A driving control system and method based on voice recognition

Abstract

The present invention discloses a driving control system and method based on speech recognition, which belongs to the field of control technology. The operator's voice command data is collected through a microphone array, and the voice command data is denoised to obtain the denoised voice command data. Then, a deep learning algorithm is used to recognize the denoised voice command data, determine text command data, and parse the text command data to determine driving control keywords, and convert the driving control keywords into word vectors. Finally, based on the word vectors and driving status information, a reinforcement learning algorithm is used to determine driving control parameters. Based on the driving control parameters, the driving is controlled to perform operations, which can achieve accurate and rapid control of the driving, is applicable to a variety of scenarios, and improves driving control efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of control technology, and in particular relates to a vehicle driving control system and method based on speech recognition. Background Art

[0002] A crane generally refers to a gantry crane, also known as a bridge crane. This type of lifting equipment is installed horizontally in workshops, warehouses, and open-air storage yards to lift materials at a certain height. It consists of a trolley (bridge) and a small trolley. The trolley runs longitudinally along tracks laid on elevated platforms on both sides, while the small trolley runs horizontally on tracks on the gantry, forming a rectangular working area. Driving cranes offer high lifting capacity and high operational efficiency, making them widely used in machinery manufacturing, metallurgy, logistics, and other fields. They can be equipped with various lifting devices, such as hooks, grabs, and electromagnetic suction cups, to meet diverse material handling needs. Compared to other types of cranes, gantry cranes offer advantages such as wide coverage, high efficiency, and flexible operation, making them an indispensable piece of equipment in modern production. As a crucial material handling device in industrial production, the safety and efficiency of gantry crane operations are crucial. Traditional gantry crane control systems rely primarily on manual operation or simple remote control, which can be complex, inefficient, and pose safety risks. In recent years, with the rapid development of voice recognition technology, its application in industrial control has gradually attracted attention. However, the existing crane control system based on speech recognition still has shortcomings in recognition accuracy, anti-interference ability and system stability. Summary of the Invention

[0003] The present invention provides a crane control system and method based on voice recognition, which is used to solve the problems in the prior art that traditional crane control systems mainly rely on manual operation or simple remote control operation, and have complex operation and low efficiency.

[0004] In one aspect, the present invention provides a vehicle driving control system based on speech recognition, comprising:

[0005] A voice acquisition module is used to collect the operator's voice command data through a microphone array and perform denoising on the voice command data to obtain denoised voice command data;

[0006] A speech recognition module, configured to recognize the denoised speech instruction data using a deep learning algorithm to determine text instruction data;

[0007] An instruction parsing module, configured to parse the text instruction data, determine driving control keywords, and convert the driving control keywords into word vectors;

[0008] A reinforcement learning module, configured to determine driving control parameters using a reinforcement learning algorithm based on the word vectors and driving state information;

[0009] The driving control module is used to control the driving operation based on the driving control parameters and complete the driving control based on voice recognition.

[0010] In some possible implementations, the speech recognition module includes a speech recognition model deployment submodule, a speech data feature extraction submodule, and a speech feature recognition submodule;

[0011] The speech recognition model deployment submodule is used to pre-deploy a speech recognition model for speech recognition using a deep learning algorithm;

[0012] The voice data feature extraction submodule is used to extract features from the denoised voice command data to obtain voice features;

[0013] The speech feature recognition submodule is used to schedule the speech recognition model pre-deployed by the speech recognition model deployment submodule to recognize the speech features and determine the text instruction data.

[0014] In some possible implementations, a speech recognition model for speech recognition is pre-deployed using a deep learning algorithm, including:

[0015] Initializing hyperparameters of the deep learning algorithm using a chaotic mapping sequence to determine a plurality of different parameter vectors; wherein the parameter vectors include all or part of the hyperparameters to be optimized of the deep learning algorithm;

[0016] Obtaining a loss function value for each parameter vector, and determining the fitness corresponding to the parameter vector based on the loss function value;

[0017] Determining an optimal parameter vector and a worst parameter vector according to the fitness corresponding to the parameter vector;

[0018] Based on the optimal parameter vector, a variable spiral position selection mechanism is used to perform an initialization search on the parameter vector, and a parameter vector after the initialization search is determined;

[0019] For the parameter vector after the initialization search, according to the worst parameter vector, an adaptive probability response mechanism is used to perform a local and global balanced search on the parameter vector to determine the parameter vector after the balanced search;

[0020] Determine whether the current number of training times is greater than or equal to the preset maximum number of training times. If so, re-determine the optimal parameter vector based on the parameter vector after the balanced search; otherwise, return to the step of obtaining fitness;

[0021] Based on the re-determined optimal parameter vector, final hyperparameters of the deep learning algorithm are determined, and a speech recognition model for speech recognition is deployed based on the final hyperparameters of the deep learning algorithm.

