fine granularity and scene adaptability of cost estimation; finally, the generated initial cost tensor fully reflects the comprehensive cost distribution of each supplier under the predicted demand, which ultimately improves the accuracy and decision rationality of the cost term in subsequent multi-objective optimization.

[0039] In this embodiment, the supplier capability vector is first dimensionally replicated to align with the demand prediction tensor, and then tensor spliced to form a joint input tensor. The supplier capability vector is a 7-dimensional vector (including 3 relative performance indicators and 4 certification level encodings), while the demand prediction tensor is a three-dimensional structure with dimensions defined as (number of future time slices × number of material types × demand indicator dimension), where the demand indicator dimension is consistent with the feature dimension of the material demand vector (including material specifications, quantity, and urgency-related quantitative indicators). To achieve dimensional alignment, this embodiment adopts a dynamic dimension expansion strategy: first, add two new dimensions (time dimension and material type dimension) to the supplier capability vector, where the time dimension has the same length as the number of future time slices of the demand prediction tensor, and the material type dimension has the same length as the number of material types of the demand prediction tensor; then, through replication, the original 7-dimensional supplier capability vector is expanded along the newly added time dimension and material type dimension, so that the dimension of the expanded supplier capability tensor is (number of future time slices × number of material types × 7), which completely matches the first two dimensions of the demand prediction tensor. Subsequently, the expanded supplier capability tensor and the demand prediction tensor are spliced along the third dimension (feature dimension) to form a joint input tensor with dimensions (number of future time slices × number of material types × (7 + demand indicator dimension)), which contains both supplier capability features and demand prediction features, providing complete input for subsequent cost association analysis.

[0040] Secondly, the joint input tensor is input into the cost calculation network containing convolutional layers and fully connected layers to extract deep cost association features. The convolutional layer part of the cost calculation network contains 2 three-dimensional convolutional layers, the first convolutional layer uses multi-scale convolutional kernels (sizes are 3×3×3, 3×3×5), where the first two dimensions correspond to time and material type, and the third dimension corresponds to the feature dimension, capturing time-material-feature associations of different ranges (such as cost association of similar materials in short-term time, cross-influence of different materials in medium-term time) through multi-scale convolution. After convolution, the features are standardized by the Batch Normalization layer and more feature details are preserved using the Leaky ReLU activation function. The second convolutional layer uses a 1×1×1 convolutional kernel to compress the dimensions and fuse the channels of the output features of the first layer, strengthening key cost association features (such as the cost sensitivity association of historical delivery rate in supplier capability and demand urgency). The feature tensor output by the convolutional layer is then flattened into a one-dimensional feature vector and input into the fully connected layer: the first fully connected layer maps the flattened features to a higher-dimensional space (dimension is twice the dimension of the convolutional layer output feature), further mining non-linear cost associations; the second fully connected layer prevents overfitting through the dropout mechanism (dropout rate is 0.3), and outputs deep cost association features with dimensions matching the target cost feature dimension.

[0041] Finally, the initial cost tensor for different suppliers and different materials is outputted through the regression output layer of the cost calculation network. The regression output layer adopts an improved multi-dimensional regression design, which takes the deep cost correlation features outputted by the fully connected layer as input, and the number of neurons in the output layer is set as “supplier number x future time slice number x material type number”, and each neuron corresponds to the cost prediction value of a specific supplier, a specific time slice and a specific material. To ensure the structure of the output tensor, a dimension constraint matrix is introduced in the calculation of the output layer, and multi-dimensional cost mapping is realized through the following formula: , is the initial cost tensor, with dimensions of (supplier number x future time slice number x material type number); is the weight matrix of the output layer, with dimensions matching the deep cost correlation features and the output dimensions; is the deep cost correlation feature vector outputted by the fully connected layer; is the bias vector; is the dimension constraint matrix (element value is 1 or 0), which is used to shield unreasonable cost correlation (such as setting the position element corresponding to the material that the supplier cannot supply to 0); “x” represents matrix multiplication, and “⊙” represents element-level multiplication. Through the formula, the regression output layer can output the cost data of multiple suppliers at different time slices for different materials, and ensure the rationality of cost prediction through the dimension constraint matrix. The final generated initial cost tensor completely contains the cost characteristics of each supplier, providing accurate cost input for subsequent constraint optimization solution.

