Method and device for checking compliance of power guarantee scheme based on AI large model

Abstract

The invention relates to the technical field of scheme auditing, and discloses a method and device for auditing the compliance of a power guarantee scheme based on an AI large model, and the method comprises the following steps: collecting first comprehensive information and second comprehensive information of a current time point and n previous continuous time points; the first comprehensive information is coded into first graph structure data, and the second comprehensive information is coded into second graph structure data; the first graph structure data and the second graph structure data are input into a scheme auditing model, the scheme auditing model comprises a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a first feature fusion layer, a first output layer and a second output layer, and the scheme auditing model outputs a representation auditing result; and decoding the output result of the scheme auditing model to obtain the problem of non-compliance. In the production process of some special drugs, the influence of the fermentation process on the electric power is considered, misjudgment is prevented, and therefore compliance checking of the electric power guarantee scheme is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of scheme auditing, and more specifically, to a method and device for auditing the compliance of a power supply protection scheme based on an AI big model. Background Art

[0002] In the pharmaceutical production process, maintaining specific humidity, temperature and cleanliness levels is a key factor in ensuring the quality and safety of pharmaceuticals. In order to ensure that these environmental conditions continue to meet the standards, the stability and compliance of the power supply become particularly important. This involves reviewing the compliance of the power supply plan.

[0003] Pharmaceutical production has strict requirements on environmental conditions (such as temperature and humidity), and the instability of power supply will directly affect the maintenance of these conditions. Power outages or fluctuations may cause shutdown or performance fluctuations of environmental control systems (such as HVAC systems and humidity control equipment), which in turn may cause environmental conditions to deviate from the standard range.

[0004] Pharmaceutical production is typically a continuous process and interruptions in production can result in the scrapping of batches of product. Line interruptions or equipment failures due to power failures can result in large amounts of material wasted, non-compliant products, or even the need to destroy entire batches.

[0005] All environmental parameters in the production process must be accurately controlled and recorded with the support of a compliant power solution. Stable power supply ensures the accuracy and real-time data of the production monitoring system, thereby supporting quality assurance and auditing of the production process.

[0006] In the production process of some special drugs such as antibiotics, fermentation is one of the key steps, involving the cultivation and metabolism of microorganisms. The operation of the fermenter includes stirring, heating and cooling, and the power load requirements of these operations will change with the different stages of the fermentation process, resulting in periodic fluctuations.

[0007] In the early stage of fermentation, the liquid viscosity is low and the power required for stirring is small. As fermentation progresses, the cell density increases, the liquid viscosity also increases, and the power required for stirring increases. This change will cause the load fluctuation of the stirring motor.

[0008] At the beginning and end of fermentation, external heating may be required to maintain a constant system temperature. However, at the peak of fermentation, microbial activity intensifies and the heat generated by metabolism increases significantly. In this case, the operating frequency of the heating system can be reduced, and even external cooling may be required to dissipate the excess heat, which reduces the power burden on the heating system.

[0009] During the peak fermentation period, especially when bacterial cells are densely multiplying and metabolically active, the heat released may require the cooling system to run at a high load to maintain the appropriate temperature. At this time, the power consumption of the cooling equipment will increase, while its power demand will decrease during the period when the cooling burden is lighter. Summary of the invention

[0010] The present invention provides a method and device for compliance review of power conservation plans based on an AI large model, which solves the technical problem in the related technology that at different stages of the antibiotic production process, various parameters will have some periodic fluctuations within a normal range, which may lead to the power conservation plan being easily misjudged as non-compliant.

[0011] The present invention provides a method for compliance review of a power supply protection scheme based on an AI large model, comprising the following steps:

[0012] Step 100, collecting No. 1 comprehensive information and No. 2 comprehensive information including the current time point and a total of n consecutive time points before, No. 1 comprehensive information including power usage data, environmental data, equipment operation status data, HVAC system operation parameters, backup power data, regulations and compliance standards, No. 2 comprehensive information including power usage data, environmental data, equipment operation status data, HVAC system operation parameters, backup power data, fermentation status data;

[0013] Step 200, encoding the first comprehensive information into the first graph structure data, and encoding the second comprehensive information into the second graph structure data;