[0022] In some possible implementations, based on the optimal parameter vector, a variable spiral position selection mechanism is used to perform an initialization search on the parameter vector, and determining the parameter vector after the initialization search includes:

[0023] Based on the current number of training times, the hyperbolic tangent function is used to determine the adaptive inertia weight;

[0024] weighting the optimal parameter vector using the adaptive inertia weight to determine a weighted optimal parameter vector;

[0025] For any parameter vector, obtaining a first difference vector between the optimal parameter vector and the parameter vector;

[0026] For any parameter vector, a variable spiral function is used to perform position transformation on the parameter vector to determine the parameter vector after position transformation;

[0027] The parameter vector after the initialization search is determined according to the weighted optimal parameter vector, the first difference vector, and the parameter vector after the position transformation.

[0028] In some possible implementations, for the parameter vector after the initialization search, performing a local and global balanced search on the parameter vector using an adaptive probabilistic response mechanism based on the worst parameter vector, and determining the parameter vector after the balanced search includes:

[0029] Obtain the fitness corresponding to each parameter vector after the initialization search, and obtain the degree of group dispersion based on the fitness corresponding to each parameter vector after the initialization search;

[0030] Obtaining a local development impact factor and a global development impact factor according to the degree of dispersion of the group;

[0031] Obtaining a single search rate according to the fitness corresponding to the parameter vector after each initialization search, and determining an average search rate during multiple training processes according to the single search rate;

[0032] Determining a local development threshold parameter and a global development threshold parameter according to the average search speed;

[0033] Determining a local development probability based on the local development influencing factor and the local development threshold parameter, and performing local development using a location transfer strategy based on the local development probability to determine a parameter vector after the local development;

[0034] Determining a global development probability based on the global development influencing factor and the global development threshold parameter, and performing global development using a chaotic global search strategy based on the global development probability to determine a parameter vector after global development;

[0035] For any parameter vector after initialization search, if the parameter vector has not been globally developed or locally developed, the parameter vector after the original initialization search is directly used as the parameter vector after the equilibrium search;

[0036] If the parameter vector is globally developed or locally developed, the parameter vector after global development or the parameter vector after local development is used as the parameter vector after equilibrium search;

[0037] If the parameter vector is globally developed and locally developed, the annealing simulation algorithm is used to select the parameter vector after global development or the parameter vector after local development as the parameter vector after the equilibrium search.

[0038] In some possible implementations, based on the local development probability, a position transfer strategy is used to perform local development, and a parameter vector after local development is determined, including:

[0039] Randomly generate a first decision factor between the interval (0, 1), and determine whether the first decision factor is less than the local development probability. If so, adopt a position transfer strategy to perform local development and determine a parameter vector after local development. Otherwise, do not perform local development.

[0040] Use the position transfer strategy for local development and determine the parameter vector after local development, including:

[0041] For any parameter vector after the initialization search, determine the transition probabilities corresponding to other parameter vectors according to the fitness corresponding to the parameter vector after the initialization search;

[0042] Determining a target collaboration vector for the parameter vector after the initialization search based on the transition probabilities corresponding to the other parameter vectors;

[0043] According to the target collaboration vector, the parameter vector after the initialization search is locally developed to determine the parameter vector after the local development.

[0044] In some possible implementations, based on the global development probability, a chaotic global search strategy is used to perform global development, and a parameter vector after global development is determined, including:

[0045] Randomly generate a second decision factor between the interval (0, 1), and determine whether the second decision factor is less than the global development probability. If so, use the chaotic global search strategy to perform global development and determine the parameter vector after global development. Otherwise, do not perform global development.

[0046] Use the chaos global search strategy for global development and determine the parameter vector after global development, including:

[0047] Get the chaotic mapping factor in the current training process;

[0048] Determine the search center vector according to the search upper limit vector and the search lower limit vector;

[0049] According to the chaotic mapping factor, the parameter vector after the initialization search is globally developed in combination with the sine function to determine the parameter vector after the global development.

[0050] In some possible implementations, performing feature extraction on the denoised voice command data to obtain voice features includes: extracting MFCC of the denoised voice command data to obtain voice features.

[0051] In some possible implementations, based on the word vector and the driving state information, a reinforcement learning algorithm is used to determine driving control parameters, including:

[0052] Collecting driving status information; wherein the status information includes position, posture and speed;

[0053] The word vector and the driving state information are constructed into a current state space, and the current state space is used as the input of the reinforcement learning algorithm to determine the selected action and obtain the driving control parameters;

[0054] Among them, the action is selected from the action space, each action represents a driving control parameter, and the action space should include all driving control parameters.

[0055] In another aspect, the present invention provides a vehicle control method based on speech recognition, comprising:

[0056] The operator's voice command data is collected through a microphone array, and the voice command data is subjected to denoising processing to obtain denoised voice command data;

[0057] Using a deep learning algorithm to recognize the denoised voice command data to determine text command data;

[0058] Parsing the text instruction data to determine driving control keywords, and converting the driving control keywords into word vectors;

[0059] Determining driving control parameters using a reinforcement learning algorithm based on the word vector and driving state information;

[0060] Based on the driving control parameters, the driving operation is controlled to complete the driving control based on voice recognition.

