Smart city operation decision-making method and system based on big data

Abstract

The invention provides a smart city operation decision-making method and system based on big data, and relates to the field of smart cities, and the method comprises the steps: obtaining a multi-source data set, and generating a deep belief network according to the multi-source data set; performing noise feature extraction on the data through a deep belief network to obtain a noise feature set; calculating a scalar probability value for the noise feature set, and obtaining a target data set according to the scalar probability value; performing data fusion on the target data set through principal component analysis to obtain fused data; performing decision strategy analysis on the fused data to obtain a plurality of decision strategy data sets; performing strategy optimization on the plurality of decision strategy data sets through a multi-objective optimization algorithm to obtain a plurality of optimization strategy data sets; and generating a city operation scheme according to the plurality of optimization strategy data sets. According to the invention, through advanced technologies such as a deep belief network, principal component analysis and a multi-objective optimization algorithm, a set of complete smart city operation decision-making method is formed.

Description

Technical Field

[0001] The present invention relates to the field of smart cities, and in particular to a smart city operation decision-making method and system based on big data. Background Art

[0002] With the rapid development of urbanization and the continuous advancement of information technology, smart cities have become a key direction for urban development today. In the construction and operation of smart cities, how to effectively utilize big data for decision-making and improve urban operational efficiency and residents' quality of life is a major challenge currently faced.

[0003] Traditional urban operations decision-making methods are often based on limited data and empirical judgment, making them unable to fully and accurately reflect the complexity and dynamic nature of cities. The emergence of big data technology has provided richer and more accurate data support for urban operations decision-making. However, how to extract effective information from massive amounts of data and conduct scientific decision-making analysis remains a pressing issue. Summary of the Invention

[0004] The technical problem to be solved by the present invention is a smart city operation decision-making method and system based on big data. The present invention forms a complete smart city operation decision-making method through advanced technologies such as deep belief network, principal component analysis, and multi-objective optimization algorithm.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] In a first aspect, a smart city operation decision-making method based on big data is provided, the method comprising:

[0007] Acquire multi-source datasets and generate a deep belief network based on the multi-source datasets;

[0008] The noise feature of the data is extracted through the deep belief network to obtain the noise feature set;

[0009] Calculate the scalar probability value for the noise feature set, and obtain the target data set based on the scalar probability value;

[0010] The target data set is fused through principal component analysis to obtain fused data;

[0011] Perform decision strategy analysis on the fused data to obtain multiple decision strategy data sets;

[0012] Optimize multiple decision strategy data sets through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets;

[0013] Generate city operation plans based on multiple optimization strategy datasets.

[0014] Furthermore, a multi-source dataset is obtained, and a deep belief network is generated based on the multi-source dataset, including:

[0015] Acquire a multi-source data set, and preprocess the acquired multi-source data set;

[0016] Initialize the deep belief network parameter values based on the preprocessed data. In each iteration, randomly extract a portion of samples from the preprocessed data set, calculate the gradient of the loss function based on the extracted samples, and update the deep belief network parameters to minimize the loss function. Repeat the steps until the preset number of iterations is reached to obtain the deep belief network. The loss function is the gap between the predicted value and the actual value, which is used to measure the similarity between the data generated by the deep belief network and the real data.

[0017] The deep belief network is tuned to obtain a tuned deep belief network.

[0018] Furthermore, the noise feature of the data is extracted through the deep belief network to obtain the noise feature set, including:

[0019] The deep belief network extracts noise features by analyzing the activation values of the hidden layer to extract noise features;

[0020] The extracted noise features are integrated to form a noise feature set.

[0021] Furthermore, a scalar probability value is calculated for the noise feature set, and a target data set is obtained based on the scalar probability value, including:

[0022] By extracting the noise feature set, multiplying the feature value of each sample by the corresponding weight and summing them, adding the intercept term, and obtaining the linear combination value, the linear combination value is mapped to the probability space through the logistic function to obtain the scalar probability value;

[0023] Set a threshold and compare the scalar probability value with the preset threshold to obtain the target data set.

[0024] Furthermore, the target dataset is fused through principal component analysis to obtain fused data, including:

[0025] Preprocess the target data set and calculate the covariance matrix of the target data set;

[0026] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and corresponding eigenvectors;

[0027] According to the size of the eigenvalue, the principal component is selected, and the original data is projected onto the selected principal component to obtain the coordinates of each data point in the new coordinate system to realize data projection;

[0028] The projected data are fused to obtain fused data.

