Enterprise short-term load prediction method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of load prediction, and discloses an enterprise short-term load prediction method and system based on artificial intelligence. The method comprises the following steps: carrying out key generation and identity authentication based on a trusted mechanism; based on the enterprise server, compressing, packaging, encrypting and signing the real-time load data and the real-time external data, and uploading the real-time load data and the real-time external data to the cloud data center; performing initialization based on a cloud data center; signature verification, decryption and data analysis are carried out; carrying out data dimension reduction; on the basis of the cloud data center, index weight generation is carried out, and the enterprise short-term load prediction model is adjusted; carrying out enterprise short-term load prediction and enterprise supply chain optimization; and encrypting the real-time enterprise supply chain optimization strategy, and sending the encrypted real-time enterprise supply chain optimization strategy to a corresponding enterprise server. According to the method, the problems of low safety, low prediction precision, poor real-time performance, poor adaptability, poor comprehensiveness and lack of countermeasures in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of load forecasting, and particularly relates to an enterprise short-term load forecasting method and system based on artificial intelligence. Background Art

[0002] In today's enterprise operations, accurate short-term load forecasting is crucial for resource allocation, cost control, and production efficiency improvement. With the advancement of Industry 4.0 and the development of intelligent manufacturing, enterprises have higher and higher requirements for the accuracy and real-time performance of load forecasting. However, the existing enterprise short-term load forecasting technologies have the following limitations:

[0003] 1) Insufficient data security and privacy protection: Existing load forecasting systems often lack effective data encryption and identity authentication mechanisms, resulting in the risk of leakage of enterprise key data during transmission; insufficient protection of enterprise sensitive information may increase the operation risk of enterprises due to data security issues.

[0004] 2) Low accuracy of prediction models: Traditional prediction models mostly rely on linear assumptions and are difficult to accurately capture the non-linear relationships and complex dynamics in load data; the models do not sufficiently consider external influencing factors, resulting in a large deviation between the prediction results and the actual load.

[0005] 3) Poor real-time performance: When existing technologies perform load forecasting, they often require a long time for data analysis and model operation, making it difficult to meet the requirements of real-time forecasting; for emergencies and rapidly changing load demands, the system responds sluggishly and cannot adjust supply chain decisions in a timely manner.

[0006] 4) Poor model adaptability: Existing load forecasting models are often designed for specific scenarios and lack adaptability to the characteristics of different industries and enterprises; it is difficult to update and maintain the models, making it difficult to keep up with the changes in enterprise production and operations.

[0007] 5) Poor comprehensiveness: Existing technologies often ignore the influence of external data (such as weather, market trends, etc.) on load forecasting, resulting in incomplete prediction results.

[0008] 6) Lack of countermeasures: Existing load forecasting technologies often only provide prediction results and lack enterprise supply chain optimization strategies for the prediction results, limiting their value in practical applications. Summary of the Invention

[0009] In order to solve the problems of low security, low prediction accuracy, poor real-time performance, poor adaptability, poor comprehensiveness, and lack of countermeasures existing in the prior art, the purpose of the present invention is to provide an enterprise short-term load forecasting method and system based on artificial intelligence.

[0010] The technical solution adopted by the present invention is as follows:

[0011] An enterprise short-term load forecasting method based on artificial intelligence, comprising the following steps:

[0012] Based on a trusted institution, generate keys and authenticate the identities of a number of enterprise servers to obtain the public-private key pairs and signature information of each enterprise server. Return the private keys in the public-private key pairs and the signature information to the corresponding enterprise servers, and send the public keys in the public-private key pairs to the cloud data center;

[0013] Based on the enterprise servers, compress and package the real-time load data and real-time external data to obtain real-time upload data packets. Encrypt and sign the real-time upload data packets according to the private keys and signature information to obtain encrypted real-time upload data packets and real-time signature data, and upload them to the cloud data center;

[0014] Based on the cloud data center, perform initialization. According to a number of historical load data and corresponding historical external data, use artificial intelligence algorithms to construct an index weight generation model, an enterprise short-term load forecasting model, and an enterprise supply chain optimization model;

[0015] Based on the cloud data center, call the trusted institution to verify the signature of the real-time signature data. After the signature verification passes, decrypt the encrypted real-time upload data packets according to the public keys of the corresponding enterprise servers to obtain decrypted real-time upload data packets, and perform data parsing on the decrypted real-time upload data packets to obtain parsed real-time load data and parsed real-time external data;

[0016] Based on the cloud data center, perform data dimensionality reduction on the parsed real-time load data and parsed real-time external data to obtain reduced-dimensional real-time load data and corresponding real-time key load indicators, as well as reduced-dimensional real-time external data and corresponding real-time key external indicators;

[0017] Based on the cloud data center, according to the real-time key load indicators and real-time key external indicators, use the index weight generation model to generate index weights to obtain real-time load index weights and real-time external index weights, and adjust the enterprise short-term load forecasting model according to the real-time load index weights and real-time external index weights to obtain an adjusted enterprise short-term load forecasting model;

[0018] Based on the cloud data center, according to the reduced-dimensional real-time load data and reduced-dimensional real-time external data, use the adjusted enterprise short-term load forecasting model to perform enterprise short-term load forecasting to obtain real-time enterprise short-term load forecasting results, and use the enterprise supply chain optimization model according to the real-time enterprise short-term load forecasting results to optimize the enterprise supply chain to obtain real-time enterprise supply chain optimization strategies;

[0019] Based on the cloud data center, encrypt the real-time enterprise supply chain optimization strategy according to the public key of the corresponding enterprise server to obtain the encrypted real-time enterprise supply chain optimization strategy, and send the encrypted real-time enterprise supply chain optimization strategy to the corresponding enterprise server.