[0061] The present invention provides a driving control system and method based on speech recognition. The method collects the operator's voice command data through a microphone array, denoises the voice command data, and obtains the denoised voice command data. A deep learning algorithm is then used to recognize the denoised voice command data, determine text command data, and parse the text command data to determine driving control keywords, and convert the driving control keywords into word vectors. Finally, based on the word vectors and driving status information, a reinforcement learning algorithm is used to determine driving control parameters. Based on the driving control parameters, the driving operation is controlled, which can achieve accurate and rapid control of driving, is applicable to a variety of scenarios, and improves driving control efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0063] Figure 1 A schematic structural diagram of a vehicle control system based on speech recognition provided by an embodiment of the present invention.

[0064] Figure 2 A flow chart of a vehicle control method based on speech recognition provided by an embodiment of the present invention.

[0065] Among them, 101-speech acquisition module, 102-speech recognition module, 103-command parsing module, 104-reinforcement learning module, and 105-driving control module.

[0066] The above drawings illustrate specific embodiments of the present invention, which will be described in more detail below. These drawings and the accompanying description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0067] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0068] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0069] like Figure 1 As shown, an embodiment of the present invention provides a vehicle driving control system based on speech recognition, comprising:

[0070] The voice acquisition module 101 is configured to acquire the operator's voice command data through a microphone array and perform denoising on the voice command data to obtain denoised voice command data;

[0071] A speech recognition module 102 is configured to recognize the denoised speech instruction data using a deep learning algorithm to determine text instruction data;

[0072] The voice features of the voice instruction data after denoising may be extracted first, and then the voice features may be recognized to determine the text instruction data.

[0073] An instruction parsing module 103 is configured to parse the text instruction data, determine driving control keywords, and convert the driving control keywords into word vectors;

[0074] Driving control keywords refer to pre-set control keywords, such as forward, backward, descend, ascend, and various distance digital data.

[0075] A reinforcement learning module 104 is configured to determine driving control parameters using a reinforcement learning algorithm based on the word vector and driving state information;

[0076] In the case of aging of the driving motor or inaccurate instructions from the commander (such as not stating the forward distance), the driving vehicle may not be able to accurately reach the working position. The embodiment of the present invention determines the driving control parameters through a reinforcement learning algorithm, which can not only achieve effective control of the driving, but also select more accurate driving control parameters by learning long-term habits, so that the driving vehicle can accurately reach the working position in fixed working scenarios, effectively improving the driving operation efficiency.

[0077] The driving control module 105 is used to control the driving operation based on the driving control parameters and complete the driving control based on voice recognition.

[0078] In some possible implementations, the speech recognition module includes a speech recognition model deployment submodule, a speech data feature extraction submodule, and a speech feature recognition submodule;

[0079] The speech recognition model deployment submodule is used to pre-deploy a speech recognition model for speech recognition using a deep learning algorithm;

[0080] The voice data feature extraction submodule is used to extract features from the voice command data after denoising to obtain voice features; denoising is a relatively conventional technical means and will not be described in detail in the embodiment of the present invention.

[0081] The speech feature recognition submodule is used to schedule the speech recognition model pre-deployed by the speech recognition model deployment submodule to recognize the speech features and determine the text instruction data.

[0082] Deep learning algorithms generally cannot be used directly for speech recognition; they must first be trained on historical data to obtain a speech recognition model for speech recognition. Traditionally, gradient descent methods are used to update the hyperparameters of deep learning algorithms, which can cause the search process to fall into a local optimum, ultimately rendering the trained speech recognition model ineffective for speech recognition. Therefore, embodiments of the present invention provide the following pre-deployment algorithm to achieve global training while ensuring both training speed and precision, ultimately guaranteeing speech recognition accuracy.

[0083] In some possible implementations, a speech recognition model for speech recognition is pre-deployed using a deep learning algorithm, including:

[0084] A1. Initializing the hyperparameters of the deep learning algorithm using a chaotic mapping sequence to determine a plurality of different parameter vectors; wherein the parameter vectors include all or part of the hyperparameters to be optimized of the deep learning algorithm,

[0085] The above step A1 indicates that all or part of the hyperparameters of the deep learning algorithm can be adjusted during the training process, so that different optimization goals can be set according to actual needs.

[0086] The deep learning algorithm can be set up as a CNN-CTC algorithm. CTC (Connectionist Temporal Classification) uses a learning algorithm that directly calculates the posterior probability of the overall sequence output. This allows a single neural network to transform a speech signal into a corresponding sequence of words or other text units. Within the end-to-end CTC framework, convolutional neural networks (CNNs) have greater application potential. CTC combines all feature vectors of the entire speech signal into a large feature map as input. By stacking convolutional and pooling layers in a CNN (Convolutional Neural Network), the "receptive field" of the output layer neurons increases with the number of layers, enabling the network's final output to capture more contextual information between speech frames. Furthermore, regarding the selection of modeling units, CTC introduces "blank" labels, eliminating the need for strict classification of each speech frame. Therefore, more granular modeling units, such as syllables and Chinese characters, can be selected. Moreover, the convolution and downsampling operations unique to CNN not only compress the input feature maps but also integrate and filter the speech frame data, which can extract more essential features from the speech signal of large-granularity modeling units containing more complex information.