[0014] Step 300, inputting the No. 1 graph structure data and the No. 2 graph structure data into a solution review model, the solution review model comprising a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a first feature fusion layer, a first output layer, and a second output layer, wherein the first hidden layer is used to input the No. 1 graph structure data, output a first hidden state to the second hidden layer, the second hidden layer outputs the second hidden layer to the first feature fusion layer, the third hidden layer is used to input the No. 2 graph structure data, output a third hidden state to the fourth hidden layer, the fourth hidden layer outputs a fourth hidden state to the first feature fusion layer and the second output layer, the second output layer outputs a deviation value, the first feature fusion layer outputs a first fusion state to the first output layer, and the first output layer outputs a review result;

[0015] Step 400, decoding the output results of the solution audit model to obtain the existing non-compliance issues.

[0016] In a preferred embodiment, the power usage data includes: power load, voltage, and current;

[0017] Environmental data include: temperature, humidity, cleanliness;

[0018] Equipment operation status data includes: start and stop status of each production equipment;

[0019] Backup power data includes: backup battery voltage, remaining power, backup generator start / stop status, and load;

[0020] Regulations and compliance standards: power-related standards, backup power equipment specifications, environmental control-related regulations, power quality standards;

[0021] Fermentation status data include: liquid viscosity, fermentation temperature, fermentation stage;

[0022] In a preferred embodiment, the operating parameters of the HVAC system include: production temperature, storage area temperature, supply air temperature, production humidity, storage area humidity, supply air humidity, production positive pressure value, supply air static pressure, return air static pressure, dust particle count, microbial limit, supply air volume, and return air volume.

[0023] In a preferred embodiment, the fermentation stages include: an initial stage, a peak stage, and an end stage.

[0024] In a preferred embodiment, the No. 1 graph structure data includes nodes and edges connecting the nodes, a node represents a power system or a backup power system or an environment or a production device or a regulation and compliance standard;

[0025] There are edges between the node representing the power system and the node representing the backup power system, there are edges between all nodes representing production equipment and the node representing the environment, there are edges between all nodes representing production equipment and the nodes representing the power system and the nodes representing the backup power system, there are edges between all nodes representing regulations and compliance standards and the remaining nodes except those representing regulations and compliance standards;

[0026] The No. 1 comprehensive information of each node is encoded into No. 1 sequence data, the No. 1 sequence data includes n sequence units, and the t-th sequence unit represents the No. 1 comprehensive information of all nodes at the t-th moment.

[0027] In a preferred embodiment, the second graph structure data includes nodes and edges connecting the nodes, a node represents a power system or a backup power system or an environment or a production equipment or a regulation and compliance standard or fermentation information;

[0028] There are edges between the node representing the power system and the node representing the backup power system, there are edges between all nodes representing production equipment and the node representing the environment, there are edges between all nodes representing production equipment and the nodes representing the power system and the nodes representing the backup power system, there are edges between all nodes representing fermentation information and production equipment, there are edges between all nodes representing regulations and compliance standards and the remaining nodes except those representing regulations and compliance standards;

[0029] The second comprehensive information of each node is encoded into second sequence data, the second sequence data includes n sequence units, and the tth sequence unit represents the second comprehensive information of all nodes at the tth moment.

[0030] In a preferred embodiment, the first comprehensive information and the second comprehensive information are input into the model after feature engineering; the first comprehensive information and the second comprehensive information include text modality, and the WordEmbedding algorithm is used as a feature engineering method for the information in the text modality.

[0031] In a preferred embodiment, the loss function during training of the solution review model is as follows:

[0032]

[0033] LOSS = LOSS1 + LOSS2;

[0034] Where LOSS1 represents the first loss value, N1 represents the total number of training samples, is the true classification label of the q1th training sample, which takes a value of 0 or 1, 1 means that the problem is compliant, and 0 means that the problem is not compliant. is the classification label of the q1th training sample predicted by the solution review model, ln represents the logarithmic function with the natural constant e as the base, LOSS2 represents the second loss value, LOSS represents the total loss value, N2 represents the total number of training samples, represents the actual deviation values ​​of the maintenance personnel of the q2th sample, Represents the deviation values ​​of the q2th sample output by the model.