[0020] Furthermore, the real-time load data includes real-time order volume, real-time purchase volume, real-time production volume, real-time inventory volume, real-time distribution volume, and real-time sales volume;

[0021] The real-time external data includes real-time weather data, real-time traffic data, real-time raw material prices, real-time promotion data, real-time price fluctuation data, and real-time competitor data.

[0022] Furthermore, use the PCA method to reduce the dimension of the parsed real-time load data and the parsed real-time external data.

[0023] Furthermore, based on the cloud data center, perform initialization. According to a number of historical load data and the corresponding historical external data, use artificial intelligence algorithms to construct an index weight generation model, an enterprise short-term load prediction model, and an enterprise supply chain optimization model, including the following steps:

[0024] Based on the cloud data center, collect a number of historical load data and the corresponding historical external data, and preprocess the number of historical load data and the corresponding historical external data respectively to obtain a number of preprocessed historical load data and the corresponding preprocessed historical external data;

[0025] According to a number of preprocessed historical load data and the corresponding preprocessed historical external data, use the swarm intelligence optimization algorithm to construct an index weight generation model, and generate a number of historical load index weights and the corresponding historical external index weights;

[0026] According to a number of historical load data and the corresponding historical load index weights, and historical external data and the corresponding historical external index weights, use the deep learning algorithm to construct an enterprise short-term load prediction model, and generate a number of historical enterprise short-term load prediction results;

[0027] According to a number of historical enterprise short-term load prediction results, use the reinforcement learning algorithm to construct an enterprise supply chain optimization model, and store the generated number of historical enterprise supply chain optimization experiences in the experience replay pool of the enterprise supply chain optimization model.

[0028] Furthermore, the index weight generation model is constructed based on the IWOA algorithm.

[0029] Furthermore, the enterprise short-term load prediction model is constructed based on the Attention-LSTM-Elman algorithm.

[0030] Furthermore, the enterprise supply chain optimization model is constructed based on the DQN algorithm.

[0031] Furthermore, the enterprise short-term load forecasting model includes an input layer constructed based on the LSTM algorithm, several hidden layers, an attention layer constructed based on the Attention mechanism, and an output layer constructed based on the Elman algorithm. The input layer, the attention layer, several hidden layers, and the output layer are connected in sequence.

[0032] Furthermore, according to the real-time key load indicators and real-time key external indicators, use the indicator weight generation model to generate indicator weights, obtain the real-time load indicator weights and real-time external indicator weights, and adjust the enterprise short-term load forecasting model according to the real-time load indicator weights and real-time external indicator weights to obtain the adjusted enterprise short-term load forecasting model, including the following steps:

[0033] Take the indicator weight values corresponding to the real-time key load indicators and real-time key external indicators as the optimization target, use the indicator weight generation model to generate indicator weights, and obtain the real-time load indicator weights and real-time external indicator weights;

[0034] According to the real-time key load indicators and real-time key external indicators, adjust the number of neurons in the input layer of the enterprise short-term load forecasting model and the neuron weights between the input layer and the attention layer to obtain the adjusted input layer;

[0035] According to the real-time load indicator weights and real-time external indicator weights, adjust the attention weights of the attention layer to obtain the adjusted attention layer and adjusted attention weights, and obtain the adjusted enterprise short-term load forecasting model according to the adjusted input layer, adjusted attention layer, and adjusted attention weights.

[0036] An enterprise short-term load forecasting system based on artificial intelligence is used to implement the enterprise short-term load forecasting method. The system includes a cloud data center, a trusted institution, and several enterprise servers. The several enterprise servers are respectively communicatively connected to the cloud data center and the trusted institution. The cloud data center is communicatively connected to the trusted institution, and the cloud data center includes an initialization unit, a data reception unit, a data dimensionality reduction unit, a weight generation and model adjustment unit, a short-term load forecasting and supply chain optimization unit, and a data encryption and transmission unit.

[0037] The beneficial effects of the present invention are:

[0038] The present invention discloses an artificial intelligence-based short-term load forecasting method and system for enterprises, and proposes a data security scheme combining key generation and identity authentication, ensuring the security of data transmission between enterprise servers and cloud data centers, effectively preventing data leakage and unauthorized access, and significantly enhancing data security; by using artificial intelligence algorithms to construct an index weight generation model, an enterprise short-term load forecasting model, and an enterprise supply chain optimization model, the enterprise short-term load forecasting model can more accurately capture complex relationships and non-linear characteristics in the data, thereby improving the accuracy of forecasting; the index weight generation model can adaptively adjust the model parameters of the enterprise short-term load forecasting model to adapt to the changing data structure and improve the adaptability to the characteristics of different industries and enterprises; the cloud data center realizes the decryption verification and dimensionality reduction processing of real-time data, ensuring the real-time nature of forecasting. This real-time data processing ability enables enterprises to quickly respond to load changes and timely adjust supply chain decisions. For emergencies and rapidly changing load demands, the system responds quickly; combined with the enterprise supply chain optimization model, it provides a complete set of optimization strategies for enterprises. This integrated optimization strategy can help enterprises better cope with forecasting results and achieve the optimal allocation of resources; considering the impact of external data on load forecasting, it improves the comprehensiveness of forecasting and affects the formulation of enterprise supply chain optimization strategies.

[0039] Other beneficial effects of the present invention will be further described in the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flowchart of the artificial intelligence-based short-term load forecasting method for enterprises in the present invention.