[0087] Initializing the hyperparameters of the deep learning algorithm using a chaotic mapping sequence may include:

[0088] For each hyperparameter of the deep learning algorithm, you can first randomly initialize it between the upper and lower limits of the hyperparameter (for example, the upper limit of the hyperparameter in the embodiment of the present invention is generally 1, and the lower limit is generally 0), and then encode the initialized hyperparameter into a vector to obtain a basic parameter vector;

[0089] Based on the basic parameter vector, a chaotic mapping sequence is used to obtain multiple different parameter vectors:

[0090]

[0091] in, Indicates the first i The parameter vector d dimensional hyperparameters, d =1,2,…,D, D represents the total dimension of hyperparameters, and when i=1, Indicates the first i The first of the basic parameter vectors d dimensional hyperparameters, Indicates the first i +1 parameter vector d dimensional hyperparameters, represents the first chaotic mapping parameter and is set to 0.4; represents the second chaotic mapping parameter and is set to 0.3; represents the sine function, represents pi, represents the first random number between the interval (0,1), represents the second random number between the interval (0,1), represents the third random number between the interval (0,1), Represents the fourth random number in the interval (0,1).

[0092] Compared to the existing Tent chaotic mapping method, the chaotic mapping method provided by the present invention can ensure that the parameter vectors in the population are extremely evenly distributed within the solution space. This even distribution helps ensure that the relative distances between parameter vectors are large, thus enabling the algorithm to better explore the search space. This helps accelerate the algorithm's convergence process and increases the probability of finding a solution within a limited number of iterations.

[0093] A2. Obtain the loss function value of each parameter vector, and determine the fitness corresponding to the parameter vector based on the loss function value;

[0094] For example, the fitness of the parameter vector can be obtained by adding the loss function value to a preset constant term (such as 0.001) and then taking the inverse.

[0095] A3. Determine the optimal parameter vector and the worst parameter vector according to the fitness corresponding to the parameter vector;

[0096] The greater the fitness, the closer the parameter vector is to the optimal solution in the solution space. Therefore, the parameter vector with the largest fitness is the optimal parameter vector, and the parameter vector with the smallest fitness is the worst parameter vector.

[0097] A4. Based on the optimal parameter vector, a variable spiral position selection mechanism is used to perform an initialization search on the parameter vector to determine the parameter vector after the initialization search;

[0098] A5. For the parameter vector after the initialization search, based on the worst parameter vector, an adaptive probability response mechanism is used to perform a local and global balanced search on the parameter vector to determine the parameter vector after the balanced search;

[0099] A6. Determine whether the current number of training times is greater than or equal to the preset maximum number of training times. If so, re-determine the optimal parameter vector based on the parameter vector after the balanced search. Otherwise, return to the step of obtaining the fitness.

[0100] A7. Determine final hyperparameters of the deep learning algorithm based on the re-determined optimal parameter vector, and deploy a speech recognition model for speech recognition based on the final hyperparameters of the deep learning algorithm.

[0101] Optionally, after each search of the parameter vector, an out-of-bounds process may be performed on the parameter vector to ensure the validity of the parameter vector.

[0102] In some possible implementations, based on the optimal parameter vector, a variable spiral position selection mechanism is used to perform an initialization search on the parameter vector, and determining the parameter vector after the initialization search includes:

[0103] Based on the current number of training times, the hyperbolic tangent function is used to determine the adaptive inertia weight:

[0104]

[0105] in, represents the adaptive inertia weight, represents the maximum value of the adaptive inertia weight, represents the minimum value of the adaptive inertia weight, represents the hyperbolic tangent function, represents the maximum number of training times, Indicates the current number of training times;

[0106] In order to reduce the probability of the algorithm entering the local optimal solution, the inertia weight provided in the embodiment of the present invention introduces a hyperbolic tangent function, which not only prevents the algorithm from entering the local optimal state too early in the early iteration, but also enhances the local search capability of the algorithm in the middle and late stages, and can more accurately find the global optimal solution, balancing the global and local search capabilities of the group.