[0029] Furthermore, the fused data is analyzed for decision strategies to obtain multiple decision strategy data sets, including:

[0030] Initialize the decision strategy network and environment model based on the characteristics of the fused data;

[0031] Collect data samples by performing multiple interactions in the environment model through the decision-making policy network;

[0032] Calculate the advantage function value based on each collected data sample;

[0033] Update the strategy model based on the advantage function value and the preset cropping ratio;

[0034] Iteratively generate data samples based on the updated policy network to obtain iterative optimization;

[0035] Based on iterative optimization, multiple decision strategy data sets are generated.

[0036] Furthermore, multiple decision-making strategy data sets are optimized through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets and generate urban operation plans, including:

[0037] Extracting strategy features from each decision strategy data set to obtain a strategy feature set for each decision strategy data set, and inputting the data into a multi-objective optimization algorithm for optimization target matching to obtain an optimization target set;

[0038] According to the optimization target set, the optimization variables of the optimization target set are matched to obtain the optimization variable set;

[0039] According to the optimization variable set, the strategy feature set of each decision strategy data set is feature encoded by a multi-objective optimization algorithm to obtain the encoded feature set of each decision strategy data set;

[0040] According to the coding feature set, the coding feature set of each decision strategy data set is iteratively optimized through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets;

[0041] Generate city operation plans based on multiple optimization strategy datasets.

[0042] The second aspect is a smart city operation decision-making system based on big data, including:

[0043] An acquisition module is used to acquire multi-source data sets and generate a deep belief network based on the multi-source data sets; extract noise features from the data through the deep belief network to obtain a noise feature set; calculate a scalar probability value for the noise feature set, and acquire a target data set based on the scalar probability value;

[0044] The processing module is used to perform data fusion on the target data set through principal component analysis to obtain fused data; perform decision strategy analysis on the fused data to obtain multiple decision strategy data sets;

[0045] The optimization module is used to optimize multiple decision-making strategy data sets through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets; and generate urban operation plans based on the multiple optimized strategy data sets.

[0046] According to a third aspect, a computing device includes:

[0047] one or more processors;

[0048] The storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0049] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0050] The above solution of the present invention includes at least the following beneficial effects:

[0051] By generating a deep belief network to extract noise features from multi-source data sets, the noise and outliers in the data are effectively removed, and the quality and availability of the data are significantly improved; the scalar probability value is calculated for the noise feature set; the target data set is fused using principal component analysis, fully considering the correlation and uncertainty between the data, making the fused data more comprehensive and accurate; decision-making strategy analysis is performed based on the fused data to generate multiple decision-making strategy data sets, and the decision-making strategy data sets are optimized through a multi-objective optimization algorithm, which improves the overall quality and comprehensive benefits of the decision; the urban operation plan generated based on the optimized strategy data set can accurately analyze and reasonably allocate urban resources, avoid waste and idleness of resources, and improve resource utilization efficiency; by continuously optimizing decision-making strategies, the present invention can support smart cities in maintaining efficient, stable and sustainable operation while developing rapidly. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a smart city operation decision-making method based on big data provided by an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram of a smart city operation decision-making system based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention may be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0055] like Figure 1 As shown, an embodiment of the present invention proposes a smart city operation decision-making method based on big data, which includes the following steps:

[0056] Step 11, obtaining a multi-source dataset and generating a deep belief network based on the multi-source dataset;

[0057] Step 12: extract noise features from the data using a deep belief network to obtain a noise feature set;

[0058] Step 13: Calculate a scalar probability value for the noise feature set, and obtain a target data set based on the scalar probability value;

[0059] Step 14: perform data fusion on the target data set through principal component analysis to obtain fused data;

[0060] Step 15: Analyze the fused data for decision strategies to obtain multiple decision strategy data sets;

[0061] Step 16, performing strategy optimization on the multiple decision strategy data sets through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets;

[0062] Step 17: Generate a city operation plan based on multiple optimization strategy data sets.