[0042] As an improvement of the above embodiment, the initial cost tensor and the supplier capability vector are taken as input parameters, and a constraint optimization problem with the objective function of minimizing cost and maximizing supply stability is solved, and an optimal supplier allocation strategy related to the procurement of the engineering project is outputted, including the following sub-steps: A first objective function with the objective of minimizing total procurement cost is constructed according to the initial cost tensor; A second objective function with the objective of maximizing supply stability is constructed according to the historical delivery rate and capacity utilization rate in the supplier capability vector; The first objective function and the second objective function are combined into a single comprehensive objective function through weighted summation, and non-negativity constraints of supplier allocation proportion and demand complete allocation constraints are introduced, forming the constraint optimization problem; The constraint optimization problem is solved using a linear programming algorithm, and the procurement allocation proportion for each supplier is outputted as the optimal supplier allocation strategy.

[0043] In this embodiment, a first objective function is first constructed based on the initial cost tensor, aiming to minimize the total procurement cost. This transforms cost data from multiple suppliers, materials, and time periods into an optimizable mathematical expression. Next, a second objective function is constructed based on the historical delivery rate and capacity utilization rate in the supplier capability vector, aiming to maximize supply stability, thus achieving quantitative modeling of supplier performance reliability. Then, the first and second objective functions are combined into a single comprehensive objective function through a weighted summation. Introducing non-negativity constraints on supplier allocation ratios and complete demand allocation constraints improves the feasibility and constraint completeness of the optimization model in real-world procurement scenarios. Finally, a linear programming algorithm is used to solve this constrained optimization problem, outputting the procurement allocation ratio for each supplier as the optimal supplier allocation strategy, ultimately improving the synergy between cost control and supply chain stability in procurement decisions. Therefore, this embodiment, by constructing a dual-objective fusion constrained optimization model and using linear programming for solution, achieves a joint optimal allocation of cost and supply stability while meeting engineering requirements.

[0044] Firstly, a first objective function is constructed based on the initial cost tensor, aiming to minimize the total procurement cost. The initial cost tensor is a three-dimensional structure, with its dimensions defined as (number of suppliers S × number of future time slices T × number of material types M), where each element... Representing the The supplier in the first The time slice is the first The unit cost of supplying this type of material. When constructing the first objective function, it is necessary to consider the supplier allocation ratio variable to be solved—let... For the first The supplier in the first The time slice for the first The procurement allocation ratio of a certain type of material (i.e., the proportion of that material's demand that the supplier undertakes for that time slice), and at the same time, the first... The first time slice Total demand for this type of material (Extracted from demand forecast tensor). To minimize total procurement cost, the first objective function is designed as the minimum sum of "unit cost × allocation ratio × total demand" for all suppliers, all time slices, and all materials, as shown in the following formula: , Let the first objective function be (total procurement cost). , , These are indexes for supplier, time slice, and material type, respectively. Taken directly from the initial cost tensor. The forecasting dimension related to "material quantity" is taken from the demand forecasting tensor. For the variables to be optimized, the formula fully covers the procurement cost of the whole project cycle and all materials through triple summation, ensuring that there is no omission in cost calculation.

[0045] Secondly, a second objective function is constructed to maximize supply stability according to the historical delivery rate and capacity utilization rate in the supplier capability vector. The supplier capability vector is a 7-dimensional vector generated by claim 3, where the first dimension is the historical delivery rate of the first supplier (i.e., the proportion of past orders delivered on time), the third dimension is the capacity utilization rate of the first supplier (i.e., the ratio of actual capacity to maximum capacity, the closer the value to 1, the more saturated the capacity and the lower the supply flexibility). To quantify supply stability, both indicators need to be adapted: the historical delivery rate is directly positively correlated with stability (the higher the value, the stronger the stability), and the capacity utilization rate is inversely correlated with stability (the value is too high, the weight needs to be reduced to avoid relying on suppliers with saturated capacity). Therefore, the second objective function is designed as "the maximum value of the weighted sum of historical delivery rate and the modified value of capacity utilization rate", the specific formula is as follows: , is the second objective function (supply stability), is the weight coefficient of historical delivery rate, is the weight coefficient of the capacity utilization rate correction term; is the reasonable capacity utilization rate threshold in the industry (taken from industry benchmark data, set to 0.8), is the difference of capacity utilization rate from the reasonable threshold (if , the difference is 0, to avoid excessive punishment of suppliers with unsaturated capacity); The allocation proportion variable of the first objective function is consistent, ensuring that the stability evaluation is directly linked to the actual allocation strategy, avoiding "empty stability" calculation that deviates from the allocation proportion.