[0107] The optimal parameter vector is weighted using the adaptive inertia weight, and the optimal parameter vector after weighting is determined to be: ;in, Indicates the t The optimal parameter vector during the training process;

[0108] For any parameter vector, the first difference vector between the optimal parameter vector and the parameter vector is obtained as: ;in, Indicates the t During the training j parameter vector, j =1,2,…,NP, NP represents the total number of parameter vectors, NP is an even number;

[0109] For any parameter vector, a variable spiral function is used to perform position transformation on the parameter vector, and the parameter vector after position transformation is determined to be: ;in, represents a natural constant, cos represents the cosine function, q represents the spiral shape constant, l represents a random variable spiral control factor between (-1,1);

[0110] According to the weighted optimal parameter vector, the first difference vector, and the parameter vector after position transformation, the parameter vector after initialization search is determined to be: ;in, Indicates the j The parameter vector after the initialization search, Represents the fifth random number between the interval (0,1);

[0111] The embodiment of the present invention adopts a variable spiral position selection mechanism to initialize the search for the parameter vector. Based on the position of the optimal parameter vector, combined with inertia weight and variable spiral search, it can have a larger search range in the early stage of the algorithm, and gradually search for more refined solutions around the optimal position in the later stage of the search, effectively improving the training performance of the algorithm.

[0112] In some possible implementations, for the parameter vector after the initialization search, performing a local and global balanced search on the parameter vector using an adaptive probabilistic response mechanism based on the worst parameter vector, and determining the parameter vector after the balanced search includes:

[0113] Obtain the fitness corresponding to the parameter vector after each initialization search, and according to the fitness corresponding to the parameter vector after each initialization search, obtain the degree of group dispersion: ;in, Indicates the degree of group dispersion. Indicates the n The fitness corresponding to the parameter vector after the initialization search, represents the average fitness of the parameter vector after the initialization search, n =1,2,…,NP;

[0114] According to the degree of group dispersion, the local development impact factor and the global development impact factor are obtained as follows:

[0115]

[0116]

[0117] in, represents the local development impact factor, represents the exponential function with the natural constant e as the base, Indicates the scaling factor, which can be set to 80; represents the global development impact factor;

[0118] The single search rate is obtained according to the fitness corresponding to the parameter vector after each initialization search, and the average search rate during multiple training processes is determined according to the single search rate:

[0119]

[0120]

[0121] in, represents the average search rate, It represents the single search rate corresponding to the sth time of being selected for local development or global development, N represents the number of times it was selected for local development or global development in the previous training process, represents the maximum value function, Indicates the n The parameter vector after the initialization search is selected for local development or global development for the sth time, Indicates the n The parameter vector after the initialization search is selected for local development or global development at the s-1th time;

[0122] According to the average search speed, the local development threshold parameter and the global development threshold parameter are determined as follows:

[0123]

[0124]

[0125] in, Indicates the n The local development threshold parameter corresponding to the parameter vector after the initialization search, Indicates the n The global development threshold parameter corresponding to the parameter vector after the initialization search, represents the average search rate for global development, Average search rate for local development;

[0126] According to the local development influencing factors and the local development threshold parameters, the local development probability is determined as: ;in, Indicates the n The local development probability corresponding to the parameter vector after the initialization search.

[0127] According to the local development probability, adopting a position transfer strategy to perform local development, and determining a parameter vector after the local development;

[0128] According to the global development impact factor and the global development threshold parameter, the global development probability is determined as: ;in, Indicates the n The global development probability corresponding to the parameter vector after the initialization search;

[0129] According to the global development probability, a chaotic global search strategy is used to perform global development and determine a parameter vector after global development;

[0130] For any parameter vector after initialization search, if the parameter vector has not been globally developed or locally developed, the parameter vector after the original initialization search is directly used as the parameter vector after the equilibrium search;

[0131] If the parameter vector is globally developed or locally developed, the parameter vector after global development or the parameter vector after local development is used as the parameter vector after equilibrium search;

[0132] If the parameter vector is globally developed and locally developed, the annealing simulation algorithm is used to select the parameter vector after global development or the parameter vector after local development as the parameter vector after the equilibrium search.

[0133] Since the parameter vector after local development will not have a sharp change in position, when the fitness of the parameter vector after global development is greater than the fitness of the parameter vector after local development, the parameter vector after global development is used as the parameter vector after the balanced search; if the fitness of the parameter vector after global development is less than the fitness of the parameter vector after local development, the probability acceptance method of the annealing simulation algorithm is used to accept the parameter vector after global development as the parameter vector after the balanced search. If it is not accepted, the parameter vector after local development is used as the parameter vector after the balanced search.

[0134] Through the above selection operations, the algorithm can have greater global search capabilities in the early and middle stages, and gradually use the parameter vector after local development as the parameter vector after balanced search in the later stages of the algorithm, which indicates that the algorithm gradually turns to local search, but retains a certain global search capability.

[0135] In some possible implementations, based on the local development probability, a position transfer strategy is used to perform local development, and a parameter vector after local development is determined, including:

[0136] Randomly generate a first decision factor between the interval (0, 1), and determine whether the first decision factor is less than the local development probability. If so, adopt a position transfer strategy to perform local development and determine a parameter vector after local development. Otherwise, do not perform local development.