[0063] In an embodiment of the present invention, the acquisition of multi-source data sets integrates data from different channels and dimensions, fully explores the value of data, and generates a deep belief network; by generating a deep belief network, noise features are extracted to mine valuable information hidden behind the noise from complex data; the scalar probability value of the noise feature set is calculated and the target data set is obtained, and the data is accurately screened and located, thereby improving the quality and availability of the data; the target data sets from different data sources are organically combined through principal component analysis, and the correlation and uncertainty between the data are fully considered; decision strategy analysis is performed on the fused data to generate multiple decision strategy data sets, and the most suitable decision plan is selected based on the actual situation and development needs of the city; multiple decision strategy data sets are optimized through a multi-objective optimization algorithm; based on multiple optimization strategy data sets, urban resources are accurately analyzed and reasonably allocated to achieve efficient resource utilization and generate scientific, reasonable and efficient urban operation plans.

[0064] In a preferred embodiment of the present invention, the above step 11 may include:

[0065] Step 111: obtaining a multi-source data set and preprocessing the obtained multi-source data set;

[0066] Step 112: Initialize the deep belief network parameter values based on the preprocessed data. In each iteration, randomly extract a portion of samples from the preprocessed data set, calculate the gradient of the loss function based on the extracted samples, and update the deep belief network parameters to minimize the loss function. Repeat the steps until a preset number of iterations are reached to obtain a deep belief network. The loss function is the difference between the predicted value and the actual value.

[0067] Step 113: tune the deep belief network to obtain a tuned deep belief network.

[0068] In an embodiment of the present invention, multi-source datasets are preprocessed, standardized, and normalized to make the data more accurate and reliable. Fusion of these datasets provides richer features for the deep belief network, enabling the model to learn more complex patterns and regularities. By initializing the deep belief network parameters and calculating the gradient of the loss function to update the parameters, the model can gradually adapt to the data distribution, thereby improving the model's predictive accuracy. By properly setting the number of iterations, model quality can be guaranteed and training efficiency improved.

[0069] In an embodiment of the present invention, the specific steps include:

[0070] Step 111 , determining different sources for obtaining data, collecting data according to the determined data sources, integrating and preprocessing the data from different data sources to obtain a preprocessed multi-source data set.

[0071] Step 112, based on the preprocessed multi-source data set, initialize the parameter values of the deep belief network. In each iteration, randomly extract a portion of samples from the preprocessed training set to form a mini-batch of data; use the extracted mini-batch data to calculate the gradient of the loss function with respect to the parameters of the deep belief network; update the parameters of the deep belief network to minimize the loss function; repeat the above steps of random sampling and parameter updating until the preset number of iterations is reached to obtain the deep belief network.

[0072] Step 113, according to the specific task type of the deep belief network, during the tuning process, after each model tuning is completed, the validation set is used to evaluate the tuned model, and the specific values of the tuning indicators during each evaluation are recorded in detail. Pay attention to the changing trend of the tuning indicators as the tuning process progresses, select the optimal model as the tuned deep belief network, and use the test set for the final test to obtain a tuned deep belief network with good performance.

[0073] In a preferred embodiment of the present invention, the above step 12 may include:

[0074] Step 121, the deep belief network extracts noise features by analyzing activation values of the hidden layer to extract noise features;

[0075] In step 122, the extracted noise features are integrated to form a noise feature set.

[0076] In an embodiment of the present invention, the deep belief network extracts noise features by analyzing the data activation values, which helps us to more deeply analyze the distribution, type characteristics and intensity of the noise in the data. With the accurate grasp of the noise features, it can provide more accurate and effective guidance for the subsequent data processing process and model training. Integrating the extracted noise features to form a noise feature set can enhance the robustness of the features, so that the features remain stable and reliable under different data environments and interferences, and can provide additional supervision information for model training, thereby improving the generalization ability of the model.

[0077] In an embodiment of the present invention, the specific steps include:

[0078] In step 121, according to the deep belief network, activation values refer to the outputs of the hidden layer. The mean, variance, or higher-order moments of the activation values are calculated. Abnormal statistics indicate noise. By observing how activation values change under different inputs, areas with significant changes may correspond to noise characteristics. In specific implementations, forward propagation is performed on a batch of input data, the activation values of the hidden layer are recorded, and statistics are used to screen for abnormal characteristics.