[0046] Then, the first objective function and the second objective function are combined into a single comprehensive objective function through weighted summation, and non-negativity constraints and demand complete allocation constraints of supplier allocation proportion are introduced to form the constraint optimization problem. When combining the comprehensive objective function, weight coefficients (cost priority weight) and (stability priority weight) are introduced, the specific formula is as follows: , is the comprehensive objective function (through the combination logic of "maximizing stability contribution and minimizing cost expenditure", the double objective optimization is converted into single objective optimization); and The value of can be dynamically adjusted according to the engineering project demand. In terms of the constraint condition, firstly, the non-negativity constraint is introduced: (all the allocation proportions of the suppliers, time slices and materials cannot be negative, which conforms to the actual procurement logic); and secondly, the demand complete allocation constraint is introduced: (the sum of the allocation proportions of each time slice and each type of material is 1, which ensures that the demand of the material in the time slice is completely covered without demand omission). Through the "one comprehensive objective function + two constraint conditions", the constraint optimization problem is completely constituted, and the optimization direction is clear and conforms to the actual demand of the engineering procurement.

[0047] Finally, the linear programming algorithm is used to solve the constraint optimization problem, and the procurement allocation proportion of each supplier is output as the optimal supplier allocation strategy. When solving, the dimension information (the specific number of S, T and M) of the comprehensive objective function , the constraint condition and the variable is input into the linear programming solver (such as the simplex method solver), and the solver will automatically traverse the feasible region of the variable to find the value set of that satisfies all the constraint conditions and makes maximum. After the solving is completed, the output result is a three-dimensional allocation proportion matrix (the dimension is consistent with ), which clearly shows the specific allocation proportion of each supplier to each type of material in each time slice (for example, "the allocation proportion of supplier A to the steel bar in the second time slice is 0.6"). The allocation proportion matrix is associated and matched with the supplier name, material type and time slice information to form a structured optimal supplier allocation strategy document, and the supply range, allocation proportion, corresponding cost and stability contribution value of each supplier are marked in the document to ensure that the procurement personnel can directly carry out procurement work according to the strategy. After obtaining the optimal supplier allocation strategy, firstly, the feasibility is checked in combination with the real-time demand of the engineering project and the current production capacity state of the supplier, and it is confirmed that there is no problem such as over-production capacity allocation of the supplier and supply fault of the key material; after the checking is passed, the strategy is converted into a structured procurement execution document, which clearly shows the material supply category, quantity, delivery time and cost details of each supplier; then, the strategy is synchronized to the enterprise procurement management system and the supplier collaboration platform, the procurement invitation is pushed to the corresponding supplier, the internal procurement department and the project construction department are synchronized with the allocation scheme to connect the subsequent stock preparation and construction plan; meanwhile, a strategy execution monitoring mechanism is established to track the supplier performance in real time, and if there is a delay in delivery, quality fluctuation or the like, the reserved alternative allocation scheme is dynamically adjusted to ensure the stable progress of the engineering project procurement link.

[0048] Referring to Figure 2 , it is a structural schematic diagram of an engineering enterprise procurement intelligent management device provided by an embodiment of the present application. The engineering enterprise procurement intelligent management device comprises: The acquisition module 10 is configured to acquire a procurement demand text and a supplier information text of an engineering project; The feature extraction module 11 is configured to perform feature extraction on the procurement demand text to obtain a material demand vector; The analysis module 12 is configured to analyze the supplier information text to obtain a supplier capability vector; The prediction module 13 is configured to input the material demand vector into a demand prediction model to output a demand prediction tensor of a next stage; The generation module 14 is configured to input the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers; The solving module 15 is configured to input the initial cost tensor and the supplier capability vector as input parameters, and solve a constraint optimization problem with a cost minimization and a supplier stability maximization as objective functions, to output an optimal supplier allocation strategy related to the procurement of the engineering project.