[0041] Figure 2 is a block diagram of the structure of the artificial intelligence-based short-term load forecasting system for enterprises in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0042] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0043] Embodiment 1:

[0044] As Figure 1 shown, this embodiment provides an artificial intelligence-based short-term load forecasting method for enterprises, including the following steps:

[0045] S1: Based on a trusted institution, perform key generation and identity authentication on a number of enterprise servers to obtain the public-private key pairs and signature information of each enterprise server, return the private key in the public-private key pair and the signature information to the corresponding enterprise server, and send the public key in the public-private key pair to the cloud data center, including the following steps:

[0046] S1-1: Based on a trusted institution, perform key initialization to obtain public parameters, a master key, and an initial key;

[0047]

[0048] where GP are the public parameters; MSK is the master key; PK is the initial key; a is a random number in the integer domain Z p ; H 1 , H 2 , H 3 , H 4 , H 5 , H 6 , H u are all target hash functions; g, g 1 , g a are all random numbers of the generators of the cyclic group G; e(g, g) a is the bilinear mapping of the random number g;

[0049] S1-2: Collect the attribute information V u of the enterprise server and the entity ID, and use the asymmetric encryption algorithm to generate keys for the enterprise server based on the attribute information, public parameters, master key, and initial key, to obtain the corresponding public-private key pair;

[0050] SK u = {MSK, V u , K = g a g ab , L u = g b , (K u = H 3 (V u )) b )}

[0051]

[0052] where SK u is the private key of the enterprise server u; b is a random number in the integer domain Z p ; L u , K u are the private key parameters of the enterprise server u; H 3 is the target hash function of the public parameter GP; u is the enterprise server indicator; MSK is the master key; PK is the initial key; PK u is the public key of the enterprise server u; g b , g a , g ab are random numbers of the generators of the cyclic group G; V u is the attribute information of the enterprise server u;

[0053] S1-3: According to the public-private key pair and the corresponding entity ID, use the digital identity authentication method to perform identity registration and obtain the signature information of the corresponding enterprise server;

[0054] The formula is:

[0055]

[0056] In the formula, k' is a random number; K u is the registration parameter of enterprise server u; KID u is the registration ID of enterprise server u; KID u and the corresponding K u constitute the signature information {K u , KID u}; H 1 is the target hash function; ID u is the entity ID of enterprise server u; is the prime order; P is the prime field base point;

[0057] S2: Based on the enterprise server, compress and package the real-time load data and real-time external data to obtain the real-time upload data packet. According to the private key and the signature information, encrypt and sign the real-time upload data packet to obtain the encrypted real-time upload data packet and the real-time signature data, and upload them to the cloud data center;

[0058] The real-time load data includes real-time order volume, real-time purchase volume, real-time production volume, real-time inventory volume, real-time distribution volume, and real-time sales volume;

[0059] The real-time external data includes real-time weather data, real-time traffic data, real-time raw material prices, real-time promotion data, real-time price fluctuation data, and real-time competitor data;

[0060] The formula is:

[0061] M u =E(SK u , m u )

[0062] In the formula, M u is the encrypted real-time upload data packet of enterprise server u; E(*) is the asymmetric encryption function; m u is the real-time upload data packet of enterprise server u; SK u is the private key of enterprise server u; u is the enterprise server indicator;

[0063] The formula is:

[0064]

[0065] where r' is a random number; is a prime order; P is a prime field base point; H 2 is a target hash function; K u is the registration parameter of enterprise server u in the signature information {K u , KID u}; KID u is the registration ID of enterprise server u in the signature information {K u , KID u}; ID u is the entity ID of enterprise server u; the real-time signature data formed is {ID u , M u , γ' = {K u , R u , B u}}; R u , B u , γ' are all signature parameters of enterprise server u;

[0066] S3: Based on the cloud data center, perform initialization. According to a number of historical load data and corresponding historical external data, use artificial intelligence algorithms to construct an index weight generation model, an enterprise short-term load forecasting model, and an enterprise supply chain optimization model, including the following steps:

[0067] S3-1: Based on the cloud data center, collect a number of historical load data and corresponding historical external data, and preprocess the number of historical load data and corresponding historical external data respectively to obtain a number of preprocessed historical load data and corresponding preprocessed historical external data;

[0068] S3-2: According to a number of preprocessed historical load data and corresponding preprocessed historical external data, use the improved whale optimization algorithm (Improved Whale Optimization Algorithm, IWOA) of the swarm intelligence optimization algorithm to construct an index weight generation model, and generate a number of historical load index weights and corresponding historical external index weights;

[0069] S3-3: According to a number of historical load data and corresponding historical load index weights, as well as historical external data and corresponding historical external index weights, use the Attention-Long Short-Term Memory (LSTM)-Elman algorithm of the deep learning algorithm to construct an enterprise short-term load forecasting model, and generate a number of historical enterprise short-term load forecasting results;

[0070] The enterprise short-term load forecasting model includes an input layer constructed based on the LSTM algorithm, several hidden layers, an attention layer constructed based on the Attention mechanism, and an output layer constructed based on the Elman algorithm. The input layer, attention layer, several hidden layers, and output layer are connected in sequence;

[0071] The selection of the enterprise short-term load forecasting model depends on various factors, including the characteristics of the data, the accuracy requirements of the prediction, computing resources, and the actual application scenario, etc. The LSTM network is a special recurrent neural network. Its special cell state design makes it perform excellently in processing and predicting time-series data, capable of capturing long-term dependencies, suitable for data with time correlations, and showing high accuracy in many time-series prediction tasks. Therefore, choosing the LSTM network as the network framework of the enterprise short-term load forecasting model improves the accuracy and efficiency of the enterprise short-term load forecasting. The Attention mechanism sets corresponding attention weights for each input data feature quantity, improving the ability to focus on the enterprise short-term load situation and further enhancing the prediction accuracy. The structure of the Elman network is simple and can train its own neuron structure according to the preset label encoding format and data, acting as a stable classifier and improving the model training efficiency;