[0137] Use the position transfer strategy for local development and determine the parameter vector after local development, including:

[0138] For any parameter vector after the initialization search, determine the transition probabilities corresponding to other parameter vectors according to the fitness corresponding to the parameter vector after the initialization search;

[0139]

[0140]

[0141] in, Indicates the m The parameter vector after the initialization search is k The transition probability of the parameter vector after the initialization search, m =1,2,…,NP, k =1,2,…,NP, and m Not equal to k ; Indicates the t During the training k The transfer strength factor of the parameter vector after the initialization search, Indicates the t During the training m The transfer strength factor of the parameter vector after the initialization search, Indicates the t During the training h The transfer strength factor of the parameter vector after the initialization search, h =1,2,…,NP, represents the attenuation factor between the interval (0,1), Indicates the t -1 training session m The transfer strength factor of the parameter vector after the initialization search, A finer parameter representing the transfer intensity factor, Indicates the t During the training m The fitness of the parameter vector after the initialization search;

[0142] 、 The calculation process and The calculation process of is similar and will not be described here.

[0143] Determining a target collaboration vector for the parameter vector after the initialization search based on the transition probabilities corresponding to the other parameter vectors;

[0144] According to the transition probability corresponding to the other parameter vectors, the roulette strategy can be used to m The parameter vector after the initialization search determines the target collaboration vector;

[0145] According to the target collaboration vector, the parameter vector after the initialization search is locally developed, and the parameter vector after the local development is determined to be:

[0146]

[0147] in, Indicates the t During the training m The parameter vector after the initialization search, Indicates the m The parameter vector after local development, represents the search step length, express The target collaboration vector, express and The Euclidean distance between

[0148] The embodiment of the present invention adopts a position transfer strategy for local development, which can effectively enable the parameter vector to learn information from other positions and complete local information interaction, thereby realizing local search. At the same time, during the search process, there will be a greater probability of searching for a better position, while retaining the possibility of searching for some intersection positions. It can improve the global search capability in the early and middle stages of the algorithm, and improve the search precision in the later stages of the algorithm.

[0149] In some possible implementations, based on the global development probability, a chaotic global search strategy is used to perform global development, and a parameter vector after global development is determined, including:

[0150] Randomly generate a second decision factor between the interval (0, 1), and determine whether the second decision factor is less than the global development probability. If so, use the chaotic global search strategy to perform global development and determine the parameter vector after global development. Otherwise, do not perform global development.

[0151] Use the chaos global search strategy for global development and determine the parameter vector after global development, including:

[0152] Get the chaotic mapping factor in the current training process:

[0153]

[0154] in, Indicates the t The chaotic mapping factor in the training process is randomly generated from the interval (0, 1) at the beginning of the training. Indicates the t +1 chaotic mapping factor during training, represents a constant term between the interval (0,1) and is set to 0.7;

[0155] According to the search upper limit vector and the search lower limit vector, the search center vector is determined as: ;in, represents the search center vector, represents the search upper limit vector, Represents the search lower limit vector;

[0156] The search upper limit vector is used to represent the vector composed of the upper limits of the hyperparameters to be trained, and the search lower limit vector is used to represent the vector composed of the lower limits of the hyperparameters to be trained; it is worth noting that the hyperparameters represented by each dimension of the quantity pointed to in the embodiment of the present invention should be fixed to facilitate encoding and decoding.

[0157] According to the chaotic mapping factor, the parameter vector after the initialization search is globally developed in combination with the sine function, and the parameter vector after the global development is determined to be:

[0158]

[0159] in, Indicates the t During the training g The parameter vector after the initialization search, Indicates the g The parameter vector after global development, means (0, ), express and The Euclidean distance between express and The Euclidean distance between .

[0160] The embodiment of the present invention adopts a chaotic global search strategy for global development, which can provide the algorithm with a powerful global search capability and help the algorithm escape from the local optimal solution.

[0161] Through the above-mentioned training process, the algorithm provided by the embodiment of the present invention can provide a relatively powerful global search capability in the early and middle stages of training, and in the later stages of the algorithm, it provides a relatively powerful local search capability, but retains the global search capability, so that the algorithm can still perform global search, thereby improving the training effect of the algorithm and ultimately ensuring the accuracy of speech recognition.

[0162] In some possible implementations, feature extraction is performed on the denoised voice command data to obtain voice features, including: extracting MFCC (Mel-Fequency Cepstral Coefficients) of the denoised voice command data to obtain voice features.

[0163] It is worth noting that the above MFCC is only a preferred example provided by the embodiment of the present invention. Any time domain feature or frequency domain feature of speech features can also be used as speech features to achieve similar effects.

[0164] In some possible implementations, based on the word vector and the driving state information, a reinforcement learning algorithm is used to determine driving control parameters, including:

[0165] Collecting driving status information; wherein the status information includes position, posture and speed;

[0166] The word vector and the driving state information are constructed into a current state space, and the current state space is used as the input of the reinforcement learning algorithm to determine the selected action and obtain the driving control parameters;

[0167] Optionally, the reinforcement learning algorithm can be set to the DQN (Deep Q-Network) algorithm.

[0168] Among them, the action is selected from the action space, each action represents a driving control parameter, and the action space should include all driving control parameters.

[0169] After the vehicle is controlled according to the driving control parameters, if no new voice commands are generated, the word vectors in the new state space are all set to 0.