[0035] In a preferred embodiment, whether there is a non-compliance problem is determined as follows: when the probability value is greater than 0.5, it is determined that the non-compliance problem exists.

[0036] The beneficial effect of the present invention is that the present invention takes into account the impact of the fermentation process on electricity in the production process of some special drugs, such as the production process of antibiotics, resulting in fluctuations in electricity within a normal range, thereby preventing misjudgment and making the compliance review of the power protection plan more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of a method for compliance review of a power supply protection plan based on an AI big model of the present invention. DETAILED DESCRIPTION

[0038] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0039] At least one embodiment of the present invention discloses a method for compliance review of a power protection solution based on an AI big model, such as Figure 1 As shown, the following steps are included:

[0040] Step 100, collecting No. 1 comprehensive information and No. 2 comprehensive information including the current time point and a total of n consecutive time points before, No. 1 comprehensive information including power usage data, environmental data, equipment operation status data, HVAC system operation parameters, backup power data, regulations and compliance standards, No. 2 comprehensive information including power usage data, environmental data, equipment operation status data, HVAC system operation parameters, backup power data, fermentation status data;

[0041] Power usage data includes: power load, voltage, and current;

[0042] Environmental data include: temperature, humidity, cleanliness;

[0043] Equipment operation status data includes: start and stop status of each production equipment;

[0044] Backup power data includes: backup battery voltage, remaining power, backup generator start / stop status, and load;

[0045] Regulations and compliance standards: power-related standards, backup power equipment specifications, environmental control-related regulations, power quality standards;

[0046] Fermentation status data include: liquid viscosity, fermentation temperature, fermentation stage;

[0047] In one embodiment of the present invention, the operating parameters of the HVAC system include: production temperature, storage area temperature, supply air temperature, production humidity, storage area humidity, supply air humidity, production positive pressure value, supply air static pressure, return air static pressure, dust particle count, microbial limit, supply air volume, and return air volume.

[0048] In one embodiment of the present invention, the fermentation stages include: an initial stage, a peak stage, and an end stage.

[0049] Step 200, encoding the first comprehensive information into the first graph structure data, and encoding the second comprehensive information into the second graph structure data;

[0050] The No. 1 graph structure data includes nodes and edges connecting the nodes, where a node represents a power system or a backup power system or an environment or a production device or a regulation and compliance standard;

[0051] There are edges between the node representing the power system and the node representing the backup power system, there are edges between all nodes representing production equipment and the node representing the environment, there are edges between all nodes representing production equipment and the nodes representing the power system and the nodes representing the backup power system, there are edges between all nodes representing regulations and compliance standards and the remaining nodes except those representing regulations and compliance standards;

[0052] Encode the No. 1 comprehensive information of each node into No. 1 sequence data, the No. 1 sequence data includes n sequence units, and the t-th sequence unit represents the No. 1 comprehensive information of all nodes at the t-th moment;

[0053] The No. 2 graph structure data includes nodes and edges connecting the nodes, where a node represents a power system or a backup power system or an environment or a production equipment or a regulation and compliance standard or fermentation information;

[0054] There are edges between the node representing the power system and the node representing the backup power system, there are edges between all nodes representing production equipment and the node representing the environment, there are edges between all nodes representing production equipment and the nodes representing the power system and the nodes representing the backup power system, there are edges between all nodes representing fermentation information and production equipment, there are edges between all nodes representing regulations and compliance standards and the remaining nodes except those representing regulations and compliance standards;

[0055] Encode the No. 2 comprehensive information of each node into No. 2 sequence data, the No. 2 sequence data includes n sequence units, and the t-th sequence unit represents the No. 2 comprehensive information of all nodes at the t-th moment;

[0056] Step 300, inputting the No. 1 drawing structure data and the No. 2 drawing structure data into the scheme review model, and the scheme review model outputs the review result;

[0057] The scheme review model includes a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a first feature fusion layer, a first output layer and a second output layer, wherein the first hidden layer is used to input the No. 1 graph structure data, output the first hidden state to the second hidden layer, the second hidden layer outputs the second hidden layer to the first feature fusion layer, the third hidden layer is used to input the No. 2 graph structure data, output the third hidden state to the fourth hidden layer, the fourth hidden layer outputs the fourth hidden state to the first feature fusion layer and the second output layer, the second output layer outputs a deviation value, the first feature fusion layer outputs the first fusion state to the first output layer, and the first output layer outputs a review result;

[0058] In one embodiment of the present invention, the first comprehensive information and the second comprehensive information are input into the model after feature engineering; the first comprehensive information and the second comprehensive information include text modality, and the WordEmbedding algorithm is used as a feature engineering method for the information of the text modality.