[0049] Compared with the prior art, the embodiment has the following beneficial effects: The procurement demand text and the supplier information text of the engineering project are acquired first, the material demand vector accurately representing the material demand is obtained by performing feature extraction on the procurement demand text, and the supplier capability vector comprehensively reflecting the supplier capability is obtained by analyzing the supplier information text; then the material demand vector is input into the demand prediction model to output the demand prediction tensor of the next stage, the demand prediction is performed by the model instead of relying on the existing static data, and the engineering demand change is dynamically reflected; subsequently, the supplier capability vector and the demand prediction tensor are input into the cost calculation network to generate the initial cost tensor for different suppliers, the cost is calculated in combination with the supplier capability and the dynamic demand, and the limitation of the local cost estimation in the prior art is avoided; finally, the initial cost tensor and the supplier capability vector are input as input parameters, the constraint optimization problem with the cost minimization and the supplier stability maximization as objective functions is solved, and the optimal supplier allocation strategy is finally output by the double-objective optimization instead of the single sorting selection. Therefore, the embodiment can cope with the dynamic demand change of the engineering project, realize the collaborative decision-making considering the cost and the supplier stability, and thus realize the cost minimization and the supplier stability maximization simultaneously in the supplier allocation.

[0050] As an improvement of the above embodiment, the feature extraction module is specifically configured to: perform field segmentation processing based on an engineering material dictionary on the procurement demand text to obtain a segmentation sequence with engineering field semantics; extract a material specification entity, a material quantity entity and a demand urgency entity from the segmentation sequence by using a named entity recognition model; The material specification entity, the material quantity entity and the demand urgency entity are mapped to a unified vector space respectively, and vector splicing is performed to generate a material demand vector.

[0051] It can be understood that the embodiment of the engineering enterprise procurement intelligent management device can correspond to the above-mentioned engineering enterprise procurement intelligent management method embodiments, and details are not repeated here.

[0052] Referring to Figure 3 is a schematic diagram of an engineering enterprise procurement intelligent management device provided by an embodiment of the present application. The engineering enterprise procurement intelligent management device of this embodiment comprises a processor 100, a memory 101, and a computer program, such as an engineering enterprise procurement intelligent management program, stored in the memory 101 and executable on the processor 100. The processor 100 implements the steps in each of the above-mentioned method embodiments when executing the computer program. Alternatively, the processor 100 implements the functions of each module / unit in each of the above-mentioned device embodiments when executing the computer program.

[0053] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the engineering enterprise procurement intelligent management device.

[0054] The engineering enterprise procurement intelligent management device can be a desktop computer, a notebook computer, a palm computer, a cloud server, and other computing devices. The engineering enterprise procurement intelligent management device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the schematic diagram is only an example of the engineering enterprise procurement intelligent management device and does not constitute a limitation on the engineering enterprise procurement intelligent management device, and can include more or fewer components than the diagram, or combine certain components, or different components, for example, the engineering enterprise procurement intelligent management device can also include an input / output device, a network access device, a bus, etc.

[0055] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or can also be any conventional processor. The processor is a control center of the engineering enterprise procurement intelligent management device, and connects various parts of the engineering enterprise procurement intelligent management device through various interfaces and lines.

[0056] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the engineering enterprise procurement intelligent management device by running or executing the computer program and / or modules 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. The program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), 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 media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0057] The modules / units of the engineering enterprise procurement intelligent management device integration, if implemented in the form of software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When the processor executes the computer program, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer-readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code. It should be noted that the contents included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0058] It should be noted that the above-described device embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0059] The above is the preferred embodiment of the present application. It should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which are also considered within the scope of protection of the present application.

Claims

1. An intelligent management method for engineering enterprise procurement, characterized in that, The method comprises the following steps: obtaining a procurement demand text and a supplier information text of an engineering project; performing feature extraction on the procurement demand text to obtain a material demand vector; analyzing the supplier information text to obtain a supplier capability vector; inputting the material demand vector into a demand prediction model to output a demand prediction tensor of a next stage; inputting the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers; taking the initial cost tensor and the supplier capability vector as input parameters, and solving a constraint optimization problem with a target function of minimizing cost and maximizing supply stability to output an optimal supplier allocation strategy related to the procurement of the engineering project.