[0079] In step 122, the noise features of different layers or different samples are concatenated by dimension to form a high-dimensional feature vector. The features are compressed by statistics (such as mean, maximum value) or attention mechanism. The noise features are extracted for each sample to form a feature vector. The feature vectors of all samples are stacked row by row to form a feature matrix, thereby forming a noise feature set.

[0080] In a preferred embodiment of the present invention, the above step 13 may include:

[0081] Step 131, by extracting the noise feature set, the feature values x1, x2..., x m and the corresponding weights β1,β2...β m Perform weighted summation and combine with intercept term β0, through Calculate the scalar probability value corresponding to each feature data, where P is the scalar probability value, x i is the characteristic value of the sample, β0 is the intercept term, β i is related to each feature x i The corresponding weight, i is the index;

[0082] Step 132 : Setting a threshold, comparing the scalar probability value with the preset threshold to obtain a target data set.

[0083] In an embodiment of the present invention, by calculating the scalar probability value corresponding to each feature data and setting a threshold for comparison, the target data set that meets specific conditions can be accurately screened out from the original data, thereby improving the accuracy and efficiency of data screening. The calculation of the scalar probability value and the threshold comparison process can effectively identify and remove noise and outliers in the data. By setting a reasonable threshold, data with a scalar probability value lower or higher than the threshold is regarded as abnormal data and eliminated, which can ensure the purity of the target data set and improve the quality and reliability of the data; the target data set is high-quality data that has been screened and optimized, which improves the training effect of the model. The high-quality target data set helps the model learn the true distribution of the data and enhances the generalization ability of the model. In the process of calculating the scalar probability value, parameters such as eigenvalues, intercept terms and weights are involved. By analyzing and adjusting these parameters, we can gain an in-depth understanding of the degree of influence of each feature on the target result, thereby optimizing the model parameters in a targeted manner. By setting a reasonable threshold and comparing the scalar probability values, we can promptly discover potential risk signals in the data.

[0084] In an embodiment of the present invention, the specific steps include:

[0085] Step 131: By extracting the noise feature set, multiplying the eigenvalue of each sample by the corresponding weight and summing them, adding the intercept term, obtaining the linear combination value, mapping the linear combination value to the probability space through the logic function, and obtaining the scalar probability value. The process includes: setting the weight and intercept, traversing the samples, calculating the linear combination value, substituting the linear combination value into the logic function, obtaining the probability value, and obtaining the probability value vector of all samples.

[0086] Step 132: Set a threshold θ∈[0,1] based on business requirements or data characteristics. For strict noise filtering, set θ=0.8; for looser filtering, set θ=0.5. Determine the optimal threshold through ROC curve analysis. For each sample's probability value, if the probability value ≥ θ, retain the sample; otherwise, discard it. The filtered samples constitute the target dataset.

[0087] In a preferred embodiment of the present invention, the above step 14 may include:

[0088] Step 141, preprocessing the target data set and calculating the covariance matrix of the target data set;

[0089] Step 142, performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors;

[0090] Step 143 , selecting a principal component based on the size of the eigenvalue, and projecting the original data onto the selected principal component, obtaining the coordinates of each data point in the new coordinate system to achieve data projection;

[0091] Step 144 , fusing the projected data to obtain fused data.

[0092] In an embodiment of the present invention, through preprocessing steps such as data cleaning and standardization, missing values, outliers, etc. in the target data set can be removed to improve the quality and consistency of the data. Eigenvalue decomposition is the core step of PCA. By decomposing the covariance matrix, a set of orthogonal eigenvectors can be obtained. These vectors constitute a new coordinate system, namely the principal component space. The size of the eigenvalue reflects the amount of variance explained by each principal component. By selecting a sufficient number of principal components, we can reduce the dimension of the data without significantly losing information. These principal components contain most of the information in the original data. The data projected onto the principal component space removes the redundancy and noise in the original data, making the data more concise and accurate. By fusing the projected data, we can integrate information from multiple data sources into a unified data set. This helps us understand the characteristics and laws of the data more comprehensively.

[0093] In an embodiment of the present invention, the specific steps include:

[0094] Step 141: Preprocess the target dataset, including data cleaning and standardization, and calculate the covariance matrix of the target dataset. The covariance matrix reflects the degree of linear correlation between different features.