[0072] S3-4: According to several historical enterprise short-term load forecasting results, use the Deep Q Network (DQN) algorithm of the reinforcement learning algorithm to construct an enterprise supply chain optimization model, and store the generated several historical enterprise supply chain optimization experiences in the experience replay pool of the enterprise supply chain optimization model, including the following steps:

[0073] S3-4-1: Take the generation of the enterprise supply chain optimization plan as the simulation environment of the DQN algorithm, and construct an intelligent agent and an experience replay pool;

[0074] S3-4-2: Define the state space of the DQN algorithm according to each supply chain state type corresponding to the historical enterprise short-term load forecasting results, and the parameters of the state space correspond to each supply chain state. For example, enterprise short-term load states such as excessive order volume, insufficient procurement volume, insufficient production volume, excessive inventory, traffic congestion, rising raw material prices, etc.;

[0075] S3-4-3: Define the action space of the DQN algorithm according to the actions that need to be output by the enterprise supply chain optimization strategy; for example, reducing the order volume, increasing the procurement volume, increasing the production volume, performing inventory resource scheduling, reducing the number of transport vehicles, staggering peak travel, etc.;

[0076] S3-4-4: Define the reward function of the DQN algorithm according to the possible influence situations of each action in the action space, which is used to evaluate the quality or influence of the action;

[0077] S3-4-5: Construct the input layer, several hidden layers, and the output layer of the deep Q-network, connect the input layer to the state space, and connect the output layer to the action space;

[0078] S3-4-6: Based on the state space, action space, and reward function, optimize and train the deep Q-network and the agent according to several historical short-term enterprise load forecasting results, construct an enterprise supply chain optimization model, and store the generated several historical enterprise supply chain optimization experiences in the experience replay pool of the enterprise supply chain optimization model;

[0079] S4: Based on the cloud data center, call a trusted institution to verify the signature of the real-time signature data. After the signature verification passes, decrypt the encrypted real-time upload data packet according to the public key of the corresponding enterprise server to obtain the decrypted real-time upload data packet, and perform data parsing on the decrypted real-time upload data packet to obtain the parsed real-time load data and the parsed real-time external data;

[0080] The formula is:

[0081] β u B u P = β u H 2 (R u , M u , ID u , K u )R u + β u K u + β u H 1 (ID u , K u )PK u

[0082] In the formula, β u is the signature verification parameter of enterprise server u; PK u is the public key of enterprise server u; if the left side of the equation is equal to the right side, the signature verification passes;

[0083] The formula is:

[0084] m' u = E - (PK u , M u )

[0085] In the formula, m' u is the decrypted real-time upload data packet of terminal server u; E - (*) is the asymmetric decryption function; PK u is the public key of enterprise server u; M uThe encrypted real-time upload data packet for enterprise server u;

[0086] S5: Based on the cloud data center, use the Principal Component Analysis (PCA) method to reduce the dimension of the parsed real-time load data and the parsed real-time external data, obtaining the reduced-dimension real-time load data and the corresponding real-time key load indicators, as well as the reduced-dimension real-time external data and the corresponding real-time key external indicators, including the following steps:

[0087] S5-1: Perform matrix transformation on the parsed real-time load data to obtain the corresponding real-time load data matrix X = [x 1 ,x 2 ,...x p ,...,x n T , where x p is the p-th row vector of the parsed real-time load data, p is the row vector indicator, and n is the total number of parsed real-time load data; the initial row vector of the real-time load data matrix is the parsed real-time load data, and the initial column vector of the real-time load data matrix is the real-time load indicator data, including real-time order volume, real-time purchase volume, real-time production volume, real-time inventory volume, real-time distribution volume, and real-time sales volume;

[0088] S5-2: Perform standardization processing on the real-time load data matrix to obtain the corresponding standardized real-time load data matrix;

[0089] The formula is:

[0090]

[0091] In the formula, X' is the standardized real-time load data matrix; μ is the mean of the real-time load data matrix; σ is the variance of the real-time load data matrix;

[0092] S5-3: Obtain the covariance matrix of the standardized real-time load data matrix, and based on the standardized real-time load data matrix and the covariance matrix, obtain the corresponding alternative principal component matrix; the alternative row vector of the alternative principal component matrix is the parsed real-time load data, and the alternative column vector of the alternative principal component matrix is the real-time load alternative indicator data;

[0093] The formula is:

[0094]

[0095] In the formula, D is the covariance matrix of the standardized real-time load data matrix; Y is the alternative principal component matrix; P is the transformation matrix; E is the unit eigenvector matrix; n is the total number of parsed real-time load data; ​

[0096] Y = PX'

[0097] Where Y is the alternative principal component matrix; P is the transformation matrix; X' is the real-time load data matrix after standardization processing;

[0098] S5-4: Take the first 90% of the cumulative contribution rate of variance in the alternative principal component matrix Y = [y 1 , y 2 ,... y l ,..., y L as several alternative column vectors y' with the highest cumulative contribution rate of variance, and use them as the corresponding several principal component column vectors to obtain the reduced-dimensional real-time load data matrix Y' = [y' l , y' 1 ,... y' 2 ,..., y l ,..., y K ; The key row vector of the reduced-dimensional real-time load data matrix is the reduced-dimensional real-time load data, and the key column vector of the reduced-dimensional real-time load data matrix is the real-time key load index;