2. The intelligent management method for engineering enterprise procurement of claim 1, wherein, The feature extraction on the procurement demand text to obtain a material demand vector comprises the following sub-steps: performing domain-based word segmentation on the procurement demand text based on an engineering material dictionary to obtain a word segmentation sequence with engineering domain semantics; extracting material specification entities, material quantity entities and demand urgency entities from the word segmentation sequence using a named entity recognition model; mapping the material specification entities, material quantity entities and demand urgency entities to a unified vector space respectively, and performing vector splicing to generate a material demand vector.

3. The intelligent management method for engineering enterprise procurement of claim 2, wherein, The analysis of the supplier information text to obtain a supplier capability vector comprises the following sub-steps: structurally extracting historical delivery rate, quality pass rate and capacity utilization rate data of the supplier from the supplier information text; comparing and calculating the extracted historical delivery rate, quality pass rate and capacity utilization rate data with industry benchmark data to obtain relative performance indicators; vector fusing the relative performance indicators and the authentication level code of the supplier to generate the supplier capability vector.

4. The intelligent management method for engineering enterprise procurement of claim 3, wherein, The inputting of the material demand vector into a demand prediction model to output a demand prediction tensor of a next stage comprises the following sub-steps: inputting the material demand vector into the demand prediction model with an encoder-decoder structure composed of a gating cycle unit to obtain a hidden state sequence; applying an attention mechanism to the hidden state sequence to calculate weighted time-series context features; inputting the weighted time-series context features into a fully connected output layer to generate initial demand predictions for multiple future time slices; stacking the initial demand predictions for the multiple future time slices along the time dimension to form the demand prediction tensor of the next stage.

5. The intelligent management method for engineering enterprise procurement of claim 4, wherein, The inputting of the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers comprises the following sub-steps: dimensionally expanding and replicating the supplier capability vector to align its dimensions with the demand prediction tensor, and performing tensor splicing to form a joint input tensor; inputting the joint input tensor into the cost calculation network containing convolutional layers and fully connected layers to extract deep cost correlation features; simultaneously outputting the initial cost tensor for different suppliers and different materials through the regression output layer of the cost calculation network.

6. The intelligent management method for engineering enterprise procurement of claim 5, wherein, The initial cost tensor and the supplier capability vector are taken as input parameters, and a constrained optimization problem with a cost minimization and a supply stability maximization as objective functions is solved, and an optimal supplier allocation strategy related to procurement of the engineering project is output, including the following sub-steps: A first objective function with a procurement total cost minimization as a target is constructed according to the initial cost tensor; A second objective function with a supply stability maximization as a target is constructed according to historical delivery rates and capacity utilization rates in the supplier capability vector; The first objective function and the second objective function are combined into a single comprehensive objective function through weighted summation, and non-negativity constraints of supplier allocation proportions and demand complete allocation constraints are introduced to form the constrained optimization problem; A linear programming algorithm is used to solve the constrained optimization problem, and procurement allocation proportions for each supplier are output as the optimal supplier allocation strategy.

7. An intelligent management device for engineering enterprise procurement, characterized in that, It comprises: An acquisition module for acquiring procurement demand text and supplier information text of an engineering project; A feature extraction module for feature extraction on the procurement demand text to obtain a material demand vector; An analysis module for analyzing the supplier information text to obtain a supplier capability vector; A prediction module for inputting the material demand vector into a demand prediction model to output a next-stage demand prediction tensor; A generation module for inputting the supplier capability vector and the demand prediction tensor into a cost calculation network to generate an initial cost tensor for different suppliers; A solving module for taking the initial cost tensor and the supplier capability vector as input parameters, and solving a constrained optimization problem with a cost minimization and a supply stability maximization as objective functions, and outputting an optimal supplier allocation strategy related to procurement of the engineering project.

8. The intelligent procurement management apparatus for an engineering enterprise according to claim 7, wherein The feature extraction module is specifically configured to: Perform domain word segmentation processing based on an engineering material dictionary on the procurement demand text to obtain a word segmentation sequence with engineering field semantics; Extract material specification entities, material quantity entities, and demand urgency entities from the word segmentation sequence using a named entity recognition model; Map the material specification entities, material quantity entities, and demand urgency entities to a unified vector space respectively, and perform vector splicing to generate a material demand vector.

9. An intelligent management device for engineering enterprise procurement, characterized by, The computer readable storage medium comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the engineering enterprise procurement intelligent management method according to any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the engineering enterprise procurement intelligent management method according to any one of claims 1 to 6 when the computer program runs.

Images