[0095] Step 142: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors. The eigenvalues represent the amount of variance explained by each principal component, while the eigenvectors indicate the direction of each principal component. By decomposing the covariance matrix, a set of orthogonal eigenvectors can be obtained. These vectors form a new coordinate system, namely the principal component space.

[0096] Step 143: Based on the size of the eigenvalues, principal components are selected. The number of principal components selected depends on the desired proportion of explained variance (e.g., 90%, 95%, etc.). Principal components retain the main information in the data while removing redundancy and noise. By selecting a sufficient number of principal components, the dimensionality of the data can be reduced without significantly losing information. The original data is projected onto the selected principal components. Through projection, a representation of each data point in the new coordinate system can be obtained. These representations are low-dimensional approximations of the original data, and the coordinates of each data point in the new coordinate system are obtained.

[0097] Step 144: The projected data are fused by simple averaging.

[0098] In a preferred embodiment of the present invention, the above step 15 may include:

[0099] Step 151, initializing the decision strategy network and the environment model according to the characteristics of the fused data;

[0100] Step 152, the decision strategy network is used to interact multiple times in the environment model to collect data samples; according to each collected data sample, Calculate the advantage function value, where is the advantage function value at time step t, k is the time step index, γ is the discount coefficient, the discount coefficient is 0 or 1, r t+k is the immediate reward obtained at time step t+k, n is a fixed number of steps, is the estimated value of the state value function at time step t+n, is the estimated value of the state value function at time step t;

[0101] Step 153: updating the policy model based on the advantage function value and the preset cropping ratio;

[0102] Step 154, iteratively generating data samples according to the updated policy network to obtain iterative optimization;

[0103] Step 155: Generate multiple decision strategy data sets based on iterative optimization.

[0104] In this embodiment of the present invention, the decision policy network and environment model are initialized based on the characteristics of the fused data, accurately matching the data distribution and characteristics of the actual problem. This precise initialization improves the efficiency and quality of decision policy optimization. By repeatedly interacting with the environment model to collect data samples, the decision policy network can efficiently obtain a large amount of effective decision-related information, improve the quality and utilization of data samples, and accelerate the decision policy optimization process. The advantage function value is calculated, providing a clear optimization direction for updating the policy model. The advantage function value measures the superiority or inferiority of taking a specific action at a certain time step relative to the average level, improving the efficiency and effectiveness of policy optimization. Updating the policy model based on the advantage function value and a preset clipping ratio ensures stable updates of the policy model, helping to improve its robustness and reliability. Iteratively generating data samples based on the updated policy network enables iterative optimization of the decision policy. This iterative optimization process enables the decision policy to continuously approach the optimal solution, improving its performance and quality. Generating multiple decision policy datasets based on iterative optimization provides diverse decision options for practical applications. These diverse decision policy datasets can be used for comparative analysis and risk assessment, helping decision makers better understand the advantages and disadvantages of different strategies and make more informed decisions. Multiple decision strategy datasets enhance the adaptability and robustness of decisions in different scenarios.

[0105] In an embodiment of the present invention, the specific steps include:

[0106] Step 151: Analyze the distribution, dimension and dynamic characteristics of the fused data, construct a neural network whose input is the environment state and output is the action probability distribution, initialize the network parameters (such as weights and biases), and use random initialization or pre-training based on domain knowledge to define the state transfer function and reward function of the environment. The environment model is a real environment simulator.

[0107] Step 152: In the initial state of the environment, the policy network generates an action based on the current state. If the policy is a random policy, the action is sampled from the probability distribution. If the policy is a deterministic policy, the action is directly output and executed. The environmental model returns the next state and immediate reward. The state, action, reward, and next state during the interaction process are recorded as data samples. The above process is repeated to collect multiple trajectories. Each trajectory contains samples of multiple time steps. Based on the samples, the advantage function value is calculated by the formula.