[0099] The formula is:

[0100]

[0101] Where G is the cumulative contribution rate of variance; λ l is the variance of the l-th alternative principal component y l ; l is the alternative principal component index; L is the total number of alternative principal components; K is the total number of principal components;

[0102] S5-5: Perform inverse matrix transformation on the reduced-dimensional real-time load data matrix to obtain the reduced-dimensional real-time load data;

[0103] S5-6: Use the PCA method to reduce the dimension of the parsed real-time external data to obtain the reduced-dimensional real-time external data and the corresponding real-time key external indicators;

[0104] S6: Based on the cloud data center, according to the real-time key load indicators and real-time key external indicators, use the index weight generation model to generate index weights, obtain the real-time load index weights and real-time external index weights, and adjust the enterprise short-term load forecasting model according to the real-time load index weights and real-time external index weights to obtain the adjusted enterprise short-term load forecasting model, including the following steps:

[0105] S6-1: Take the index weight values corresponding to the real-time key load indicators and real-time key external indicators as the optimization objectives, use the index weight generation model to generate index weights, and obtain the real-time load index weights and real-time external index weights, including the following steps:

[0106] S6-1-1: Take the index weight values corresponding to the real-time key load index and the real-time key external index as the optimization objectives;

[0107] S6-1-2: Initialize the algorithm parameters of the IWOA optimization algorithm, and use the chaotic mapping sequence to initialize the IWOA population;

[0108] The chaotic mapping sequence is the Tent-Logistic-Cosine chaotic mapping sequence, and the formula of the chaotic mapping sequence is:

[0109]

[0110] In the formula, x i+1 is the initial position of the whale individual generated by the Tent-Logistic-Cosine chaotic mapping sequence; x i is the initial position of the randomly generated whale population; r is a preset parameter; i is the whale individual indicator; Using the Tent-Logistic-Cosine chaotic mapping sequence to generate the initial population, compared with the randomly distributed population, the initial position distribution of the improved whale population is more uniform, expanding the search range of the whale population in space, increasing the diversity of the group positions, and improving the defect that the algorithm is prone to fall into local extrema to a certain extent, thereby improving the optimization efficiency of the algorithm;

[0111] S6-1-3: Calculate the fitness value of the whale individuals in the IWOA population, and retain the optimal whale individual according to the fitness value of the whale individuals;

[0112] The formula for the fitness value is:

[0113]

[0114] In the formula, fit is the fitness function; E is the index error function; y n' is the true weight value of the n'-th index; y n * ' is the ideal weight value of the n'-th index; n' is the index indicator; N is the total number of indexes;

[0115] S6-1-4: Randomly generate p'. If p'<0.5 and |A|<1, perform the behavior of surrounding the prey and update the position of the IWOA population. If p'<0.5 and |A|≥1, perform the behavior of searching for the prey and update the position of the IWOA population. If p'≥0.5, perform the bubble net attack behavior and update the position of the IWOA population; where p' is the update parameter and A is the step size coefficient optimized by introducing the convergence factor;

[0116] The formula for the behavior of surrounding the prey is:

[0117] X 1 (t + 1) = X * (t) - AD

[0118] Where X 1 (t + 1) is the position of the whale individual updated for the behavior of surrounding the prey; X * (t) is the optimal position of the whale individual; D is the distance between the current whale individual and the optimal whale individual, that is, the search step size; A = 2ar - a, where a is the convergence factor that decreases from 2 to 0;

[0119] The convergence factor is used to improve the search step size of the traditional whale optimization algorithm, and the formula for the convergence factor is:

[0120]

[0121] Where a is the convergence factor; tanh(.) is the hyperbolic tangent function; t, t max are the current iteration number and the maximum iteration number respectively; a max , a min are the maximum and minimum values of the convergence factor respectively; λ is the decreasing rate parameter, k' is the decreasing period parameter, λ = -2π, k' = π;

[0122] In the early stage of iteration, the value of a is larger, and the updated A is also larger, |A| ≥ 1, making the IWOA algorithm stay in the behavior of searching for prey for a long time in the early stage of iteration, enhancing the global search ability of the algorithm. In the later stage of iteration, the value of a is smaller, and the updated A is also smaller, |A| < 1, making the IWOA algorithm stay in the behavior of surrounding the prey for a long time in the later stage of iteration, enhancing the local surrounding ability of the algorithm and improving the ability of local hunting;

[0123] The formula for the behavior of searching for prey is:

[0124] X 2 (t + 1) = X rand (t) - AD

[0125] Where X 2 (t + 1) is the position of the whale individual updated for the behavior of searching for prey; X rand (t) is the position of a randomly selected whale individual from the IWOA population;

[0126] The formula for the bubble net attack behavior is:

[0127] X 3 (t + 1) = D'c bl cos(2πl) + X * (t)

[0128] Where X 3(t + 1) is the position of the whale individual updated for the Paopao.com attack behavior; D' is the distance between the current whale individual and the prey, calculated in the same way as the search step D above; b is a constant defining the spiral equation, b = 1; l is a random number between [-1, 1];

[0129] S6-1-5: Based on the updated IWOA population, perform dynamic reverse learning to obtain each whale individual in the updated IWOA population, that is, the reverse solution corresponding to the forward solution, and calculate the fitness values of the forward solution and the reverse solution. According to the fitness values of the forward solution and the reverse solution, update the optimal whale individual;

[0130] The formula for the dynamic reverse learning strategy is:

[0131] X' i X'(t) = k(a(t) + b(t)) - X(t) i (t)