[0059] In one embodiment of the present invention, the calculation formula of the first hidden layer is as follows:

[0060]

[0061] where k v,t The first hidden state of the node of the vth node of the tth sequence unit of the first sequence data, represents the aggregation coefficient of the vth node of the tth sequence unit of the first sequence data, M v represents the set of nodes that have edges with the vth node of the tth sequence unit of the first sequence data, σ represents the activation function, and W k represents the graph weight coefficient;

[0062]

[0063] Z v,t =W z x v,t ;

[0064] Z u,t =W z x u,t ;

[0065] where x v,t and x u,t Respectively represent the node features of the vth and uth nodes of the tth sequence unit of the first sequence data, Z v,t and Z v,t Respectively represent the linear transformation features of the vth and uth nodes of the tth sequence unit of the first sequence data, W z represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transpose, exp represents the exponential function with natural constant as the base, LeakyRelu represents the modified linear unit function, M v Represents the set of nodes that have edges with the vth node.

[0066] In one embodiment of the present invention, the calculation formula of the second hidden layer is as follows:

[0067] u (t) =σ(W u X (t) +W u H (t-1) +b u );

[0068] r (t) =σ(W r X (t) +W r H (t-1) +b r );

[0069] C (t) =tanh(W c X (t) +W c r (t) ⊙H (t-1) +b c );

[0070] H (t) =(1-u (t) )⊙C (t) +u (t) ⊙H (t-1) ;

[0071] Among them, W u , W r , W c represents the weight parameter of the first, second and third sequences, b u , b r , b c represents the first, second, and third sequence bias parameters, ⊙ represents the dot product, u 9t) 、r (t) and C (t) Represent the first, second, and third intermediate states respectively, where X t =∑ v∈N k v,t , k v,t Represents the second hidden state of the vth node of the tth sequence unit of the first sequence data, ∑ v∈N k v,t represents the concatenation of the second hidden states of all nodes of the t-th sequence unit, N represents the set of all nodes, H (t) and H (t-1) They represent the t-th and t-1-th second hidden states respectively, n≥t≥1, n represents the total number of sequence units of the first sequence data, and when t=1, H (t-1) =X (t) , tanh is the hyperbolic tangent function, and σ represents the S-type function.

[0072] In one embodiment of the present invention, the calculation formula of the third hidden layer is as follows:

[0073]

[0074] Where L v,trepresents the third hidden state of the vth node of the tth sequence unit of the second sequence data, W L represents the graph weight parameter, X v,t and X u,t They represent the node features of the vth and uth nodes of the tth sequence unit of the second sequence data, respectively. (v) represents the set of nodes directly connected to the vth node, M represents M (v) The total number of nodes in ;

[0075] R v,u,t =tanh(δ*X v,t T *X u,t +1)

[0076] Where R v,u,t It represents the aggregation coefficient between the vth and uth nodes of the tth sequence unit of the second sequence data, tanh represents the hyperbolic tangent function, T represents the transpose, and δ represents an adjustable parameter; the default value of δ is 1 / N, and N is the dimension of the node feature.

[0077] In one embodiment of the present invention, the calculation formula of the fourth hidden layer is as follows:

[0078] Long short-term memory network, the long short-term memory network includes a forget gate, an input gate, a cell state, and an output gate. The t-th output f of the forget gate t for:

[0079] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0080] Among them, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate, [h t-1 ,x t ] is the t-1th hidden state h t-1 and x t splicing, σ is an S-type function; Where L v,t represents the third hidden state of the vth node of the tth sequence unit of the second sequence data, M (v) represents the set of nodes directly connected to the vth node;

[0081] The t-th output i of the input gate t and the tth candidate cell state The calculation method is:

[0082] i t=σ(W i ·[h t-1 ,x t ]+b i );

[0083]

[0084] Among them, W i is the weight matrix of the input gate, b i is the bias term of the input gate, tanh is the hyperbolic tangent activation function, and W C is the weight matrix for calculating the candidate cell state, b C is the bias term of the candidate cell state;

[0085] The tth cell state C t is the state C of the t-1th cell t-1 and i t Weighted synthesis, the calculation method is:

[0086]

[0087] Among them, * represents element-by-element multiplication operation;

[0088] The tth activation value o of the output gate t and the tth fourth hidden state h t The calculation method is:

[0089] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0090] h t =o t *tanh(C t );

[0091] Among them, W o is the weight matrix of the output gate, b o is the bias term of the output gate.