[0108] In step 152, the value of the discount coefficient γ directly affects the calculation logic of the advantage function. According to the value range of γ, the following two algorithm modes are defined:

[0109] When γ = 0, it is a single-step TD algorithm, which only considers the immediate reward r t+0 , ignoring future rewards, the formula degenerates into a single-step time difference error of It is suitable for short-term decision optimization that requires fast response, such as real-time adjustment of traffic lights). The model quickly iterates strategies through immediate feedback. is the advantage function value at time step t, is the estimated value of the state value function at time step t, r t It’s an immediate reward;

[0110] When γ = 1, it is a multi-step TD or Monte Carlo approximation algorithm, which accumulates the estimated value of the reward and state value function for the next n steps. When γ = 1, the formula is simplified to This form combines the "full round return" idea of the Monte Carlo method, where the "full round return" idea is specifically truncated by a fixed number of steps n, but through the value function V φ Estimate unobserved states in advance to avoid relying on complete episodes. This approach is suitable for scenarios that require balancing long-term benefits and computational efficiency, such as energy scheduling optimization. Variance can be reduced by truncating with a fixed number of steps.

[0111] Step 153: According to the advantage function value, set the clipping ratio to between 0.1 and 0.3, and limit the range of the probability ratio of the new and old strategies to [1-0.3, 1+0.3]. If the probability ratio of the new and old strategies exceeds this range, clipping is performed to avoid over-update. For each sample, the probability ratio and clipping ratio are calculated, and the strategy parameters are updated to maximize the objective function.

[0112] In step 154, the updated policy network is used to interact with the environment model again to generate new data samples, and step 152 is repeated to collect samples of multiple trajectories. The new samples are used to calculate the advantage function value and the policy network is updated again.

[0113] Step 155: After the iterative optimization is completed, the final policy network is used to run multiple times in the environment. Each run generates a trajectory, recording the state, action, and reward sequence to form a decision policy dataset. The dataset contains a variety of policy behaviors and reflects the performance of the policy under different initial conditions.

[0114] In a preferred embodiment of the present invention, the above steps 16 to 17 may include:

[0115] Step 161: extracting strategy features from each decision strategy data set to obtain a strategy feature set for each decision strategy data set, and inputting the data into a multi-objective optimization algorithm for optimization target matching to obtain an optimization target set;

[0116] Step 162: performing optimization variable matching on the optimization target set according to the optimization target set to obtain an optimization variable set;

[0117] Step 163: feature encoding the strategy feature set of each decision strategy data set using a multi-objective optimization algorithm based on the optimization variable set to obtain a coded feature set of each decision strategy data set;

[0118] Step 164, iteratively optimizing the encoding feature set of each decision strategy data set using a multi-objective optimization algorithm based on the encoding feature set to obtain multiple optimized strategy data sets;

[0119] Step 17: Generate a city operation plan based on multiple optimization strategy data sets.

[0120] In an embodiment of the present invention, strategy feature extraction is performed on each decision strategy dataset, accurately identifying the core features of each decision strategy. Accurate strategy feature extraction helps avoid missing important information during the optimization process and improves the accuracy and effectiveness of optimization results. The strategy feature set is input into a multi-objective optimization algorithm for optimization goal matching, resulting in an optimization target set, which clarifies the optimization direction and objectives. The multi-objective optimization algorithm can determine the performance of each decision strategy dataset on these objectives based on the strategy features, allowing for targeted optimization. This clear optimization goal orientation makes the optimization process more focused and improves optimization efficiency and effectiveness. Optimization variable matching is performed on the optimization target set based on the optimization target set to obtain an optimization variable set, which can rationally identify the key factors influencing the optimization target. Proper optimization variable matching helps improve the relevance and effectiveness of optimization and avoid wasting resources on unimportant factors. Based on the optimization variable set, the multi-objective optimization algorithm performs feature encoding on the strategy feature set of each decision strategy dataset, generating an encoded feature set. Feature encoding converts complex strategy features into computer-processable numerical form, facilitating subsequent optimization calculations. Efficient feature encoding improves data processing efficiency and accuracy, providing better input data for the optimization algorithm. Based on the coded feature set, a multi-objective optimization algorithm is used to iteratively optimize the coded feature set of each decision strategy dataset, resulting in multiple optimized strategy datasets. This iterative optimization process continuously refines the decision strategy, achieving a better balance across multiple optimization objectives. This continuous iterative optimization refines the decision strategy and improves the scientific and rational nature of the decision. Based on these multiple optimized strategy datasets, a city operation plan is generated. High-quality city operation plans help improve city management and operational efficiency, enhance city competitiveness, and enhance the quality of life for residents.