[0132] In the formula, X' i (t) and X i (t) are the reverse solution position and the forward solution position of the i-th whale individual respectively; a(t) and b(t) are the upper bound and the lower bound of the current IWOA population in the search dimension respectively; k is a decreasing inertia factor, k = 0.9 - 0.5t / t max ; t and t max are the current iteration number and the maximum iteration number respectively; reduce the search blind spot and more effectively avoid the algorithm from premature convergence and falling into the local optimal value;

[0133] S6-1-6: Determine whether the iteration number meets the requirements or whether the optimal fitness value corresponding to the updated optimal whale individual meets the requirements. If so, output the position of the global optimal solution corresponding to the updated optimal whale individual to obtain the optimal real-time load index weight and the optimal real-time external index weight. Otherwise, perform the next update of the IWOA population;

[0134] S6-2: According to the real-time key load index and the real-time key external index, adjust the number of neurons in the input layer of the enterprise short-term load forecasting model and the neuron weights between the input layer and the attention layer to obtain the adjusted input layer;

[0135] S6-3: According to the real-time load index weight and the real-time external index weight, adjust the attention weights of the attention layer to obtain the adjusted attention layer and the adjusted attention weights. And according to the adjusted input layer, the adjusted attention layer and the adjusted attention weights, obtain the adjusted enterprise short-term load forecasting model;

[0136] S7: Based on the cloud data center, according to the dimension-reduced real-time load data and the dimension-reduced real-time external data, use the adjusted short-term enterprise load forecasting model to conduct short-term enterprise load forecasting, obtain the real-time short-term enterprise load forecasting result, and according to the real-time short-term enterprise load forecasting result, use the enterprise supply chain optimization model to optimize the enterprise supply chain, and obtain the real-time enterprise supply chain optimization strategy, including the following steps:

[0137] S7-1: Based on the cloud data center, according to the dimension-reduced real-time load data and the dimension-reduced real-time external data, use the adjusted short-term enterprise load forecasting model to conduct short-term enterprise load forecasting, obtain the real-time short-term enterprise load forecasting result. For example, in this embodiment, the real-time short-term enterprise load forecasting result is insufficient purchase quantity and traffic congestion;

[0138] S7-2: According to the real-time short-term enterprise load forecasting result, use the enterprise supply chain optimization model to optimize the enterprise supply chain, and obtain the real-time enterprise supply chain optimization strategy, including the following steps:

[0139] S7-2-1: According to the dimension-reduced real-time load data and the dimension-reduced real-time external data, match a number of historical enterprise supply chain optimization experiences in the experience replay pool;

[0140] S7-2-2: According to a number of historical enterprise supply chain optimization experiences, update the action space of the enterprise supply chain optimization model to obtain the updated action space A' = [a' 1 ,..., a' j" ,..., a' I , where a' j" is the updated j-th action value, j is the action indicator; I is the total number of action space dimensions;

[0141] S7-2-3: According to the real-time short-term enterprise load forecasting result, update the state space of the enterprise supply chain optimization model to obtain the updated state space S' = [s' 1 ,..., s' i" ,..., s' I' , where s' i" is the updated i-th state value, i is the state indicator, and I' is the total number of state space dimensions;

[0142] S7-2-4: Use the updated state space S' = [s' 1 ,..., s' i' ,..., s' I as the input of the enterprise supply chain optimization model, and use the deep Q-network to generate the updated action space A' = [a' 1 ,..., a' j' ,..., a' IThe Q-values of each possible action in

[0143] S7-2-5: Use the reward function to obtain the reward values of each possible action in the updated action space, and update the Q-values of the possible actions according to the reward values to obtain the updated Q-values of the possible actions;

[0144] The formula is:

[0145] Q(s' p' ,a' p' ) = (1 - α")·Q(s p' ,a p' ) + α"·(R(s p' ,a p' ,s' p' ) + γ·Q max (s p' ,a p' ))

[0146] In the formula, Q(s' p' ,a' p' ) is the updated Q-value corresponding to the updated state value s' p' and the updated action value a' p' ; Q(s p' ,a p' ) is the predicted Q-value corresponding to the state value s p' and the action value a p' ; α" is the learning rate; Q max (s p' ,a p' ) is the highest predicted Q-value;

[0147] S7-2-6: Repeat the above steps until the iteration number threshold is reached. Use the greedy strategy to take the possible action corresponding to the highest updated Q-value as the execution action, and output the execution action as the real-time enterprise supply chain optimization strategy. In this embodiment, the real-time enterprise supply chain optimization strategy includes increasing the procurement volume, reducing the number of transport vehicles, and staggering trips;

[0148] S8: Based on the cloud data center, encrypt the real-time enterprise supply chain optimization strategy according to the public key of the corresponding enterprise server to obtain the encrypted real-time enterprise supply chain optimization strategy, and send the encrypted real-time enterprise supply chain optimization strategy to the corresponding enterprise server;

[0149] The formula is:

[0150] MM u = E(PK u ,mm u )

[0151] In the formula, MM uThe encrypted real-time enterprise supply chain optimization strategy for enterprise server u; E(*) is an asymmetric encryption function; mm u The real-time enterprise supply chain optimization strategy for enterprise server u; PK u The public key of enterprise server u; u is the enterprise server indicator.