[0092] In one embodiment of the present invention, the calculation formula of the second output layer is as follows:

[0093]

[0094] Where y2 represents a vector, and the values ​​of the vector represent the power load fluctuation deviation value, voltage fluctuation deviation value, temperature fluctuation deviation value, humidity fluctuation deviation value, and standby power system load fluctuation deviation value, respectively. t represents the tth fourth hidden state, represents the concatenation of the 1st to the nth fourth hidden states, n represents the total number of sequence units of the second sequence data, is the second weight parameter, is the second bias parameter and σ represents the S-type function.

[0095] In one embodiment of the present invention, the calculation formula of the first feature fusion layer is as follows:

[0096]

[0097] Fusion indicates the first fusion state, H (n) represents the nth second hidden state, h n represents the nth fourth hidden state, Concat represents the concatenation function, W Fusion represents the sum weight matrix, b Fusion represents the summed bias parameter and σ represents the S-type function.

[0098] In one embodiment of the present invention, the calculation formula of the first output layer is as follows:

[0099]

[0100] Where y1 represents a vector, the i-th value of the vector represents the probability of the i-th non-compliant problem, and Fusion represents the first fusion state. is the first weight parameter, is the first bias parameter and σ represents the sigmoid function.

[0101] In one embodiment of the present invention, when the probability value is greater than 0.5, it is determined that the non-compliance problem occurs.

[0102] In one embodiment of the present invention, non-compliance issues include: excessive power load fluctuations, excessive temperature fluctuations, excessive humidity fluctuations, excessive cleanliness fluctuations, frequent power outages, excessive HVAC system load fluctuations, long backup power system switching time, overload during backup power system operation, etc.

[0103] In one embodiment of the present invention, the loss function during training of the solution review model is as follows:

[0104]

[0105] LOSS = LOSS1 + LOSS2;

[0106] Where LOSS1 represents the first loss value, N1 represents the total number of training samples, is the true classification label of the q1th training sample, which takes a value of 0 or 1, 1 means that the problem is compliant, and 0 means that the problem is not compliant. is the classification label of the q1th training sample predicted by the solution review model, ln represents the logarithmic function with the natural constant e as the base, LOSS2 represents the second loss value, LOSS represents the total loss value, N2 represents the total number of training samples, represents the actual deviation values ​​of the maintenance personnel of the q2th sample, Represents the deviation values ​​of the q2th sample output by the model.

[0107] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for compliance review of power supply protection scheme based on AI big model, characterized in that: The following steps are involved: Step 100, collecting No. 1 comprehensive information and No. 2 comprehensive information including the current time point and a total of n consecutive time points before, No. 1 comprehensive information including power usage data, environmental data, equipment operation status data, HVAC system operation parameters, backup power data, regulations and compliance standards, No. 2 comprehensive information including power usage data, environmental data, equipment operation status data, HVAC system operation parameters, backup power data, fermentation status data; Step 200, encoding the first comprehensive information into the first graph structure data, and encoding the second comprehensive information into the second graph structure data; Step 300, inputting the No. 1 graph structure data and the No. 2 graph structure data into a solution review model, the solution review model comprising a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer, a first feature fusion layer, a first output layer, and a second output layer, wherein the first hidden layer is used to input the No. 1 graph structure data, output a first hidden state to the second hidden layer, the second hidden layer outputs the second hidden layer to the first feature fusion layer, the third hidden layer is used to input the No. 2 graph structure data, output a third hidden state to the fourth hidden layer, the fourth hidden layer outputs a fourth hidden state to the first feature fusion layer and the second output layer, the second output layer outputs a deviation value, the first feature fusion layer outputs a first fusion state to the first output layer, and the first output layer outputs a review result; Step 400, decoding the output results of the solution audit model to obtain the existing non-compliance issues.