[0121] In an embodiment of the present invention, the specific steps include:

[0122] Step 161: Clean each decision strategy dataset and extract key features based on the nature of the decision strategy. Integrate the features of each decision strategy dataset into a strategy feature set, define the core objectives of urban operations, and convert them into quantifiable optimization objectives. Use a multi-objective optimization algorithm to match the strategy feature set with the optimization objectives to generate an optimization target set.

[0123] Step 162: Analyze the optimization objective set and identify key variables that influence achievement. For example, if the optimization objective is "reducing traffic congestion," key variables might include "signal timing," "lane allocation ratio," and "public transportation frequency." If the optimization objective is "reducing energy consumption," key variables might include "equipment runtime," "energy purchase price," and "energy storage system capacity." Combine all key variables into an optimization variable set, defining a range or constraints for each optimization variable. For example, the range for signal timing is 30 to 120 seconds, and the energy purchase price must be "no more than 120% of the market average."

[0124] Step 163: Convert the strategy feature set into an encoding format suitable for processing by the multi-objective optimization algorithm to facilitate subsequent iterative optimization. Real number encoding is used, retaining the feature values directly, using integer encoding or one-hot encoding. For example, the lane allocation ratio "high / medium / low" can be encoded as {0, 1, 2}. Encode each strategy feature set using one-hot encoding or label encoding to generate an encoded feature set.

[0125] Associate the encoding feature set with the optimization variable set, ensuring that each encoding feature corresponds to one or more optimization variables.

[0126] Step 164: Select a suitable multi-objective optimization algorithm, consider the algorithm's convergence, diversity, and computational efficiency, randomly generate a set of initial solutions, and calculate the objective function value of each solution based on the optimization target set. The objective function value is to minimize the cost and maximize the benefit. For example, for the traffic congestion optimization objective function, the optimization goal is to minimize the average travel time and minimize the proportion of congested sections. Calculate the objective function value of each solution, where N is the total number of vehicles in the road network, L j is the length of the vehicle j’s travel path, is the average speed of vehicle j, S j is the set of traffic lights that vehicle j passes through, δ jk If vehicle j stops at traffic light k, it is 1, otherwise it is 0, T r (x) is the red light duration of the traffic light, M is the total number of road sections in the road network, v i (x)) is the average speed of vehicles on road section i, v 阈值is the congestion speed threshold, and f(x) is the minimum average travel time and the minimum congested road section ratio.

[0127] Perform non-dominated sorting based on the objective function value to determine the quality of the solution, select, crossover and mutate: retain high-quality solutions through selection operations, generate new solutions through crossover and mutation operations, repeat the above steps until the termination condition is met, decode the iteratively optimized encoded feature set into the original feature space, and generate an optimization strategy data set.

[0128] Step 17: Conduct a comprehensive analysis of multiple optimization strategy datasets, evaluate each strategy's performance under different optimization objectives, rank the strategies using a weighted summation, and select the strategy with the best overall performance. Based on the actual city operations, develop a feasible operational plan. For example, if the optimization objective is "reducing traffic congestion," solutions might include "adjusting traffic light timing," "adding tidal lanes," and "promoting shared mobility." If the optimization objective is "reducing energy consumption," solutions might include "upgrading energy-saving equipment," "optimizing energy procurement strategies," and "building a distributed energy storage system."

[0129] like Figure 2 As shown, an embodiment of the present invention further provides a smart city operation decision-making system based on big data, comprising:

[0130] An acquisition module is used to acquire multi-source data sets and generate a deep belief network based on the multi-source data sets; extract noise features from the data through the deep belief network to obtain a noise feature set; calculate a scalar probability value for the noise feature set, and acquire a target data set based on the scalar probability value;

[0131] The processing module is used to perform data fusion on the target data set through principal component analysis to obtain fused data; perform decision strategy analysis on the fused data to obtain multiple decision strategy data sets;

[0132] The optimization module is used to optimize multiple decision-making strategy data sets through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets; and generate urban operation plans based on the multiple optimized strategy data sets.