[0152] Embodiment 2:

[0153] As Figure 2 shown, this embodiment provides an enterprise short-term load forecasting system based on artificial intelligence for implementing the enterprise short-term load forecasting method. The system includes a cloud data center, a trusted institution, and several enterprise servers. The several enterprise servers are respectively communicatively connected to the cloud data center and the trusted institution. The cloud data center is communicatively connected to the trusted institution, and the cloud data center includes an initialization unit, a data receiving unit, a data dimensionality reduction unit, a weight generation and model adjustment unit, a short-term load forecasting and supply chain optimization unit, and a data encryption and sending unit;

[0154] The trusted institution is used to generate keys and authenticate the identities of several enterprise servers, obtain the public-private key pairs and signature information of each enterprise server, return the private key and signature information in the public-private key pairs to the corresponding enterprise servers, and send the public keys in the public-private key pairs to the cloud data center;

[0155] The enterprise server is used to compress and package the real-time load data and real-time external data to obtain a real-time upload data packet, encrypt and sign the real-time upload data packet according to the private key and signature information to obtain an encrypted real-time upload data packet and real-time signature data, and upload them to the cloud data center;

[0156] The initialization unit is used to perform initialization, and construct an index weight generation model, an enterprise short-term load forecasting model, and an enterprise supply chain optimization model according to several historical load data and corresponding historical external data using artificial intelligence algorithms;

[0157] The data receiving unit is used to call the trusted institution to verify the signature of the real-time signature data. After the signature verification passes, decrypt the encrypted real-time upload data packet according to the public key of the corresponding enterprise server to obtain a decrypted real-time upload data packet, and perform data parsing on the decrypted real-time upload data packet to obtain parsed real-time load data and parsed real-time external data;

[0158] The data dimensionality reduction unit is used to perform data dimensionality reduction on the parsed real-time load data and parsed real-time external data to obtain reduced-dimensional real-time load data and corresponding real-time key load indicators, and reduced-dimensional real-time external data and corresponding real-time key external indicators;

[0159] A weight generation and model adjustment unit, which is used to generate real-time load index weights and real-time external index weights by using an index weight generation model according to real-time key load indicators and real-time key external indicators, and adjust the enterprise short-term load forecasting model according to the real-time load index weights and real-time external index weights to obtain an adjusted enterprise short-term load forecasting model;

[0160] A short-term load forecasting and supply chain optimization unit, which is used to perform enterprise short-term load forecasting by using the adjusted enterprise short-term load forecasting model according to the dimension-reduced real-time load data and dimension-reduced real-time external data to obtain a real-time enterprise short-term load forecasting result, and perform enterprise supply chain optimization by using the enterprise supply chain optimization model according to the real-time enterprise short-term load forecasting result to obtain a real-time enterprise supply chain optimization strategy;

[0161] A data encryption and sending unit, which is used to encrypt the real-time enterprise supply chain optimization strategy according to the public key of the corresponding enterprise server to obtain an encrypted real-time enterprise supply chain optimization strategy, and send the encrypted real-time enterprise supply chain optimization strategy to the corresponding enterprise server.

[0162] The present invention discloses an enterprise short-term load forecasting method and system based on artificial intelligence, and proposes a data security scheme combining key generation and identity authentication, which ensures the security of data transmission between the enterprise server and the cloud data center, effectively prevents data leakage and unauthorized access, and significantly improves data security; by using artificial intelligence algorithms to construct an index weight generation model, an enterprise short-term load forecasting model and an enterprise supply chain optimization model, the enterprise short-term load forecasting model can more accurately capture the complex relationships and non-linear characteristics in the data, thereby improving the prediction accuracy; the index weight generation model can adaptively adjust the model parameters of the enterprise short-term load forecasting model to adapt to the changing data structure and improve the adaptability to the characteristics of different industries and different enterprises; the cloud data center realizes the decryption verification and dimension reduction processing of real-time data, ensuring the real-time nature of the prediction. This real-time data processing ability enables enterprises to quickly respond to load changes and timely adjust supply chain decisions. For emergencies and rapidly changing load demands, the system responds quickly; it combines an enterprise supply chain optimization model to provide a complete set of optimization strategies for enterprises. This integrated optimization strategy can help enterprises better respond to prediction results and achieve the optimal allocation of resources; it considers the impact of external data on load forecasting, improves the comprehensiveness of forecasting, and affects the formulation of enterprise supply chain optimization strategies.

[0163] The present invention is not limited to the above optional embodiments, and any person can obtain other various forms of products under the inspiration of the present invention. The above specific embodiments should not be construed as limiting the protection scope of the present invention, and the protection scope of the present invention shall be defined by the claims, and the specification can be used to interpret the claims.

Claims

1. An enterprise short-term load forecasting method based on artificial intelligence, characterized by: The steps include: Based on the trusted institution, key generation and identity authentication are performed on several enterprise servers to obtain the public-private key pair and signature information of each enterprise server, and the private key and signature information in the public-private key pair are returned to the corresponding enterprise server, and the public key in the public-private key pair is sent to the cloud data center; Based on the enterprise server, the real-time load data and real-time external data are compressed and packaged to obtain a real-time upload data packet. The real-time upload data packet is encrypted and signed according to the private key and signature information to obtain the encrypted real-time upload data packet and real-time signature data, and then uploaded to the cloud data center; Initialization is performed based on the cloud data center. Based on a number of historical load data and corresponding historical external data, an artificial intelligence algorithm is used to build an indicator weight generation model, an enterprise short-term load forecasting model, and an enterprise supply chain optimization model. Based on the cloud data center, a trusted institution is called to perform signature verification on the real-time signature data. After the signature verification is passed, the encrypted real-time upload data packet is decrypted according to the public key of the corresponding enterprise server to obtain the decrypted real-time upload data packet, and the decrypted real-time upload data packet is parsed to obtain the parsed real-time load data and parsed real-time external data; Based on the cloud data center, the parsed real-time load data and the parsed real-time external data are subjected to data dimensionality reduction to obtain the reduced-dimensional real-time load data and the corresponding real-time key load indicators, as well as the reduced-dimensional real-time external data and the corresponding real-time key external indicators; Based on the cloud data center, according to the real-time key load indicators and the real-time key external indicators, the indicator weight generation model is used to generate the indicator weight, and the real-time load indicator weight and the real-time external indicator weight are obtained. According to the real-time load indicator weight and the real-time external indicator weight, the enterprise short-term load forecasting model is adjusted to obtain the adjusted enterprise short-term load forecasting model; Based on the cloud data center, according to the real-time load data after dimensionality reduction and the real-time external data after dimensionality reduction, the adjusted enterprise short-term load forecasting model is used to perform enterprise short-term load forecasting, and the real-time enterprise short-term load forecasting result is obtained. According to the real-time enterprise short-term load forecasting result, the enterprise supply chain optimization model is used to optimize the enterprise supply chain, and the real-time enterprise supply chain optimization strategy is obtained; Based on the cloud data center, the real-time enterprise supply chain optimization strategy is encrypted according to the public key of the corresponding enterprise server to obtain the encrypted real-time enterprise supply chain optimization strategy, and the encrypted real-time enterprise supply chain optimization strategy is sent to the corresponding enterprise server.