2. According to claim 1, a method for compliance review of power supply protection scheme based on AI big model is characterized in that: Power usage data includes: power load, voltage, and current; Environmental data include: temperature, humidity, cleanliness; Equipment operation status data includes: start and stop status of each production equipment; Backup power data includes: backup battery voltage, remaining power, backup generator start / stop status, and load; Regulations and compliance standards: power-related standards, backup power equipment specifications, environmental control-related regulations, power quality standards; Fermentation status data include: liquid viscosity, fermentation temperature, fermentation stage; The operating parameters of the HVAC system include: production temperature, storage area temperature, supply air temperature, production humidity, storage area humidity, supply air humidity, production positive pressure value, supply air static pressure, return air static pressure, dust particle count, microbial limit, supply air volume, and return air volume.

3. According to claim 2, a method for compliance review of power supply protection scheme based on AI big model is characterized in that: The fermentation stages include: initial stage, peak stage and final stage.

4. According to claim 1, a method for compliance review of power supply protection scheme based on AI big model is characterized in that: The No. 1 graph structure data includes nodes and edges connecting the nodes, where a node represents a power system or a backup power system or an environment or a production device or a regulation and compliance standard; There are edges between the node representing the power system and the node representing the backup power system, there are edges between all nodes representing production equipment and the node representing the environment, there are edges between all nodes representing production equipment and the nodes representing the power system and the nodes representing the backup power system, there are edges between all nodes representing regulations and compliance standards and the remaining nodes except those representing regulations and compliance standards; The No. 1 comprehensive information of each node is encoded into No. 1 sequence data, the No. 1 sequence data includes n sequence units, and the t-th sequence unit represents the No. 1 comprehensive information of all nodes at the t-th moment.

5. According to claim 1, a method for compliance review of power supply protection scheme based on AI big model is characterized in that: The No. 2 graph structure data includes nodes and edges connecting the nodes, where a node represents a power system or a backup power system or an environment or a production equipment or a regulation and compliance standard or fermentation information; There are edges between the node representing the power system and the node representing the backup power system, there are edges between all nodes representing production equipment and the node representing the environment, there are edges between all nodes representing production equipment and the nodes representing the power system and the nodes representing the backup power system, there are edges between all nodes representing fermentation information and production equipment, there are edges between all nodes representing regulations and compliance standards and the remaining nodes except those representing regulations and compliance standards; The second comprehensive information of each node is encoded into second sequence data, the second sequence data includes n sequence units, and the tth sequence unit represents the second comprehensive information of all nodes at the tth moment.

6. According to claim 1, a method for compliance review of power supply protection scheme based on AI big model is characterized in that: After feature engineering, the first comprehensive information and the second comprehensive information are input into the model; the first comprehensive information and the second comprehensive information include text modality, and the Word Embedding algorithm is used as a feature engineering method for the information of the text modality.

7. According to claim 1, a method for compliance review of power supply protection scheme based on AI big model is characterized in that: The loss function of the solution review model training is as follows: LOSS = LOSS1 + LOSS2; Where LOSS1 represents the first loss value, N1 represents the total number of training samples, is the true classification label of the q1th training sample, which takes a value of 0 or 1, 1 means that the problem is compliant, and 0 means that the problem is not compliant. is the classification label of the q1th training sample predicted by the solution review model, ln represents the logarithmic function with the natural constant e as the base, LOSS2 represents the second loss value, LOSS represents the total loss value, N2 represents the total number of training samples, represents the actual deviation values ​​of the maintenance personnel of the q2th sample, Represents the deviation values ​​of the q2th sample output by the model.

8. The method for compliance review of power supply protection scheme based on AI big model according to claim 1 is characterized in that: The determination of whether there is a non-compliant problem is: when the probability value is greater than 0.5, it is determined that the non-compliant problem exists.

9. A device for compliance review of power supply protection scheme based on AI big model, characterized in that: It includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement a method for compliance review of a power supply protection plan based on an AI large model as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, a method for compliance review of a power supply protection plan based on an AI large model as described in any one of claims 1-8 is implemented.

Images