[0170] The crane control parameters may include the forward rotation number parameter and the reverse rotation number parameter of the crane displacement motor, and may include the forward rotation number parameter and the reverse rotation number parameter of the spreader motor. Thus, the crane can be controlled according to the crane control parameters.

[0171] Optionally, a pressure sensor can be installed on the hook, and the operating speed of the crane can be adjusted according to the pressure sensing data. For example, when the crane is empty or lightly loaded, the speed can be appropriately increased to improve work efficiency, while when approaching the target position or lifting heavy objects, the speed needs to be reduced to achieve smooth docking and precise alignment.

[0172] After executing the action, an experience pool should be built, and then the network parameters of the policy network in the reinforcement learning algorithm should be updated based on the experience pool so that the policy network can perform more accurate actions.

[0173] The present invention provides a driving control system based on speech recognition. It collects the operator's voice command data through a microphone array, and after denoising the voice command data, obtains the denoised voice command data. Then, it uses a deep learning algorithm to recognize the denoised voice command data, determines the text command data, and parses the text command data to determine the driving control keywords, and converts the driving control keywords into word vectors. Finally, based on the word vectors and the driving status information, a reinforcement learning algorithm is used to determine the driving control parameters. Based on the driving control parameters, the driving is controlled to perform operations, which can achieve accurate and rapid control of the driving, is suitable for a variety of scenarios, and improves the driving control efficiency.

[0174] like Figure 2 As shown, an embodiment of the present invention provides a driving control method based on speech recognition, comprising:

[0175] S201, collecting operator's voice command data through a microphone array, and performing denoising processing on the voice command data to obtain denoised voice command data;

[0176] S202: using a deep learning algorithm to recognize the denoised voice command data to determine text command data;

[0177] S203, parsing the text instruction data, determining driving control keywords, and converting the driving control keywords into word vectors;

[0178] S204, determining driving control parameters using a reinforcement learning algorithm based on the word vector and driving state information;

[0179] S205: Based on the driving control parameters, control the driving to perform operations, and complete the driving control based on voice recognition.

[0180] A driving control method based on speech recognition provided in an embodiment of the present invention can be applied to the above-mentioned system technical solution. Its principles and beneficial effects are similar and will not be described in detail here.

[0181] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0182] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0183] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0184] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0185] Those skilled in the art will understand that all or part of the steps in implementing the above facts and methods can be completed by instructing relevant hardware through a program, and the program involved or the program can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: the corresponding method steps are then brought out, and the storage medium can be ROM / RAM, a disk, an optical disk, etc.

[0186] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A vehicle control system based on speech recognition, characterized in that: include: A voice acquisition module is used to collect the operator's voice command data through a microphone array and perform denoising on the voice command data to obtain denoised voice command data; A speech recognition module, configured to recognize the denoised speech instruction data using a deep learning algorithm to determine text instruction data; An instruction parsing module, configured to parse the text instruction data, determine driving control keywords, and convert the driving control keywords into word vectors; A reinforcement learning module, configured to determine driving control parameters using a reinforcement learning algorithm based on the word vectors and driving state information; A driving control module, used to control the driving operation based on the driving control parameters and complete the driving control based on voice recognition; The speech recognition module includes a speech recognition model deployment submodule, a speech data feature extraction submodule and a speech feature recognition submodule; The speech recognition model deployment submodule is used to pre-deploy a speech recognition model for speech recognition using a deep learning algorithm; The voice data feature extraction submodule is used to extract features from the denoised voice command data to obtain voice features; The speech feature recognition submodule is used to schedule the speech recognition model pre-deployed by the speech recognition model deployment submodule to recognize the speech features and determine the text instruction data; Pre-deployed speech recognition models for speech recognition using deep learning algorithms, including: Initializing hyperparameters of the deep learning algorithm using a chaotic mapping sequence to determine a plurality of different parameter vectors; wherein the parameter vectors include all or part of the hyperparameters to be optimized of the deep learning algorithm; Obtaining a loss function value for each parameter vector, and determining the fitness corresponding to the parameter vector based on the loss function value; Determining an optimal parameter vector and a worst parameter vector according to the fitness corresponding to the parameter vector; Based on the optimal parameter vector, a variable spiral position selection mechanism is used to perform an initialization search on the parameter vector, and a parameter vector after the initialization search is determined; For the parameter vector after the initialization search, according to the worst parameter vector, an adaptive probability response mechanism is used to perform a local and global balanced search on the parameter vector to determine the parameter vector after the balanced search; Determine whether the current number of training times is greater than or equal to the preset maximum number of training times. If so, re-determine the optimal parameter vector based on the parameter vector after the balanced search; otherwise, return to the step of obtaining fitness; Based on the re-determined optimal parameter vector, final hyperparameters of the deep learning algorithm are determined, and a speech recognition model for speech recognition is deployed based on the final hyperparameters of the deep learning algorithm.