2. According to the method of claim 1, the method is characterized by: The real-time load data includes real-time order volume, real-time purchase volume, real-time production volume, real-time inventory volume, real-time distribution volume and real-time sales volume; The real-time external data includes real-time weather data, real-time traffic data, real-time raw material prices, real-time promotion data, real-time price fluctuation data and real-time competitive product data.

3. The method for enterprise short-term load forecasting based on artificial intelligence according to claim 1 is characterized in that: The PCA method is used to reduce the dimension of the parsed real-time load data and the parsed real-time external data.

4. The method for enterprise short-term load forecasting based on artificial intelligence according to claim 1 is characterized in that: Based on the cloud data center, initialization is performed, and according to a number of historical load data and corresponding historical external data, an artificial intelligence algorithm is used to build an indicator weight generation model, an enterprise short-term load forecasting model, and an enterprise supply chain optimization model, including the following steps: Based on the cloud data center, a number of historical load data and corresponding historical external data are collected, and the number of historical load data and the corresponding historical external data are preprocessed respectively to obtain a number of preprocessed historical load data and corresponding preprocessed historical external data; According to a number of pre-processed historical load data and corresponding pre-processed historical external data, a swarm intelligence optimization algorithm is used to construct an index weight generation model, and generate a number of historical load index weights and corresponding historical external index weights; Based on several historical load data and corresponding historical load index weights, as well as historical external data and corresponding historical external index weights, a deep learning algorithm is used to build an enterprise short-term load forecasting model and generate several historical enterprise short-term load forecasting results; According to several historical enterprise short-term load forecast results, a reinforcement learning algorithm is used to build an enterprise supply chain optimization model, and the generated several historical enterprise supply chain optimization experiences are stored in the experience replay pool of the enterprise supply chain optimization model.

5. The method for enterprise short-term load forecasting based on artificial intelligence according to claim 4 is characterized in that: The indicator weight generation model is constructed based on the IWOA algorithm.

6. The method for enterprise short-term load forecasting based on artificial intelligence according to claim 5 is characterized by: The enterprise short-term load forecasting model is constructed based on the Attention-LSTM-Elman algorithm.

7. The method for enterprise short-term load forecasting based on artificial intelligence according to claim 4 is characterized by: The enterprise supply chain optimization model is constructed based on the DQN algorithm.

8. The method for enterprise short-term load forecasting based on artificial intelligence according to claim 6 is characterized by: The enterprise short-term load forecasting model includes an input layer and several hidden layers constructed based on the LSTM algorithm, an attention layer constructed based on the Attention mechanism, and an output layer constructed based on the Elman algorithm. The input layer, the attention layer, the several hidden layers and the output layer are connected in sequence.

9. The method for enterprise short-term load forecasting based on artificial intelligence according to claim 8 is characterized in that: According to the real-time key load indicators and the real-time key external indicators, the indicator weight generation model is used to generate the indicator weight, and the real-time load indicator weight and the real-time external indicator weight are obtained. According to the real-time load indicator weight and the real-time external indicator weight, the enterprise short-term load forecasting model is adjusted to obtain the adjusted enterprise short-term load forecasting model, including the following steps: The indicator weight values ​​corresponding to the real-time key load indicators and the real-time key external indicators are used as optimization targets, and the indicator weight generation model is used to generate the indicator weights to obtain the real-time load indicator weights and the real-time external indicator weights; According to the real-time key load index and the real-time key external index, the number of neurons in the input layer of the enterprise short-term load forecasting model and the neuron weights between the input layer and the attention layer are adjusted to obtain an adjusted input layer; According to the real-time load indicator weight and the real-time external indicator weight, the attention weight of the attention layer is adjusted to obtain the adjusted attention layer and the adjusted attention weight. According to the adjusted input layer, the adjusted attention layer and the adjusted attention weight, the adjusted enterprise short-term load forecasting model is obtained.

10. An enterprise short-term load forecasting system based on artificial intelligence, used to implement the enterprise short-term load forecasting method according to any one of claims 1 to 9, characterized in that: The system includes a cloud data center, a trusted organization and several enterprise servers. The several enterprise servers are respectively connected to the cloud data center and the trusted organization in communication. The cloud data center is connected to the trusted organization in communication. The cloud data center includes an initialization unit, a data receiving unit, a data dimension reduction unit, a weight generation and model adjustment unit, a short-term load forecasting and supply chain optimization unit and a data encryption and sending unit.

Images