[0133] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0134] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A smart city operation decision-making method based on big data, characterized by: The method comprises: Acquire multi-source datasets and generate a deep belief network based on the multi-source datasets; The noise feature of the data is extracted through the deep belief network to obtain the noise feature set; Calculate the scalar probability value for the noise feature set, and obtain the target data set based on the scalar probability value; The target data set is fused through principal component analysis to obtain fused data; Perform decision strategy analysis on the fused data to obtain multiple decision strategy data sets; Optimize multiple decision strategy data sets through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets; Generate city operation plans based on multiple optimization strategy datasets.

2. The smart city operation decision-making method based on big data according to claim 1 is characterized in that: Obtain multi-source datasets and generate deep belief networks based on the multi-source datasets, including: Acquire a multi-source data set, and preprocess the acquired multi-source data set; Initialize the deep belief network parameter values based on the preprocessed data. In each iteration, randomly extract a portion of samples from the preprocessed data set, calculate the gradient of the loss function based on the extracted samples, and update the deep belief network parameters to minimize the loss function. Repeat the steps until the preset number of iterations is reached to obtain the deep belief network. The loss function is the difference between the predicted value and the actual value. The deep belief network is tuned to obtain a tuned deep belief network.

3. The smart city operation decision-making method based on big data according to claim 1 is characterized in that: The noise feature of the data is extracted through the deep belief network to obtain the noise feature set, including: The deep belief network extracts noise features by analyzing the activation values of the hidden layer to extract noise features; The extracted noise features are integrated to form a noise feature set.

4. The smart city operation decision-making method based on big data according to claim 1 is characterized in that: Calculate the scalar probability value for the noise feature set, and obtain the target data set based on the scalar probability value, including: By extracting the noise feature set, multiplying the feature value of each sample with the corresponding weight and summing them, adding the intercept term, and obtaining the linear combination value, the linear combination value is mapped to the probability space through the logic function to obtain the scalar probability value; Set a threshold and compare the scalar probability value with the preset threshold to obtain the target data set.

5. The smart city operation decision-making method based on big data according to claim 1 is characterized in that: The target data set is fused through principal component analysis to obtain fused data, including: Preprocess the target data set and calculate the covariance matrix of the target data set; Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and corresponding eigenvectors; According to the size of the eigenvalue, the principal component is selected, and the original data is projected onto the selected principal component to obtain the coordinates of each data point in the new coordinate system to realize data projection; The projected data are fused to obtain fused data.

6. The smart city operation decision-making method based on big data according to claim 1 is characterized in that: The fused data is analyzed for decision strategies to obtain multiple decision strategy data sets, including: Initialize the decision strategy network and environment model based on the characteristics of the fused data; Collect data samples by performing multiple interactions in the environment model through the decision-making policy network; Calculate the advantage function value based on each collected data sample; Update the strategy model based on the advantage function value and the preset cropping ratio; Iteratively generate data samples based on the updated policy network to obtain iterative optimization; Based on iterative optimization, multiple decision strategy data sets are generated.

7. The smart city operation decision-making method based on big data according to claim 1 is characterized in that: Multiple decision-making strategy data sets are optimized through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets and generate urban operation plans, including: Extracting strategy features from each decision strategy data set to obtain a strategy feature set for each decision strategy data set, and inputting the data into a multi-objective optimization algorithm for optimization target matching to obtain an optimization target set; According to the optimization target set, the optimization variables of the optimization target set are matched to obtain the optimization variable set; According to the optimization variable set, the strategy feature set of each decision strategy data set is feature encoded by a multi-objective optimization algorithm to obtain the encoded feature set of each decision strategy data set; According to the coding feature set, the coding feature set of each decision strategy data set is iteratively optimized through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets; Generate city operation plans based on multiple optimization strategy datasets.

8. A smart city operation decision-making system based on big data, which implements the method according to any one of claims 1 to 7, characterized in that: include: An acquisition module is used to acquire multi-source data sets and generate a deep belief network based on the multi-source data sets; The noise feature of the data is extracted through the deep belief network to obtain the noise feature set; Calculate the scalar probability value for the noise feature set, and obtain the target data set based on the scalar probability value; A processing module is used to fuse the target data set through principal component analysis to obtain fused data; Perform decision strategy analysis on the fused data to obtain multiple decision strategy data sets; The optimization module is used to optimize multiple decision-making strategy data sets through a multi-objective optimization algorithm to obtain multiple optimized strategy data sets; and generate urban operation plans based on the multiple optimized strategy data sets.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images