2. The vehicle control system based on voice recognition according to claim 1, characterized in that: Based on the optimal parameter vector, a variable spiral position selection mechanism is used to perform an initialization search on the parameter vector, and the parameter vector after the initialization search is determined, including: Based on the current number of training times, the hyperbolic tangent function is used to determine the adaptive inertia weight; weighting the optimal parameter vector using the adaptive inertia weight to determine a weighted optimal parameter vector; For any parameter vector, obtaining a first difference vector between the optimal parameter vector and the parameter vector; For any parameter vector, a variable spiral function is used to perform position transformation on the parameter vector to determine the parameter vector after position transformation; The parameter vector after the initialization search is determined according to the weighted optimal parameter vector, the first difference vector, and the parameter vector after the position transformation.

3. The vehicle control system based on voice recognition according to claim 2, characterized in that: For the parameter vector after the initialization search, according to the worst parameter vector, an adaptive probability response mechanism is used to perform a local and global balanced search on the parameter vector to determine the parameter vector after the balanced search, including: Obtain the fitness corresponding to each parameter vector after the initialization search, and obtain the degree of group dispersion based on the fitness corresponding to each parameter vector after the initialization search; Obtaining a local development impact factor and a global development impact factor according to the degree of dispersion of the group; Obtaining a single search rate according to the fitness corresponding to the parameter vector after each initialization search, and determining an average search rate during multiple training processes according to the single search rate; Determining a local development threshold parameter and a global development threshold parameter according to the average search speed; Determining a local development probability based on the local development influencing factor and the local development threshold parameter, and performing local development using a location transfer strategy based on the local development probability to determine a parameter vector after the local development; Determining a global development probability based on the global development influencing factor and the global development threshold parameter, and performing global development using a chaotic global search strategy based on the global development probability to determine a parameter vector after global development; For any parameter vector after initialization search, if the parameter vector has not been globally developed or locally developed, the parameter vector after the original initialization search is directly used as the parameter vector after the equilibrium search; If the parameter vector is globally developed or locally developed, the parameter vector after global development or the parameter vector after local development is used as the parameter vector after equilibrium search; If the parameter vector is globally developed and locally developed, the annealing simulation algorithm is used to select the parameter vector after global development or the parameter vector after local development as the parameter vector after the equilibrium search.

4. The vehicle control system based on voice recognition according to claim 3, characterized in that: According to the local development probability, a position transfer strategy is adopted to perform local development, and a parameter vector after local development is determined, including: Randomly generate a first decision factor between the interval (0, 1), and determine whether the first decision factor is less than the local development probability. If so, adopt the position transfer strategy to perform local development and determine the parameter vector after local development. Otherwise, do not perform local development. Use the position transfer strategy for local development and determine the parameter vector after local development, including: For any parameter vector after the initialization search, determine the transition probabilities corresponding to other parameter vectors according to the fitness corresponding to the parameter vector after the initialization search; Determining a target collaboration vector for the parameter vector after the initialization search based on the transition probabilities corresponding to the other parameter vectors; According to the target collaboration vector, the parameter vector after the initialization search is locally developed to determine the parameter vector after the local development.

5. The vehicle control system based on voice recognition according to claim 4, characterized in that: According to the global development probability, a chaotic global search strategy is used to perform global development, and a parameter vector after global development is determined, including: Randomly generate a second decision factor between the interval (0, 1), and determine whether the second decision factor is less than the global development probability. If so, use the chaotic global search strategy to perform global development and determine the parameter vector after global development. Otherwise, do not perform global development. Use the chaos global search strategy for global development and determine the parameter vector after global development, including: Get the chaotic mapping factor in the current training process; Determine the search center vector according to the search upper limit vector and the search lower limit vector; According to the chaotic mapping factor, the parameter vector after the initialization search is globally developed in combination with the sine function to determine the parameter vector after the global development.

6. The vehicle control system based on voice recognition according to claim 1, characterized in that: Feature extraction is performed on the denoised voice command data to obtain voice features, including: extracting MFCC of the denoised voice command data to obtain voice features.

7. The vehicle control system based on voice recognition according to claim 1, characterized in that: Based on the word vector and the driving status information, a reinforcement learning algorithm is used to determine the driving control parameters, including: Collecting driving status information; wherein the status information includes position, posture and speed; The word vector and the driving state information are constructed into a current state space, and the current state space is used as the input of the reinforcement learning algorithm to determine the selected action and obtain the driving control parameters; Among them, the action is selected from the action space, each action represents a driving control parameter, and the action space should include all driving control parameters.

8. A driving control method based on speech recognition, characterized in that: include: The operator's voice command data is collected through a microphone array, and the voice command data is subjected to denoising processing to obtain denoised voice command data; Using a deep learning algorithm to recognize the denoised voice command data to determine text command data; Parsing the text instruction data to determine driving control keywords, and converting the driving control keywords into word vectors; Determining driving control parameters using a reinforcement learning algorithm based on the word vector and driving state information; Based on the driving control parameters, the driving operation is controlled to complete the driving control based on voice recognition.

Images